Stochastic Finance

Nicolas Privault
Notes on Stochastic Finance
This version: April 25, 2013

http://www.ntu.edu.sg/home/nprivault/indext.html
Preface
This text is an introduction to pricing and hedging in discrete and continuous time financial models without friction (i.e. without transaction costs),
with an emphasis on the complementarity between analytical and probabilistic methods. Its contents are mostly mathematical, and also aim at making
the reader aware of both the power and limitations of mathematical models
in finance, by taking into account their conditions of applicability. The book
covers a wide range of classical topics including Black-Scholes pricing, exotic
and american options, term structure modeling and change of numeraire, as
well as models with jumps. It is targeted at the advanced undergraduate and
graduate level in applied mathematics, financial engineering, and economics.
The point of view adopted is that of mainstream mathematical finance in
which the computation of fair prices is based on the absence of arbitrage hypothesis, therefore excluding riskless profit based on arbitrage opportunities
and basic (buying low/selling high) trading. Similarly, this document is not
concerned with any prediction of stock price behaviors that belong other
domains such as technical analysis, which should not be confused with the
statistical modeling of asset prices. The text also includes 104 figures and
simulations, along with about 20 examples based on actual market data.
The descriptions of the asset model, self-financing portfolios, arbitrage and
market completeness, are first given in Chapter 1 in a simple two time-step
setting. These notions are then reformulated in discrete time in Chapter 2.
Here, the impossibility to access future information is formulated using the
notion of adapted processes, which will play a central role in the construction
of stochastic calculus in continuous time.
In order to trade efficiently it would be useful to have a formula to estimate the fair price of a given risky asset, helping for example to determine
whether the asset is undervalued or overvalued at a given time. Although
such a formula is not available, we can instead derive formulas for the pricing of options that can act as insurance contracts to protect their holders
against adverse changes in the prices of risky assets. The pricing and hedging
of options in discrete time, particularly in the fundamental example of the
"
v
N. Privault
Cox-Ross-Rubinstein model, are considered in Chapter 3, with a description
of the passage from discrete to continuous time that prepares the transition
to the subsequent chapters.
A simplified presentation of Brownian motion, stochastic integrals and
the associated Ito formula, is given in Chapter 4. The Black-Scholes model is
presented from the angle of partial differential equation (PDE) methods in
Chapter 5, with the derivation of the Black-Scholes formula by transforming
the Black-Scholes PDE into the standard heat equation wich is then solved
by a heat kernel argument. The martingale approach to pricing and hedging
is then presented in Chapter 6, and complements the PDE approach of Chapter 5 by recovering the Black-Scholes formula via a probabilistic argument.
An introduction to volatility estimation is given in Chapter 7, including historical, local, and implied volatilities. This chapter also contains a comparison
of the prices obtained by the Black-Scholes formula with option price market
data.
Exotic options such as barrier, lookback, and Asian options in continuous
asset models are treated in Chapter 8. Optimal stopping and exercise, with
application to the pricing of American options, are considered in Chapter 9.
The construction of forward measures by change of numeraire is given in
Chapter 10 and is applied to the pricing of interest rate derivatives in Chapter 12, after an introduction to the modeling of forward rates in Chapter 11,
based on material from [90]. The pricing of defaultable bonds is considered
in Chapter 13.
Stochastic calculus with jumps is dealt with in Chapter 14 and is restricted
to compound Poisson processes which only have a finite number of jumps on
any bounded interval. Those processes are used for option pricing and hedging
in jump models in Chapter 15, in which we mostly focus on risk minimizing strategies as markets with jumps are generally incomplete. Chapter 16
contains an elementary introduction to finite difference methods for the numerical solution of PDEs and stochastic differential equations, dealing with
the explicit and implicit finite difference schemes for the heat equations and
the Black-Scholes PDE, as well as the Euler and Milshtein schemes for SDEs.
The text is completed with an appendix containing the needed probabilistic
background.
The material in this book has been used for teaching in the Masters of
Science in Financial Engineering at City University of Hong Kong and at the
Nanyang Technological University in Singapore. The author thanks Ju-Yi
Yen (University of Cincinnati) for several corrections and improvements.
The cover graph represents the time evolution of the HSBC stock price
from January to September 2009, plotted on the price surface of a European
vi
"

call option on that asset, expiring on October 05, 2009, cf. 5.5.
This pdf file contains external links, and animated figures and embedded
videos in Chapters 8, 9, 11 and 14, that may require using Acrobat Reader
for viewing on the complete pdf file. Clicking on an exercise number inside
the solution section will send to the original problem text inside the file.
Conversely, clicking on the problem number sends the reader to the corresponding solution, however this feature should not be misused.
"
vii
N. Privault
viii
"
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Assets, Portfolios and Arbitrage . . . . . . . . . . . . . . . . . . . . . . . . . .

1.1 Definitions and Formalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2 Portfolio Allocation and Short-Selling . . . . . . . . . . . . . . . . . . . . .
1.3 Arbitrage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.4 Risk-Neutral Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.5 Hedging of Contingent Claims . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.6 Market Completeness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.7 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11
11
12
13
16
18
20
21
27
Discrete-Time Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1 Stochastic Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2 Portfolio Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3 Arbitrage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4 Contingent Claims . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.5 Martingales and Conditional Expectation . . . . . . . . . . . . . . . . . .
2.6 Market Completeness and Risk-Neutral Measures . . . . . . . . . . .
2.7 The Cox-Ross-Rubinstein (CRR) Market Model . . . . . . . . . . . .
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
29
29
30
33
33
35
40
42
44
Pricing and hedging in discrete time . . . . . . . . . . . . . . . . . . . . . .

3.1 Pricing of Contingent Claims . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2 Hedging of Contingent Claims - Backward Induction . . . . . . . .
3.3 Pricing of Vanilla Options in the CRR Model . . . . . . . . . . . . . .
3.4 Hedging of Vanilla Options in the CRR model . . . . . . . . . . . . .
3.5 Hedging of Exotic Options in the CRR Model . . . . . . . . . . . . . .
3.6 Convergence of the CRR Model . . . . . . . . . . . . . . . . . . . . . . . . . .
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
47
47
51
53
55
59
66
69
ix
N. Privault
4
Brownian Motion and Stochastic Calculus . . . . . . . . . . . . . . . .

4.1 Brownian Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2 Wiener Stochastic Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3 Ito Stochastic Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.4 Stochastic Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.5 Geometric Brownian Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.6 Stochastic Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . .
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
73
73
78
83
88
93
96
98
The Black-Scholes PDE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

5.1 Continuous-Time Market Model . . . . . . . . . . . . . . . . . . . . . . . . . .
5.2 Self-Financing Portfolio Strategies . . . . . . . . . . . . . . . . . . . . . . . .
5.3 Arbitrage and Risk-Neutral Measures . . . . . . . . . . . . . . . . . . . . .
5.4 Market Completeness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.5 The Black-Scholes PDE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.6 The Heat Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.7 Solution of the Black-Scholes PDE . . . . . . . . . . . . . . . . . . . . . . . .
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
105
105
105
109
111
112
120
123
126
Martingale Approach to Pricing and Hedging . . . . . . . . . . . . .

6.1 Martingale Property of the Ito Integral . . . . . . . . . . . . . . . . . . . .
6.2 Risk-neutral Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.3 Girsanov Theorem and Change of Measure . . . . . . . . . . . . . . . .
6.4 Pricing by the Martingale Method . . . . . . . . . . . . . . . . . . . . . . . .
6.5 Hedging Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
131
131
134
137
139
142
148
Estimation of Volatility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.1 Historical Volatility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.2 Implied Volatility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.3 The Black-Scholes Formula vs Market Data . . . . . . . . . . . . . . . .
7.4 Local Volatility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.5 Stochastic Volatility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.6 Volatility Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
159
159
160
162
166
170
182
183
Exotic Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.1 Generalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.2 The Reflexion Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.3 Barrier Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.4 Lookback Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.5 Asian Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
185
185
190
199
218
244
263
x
"

9
American Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.1 Filtrations and Information Flow . . . . . . . . . . . . . . . . . . . . . . . . .
9.2 Martingales, Submartingales, and Supermartingales . . . . . . . . .
9.3 Stopping Times . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.4 Perpetual American Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.5 Finite Expiration American Options . . . . . . . . . . . . . . . . . . . . . .
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
269
269
270
272
282
294
301
10 Change of Num
eraire and Forward Measures . . . . . . . . . . . . .
10.1 Notion of Numeraire . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.2 Change of Numeraire . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.3 Foreign Exchange . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.4 Pricing of Exchange Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.5 Hedging by Change of Numeraire . . . . . . . . . . . . . . . . . . . . . . . . .
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
313
313
315
323
328
330
334
11 Forward Rate Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

11.1 Short Term Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.2 Zero-Coupon Bonds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.3 Forward Rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.4 The HJM Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.5 Forward Vasicek Rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.6 Modeling Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.7 The BGM Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
339
339
342
350
356
360
364
370
373
12 Pricing of Interest Rate Derivatives . . . . . . . . . . . . . . . . . . . . . .

12.1 Forward Measures and Tenor Structure . . . . . . . . . . . . . . . . . . . .
12.2 Bond Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
12.3 Caplet Pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
12.4 Forward Swap Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
12.5 Swaption Pricing on the LIBOR . . . . . . . . . . . . . . . . . . . . . . . . . .
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
381
381
384
385
388
390
394
13 Credit Risk, CDSs and CDOs . . . . . . . . . . . . . . . . . . . . . . . . . . . .

13.1 Stochastic Default . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
13.2 Defaultable Bonds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
13.3 Credit Default Swaps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
13.4 Correlated default times . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
13.5 Merton model of credit risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
13.6 Modeling the default times . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
13.7 Collateralized debt obligations (CDOs) . . . . . . . . . . . . . . . . . . . .
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
407
407
412
413
415
426
428
433
439
"
xi
N. Privault
14 Stochastic Calculus for Jump Processes . . . . . . . . . . . . . . . . . . .
14.1 The Poisson Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
14.2 Compound Poisson Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . .
14.3 Stochastic Integrals with Jumps . . . . . . . . . . . . . . . . . . . . . . . . . .
14.4 Ito Formula with Jumps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
14.5 Stochastic Differential Equations with Jumps . . . . . . . . . . . . . .
14.6 Girsanov Theorem for Jump Processes . . . . . . . . . . . . . . . . . . . .
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
443
443
449
452
454
459
464
470
15 Pricing and Hedging in Jump Models . . . . . . . . . . . . . . . . . . . . .

15.1 Risk-Neutral Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
15.2 Pricing in Jump Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
15.3 Black-Scholes PDE with Jumps . . . . . . . . . . . . . . . . . . . . . . . . . .
15.4 Exponential Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
15.5 Self-Financing Hedging with Jumps . . . . . . . . . . . . . . . . . . . . . . .
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
473
473
474
476
478
481
484
16 Basic Numerical Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

16.1 Discretized Heat Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
16.2 Discretized Black-Scholes PDE . . . . . . . . . . . . . . . . . . . . . . . . . . .
16.3 Euler Discretization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
16.4 Milshtein Discretization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
487
487
489
493
494
Appendix: Background on Probability Theory . . . . . . . . . . . . . . . .

Probability Spaces and Events . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Probability Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Conditional Probabilities and Independence . . . . . . . . . . . . . . . . . . . .
Random Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Probability Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Expectation of a Random Variable . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Conditional Expectation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Moment Generating Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
497
497
501
502
503
505
510
517
520
522
Exercise Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Chapter 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Chapter 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Chapter 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Chapter 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Chapter 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Chapter 6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Chapter 7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Chapter 8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Chapter 9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Chapter 10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
525
525
526
527
533
546
551
560
561
578
597
xii
"


Chapter 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Chapter 12 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Chapter 13 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Chapter 14 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Chapter 15 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Background on Probability Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . .
603
612
629
635
641
645
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 647
Author index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 653
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 655
"
xiii
List of Figures
0.3
0.4
As if a whole new world was laid out before me. . . . . . . . . . . . . .

Graph of the Hang Seng index - holding a put option might be
useful here. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Sample price processes simulated by a geometric Brownian motion. . .
Infogrames stock price curve. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.1
1.2
1.3
Arbitrage - 2006 retail prices around the world for the Xbox 360. . . .
Absence of arbitrage - the Mark Six Investment Table. . . . . . . . . .
Separation of convex sets. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
14
15
18
2.1
Illustration of the self-financing condition (2.3). . . . . . . . . . . . . . . . . .
31
4.1
4.2
4.3
4.4
4.5
Sample paths of a one-dimensional Brownian motion. . . . . . . . . . . . .

Two sample paths of a two-dimensional Brownian motion. . . . . . . . .
Sample paths of a three-dimensional Brownian motion. . . . . . . . . . . .
Step function. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Geometric Brownian motion started at 1. . . . . . . . . . . . . . . . . . . . . .
76
77
77
78
96
5.1
5.2
5.3
5.4
5.5
5.6
5.7
5.8
5.9
5.10
Illustration of the self-financing condition (5.2). . . . . . . . . . . . . . . . . .

Graph of the Black-Scholes call price function with strike K = 100.
Graph of the stock price of HSBC Holdings. . . . . . . . . . . . . . . . . . . .
Path of the Black-Scholes price for a call option on HSBC. . . . . . . . .
Time evolution of the hedging portfolio for a call option on HSBC. .
Graph of the Black-Scholes put price function with strike K = 100. .
Path of the Black-Scholes price for a put option on HSBC. . . . . . . . .
Time evolution of the hedging portfolio for a put option on HSBC. .
Time-dependent solution of the heat equation. . . . . . . . . . . . . . . . . .
Option price as a function of the volatility . . . . . . . . . . . . . . . . . . . .
106
116
116
117
118
119
119
120
121
127
6.1
6.2
Drifted Brownian path. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

Delta of a European option. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
0.1
0.2
5
6
7
xv
N. Privault
6.3
6.4
6.5
6.6
6.7
6.8
6.9
Gamma of a European option. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Option price as a function of the underlying and of time to maturity
Delta as a function of the underlying and of time to maturity . . . . . .
Gamma as a function of the underlying and of time to maturity . . . .
Option price as a function of the underlying and of time to maturity
Delta as a function of the underlying and of time to maturity . . . . . .
Gamma as a function of the underlying and of time to maturity . . . .
7.1
7.2
7.3
7.4
7.5
7.6
7.7
7.8
7.9
7.10
7.11
7.12
The fugazi: its a wazy, its a woozie. Its fairy dust. . . . . . . . . . . 160
Implied volatility of Asian options on light sweet crude oil futures. . . 162
Graph of the (market) stock price of Cheung Kong Holdings. . . . . . . 162
Graph of the (market) call option price on Cheung Kong Holdings. . 163
Graph of the Black-Scholes call option price on Cheung Kong Holdings. 163
Graph of the (market) stock price of HSBC Holdings. . . . . . . . . . . . . 164
Graph of the (market) call option price on HSBC Holdings. . . . . . . . 164
Graph of the Black-Scholes call option price on HSBC Holdings. . . . 165
Graph of the (market) put option price on HSBC Holdings. . . . . . . . 165
Graph of the Black-Scholes put option price on HSBC Holdings. . . . 166
Call option price vs ALSTOM underlying. . . . . . . . . . . . . . . . . . . . . . 166
Euro / SGD exchange rate. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
8.1
8.2
8.3
8.4
8.5
8.6
8.7
8.8
Brownian motion Bt and its supremum Xt . . . . . . . . . . . . . . . . . . . . 187

A function with no last point of increase before t = 1. . . . . . . . . . . . . 187
Brownian motion Bt and its moving average. . . . . . . . . . . . . . . . . . . 189
Reflected Brownian motion with a = 1. . . . . . . . . . . . . . . . . . . . . . . 191
Probability density of the maximum of Brownian motion. . . . . . . . . . 192
Density of the supremum of geometric Brownian motion. . . . . . . . . . . 193
Joint probability density of B1 and its maximum over [0,1]. . . . . . . . . 195
Heat map of the joint probability density of B1 and its maximum
over [0,1]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
Probability density of the maximum of drifted Brownian motion. . . . . 198
Graph of the up-and-out call option price with B > K. . . . . . . . . . . 203
Graph of the up-and-out put option price (8.12) with B = 80 > K = 60. 208
Graph of the up-and-out put option price (8.13) with
K = 100 > B = 80. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
Graph of the down-and-out call option price with B < K. . . . . . . . . . 210
Graph of the down-and-out call option price with K > B. . . . . . . . . . 211
Graph of the down-and-out put option price with K > B. . . . . . . . . . 212
Delta for the up-and-out option. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
Graph of the lookback put option price. . . . . . . . . . . . . . . . . . . . . . . 224
Graph of the normalized lookback put option price. . . . . . . . . . . . . . . 229
Black-Scholes put price in the decomposition (8.32). . . . . . . . . . . . . . . 230
Function hp (, z) in the decomposition (8.32). . . . . . . . . . . . . . . . . . . 231
Graph of the lookback call option price. . . . . . . . . . . . . . . . . . . . . . . 234
Normalized lookback call option price. . . . . . . . . . . . . . . . . . . . . . . . . . 238
8.9
8.10
8.11
8.12
8.13
8.14
8.15
8.16
8.17
8.18
8.19
8.20
8.21
8.22
xvi

148
153
154
155
156
157
157
"
"
8.23
8.24
8.25
8.26
8.27
8.28
8.29
8.30
8.31
Graph of the underlying asset price. . . . . . . . . . . . . . . . . . . . . . . . . . .

Graph of the lookback call option price. . . . . . . . . . . . . . . . . . . . . . . .
Running minimum of the underlying asset price. . . . . . . . . . . . . . . . . .
Black-Scholes call price in the normalized lookback call price. . . . . . .
Function hc (, z) in the normalized lookback call option price. . . . . . .
Delta of the lookback call option. . . . . . . . . . . . . . . . . . . . . . . . . . . .
Rescaled portfolio strategy for the lookback call option. . . . . . . . . . . .
Graph of the Asian option price. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Lognormal approximation to the Asian option price. . . . . . . . . . . . . .
238
239
239
240
240
243
243
250
251
9.1
9.2
9.3
9.4
9.5
9.6
9.7
9.8
9.9
9.10
9.11
9.12
9.13
9.14
9.15
Drifted Brownian path. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Evolution of the fortune of a poker player vs number of games played.
Stopped process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Graphs of the option price by exercise at L for several values of L. .
Animated graph of the option price depending on the values of L. .
Option price as a function of L and of the underlying asset price. . . .
Path of the American put option price on the HSBC stock. . . . . . . .
Graphs of the option price by exercising at L for several values of L.
Graphs of the option prices parametrized by different values of L. . . .
Expected Black-Scholes European call price vs (x, t) 7 (x K)+ . . .
Black-Scholes put price function vs (x, t) 7 (K x)+ . . . . . . . . . . . .
Optimal frontier for the exercise of a put option. . . . . . . . . . . . . . . . .
Numerical values of the finite expiration American put price. . . . . . . .
Longstaff-Schwartz algorithm for the American put price. . . . . . . . . .
Comparison between Longstaff-Schwartz and finite differences. . . . . .
271
271
274
285
286
287
287
292
293
295
296
296
298
299
299
11.1 Graph of t 7 rt in the Vasicek model. . . . . . . . . . . . . . . . . . . . . . . .

11.2 Graphs of t 7 P (t, T ) and t 7 er0 (T t) . . . . . . . . . . . . . . . . . . .
11.3 Graph of t 7 P (t, T ) for a bond with a 2.3% coupon. . . . . . . . . . . .
11.4 Bond price graph with coupon rate 6.25%. . . . . . . . . . . . . . . . . . . . . .
11.5 Graph of T 7 f (t, T, T + ). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.6 Stochastic process of forward curves. . . . . . . . . . . . . . . . . . . . . . . . . .
11.7 Forward rate process t 7 f (t, T, S). . . . . . . . . . . . . . . . . . . . . . . . . .
11.8 Instantaneous forward rate process t 7 f (t, T ). . . . . . . . . . . . . . . . .
11.9 Forward instantaneous curve in the Vasicek model. . . . . . . . . . . . .
11.10 Forward instantaneous curve x 7 f (0, x) in the Vasicek model. . .
11.11 Short term interest rate curve t 7 rt in the Vasicek model. . . . . . .
11.12 Market example of yield curves (11.23). . . . . . . . . . . . . . . . . . . . . . .
11.13 Graph of x 7 g(x) in the Nelson-Siegel model. . . . . . . . . . . . . . . . .
11.14 Graph of x 7 g(x) in the Svensson model. . . . . . . . . . . . . . . . . . . .
11.15 Comparison of market data vs a Svensson curve. . . . . . . . . . . . . . . .
11.16 Graphs of forward rates. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.17 Forward instantaneous curve in the Vasicek model. . . . . . . . . . . . . . .
11.18 Graph of t 7 P (t, T1 ). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.19 Graph of forward rates in a two-factor model. . . . . . . . . . . . . . . . . .
340
347
348
348
352
357
360
361
362
362
363
363
364
365
365
366
366
367
369
xvii

N. Privault
11.20 Random evolution of forward rates in a two-factor model. . . . . . . . . 370
11.21 Graph of stochastic interest rate modeling. . . . . . . . . . . . . . . . . . . . . 372
12.1 Forward rates arranged according to a tenor structure. . . . . . . . . . . . 381
13.1 Different Gaussian copula graphs for = 0, = 0.85 and = 1. . . . .
13.2 Different Gaussian copula density graphs for = 0, = 0.35 and
= 0.999. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
13.3 Function x 7 1 (x) + ( r) T t/ for > r, = r, and

< r. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
13.4 A representation of CDO tranches. . . . . . . . . . . . . . . . . . . . . . . . . . . .
13.5 A Titanic-style representation of cumulative tranche losses. . . . . . . . .
13.6 Function fk (x) = min((x N k1 )+ , N pk ). . . . . . . . . . . . . . . . . . . .
13.7 Cumulative historic default rates. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
13.8 Internal Ratings-Based formula. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
418
427
434
435
437
438
439
14.1 Sample path of a Poisson process (Nt )tR+ . . . . . . . . . . . . . . . . . . . . .

14.2 Sample path of a compound Poisson process (Yt )tR+ . . . . . . . . . . . .
14.3 Sample trajectories of a gamma process. . . . . . . . . . . . . . . . . . . . . . .
14.4 Sample trajectories of a stable process. . . . . . . . . . . . . . . . . . . . . . . .
14.5 Sample trajectories of a variance gamma process. . . . . . . . . . . . . . . .
14.6 Sample trajectories of an inverse Gaussian process. . . . . . . . . . . . . . .
14.7 Sample trajectories of a negative inverse Gaussian process. . . . . . . . .
14.8 Geometric Poisson process. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
14.9 Ranking data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
14.10 Geometric compound Poisson process. . . . . . . . . . . . . . . . . . . . . . .
14.11 Geometric Brownian motion with compound Poisson jumps. . . . . .
14.12 SMRT Stock price. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
444
450
457
458
458
459
459
461
461
462
463
463
419
16.1 Divergence of the explicit finite difference method. . . . . . . . . . . . . . . . 491

16.2 Stability of the implicit finite difference method. . . . . . . . . . . . . . . . . . 493
S.1
S.2
S.3
S.4
S.5
S.6
S.7
S.8
S.9
Market data for the warrant #01897 on the MTR Corporation. . . . . 547
Price of a digital call option. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 558
Risky hedging portfolio value for a digital call option. . . . . . . . . . . . . 559
Riskless hedging portfolio value for a digital call option. . . . . . . . . . . 559
Average return by selling at the maximum vs selling at maturity T . . 562
Average return by selling at the minimum vs selling at maturity T
as a function of T . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 563
Graph of the up-and-in long forward contract price with
K = 60 < B = 80. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 567
Delta of the down-and-in long forward contract with K = 60 < B = 80. 568
Graph of the up-and-out long forward contract price with
K = 60 < B = 80. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 569
xviii
"

S.10 Delta of the up-and-out long forward contract price with
K = 60 < B = 80. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 570
S.11 Graph of the down-and-in long forward contract price with
K = 60 < B = 80. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 571
S.12 Delta of the down-and-in long forward contract with K = 60 < B = 80. 571
S.13 Graph of the down-and-out long forward contract price with
K = 60 < B = 80. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 572
S.14 Delta of the down-and-out long forward contract with
K = 60 < B = 80. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 573
S.15 Lookback call option as a function of maturity time T . . . . . . . . . . . . . 574
S.16 Cashflow data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 635
S.17 CDS price data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 636
Animated figures (work in Acrobat reader).
"
xix
Introduction
Modern mathematical finance and quantitative analysis require a strong

background in fields such as stochastic calculus, optimization, partial differential equations (PDEs) and numerical methods, or even infinite dimensional
analysis. In addition, the emergence of new complex financial instruments on
the markets makes it necessary to rely on increasingly sophisticated mathematical tools. Not all readers of this book will eventually work in quantitative
financial analysis, nevertheless they may have to interact with quantitative
analysts, and becoming familiar with the tools they employ be an advantage.
In addition, despite the availability of ready made financial calculators it still
makes sense to be able oneself to understand, design and implement such
financial algorithms. This can be particularly useful under different types of
conditions, including an eventual lack of trust in financial indicators, possible
unreliability of expert advice such as buy/sell recommendations, or other factors such as market manipulation. To some extent we would like to have some
form of control on the future behaviour of random (risky) assets, however,
since knowledge of the future is not possible, the time evolution of the prices
of risky assets will be modelled by random variables and stochastic processes.
Historical Sketch
We start with a description of some of the main steps, ideas and individuals
that played an important role in the development of the field over the last
century.
Robert Brown, botanist, 1827
Brown observed the movement of pollen particles as described in his paper
A brief account of microscopical observations made in the months of June,
July and August, 1827, on the particles contained in the pollen of plants; and
"
N. Privault
on the general existence of active molecules in organic and inorganic bodies.
Phil. Mag. 4, 161-173, 1828.
Philosophical Magazine, first published in 1798, is a journal that publishes
articles in the field of condensed matter describing original results, theories
and concepts relating to the structure and properties of crystalline materials,
ceramics, polymers, glasses, amorphous films, composites and soft matter.
Louis Bachelier, mathematician, PhD 1900
Bachelier used Brownian motion for the modelling of stock prices in his
PhD thesis Theorie de la speculation, Annales Scientifiques de lEcole Normale Superieure 3 (17): 21-86, 1900.
Albert Einstein, physicist
Einstein received his 1921 Nobel Prize in part for investigations on the
theory of Brownian motion: ... in 1905 Einstein founded a kinetic theory to
account for this movement, presentation speech by S. Arrhenius, Chairman
of the Nobel Committee, Dec. 10, 1922.
Albert Einstein, Uber

die von der molekularkinetischen Theorie der W
arme
geforderte Bewegung von in ruhenden Fl
ussigkeiten suspendierten Teilchen,
Annalen der Physik 17 (1905) 223.
Norbert Wiener, mathematician, founder of cybernetics
Wiener is credited, among other fundamental contributions, for the mathematical foundation of Brownian motion, published in 1923. In particular he
constructed the Wiener space and Wiener measure on C0 ([0, 1]) (the space of
continuous functions from [0, 1] to R vanishing at 0).
Norbert Wiener, Differential space, Journal of Mathematics and Physics of
the Massachusetts Institute of Technology, 2, 131-174, 1923.
Kiyoshi It
o (
), mathematician, Gauss prize 2006
Ito constructed the Ito integral with respect to Brownian motion, cf. It
o,
Kiyoshi, Stochastic integral. Proc. Imp. Acad. Tokyo 20, (1944). 519-524. He
also constructed the stochastic calculus with respect to Brownian motion,
which laid the foundation for the development of calculus for random processes, cf. Ito, Kiyoshi, On stochastic differential equations, Mem. Amer.
Math. Soc. 1951, (1951).
2
"

Renowned math wiz It
o, 93, dies. (The Japan Times, Saturday, Nov. 15,
2008).
Kiyoshi Ito, an internationally renowned mathematician and professor
emeritus at Kyoto University died Monday of respiratory failure at a Kyoto hospital, the university said Friday. He was 93. Ito was once dubbed
the most famous Japanese in Wall Street thanks to his contribution
to the founding of financial derivatives theory. He is known for his work
on stochastic differential equations and the Ito Formula, which laid the
foundation for the Black-Scholes model, a key tool for financial engineering. His theory is also widely used in fields like physics and biology.
Paul Samuelson, economist, Nobel Prize 1970
In 1965, Samuelson rediscovered Bacheliers ideas and proposed geometric
Brownian motion as a model for stock prices. In an interview he stated In
the early 1950s I was able to locate by chance this unknown [Bacheliers]
book, rotting in the library of the University of Paris, and when I opened it
up it was as if a whole new world was laid out before me. We refer to Rational theory of warrant pricing by Paul Samuelson, Industrial Management
Review, p. 13-32, 1965.
Fig. 0.1: [14] As if a whole new world was laid out before me.
In recognition of Bacheliers contribution, the Bachelier Finance Society was
started in 1996 and now holds the World Bachelier Finance Congress every
Click on the figure to play the video (works in Acrobat reader on the entire pdf
file).
"
3
N. Privault
2 years.
Robert Merton, Myron Scholes, economists
Robert Merton and Myron Scholes shared the 1997 Nobel Prize in economics: In collaboration with Fisher Black, developed a pioneering formula
for the valuation of stock options ... paved the way for economic valuations
in many areas ... generated new types of financial instruments and facilitated
more efficient risk management in society.
Black, Fischer; Myron Scholes (1973). The Pricing of Options and Corporate Liabilities. Journal of Political Economy 81 (3): 637-654.
The development of options pricing tools contributed greatly to the expansion
of option markets and led to development several ventures such as the Long
Term Capital Management (LTCM), founded in 1994. The fund yielded annualized returns of over 40% in its first years, but registered lost US$ 4.6
billion in less than four months in 1998, which resulted into its closure in
early 2000.
Oldrich Vasicek, economist, 1977
Interest rates behave differently from stock prices, notably due to the phenomenon of mean reversion, and for this reason they are difficult to model
using geometric Brownian motion. Vasicek was the first to suggest a meanreverting model for stochastic interest rates, based on the Ornstein-Uhlenbeck
process, in An equilibrium characterisation of the term structure, Journal
of Financial Economics 5: 177-188.
David Heath, Robert Jarrow, A. Morton
These authors proposed in 1987 a general framework to model the evolution of (forward) interest rates, known as the HJM model, see their joint paper
Bond Pricing and the Term Structure of Interest Rates: A New Methodology for Contingent Claims Valuation, Econometrica, (January 1992), Vol.
60, No. 1, pp 77-105.
Alan Brace, Dariusz Gatarek, Marek Musiela (BGM)
The BGM model is actually based on geometric Brownian motion, and it
is specially useful for the pricing of interest rate derivatives such as caps and
This has to be put in relation with the modern development of risk societies; societies increasingly preoccupied with the future (and also with safety), which generates
the notion of risk.
4
"

swaptions on the LIBOR market, see The Market Model of Interest Rate
Dynamics. Mathematical Finance Vol. 7, page 127. Blackwell 1997, by Alan
Brace, Dariusz Gatarek, Marek Musiela.
European Call and Put Options

We close this introduction with a description of European call and put options, which are at the basis of risk management. As mentioned above, an
important concern for the buyer of a stock at time t is whether its price ST
can fall down at some future date T . The buyer of the stock may seek protection from a market crash by purchasing a contract that allows him to sell
his asset at time T at a guaranteed price K fixed at time t. This contract is
called a put option with strike price K and exercise date T .
Fig. 0.2: Graph of the Hang Seng index - holding a put option might be useful here.
Definition 0.1. A (European) put option is a contract that gives its holder
the right (but not the obligation) to sell a quantity of assets at a predefined
price K called the strike price (or exercise price) and at a predefined date T
called the maturity.
In case the price ST falls down below the level K, exercising the contract
will give the holder of the option a gain equal to K ST in comparison to
those who did not subscribe the option and sell the asset at the market price
ST . In turn, the issuer of the option will register a loss also equal to K ST
(in the absence of transaction costs and other fees).
If ST is above K then the holder of the option will not exercise the option
as he may choose to sell at the price ST . In this case the profit derived from
the option is 0.
"
5
N. Privault
In general, the payoff of a (so called European) put option will be of the
form
K ST , ST K,
+
(ST ) = (K ST ) =
0,
ST K.
Two possible scenarios (ST finishing above K or below K) are illustrated
in Figure 0.3.
10
ST-K>0
9
8
7
Strike
St
K=6
5
4
ST-K<0
3
2
S0=1
0
0
0.2
0.4
|
0.6
t=0.62
0.8
T=1
Fig. 0.3: Sample price processes simulated by a geometric Brownian motion.

On the other hand, if the trader aims at buying some stock or commodity,
his interest will be in prices not going up and he might want to purchase a
call option, which is a contract allowing him to buy the considered asset at
time T at a price not higher than a level K fixed at time t.
Here, in the event that ST goes above K, the buyer of the option will
register a potential gain equal to ST K in comparison to an agent who did
not subscribe to the call option.
Definition 0.2. A (European) call option is a contract that gives its holder
the right (but not the obligation) to buy a quantity of assets at a predefined
price K called the strike and at a predefined date T called the maturity.
In general, a (European) call option is an option with payoff function
ST K, ST K,
+
(ST ) = (ST K) =
0,
ST K.
In market practice, options are often divided into a certain number n of warrants, the (possibly fractional) quantity n being called the entitlement ratio.
6
"

In order for an option contract to be fair, the buyer of the option should
pay a fee (similar to an insurance fee) at the signature of the contract. The
computation of this fee is an important issue, which is known as option
pricing.
The second important issue is that of hedging, i.e. how to manage a given
portfolio in such a way that it contains the required random payoff (K ST )+
(for a put option) or (ST K)+ (for a call option) at the maturity date T .
The next figure illustrates a sharp increase and sharp drop in asset price,
making it valuable to hold a call option during the first half of the graph,
whereas holding a put option would be recommended during the second half.
Fig. 0.4: Infogrames stock price curve.
An illustration - pricing and hedging in a binary model

We close this introduction with a simplified illustration of the pricing and
hedging technique in a binary model. Consider a risky stock price S valued
S0 = $4 at time t = 0, and taking only two possible values
$5
S1 =
$2
at time t = 1. In addition, consider an option that yields a payoff P whose
values are contingent to the data of S:
$3 if S1 = $5
P =
$0 if S1 = $2.
"
7
N. Privault
At time t = 0 we choose to invest units in the risky asset S, while keeping
$ on our bank account, meaning that we invest a total amount
S0 + $
at t = 0.
The following issues can be addressed:

a) Hedging: how to choose the portfolio allocation {, $} so that the value
S1 + $
of the portfolio matches the future payoff P at time t = 1 ?
b) Pricing: how to determine the amount S0 + $ to be invested in such a
portfolio at time t = 0 ?
Hedging means that at time t = 1 the portfolio value matches the future
payoff P , i.e.
S1 + $ = P.
This condition can be rewritten as
$3 = $5 + $
P =
$0 = $2 + $
i.e.
5 + = 3
if S1 = $5,
if S1 = $2,
which yields
2 + = 0,
= 1
$ = $2.
In other words, we buy 1 unit of the stock S at the price S0 = $4, and we
borrow $2 from the bank. The price of the option contract is given by the
portfolio value
S0 + $ = 1 $4 $2 = $2.
at time t = 0.
Conclusion: in order to deliver the random payoff P =
$3
if S1 = $5
$0
if S1 = $2.
at time t = 1, one has to:

1. receive $2 (the option price) at time t = 0,
2. borrow $ = $2 from the bank,
3. invest those $2 + $2 = $4 into the purchase of = 1 unit of stock valued

at S0 = $4 at time t = 0,
8
"

4. wait until time t = 1 to find that the portfolio value evolved into
$5 + $ = 1 $5 $2 = $3 if S1 = $5,
P =
$2 + $ = 1 $2 $2 = 0 if S1 = $2.
so that the option contract is fulfilled whatever the evolution of S.
We note that the initial amount of $2 can be turned to P = $3 (%50 profit)
... or into P = $0 (total ruin).
Thinking further
1) The expected gain of our portfolio is
IE[P ] = $3 P(P = $3) + $0 P(P = $0)
= $3 P(S1 = $5)
= $3 P(S1 = $5).
In absence of arbitrage opportunities (fair market) this expected gain IE[P ]
should equal the initial amount $2 invested in the option. In that case we
should have
IE[P ] = $3 P(S1 = $5) = $2
P(S1 = $5) + P(S1 = $2) = 1.
from which we can infer the probabilities
P(S1 = $5) = 3
(0.1)
P(S = $2) = 1 .
1
3
We see that the stock S has twice more chances to go up than to go down in
a fair market.
2) Based on the probabilities (0.1) we can also compute the expected value
IE[S1 ] of the stock at time t = 1. We find
IE[S1 ] = $5 P(S1 = $5) + $2 P(S1 = $2)
2
1
= $5 + $2
3
3
= $4
= S0 .
"
9
N. Privault
This means that, on average, no profit can be made from an investment on
the risky stock. In a more realistic model we can assume that the riskles bank
account yields an interest rate equal to r, in which case the above analysis is
modified by letting $ become $(1 + r) at time t = 1, nevertheless the main
conclusions remain unchanged.
10
"
Chapter 1
Assets, Portfolios and Arbitrage
We consider a simplified financial model with only two time instants t = 0 and
t = 1. In this simple setting we introduce the notions of portfolio, arbitrage,
completeness, pricing and hedging using the notation of [34]. A binary asset
price model is considered as an example in Section 1.7.
1.1 Definitions and Formalism

We will use the following notation. An element x
of Rd+1 is a vector
x
= (x0 , x1 , . . . , xd )
made of d+1 components. The scalar product x
y of two vectors x
, y Rd+1
is defined by
x
y = x0 y0 + x1 y1 + + xd yd .
The vector

= (0) , (1) , . . . , (d)
denotes the prices (i) > 0 at time t = 0 of d + 1 assets numbered

i = 0, 1, . . . , d.
The values S (i) > 0 at time t = 1 of assets i = 1, . . . , d are represented by
the random vector

S = S (0) , S (1) , . . . , S (d)
defined on a probability space (, F, P).
In addition we will assume that asset no 0 is a riskless asset (of savings
account type) that yields an interest rate r > 0, i.e. we have
"
N. Privault
S (0) = (1 + r) (0) .
1.2 Portfolio Allocation and Short-Selling

A portfolio based on the assets 0, 1, 2, . . . , d is viewed as a vector

= (0) , (1) , . . . , (d) Rd+1 ,
in which (i) represents the (possibly fractional) quantity of asset no i owned
by an investor, i = 0, 1, . . . , d. The price of such a portfolio is given by

=
d
X
(i) (i)
i=0
at time t = 0.
At time t = 1 the value of the portfolio has evolved into
S =
d
X
(i) S (i) .
i=0
If (0) > 0, the investor puts the amount (0) (0) > 0 on a savings account
with interest rate r, while if (0) < 0 he borrows the amount (0) (0) > 0
with the same interest rate.
For i = 1, . . . , d, if (i) > 0 then the investor buys a (possibly fractional)
quantity (i) > 0 of the asset no i, while if (i) < 0 he borrows a quantity
(i) > 0 of asset i and sells it to obtain the amount (i) (i) > 0. In the
latter case one says that the investor short sells a quantity (i) > 0 of the
asset no i.
Usually, profits are made by first buying at a lower price and then selling
at a higher price. Short-sellers apply the same rule but in the reverse time
order: first sell high, and then buy low if possible, by applying the following
procedure.
1. Borrow the asset no i.
2. At time t = 0, sell the asset no i on the market at the price (i) and
invest the amount (i) at the interest rate r > 0.
12
"

3. Buy back the asset no i at time t = 1 at the price S (i) , with hopefully
S (i) < (1 + r) (i) .
4. Return the asset to its owner, with possibly a (small) fee p > 0.
At the end of the operation the profit made on share no i equals
(1 + r) (i) S (i) p > 0,
which is positive provided S (i) < (1 + r) (i) and p > 0 is sufficiently small.
1.3 Arbitrage
As stated in the next definition, an arbitrage opportunity is the possibility
to make a strictly positive amount of money starting from 0 or even from a
negative amount. In a sense, an arbitrage opportunity can be seen as a way
to beat the market.
The short-selling procedure described in Section 1.2 represents a way to
realize an arbitrage opportunity (one can proceed similarly by simply buying
an asset instead short-selling it).
1. Borrow the amount (0) (0) > 0 on the riskless asset no 0.
2. Use the amount (0) (0) > 0 to buy the risky asset no i at time t = 0
and price (i) , for a quantity (i) = (0) (0) / (i) , i = 1, . . . , d.
3. At time t = 1, sell the risky asset no i at the price S (i) , with hopefully
S (i) > (i) .
4. Refund the amount (1 + r) (0) (0) > 0 with interest rate r > 0.
At the end of the operation the profit made is
(i) S (i) ((1 + r) (0) (0) ) = (i) S (i) + (1 + r) (0) (0)
(0) (i)
S + (1 + r) (0) (0)
(i)

(0)
= (0) (i) S (i) (1 + r) (i)

= (0)
= (i) S (i) (1 + r) (i)
The cost p of shortselling will not be taken into account in later calculations.
"
13
N. Privault
> 0,
(i)
(i)
or S (1 + r) per unit of stock invested, which is positive provided

S (i) > (i) and r is sufficiently small.
City
Tokyo
Hong Kong
Seoul
Taipei
New York
Sydney
Frankfurt
Paris
Rome
Brussels
London
Manila
Jakarta
Currency
38,800 yen
HK$2,956.67
378,533 won
NT$12,980
A$633.28
e399
e399
e399
e399.66
279.99
29,500 pesos
5,754,1676 rupiah
US$
$346
$381
$400
$404
$433
$483
$513
$513
$513
$514
$527
$563
$627
Fig. 1.1: Arbitrage - 2006 retail prices around the world for the Xbox 360.
Next, we state a mathematical formulation of the concept of arbitrage.
Definition 1.1. A portfolio Rd+1 constitutes an arbitrage opportunity if
the three following conditions are satisfied:
i)
0,
ii) S 0,
[start from 0 or even with a debt]

[finish with a non-negative amount]
iii) P( S > 0) > 0.
[a profit is made with non-zero probability]
The are many real-life examples of situations where arbitrage opportunities

can occur, such as:
- assets with different returns (finance),
- servers with different speeds (queueing, networking, computing),
- highway lanes with different speeds (driving).
In the latter two examples, the absence of arbitrage is consequence of the
fact that switching to a faster lane or server may result into congestion, thus
annihilating the potential benefit of the shift.
14
"
Fig. 1.2: Absence of arbitrage - the Mark Six Investment Table.
In the table of Figure 1.2 the absence of arbitrage opportunities is materialized by the fact that the price of each combination is found to be proportional
to its probability, thus making the game fair and disallowing any opportunity
or arbitrage that would result of betting on a more profitable combination.
In the sequel we will work under the assumption that arbitrage opportunities do not occur and we will rely on this hypothesis for the pricing of
financial instruments.
Let us give a market example of pricing by absence of arbitrage.
From March 24 to 31, 2009, HSBC issued rights to buy shares at the price
of $28. This right actually behaves like a call option since it gives the right
(with no obligation) to buy the stock at K = $28. On March 24 the HSBC
stock price finished at $41.70.
The question is: how to value the price $R of the right to buy one share?
This question can be answered by looking for arbitrage opportunities. Indeed,
there are two ways to purchase the stock:
1. directly buy the stock on the market at the price of $41.70. Cost: $41.70,
or:
2. first purchase the right at price $R and then the stock at price $28. Total
cost: $R+$28.
For an investor who owns no stock and no rights, arbitrage would be possible
in case $R + $28 < $41.70 by buying the right at a price $R, then the stock
"
15
N. Privault
at price $28, and finally selling the stock at the market price of $41.70. The
profit made by the investor would equal
$41.70 ($R + $28) > 0.
On the other hand, for an investor who owns the rights, in case $R + $28 >
$41.70, arbitrage would be possible by firt selling the right at price $R, and
then buying the stock on the market at $41.70. At time t = 1 the stock could
be sold at around $28, and profit would equal
$R + $28 $41.70 > 0.
In the absence of arbitrage opportunities, the above argument implies that
$R should satisfy
$R + $28 $41.70 = 0,
i.e. the arbitrage price of the right is given by the equation
$R = $41.70 $28 = $13.70.
(1.1)
Interestingly, the market price of the right was $13.20 at the close of the
session on March 24. The difference of $0.50 can be explained by the presence
of various market factors such as transaction costs, the time value of money,
or simply by the fact that asset prices are constantly fluctuating over time.
It may also represent a small arbitrage opportunity, which cannot be at all
excluded. Nevertheless, the absence of arbitrage argument (1.1) prices the
right at $13.70, which is quite close to its market value. Thus the absence of
arbitrage hypothesis appears as an accurate tool for pricing.
1.4 Risk-Neutral Measures

In order to use absence of arbitrage in the general context of pricing financial
derivatives, we will need the notion of risk-neutral measure.
The next definition says that under a risk-neutral (probability) measure,
the risky assets no 1, . . . , d have same average rate of return as the riskless
asset no 0.
Definition 1.2. A probability measure P on is called a risk-neutral measure if
IE [S (i) ] = (1 + r) (i) ,
i = 1, 2, . . . , d.
(1.2)
Here, IE denotes the expectation under the probability measure P . Note
that for i = 0, the condition IE [S (0) ] = (1 + r) (0) is always satisfied by
16
"

definition.
In other words, P is called risk neutral because taking risks under P
by buying a stock S (i) has a neutral effect: on average the expected yield of
the risky asset equals the riskless rate obtained by investing on the savings
account with interest rate r.
On the other hand, under a risk premium probability measure P# , the
expected return of the risky asset S (i) would be higher than r, i.e. we would
have
IE# [S (i) ] > (1 + r) (i) ,
i = 1, . . . , d.
The following result can be used to check for the existence of arbitrage opportunities, and is known as the first fundamental theorem of mathematical
finance. In the sequel we will only consider probability measures P that are
equivalent to P in the sense that
P (A) = 0
if and only if
P(A) = 0,
for all
A F.
(1.3)
Theorem 1.1. A market is without arbitrage opportunity if and only if it

admits at least one equivalent risk-neutral measure P .
Proof. For the sufficiency, given P a risk-neutral measure we have

=
d
X
i=0
(i) (i) =
d
1
1 X (i) (i)
> 0,
IE [S ] =
IE [ S]
1 + r i=0
1+r
because P ( S > 0) > 0 as P( S > 0) > 0 and P is equivalent to P, and

the condition
> 0 contradicts Definition 1.1-(i). The proof of necessity
relies on the theorem of separation of convex sets by hyperplanes Theorem 1.2
below, cf. Theorem 1.6 of [34]. It can be briefly sketched as follows. Given
two financial assets with net discounted gains X, Y and a portfolio made of
one unit of X and c unit(s) of Y , the absence of arbitrage opportunities can
be reformulated by saying that for any portfolio choice determined by c R,
we have
X + cY 0 = X + cY = 0,
P a.s.,
(1.4)
i.e. a riskless (no loss) portfolio can not entail a stricly positive gain. In other
words, if one wishes to make a strictly positive gain on the market, one has
to accept the possibility of a loss. In order to show the absence of arbitrage
opportunities implies the existence of a risk-neutral probability measure P
under which all risky investments have zero discounted return, i.e.
IEP [X] = IEP [Y ] = 0,
"
(1.5)
17
N. Privault
we apply the convex separation Theorem 1.2 to the convex set

C = (IEQ [X], IEQ [Y ]) : Q P
in R2 , where P is the family of probability measures Q on equivalent to P.
If (1.5) does not hold under any probability measure Q P then 0
/ C and
the convex separation Theorem 1.1 shows the existence of c R such that
IEQ [X] + c IEQ [Y ] 0,
Q P,
(1.6)
and IEP [X] + c IEP [Y ] > 0 for some P P. This shows that X + cY 0
a.s. while P (X + cY > 0) 6= 0, which contradicts the absence of arbitrage.

Next is a version of the separation theorem for convex sets, cf. e.g. Theorem 4.14 of [51].
Theorem 1.2. Let C1 and C2 be two disjoint convex sets in R2 . Then there
exists a, b R such that we have
y1 a + bx1
and
a + bx2 y2 ,
for all (x1 , y1 ) C1 and (x2 , y2 ) C2 (up to exchange of C1 and C2 ).
Fig. 1.3: Separation of convex sets.
1.5 Hedging of Contingent Claims

In this section we consider the notion of contingent claim, according to the
following broad definition.
Inequality (1.6) might be reversed, in this case we choose (1, c) as portfolio

allocation.
18
"

Definition 1.3. A contingent claim is any non-negative random variable
C 0.
In practice the random variable C represents the payoff of an (option) contract at time t = 1.
Referring to Definition 0.2, a European call option with maturity t = 1 on
the asset no i is a contingent claim whose the payoff C is given by
(i)
S K if S (i) K,
C = (S (i) K)+ =
0
if S (i) < K,
where K is called the strike price. The claim C is called contingent because its value may depend on various market conditions, such as S (i) > K.
A contingent claim is also called a derivative for the same reason.
Similarly, referring to Definition 0.1, a European put option with maturity
t = 1 on the asset no i is a contingent claim with payoff
K S (i) if S (i) K,
C = (K S (i) )+ =
0
if S (i) > K,
Definition 1.4. A contingent claim with payoff C is said to be attainable if
there exists a portfolio strategy such that
C = S.
When a contingent claim C is attainable, a trader will be able to:

1. at time t = 0, build a portfolio allocation = ( (0) , (1) , . . . , (d) ) Rd+1 ,
2. invest the amount

=
d
X
(i) (i)
i=0
in this portfolio at time t = 0,

S of the portfolio.
3. at time t = 1, pay the claim amount C using the value
The above shows that in order to attain the claim, an initial investment
is needed at time t = 0. This amount, to be paid by the buyer to the issuer

"
19
N. Privault
of the option (the option writer), is also called the arbitrage price of the
contingent claim C, and denoted by
(C) :=
.
(1.7)
The action of allocating a portfolio such that

C = S
(1.8)
is called hedging, or replication, of the contingent claim C.

As a rough illustration of the principle of hedging, one may buy oil-related
stocks in order to hedge oneself against a potential price rise of gasoline. In
this case, any increase in the price of gasoline that would result in a higher
value of the derivative C would be correlated to an increase in the underlying
stock value, so that the equality (1.8) would be maintained.
In case the value S exceeds the amount of the claim, i.e. if
S C,
we talk about super-hedging.
In this book we focus on hedging (i.e. replication of the contingent claim
C) and we will not consider super-hedging.
1.6 Market Completeness

Market completeness is a strong property saying that any contingent claim
can be perfectly hedged.
Definition 1.5. A market model is said to be complete if every contingent
claim C is attainable.
The next result is the second fundamental theorem of mathematical finance.
Theorem 1.3. A market model without arbitrage is complete if and only if
it admits only one risk-neutral measure.
Proof. cf. Theorem 1.40 of [34].
Theorem 1.3 will give us a concrete way to verify market completeness by

searching for a unique solution P to Equation (1.2).
20
"
1.7 Example
In this section we work out a simple example that allows us to apply Theorem 1.1 and Theorem 1.3.
We take d = 1, i.e. there is only a riskless asset no 0 and a risky asset
S (1) . In addition we choose
= { , + },
which is the simplest possible non-trivial choice of a probability space, made
of only two possible outcomes with
P({ }) > 0
and P({ + }) > 0,
in order for the setting to be non-trivial. In other words the behavior of the
market is subject to only two possible outcomes, for example, one is expecting an important binary decision of yes/no type, which can lead to two
distinct scenarios called and + .
In this context, the asset price S (1) is given by a random variable
S (1) : R
whose value depends whether the scenario , resp. + , occurs.
Precisely, we set
S (1) ( ) = a,
and
S (1) ( + ) = b,
i.e. the value of S (1) becomes equal a under the scenario , and equal to b
under the scenario + , where 0 < a < b.
The first natural question we ask is:
- are there arbitrage opportunities in such a market?
We will answer this question using Theorem 1.1, which amounts to searching
for a risk-neutral measure P . In this simple framework, any measure P on
= { , + } is characterized by the data of two numbers P ({ }) [0, 1]
and P ({ + }) [0, 1], such that
P () = P ({ }) + P ({ + }) = 1.
(1.9)
The case a = b leads to a trivial, constant market.
"
21
N. Privault
Here, saying that P is equivalent to P simply means that
P ({ }) > 0
and P ({ + }) > 0.
In addition, according to Definition 1.2 a risk-neutral measure P should

satisfy
IE [S (1) ] = (1 + r) (1) .
(1.10)
Although we should solve this equation for P , at this stage it is not yet clear
how P appears in (1.10).
In order to make (1.10) more explicit we write the expectation as
IE [S (1) ] = aP (S (1) = a) + bP (S (1) = b),
hence Condition (1.10) for the existence of a risk-neutral measure P reads
aP (S (1) = a) + bP (S (1) = b) = (1 + r) (1) .
Using the Condition (1.9) we obtain the system of two equations
aP ({ }) + bP ({ + }) = (1 + r) (1)
(1.11)
P ({ }) + P ({ + }) = 1,
with solution
P ({ }) =
b (1 + r) (1)
ba
and
P ({ + }) =
(1 + r) (1) a
.
ba
In order for a solution P to exist as a probability measure, the numbers

P ({ }) and P ({ + }) must be non-negative. In addition, for P to be
equivalent to P they should be strictly positive from (1.3).
We deduce that if a, b and r satisfy the condition
a < (1 + r) (1) < b,
(1.12)
then there exists a risk-neutral (equivalent) probability measure P which is

unique, hence by Theorems 1.1 and 1.3 the market is without arbitrage and
complete.
If a = b = (1 + r) (1) then (1.2) admits an infinity of solutions, hence the
market is without arbitrage but it is not complete. More precisely, in this
case both the riskless and risky assets yield a deterministic return rate r and
the value of the portfolio becomes
S = (1 + r)
,
22
"

at time t = 1, hence the terminal value S is deterministic and this single
value can not always match the value of a random contingent claim C that
would be allowed to take two distinct values C( ) and C( + ). Therefore,
market completeness does not hold when a = b = (1 + r) (1) .
Note that if a = (1 + r) (1) , resp. b = (1 + r) (1) , then P ({ + }) = 0,
resp. P ({ }) = 0, and P is not equivalent to P.
On the other hand, under the conditions
a < b (1 + r) (1)
or
(1 + r) (1) a < b,
(1.13)
no (equivalent) risk neutral measure exists, and as a consequence there exist

arbitrage opportunities in the market.
Let us give a financial interpretation of Conditions (1.13).
1. If (1 + r) (1) a < b, let (1) = 1 and choose (0) such that (0) (0) +
(1) (1) = 0, i.e.
(0) = (1) (1) / (0) < 0.
In particular, Condition (i) of Definition 1.1 is satisfied, and the investor
borrows the amount (0) (0) > 0 on the riskless asset and uses it to
buy one unit (1) = 1 of the risky asset. At time t = 1 she sells the
risky asset S (1) at a price at least equal to a and refunds the amount
(1 + r) (0) (0) > 0 she borrowed, with interests. Her profit is
S = (1 + r) (0) (0) + (1) S (1)
(1 + r) (0) (0) + (1) a
= (1 + r) (1) (1) + (1) a
= (1) ((1 + r) (1) + a)
0,
which satisfies Condition (ii) of Definition 1.1. In addition, Condition (iii)

of Definition 1.1 is also satisfied because
P( S > 0) = P(S (1) = b) = P({ + }) > 0.
2. If a 0 and choose (1) such that (0) (0) +
(1) (1) = 0, i.e.
(1) = (0) (0) / (1) < 0.
This means that the investor borrows a (possibly fractional) quantity
(1) > 0 of the risky asset, sells it for the amount (1) (1) , and invests this money on the riskless account for the amount (0) (0) > 0. As
mentioned in Section 1.2, in this case one says that the investor shortsells
"
23
N. Privault
the risky asset. At time t = 1 she obtains (1 + r) (0) (0) > 0 from the
riskless asset and she spends at most b to buy the risky asset and return
it to its original owner. Her profit is
S = (1 + r) (0) (0) + (1) S (1)
(1 + r) (0) (0) + (1) b
= (1 + r) (1) (1) + (1) b
= (1) ((1 + r) (1) + b)
0,
since (1) < 0. Note that here, a S (1) b became

(1) b (1) S (1) (1) a
because (1) < 0. We can check as in Part 1 above that Conditions (i)-(iii)
of Definition 1.1 are satisfied.
Under Condition (1.12) there is absence of arbitrage and Theorem 1.1 shows
that no portfolio strategy can yield S 0 and P( S > 0) > 0 starting
from (0) (0) + (1) (1) 0, although this is less simple to show directly.
Finally if a = b 6= (1 + r) (1) then (1.2) admits no solution as a probability
measure P hence arbitrage opportunities exist and can be constructed by
the same method as above.
The second natural question is:
- is the market complete, i.e. are all contingent claims attainable?
In the sequel we work under the condition
a < (1 + r) (1) < b,
under which Theorems 1.1 and 1.3 show that the market is without arbitrage
and complete since the risk-neutral measure P exists and is unique.
Let us recover this fact by elementary calculations. For any contingent
claim C we need to show that there exists a portfolio = ( (0) , (1) ) such
i.e.
that C = S,
(0)
(1 + r) (0) + (1) a = C( )
(1.14)
(0)
(1 + r) (0) + (1) b = C( + ).
These equations can be solved as
24
"

(0) =
bC( ) aC( + )
(0) (1 + r)(b a)
and (1) =
C( + ) C( )
.
ba
(1.15)
In this case we say that the portfolio ( (0) , (1) ) hedges the contingent claim
C. In other words, any contingent claim C is attainable and the market is
indeed complete. Here, the quantity
(0) (0) =
bC( ) aC( + )
(1 + r)(b a)
represents the amount invested on the riskless asset.

Note that if C( + ) C( ) then (1) 0 and there is not short selling.
This occurs in particular if C has the form C = h(S (1) ) with x 7 h(x) a
nondecreasing function, since
C( + ) C( )
ba
h(S (1) ( + )) h(S (1) ( ))
=
ba
h(b) h(a)
=
ba
0,
(1) =
thus there is no short-selling. This applies in particular to European call options with strike K, for which the function h(x) = (xK)+ is nondecreasing.
Similarly we will find that (1) 0, i.e. short-selling always occurs when h
is a nonincreasing function, which is the case in particular for European put
options with payoff function h(x) = (K x)+ .
The arbitrage price (C) of the contingent claim C is defined in (1.7) as
the initial value at t = 0 of the portfolio hedging C, i.e.
(C) =
,
(1.16)
where ( (0) , (1) ) are given by (1.15). Note that (C) cannot be 0 since this
would entail the existence of an arbitrage opportunity according to Definition 1.1.
The next proposition shows that the arbitrage price (C) of the claim can
be computed as the expected value of its payoff C under the risk-neutral
measure, after discounting at the rate 1 + r for the time value of money.
Proposition 1.1. The arbitrage price (C) =
of the contingent claim
C is given by
1
IE [C].
(1.17)
(C) =
1+r
"
25
N. Privault
Proof. We have
(C) =
= (0) (0) + (1) (1)

bC( ) aC( + )
C( + ) C( )
=
+ (1)
(1 + r)(b a)
ba

(1)
b
(1
+
r)
(1 + r) (1) a
1
C( )
+ C( + )
=
1+r
ba
ba

1
(1)
+
(1)
=
C( )P (S = a) + C( )P (S = b)
1+r
1
=
IE [C].
1+r

In the case of a European call option with strike K [a, b] we have C =
(S (1) K)+ and
((S (1) K)+ ) = (1)
(b K)a
bK
.
ba
(1 + r)(b a)
Here, ( (1) K)+ is called the intrinsic value at time 0 of the call option.
The simple setting described in this chapter raises several questions and
remarks.
Remarks
1. The fact that (C) can be obtained by two different methods, i.e. an
algebraic method via (1.15) and (1.16) and a probabilistic method from
(1.17) is not a simple coincidence. It is actually a simple example of the
deep connection that exists between probability and analysis.
In a continuous time setting, (1.15) will be replaced with a partial differential equation (PDE) and (1.17) will be computed via the Monte Carlo
method. In practice, the quantitative analysis departments of major financial institutions can be split into the PDE team and the Monte
Carlo team, often trying to determine the same option prices by two
different methods.
2. What if we have three possible scenarios, i.e. = { , o , + } and the
random asset S (1) is allowed to take more than two values, e.g. S (1)
{a, b, c} according to each scenario? In this case the system (1.11) would
be rewritten as
26
"
aP ({ }) + bP ({ o }) + cP ({ + }) = (1 + r) (1)
P ({ }) + P ({ o }) + P ({ + }) = 1,
and this system of two equations for three unknowns does not have a
unique solution, hence the market can be without arbitrage but it cannot
be complete. Completeness can be reached by adding a second risky asset,
i.e. taking d = 2, in which case we will get three equations and three
unknowns. More generally, when has n 2 elements, completeness
of the market can be reached provided we consider d risky assets with
d + 1 n. This is related to the Meta-Theorem 8.3.1 of [4] in which the
number d of traded underlying risky assets is linked to the number of
random sources through arbitrage and completeness.
Exercises
Exercise 1.1 Consider a financial model with two instants t = 0 and t = 1

and two assets:
- a riskless asset with price 0 at time t = 0 and value 1 = 0 (1 + r) at
time t = 1,
- a risky asset S with price S0 at time t = 0 and random value S1 at time
t = 1.
We assume that S1 can take only the values S0 (1 + a) and S0 (1 + b), where
1 < a < r < b. The return of the risky asset is defined as
R=
S1 S0
.
S0
a) What are the possible values of R ?

b) Show that under the probability measure P defined by
P (R = a) =
br
,
ba
P (R = b) =
ra
,
ba
the expected return IE [R] of S is equal to the return r of the riskless

asset.
c) Does there exist arbitrage opportunities in this model ? Explain why.
d) Is this market model complete ? Explain why.
e) Consider a contingent claim with payoff C given by
if R = a,
C=
if R = b.
"
27
N. Privault
Show that the portfolio (, ) defined by
=
(1 + b) (1 + a)
0 (1 + r)(b a)
and =
,
S0 (b a)
hedges the contingent claim C, i.e. show that at time t = 1 we have

1 + S1 = C.
Hint: distinguish two cases R = a and R = b.
f) Compute the arbitrage price (C) of the contingent claim C using , 0 ,
, and S0 .
g) Compute IE [C] in terms of a, b, r, , .
h) Show that the arbitrage price (C) of the contingent claim C satisfies
(C) =
1
IE [C].
1+r
(1.18)
i) What is the interpretation of Relation (1.18) above ?

j) Let C denote the payoff at time t = 1 of a put option with strike K = $11
on the risky asset. Give the expression of C as a function of S1 and K.
k) Letting 0 = S0 = 1, r = 5% and a = 8, b = 11, compute the portfolio
(, ) hedging the contingent claim C.
l) Compute the arbitrage price (C) of the claim C.
Here, is the (possibly fractional) quantity of asset and is the quantity held of
asset S.
28
"
Chapter 2
Discrete-Time Model
A basic limitation of the two time step model considered in Chapter 1 is that it
does not allow for trading until the end of the time period is reached. In order
to be able to re-allocate the portfolio over time we need to consider a discretetime financial model with N + 1 time instants t = 0, 1, . . . , N . The practical
importance of this model lies also in its direct computer implementability.
2.1 Stochastic Processes

A stochastic process on a probability space (, F, P) is a family (Xt )tT of
random variables Xt : R indexed by a set T . Examples include:
the two-instant model: T = {0, 1},
the discrete-time model with finite horizon: T = {0, 1, 2, . . . , N },
the discrete-time model with infinite horizon: T = N,
the continuous-time model: T = R+ .
For real-world examples of stochastic processes one can mention:
the time evolution of a risky asset - in this case Xt represents the price of
the asset at time t T .
the time evolution of a physical parameter - for example, Xt represents a
temperature observed at time t T .
In this chapter we will focus on the finite horizon discrete-time model with
T = {0, 1, 2, . . . , N }.
Here the vector
"
N. Privault
= (0) , (1) , . . . , (d)
denotes the prices at time t = 0 of d + 1 assets numbered 0, 1, . . . , d.

The random vector
(0)
(1)
(d)
St = St , St , . . . , St
on denotes the values at time t = 1, 2, . . . , N of assets 0, 1, . . . , d, and forms

a stochastic process (St )t=0,1,...,N with S0 =
.
Here we still assume that asset 0 is a riskless asset (of savings account
type) yielding an interest rate r, i.e. we have
(0)
St
= (1 + r)t (0) ,
t = 0, 1, . . . , N.
2.2 Portfolio Strategies

(i)
A portfolio strategy is a stochastic process (t )t=1,...,N Rd+1 where t
denotes the (possibly fractional) quantity of asset i held in the portfolio over
the period (t 1, t], t = 1, 2, . . . , N .
Note that the portfolio allocation

(0)
(1)
(d)
t = (t , t , . . . , t )
remains constant over the period (t 1, t] while the stock price changes from
St1 to St over this period.
In other terms,
(i)
(i)
t St1
represents the amount invested in asset i at the beginning of the time period
(t 1, t], and
(i) (i)
t St
represents the value of this investment at the end of (t 1, t], t = 1, 2, . . . , N .
The value of the porfolio at the beginning of the time period (t 1, t] is
t St1 =
d
X
(i)
(i)
t St1 ,
i=0
when the market opens at time t 1, and becomes

30
"
Discrete-Time Model
t St =
d
X
(i)
(i)
t St
(2.1)
i=0
at the end of (t 1, t], i.e. when the market closes, t = 1, 2, . . . , N .

At the beginning of the next trading period (t, t + 1] the value of the
portfolio becomes
d
X
(i)
(i)
t+1 St =
t+1 St .
(2.2)
i=0
Note that the stock price St is assumed to remain constant overnight, i.e.
from the end of (t 1, t] to the beginning of (t, t + 1].
Obviously the question arises whether (2.1) should be identical to (2.2). In
the sequel we will need such a consistency hypothesis, called self-financing,
on the portfolio strategy t .
Definition 2.1. We say that the portfolio strategy (t )t=1,...,N is self-financing
if
t St = t+1 St ,
t = 1, 2, . . . , N 1.
(2.3)
The meaning of the self-financing condition (2.3) is simply that one cannot
take any money in or out of the portfolio during the overnight transition
period at time t. In other words, at the beginning of the new trading period
(t, t+1] one should re-invest the totality of the portfolio value obtained at the
end of period (t 1, t]. The next figure is an illustration of the self-financing
condition.
Portfolio value
t St1
- t St = t+1 St
- t+1 St+1
Asset value
St1
St St
St+1
Time scale
t1
t
t t
t t+1
t+1
t+1
Portfolio allocation
@
I
@
@
I
@
@
@
Morning
Evening
Morning
@
Evening
Fig. 2.1: Illustration of the self-financing condition (2.3).
Note that portfolio re-allocation happens overnight durig which time the
portfolio global value remains the same due to the self-financing condition.
The portfolio allocation t remains the same throughout the day, however
the portfolio value changes from morning to evening due to a change in the
"
31
N. Privault
stock price. Also, 0 is not defined and its value is actually not needed in this
framework.
Of course the chosen unit of time may not be the day, and it can be replaced
by weeks, hours, minutes, or even fractions of seconds in high-frequency trading.
We will denote by
Vt := t St
the value of the portfolio at time t = 1, 2, . . . , N , with
Vt = t+1 St ,
t = 0, . . . , N 1,
by the self-financing condition (2.3), and in particular

V0 = 1 S0 .
Let also
t := (Xt(0) , Xt(1) , . . . , Xt(d) )
X
denote the vector of discounted asset prices defined as:
(i)
Xt
1
(i)
S ,
(1 + r)t t
or
t :=
X
i = 0, 1, 2, . . . , d,
1
St ,
(1 + r)t
t = 0, 1, 2, . . . , N,
t = 0, 1, 2, . . . , N.
The discounted value at time 0 of the portfolio is defined by

Vet =
1
Vt ,
(1 + r)t
t = 0, 1, 2, . . . , N.
We have
Vet =
=
1
t St
(1 + r)t
d
X
1
(i) (i)
S
(1 + r)t i=0 t t
d
X
(i)
(i)
t Xt
i=0
t,
= t X
t = 1, 2, . . . , N,
and
Ve0 = 1 X0 = 1 S0 .
32
"
Discrete-Time Model
The effect of discounting from time t to time 0 is to divide prices by (1 + r)t ,
making all prices comparable at time 0.
2.3 Arbitrage
The definition of arbitrage in discrete time follows the lines of its analog in
the two-step model.
Definition 2.2. A portfolio strategy (t )t=1,...,N constitutes an arbitrage opportunity if all three following conditions are satisfied:
i) V0 0,
ii) VN 0,
[start from 0 or even with a debt]

[finish with a non-negative amount]
iii) P(VN > 0) > 0.
[a profit is made with non-zero probability]
2.4 Contingent Claims

Recall that from Definition 1.3, a contingent claim is given by the nonnegative random payoff C of an option contract at time t = N . For example,
in the case of the European call of Definition 0.2, the payoff C is given by
C = (SN K)+ where K is called the strike (or exercise) price.
In a discrete-time setting we are able to consider path-dependent options
in addition to European type options. One can distinguish between vanilla
options whose payoff depends on the terminal value of the underlying asset,
such as simple European contracts, and exotic or path-dependent options
such as Asian, barrier, or lookback options, whose payoff may depend on the
whole path of the underlying asset price until expiration time.
The list provided below is actually very restricted and there exists many
more option types, with new ones appearing constantly on the markets.
European options
The payoff of a European call on the underlying asset no i with maturity N
and strike K is
+
(i)
C = SN K .
"
33
N. Privault
The payoff of a European put on the underlying asset no i with exercise date
N and strike K is
(i) +
C = K SN
.
Let us mention also the existence of binary, or digital options, also called
cash-or-nothing options, whose payoffs are
(i)
$1 if SN K,
(i)
C = 1[K,) SN =
0 if S (i) < K,
N
for call options, and
C=
(i)
1(,K] SN
(i)
$1 if SN K,
0 if S (i) > K,
N
for put options.
Asian options
The payoff of an Asian call option (also called average value option) on the
underlying asset no i with exercise date N and strike K is
N
C=
1 X (i)
S K
N + 1 t=0 t
!+
.
The payoff of an Asian put option on the underlying asset no i with exercise
date N and strike K is
!+
N
1 X (i)
C= K
St
.
N + 1 t=0
We refer to Section 8.5 for the pricing of Asian options in continuous time.
Barrier options
The payoff of a down-an-out barrier call option on the underlying asset no i
with exercise date N , strike K and barrier B is
34
"
Discrete-Time Model
(i)
C = SN K
+
1(
min
(i)
t=0,1,...,N
St > B
(i)
+
(i)
S K if min St > B,
N
t=0,1,...,N
if
min
t=0,1,...,N
(i)
St B.
This option is also called a Callable Bull Contract with no residual value, in
which B denotes the call price B K. It is also called a turbo warrant with
no rebate.
The payoff of an up-and-out barrier put option on the underlying asset no
i with exercise date N , strike K and barrier B is
(i)
(i) +
if max St < B,
K SN
t=0,1,...,N

+
(i)
) =
C = KSN
1(
(i)
(i)
max St < B
0
if max St B.
t=0,1,...,N
t=0,1,...,N
This option is also called a Callable Bear Contract with no residual value,
in which the call price B usually satisfies B K. See [31], [117] for recent
results on the pricing of CBBCs, also called turbo warrants. We refer the
reader to Chapter 8 for the pricing and hedging of similar exotic options in
continuous time. Barrier options in continuous time are priced in Section 8.3.
Lookback options
The payoff of a floating strike lookback call option on the underlying asset
no i with exercise date N is
(i)
C = SN
min
t=0,1,...,N
(i)
St .
The payoff of a floating strike lookback put option on the underlying asset
no i with exercise date N is

(i)
(i)
C=
max St
SN .
t=0,1,...,N
We refer to Section 8.4 for the pricing of lookback options in continuous time.
2.5 Martingales and Conditional Expectation

Before proceeding to the definition of risk-neutral probability measures in discrete time we need to introduce more mathematical tools such as conditional
expectations, filtrations, and martingales.
"
35
N. Privault
Conditional expectations
Clearly, the expected value of any risky asset or random variable is dependent
on the amount of available information. For example, the expected return on
a real estate investment typically depends on the location of this investment.
In the probabilistic framework the available information is formalized as
a collection G of events, which may be smaller than the collection F of all
available events, i.e. G F.
The notation IE[F |G] represents the expected value of a random variable F
given (or conditionally to) the information contained in G, and it is read the
conditional expectation of F given G. In a certain sense, IE[F |G] represents
the best possible estimate of F in mean square sense, given the information
contained in G.
The conditional expectation satisfies the following five properties, cf. Section 16.4 for details and proofs.
i) IE[F G | G] = G IE[F | G] if G depends only on the information contained
in G.
ii) IE[G | G] = G when G depends only on the information contained in G.
iii) IE[IE[F | H] | G] = IE[F | G] if G H, called the tower property, cf. also
Relation (16.25).
iv) IE[F | G] = IE[F ] when F does not depend on the information contained in G or, more precisely stated, when the random variable F is
independent of the -algebra G.
v) If G depends only on G and F is independent of G, then
IE[h(F, G) | G] = IE[h(x, F )]x=G .
When H = {, } is the trivial -algebra we have IE[F | H] = IE[F ], F
L1 (). See (16.25) and (16.29) for illustrations of the tower property by
conditioning with respect to discrete and continuous random variables.
The collection G is also called a -algebra, cf. Section 16.4.
36
"
Discrete-Time Model
Filtrations
The total amount of information present in the market at time t =
0, 1, . . . , N is denoted by Ft . We assume that
Ft Ft+1 ,
t = 0, 1, . . . , N 1,
which means that the amount of information available on the market increases over time.
(i)
(i)
Usually, Ft corresponds to the knowledge of the values S0 , . . . , St , i =

1, 2, . . . , d, of the risky assets up to time t. In mathematical notation we say
(i)
(i)
that Ft is generated by S0 , . . . , St , and we usually write

(i)
(i)
Ft = S0 , . . . , St , i = 1, 2, . . . , d ,
t = 0, 1, . . . , N.
The notation Ft is useful to represent a quantity of information available at
time t. Note that different agents or traders may work with distinct filtration.
For example, an insider will have access to a filtration (Gt )t=0,1,...,N larger
than the filtration (Ft )t=0,1,...,N available to an ordinary agent, in the sense
that
Ft Gt ,
t = 0, 1, 2, . . . , N.
The notation IE[F |Ft ] represents the expected value of a random variable F
given (or conditionally to) the information contained in Ft . Again, IE[F |Ft ]
denotes the best possible estimate of F in mean square sense, given the information known up to time t.
We will assume that no information is available at time t = 0, which
translates as
IE[F | F0 ] = IE[F ]
for any integrable random variable F . As above, the conditional expectation
with respect to Ft satisfies the following five properties:
i) IE[F G | Ft ] = F IE[G | Ft ] if F depends only on the information contained in Ft .
ii) IE[F | Ft ] = F when F depends only on the information known at time
t and contained in Ft .
iii) IE[IE[F | Ft+1 ] | Ft ] = IE[F | Ft ] if Ft Ft+1 (by the tower property, cf.
also Relation (6.1) below).
"
37
N. Privault
iv) IE[F | Ft ] = IE[F ] when F does not depend on the information contained
in Ft .
v) If F depends only on Ft and G is independent of Ft , then
IE[h(F, G) | Ft ] = IE[h(x, G)]x=F .
Note that by the tower property (iii) the process t 7 IE[F | Ft ] is a martingale, cf. e.g. Relation (6.1) for details.
Martingales
A martingale is a stochastic process whose value at time t+1 can be estimated
using conditional expectation given its value at time t. Recall that a process
(Mt )t=0,1,...,N is said to be Ft -adapted if the value of Mt depends only on
the information available at time t in Ft , t = 0, 1, . . . , N .
Definition 2.3. A stochastic process (Mt )t=0,1,...,N is called a discrete time
martingale with respect to the filtration (Ft )t=0,1,...,N if (Mt )t=0,1,...,N is Ft adapted and satisfies the property
IE[Mt+1 |Ft ] = Mt ,
t = 0, 1, . . . , N 1.
Note that the above definition implies that Mt Ft , t = 0, 1, . . . , N . In other

words, a random process (Mt )t=0,1,...,N is a martingale if the best possible
prediction of Mt+1 in the mean square sense given Ft is simply Mt .
As an example of the use of martingales we can mention weather forecasting. If Mt denotes the random temperature observed at time t, this process
is a martingale when the best possible forecast of tomorrows temperature
Mt+1 given information known up to time t is just todays temperature Mt ,
t = 0, 1, . . . , N 1.
Definition 2.4. A stochastic process (k )k0 is said to be predictable if k
depends only on the information in Fk1 , k 1.
Under the convention F1 = {, } we find that 0 is a constant when (k )k0
(i)
is predictable. Recall that on the other hand, the process Sk k=0,1,...,N is
(i)
adapted as Sk
i = 1, 2, . . . , d.
depends only on the information in Fk , k = 0, 1, . . . , N ,
An important property of martingales is that the martingale transform

(2.4) of a predictable process is itself a martingale, see also Proposition 6.1
38
"
Discrete-Time Model
for the continuous-time analog of the following proposition, which will be
used in the proof of Theorem 3.1 below.
Proposition 2.1. Given (Xk )k=0,1,...,N a martingale and (k )k=1,2,...,N a
square-summable predictable process, the discrete-time process (Mt )t=0,1,...,N
defined by
t
X
Mt =
k (Xk Xk1 ),
t = 0, 1, . . . , N,
(2.4)
k=1
is a martingale.
Proof. Given n, t {0, 1, . . . , N } we have
" n
#

X

IE [Mn | Ft ] = IE
k (Xk Xk1 )Ft
k=1
n
X
k=1
t
X

IE k (Xk Xk1 ) | Ft
IE [k (Xk Xk1 ) | Ft ] +
k=1
t
X
n
X
IE [k (Xk Xk1 ) | Ft ]
k=t+1
k (Xk Xk1 ) +
k=1
= Mt +
n
X
IE [k (Xk Xk1 ) | Ft ]
k=t+1
n
X
IE [k (Xk Xk1 ) | Ft ] .
k=t+1
In order to conclude to IE [Mn | Ft ] = Mt we need to show that

IE [k (Xk Xk1 ) | Ft ] = 0,
t + 1 k n.
First we note that when 0 t k 1 we have Ft Fk1 , hence by the

tower property of conditional expectations we get
IE [k (Xk Xk1 ) | Ft ] = IE [IE [k (Xk Xk1 ) | Fk1 ] | Ft ] .
Next, since the process (k )k1 is predictable, k depends only on the information in Fk1 , and using Property (ii) of conditional expectations we may
pull out k out of the expectation since it behaves as a constant parameter
given Fk1 , k = 1, 2, . . . , n. This yields
IE [k (Xk Xk1 ) | Fk1 ] = k IE [Xk Xk1 | Fk1 ] = 0
since
See here for a related discussion of martingale strategies in a particular case.
"
39
N. Privault
IE [Xk Xk1 | Fk1 ] = IE [Xk | Fk1 ] IE [Xk1 | Fk1 ]
= IE [Xk | Fk1 ] Xk1
= 0,
k = 1, 2, . . . , N,
because (Xk )k=0,1,...,N is a martingale. We conclude that

IE [k (Xk Xk1 ) | Fk1 ] = k IE [Xk Xk1 | Fk1 ] = 0,
and more generally
IE [k (Xk Xk1 ) | Ft ] = IE [IE [k (Xk Xk1 ) | Fk1 ] | Ft ] = 0,
k = t + 1, . . . , n.
2.6 Market Completeness and Risk-Neutral Measures

As in the two time step model, the concept of risk neutral measures will be
used to price financial claims under the absence of arbitrage hypothesis.
Definition 2.5. A probability measure P on is called a risk-neutral measure if under P , the expected return of each risky asset equals the return r
of the riskless asset, that is
h
i
(i)
(i)
IE St+1 Ft = (1 + r)St ,
t = 0, 1, . . . , N 1,
(2.5)
i = 0, 1, . . . , d. Here, IE denotes the expectation under P .
(i)
Since St Ft , Relation (2.5) can be rewritten in terms of asset returns as

#
" (i)
(i)
St+1 St
= r,
t = 0, 1, . . . , N 1.
IE
F
t
(i)
St
In other words, taking risks under P by buying the risky asset no i has a
neutral effect, as the expected return is that of the riskless asset. The measure
P would be represent a risk premium if we had
i
h
(i)
(i)
IE St+1 | Ft = (1 + r)St ,
t = 0, 1, . . . , N 1,
with r > r in the case of a positive premium, and r < r in the case of a
negative premium.
See also the Efficient Market Hypothesis.
40
"
Discrete-Time Model
The definition of risk-neutral probability measure can be reformulated
using the notion of martingale.
Proposition 2.2. A probability measure P on is a risk-neutral measure
(i)
if and only if the discounted price process Xt is a martingale under P , i.e.
h
i
(i)
(i)
t = 0, 1, . . . , N 1,
(2.6)
IE Xt+1 Ft = Xt ,
i = 0, 1, . . . , d.
Proof. It suffices to check that Conditions (2.5) and (2.6) are equivalent since
h
i
h
i
(i)
(i)
(i)
(i)
and St = (1 + r)t Xt ,
IE St+1 Ft = (1 + r)t+1 IE Xt+1 Ft
t = 0, 1, . . . , N 1, i = 1, 2, . . . , d.
Next we restate the first fundamental theorem of asset pricing in discrete

time, which can be used to check for the existence of arbitrage opportunities.
admits at least one equivalent risk-neutral measure.
Proof. cf. [47] and Theorem 5.17 of [34].
Next, we turn to the notion of market completeness, starting with the definition of attainability for a contingent claim.
Definition 2.6. A contingent claim with payoff C is said to be attainable
(at time N ) if there exists a portfolio strategy (t )t=1,...,N such that
C = N SN .
(2.7)
In case (t )t=1,...,N is a portfolio that attains the claim C at time N , i.e. if

(2.7) is satisfied, we also say that (t )t=1,...,N hedges the claim C. In case
(2.7) is replaced by the condition
N SN C,
we talk of super-hedging. When (t )t=1,...,N hedges C, the arbitrage price
t (C) of the claim at time t will be given by the value
t (C) = t St
of the portfolio at time t = 0, 1, . . . , N . Note that at time t = N we have
N (C) = N SN = C,
i.e. since exercise of the claim occurs at time N , the price N (C) of the claim
equals the value C of the payoff.
"
41
N. Privault
claim is attainable.
The next result can be viewed as the second fundamental theorem of asset
pricing.
it admits only one equivalent risk-neutral measure.
Proof. cf. [47] and Theorem 5.38 of [34].
2.7 The Cox-Ross-Rubinstein (CRR) Market Model

We consider the discrete time Cox-Ross-Rubinstein model [17] with N + 1
time instants t = 0, 1, . . . , N and d = 1 risky asset, also called the binomial
(0)
model. The price St of the riskless asset evolves as
(0)
St
= (0) (1 + r)t ,
Let the return of the risky asset S = S

Rt :=
St St1
,
St1
t = 0, 1, . . . , N.
(1)
be defined as
t = 1, 2, . . . , N.
In the CRR model the return Rt is random and allowed to take only two
values a and b at each time step, i.e.
Rt {a, b},
t = 1, 2, . . . , N,
with 1 < a < b. That means, the evolution of St1 to St is random and
given by
(1 + b)St1 if Rt = b
= (1 + Rt )St1 ,
t = 1, . . . , N,
St =
(1 + a)St1 if Rt = a
and
St = S0
t
Y
(1 + Rj ),
t = 0, 1, . . . , N.
j=1
Note that the price process (St )t=0,1,...,N evolves on a binary recombining (or
binomial) tree. The discounted asset price is
Xt =
St
,
(1 + r)t
t = 0, 1, . . . , N,
42
"
Discrete-Time Model
with
Xt =
1+b
Xt1
1+r
if Rt = b
1+a
Xt1
1+r
if Rt = a
and
Xt =
1 + Rt
Xt1 ,
1+r
t = 1, . . . , N,
t
t
Y
Y
S0
1 + Rj
(1 + Rj ) = X0
.
t
(1 + r) j=1
1+r
j=1
In this model the discounted value at time t of the portfolio is given by

t = t(0) 0 + t(1) Xt ,
t X
t = 1, . . . , N.
The information Ft known in the market up to time t is given by the knowledge of S1 , . . . , St , which is equivalent to the knowledge of X1 , . . . , Xt or
R1 , . . . , Rt , i.e. we write
Ft = (S1 , . . . , St ) = (X1 , . . . , Xt ) = (R1 , . . . , Rt ),
t = 0, 1, . . . , N , where as a convention F0 = {, } contains no information.
Theorem 2.3. The CRR model is without arbitrage if and only if a < r < b.
In this case the market is complete.
Proof. In order to check for arbitrage opportunities we may use Theorem 2.1
and look for a risk-neutral measure P . According to the definition of a riskneutral measure this probability P should satisfy Condition (2.5), i.e.
IE [St+1 | Ft ] = (1 + r)St ,
t = 0, 1, . . . , N 1.
Rewriting IE [St+1 | Ft ] as
IE [St+1 | Ft ] = (1 + a)St P (Rt+1 = a | Ft ) + (1 + b)St P (Rt+1 = b | Ft ),
it follows that any risk-neutral measure P should satisfy the equations
(1 + b)St P (Rt+1 = b | Ft ) + (1 + a)St P (Rt+1 = a | Ft ) = (1 + r)St
P (Rt+1 = b | Ft ) + P (Rt+1 = a | Ft ) = 1,
i.e.

bP (Rt+1 = b | Ft ) + aP (Rt+1 = a | Ft ) = r,
P (Rt+1 = b | Ft ) + P (Rt+1 = a | Ft ) = 1,
with solution
"
43
N. Privault
P (Rt+1 = b | Ft ) =
ra
ba
and
P (Rt+1 = a | Ft ) =
br
,
ba
(2.8)
t = 0, 1, . . . , N 1.
We note that (2.8) implies that the sequence of random variables (Rt )t=0,1,...,N
is independent under P . Clearly, P can be equivalent to P only if r a > 0
and b r > 0. In this case the solution P of the problem is unique by
construction, hence the market is complete by Theorem 2.2.

Note that the values of P (Rt+1 = b | Ft ) and P (Rt+1 = a | Ft ) computed
in (2.8) are non random, hence they are independent of the information contained in Ft . As a consequence, under P , the random variable Rt+1 is independent of the information Ft up to time t, which is generated by R1 , . . . , Rt .
We deduce that (R1 , . . . , RN ) form a sequence of independent and identically
distributed (i.i.d.) random variables.
In other words, Rt+1 is independent of R1 , . . . , Rt for all t = 1, . . . , N 1,
the random variables R1 , . . . , RN are independent under P , and by (2.8) we
have
ra
br
P (Rt+1 = b) =
and
P (Rt+1 = a) =
.
ba
ba
As a consequence, letting p := (r a)/(b a), when (k1 , . . . , kn ) {a, b}N +1
we have
P (R1 = k1 , . . . , RN = kn ) = (p )l (1 p )N l ,
where l, resp. N l, denotes the number of times the term a, resp. b,
appears in the sequence {k1 , . . . , kN }.
Exercises
Exercise 2.1 We consider the discrete-time Cox-Ross-Rubinstein model with
N + 1 time instants t = 0, 1, . . . , N , and the price t of the riskless asset
evolves as t = 0 (1 + r)t , t = 0, 1, . . . , N . The evolution of St1 to St is
given by
(1 + b)St1
St =
(1 + a)St1
with 1 < a < r < b. The return of the risky asset S is defined as
Rt :=
St St1
,
St1
t = 1, . . . , N,
and Ft is generated by R1 , . . . , Rt , t = 1, . . . , N .
44
"
Discrete-Time Model
a) What are the possible values of Rt ?
b) Show that under the probability measure P defined by
P (Rt+1 = a | Ft ) =
br
,
ba
P (Rt+1 = b | Ft ) =
ra
,
ba
t = 0, 1, . . . , N 1, the expected return IE [Rt+1 | Ft ] of S is equal to the

return r of the riskless asset.
c) Show that under P the process (St )t=0,...,N satisfies
IE [St+k | Ft ] = (1 + r)k St ,
"
t = 0, . . . , N k,
k = 0, . . . , N.
45
Chapter 3
Pricing and hedging in discrete time
We consider the pricing and hedging of options in a discrete time financial

model with N + 1 time instants t = 0, 1, . . . , N . Vanilla options are treated
using backward induction, and exotic options with arbitrary payoff functions
are considered using the Clark-Ocone formula in discrete time.
3.1 Pricing of Contingent Claims

Let us consider an attainable contingent claim with payoff C 0 and maturity N . Recall that by the Definition 2.6 of attainability there exists a hedging
portfolio strategy (t )t=1,2,...,N such that
N SN = C
(3.1)
at time N . Clearly, if (3.1) holds, then investing the amount

V0 = 1 S0
(3.2)
Vt = t St
(3.3)
at time t = 0, resp.
at time t = 1, 2, . . . , N , into a self-financing hedging portfolio will allow one
to hedge the option and to obtain the perfect replication (3.1) at time N .
The value (3.2)-(3.3) at time t of a self-financing portfolio strategy (t )t=1,2,...,N
hedging an attainable claim C will be called an arbitrage price of the claim
C at time t and denoted by t (C), t = 0, 1, . . . , N .
Next we develop a second approach to the pricing of contingent claims,
based on conditional expectations and martingale arguments. We will need
the following lemma.
"
N. Privault
Lemma 3.1. The following statements are equivalent:
(i) The portfolio strategy (t )t=1,...,N is self-financing.
t = t+1 X
t for all t = 1, 2, . . . , N 1.
(ii) t X
(iii) We have
Vet = Ve0 +
t
X
j X
j1 ),
j (X
t = 0, 1, . . . , N.
(3.4)
j=1
Proof. First, the self-financing condition (i)

t1 St1 = t St1 ,
t = 1, . . . , N,
is clearly equivalent to (ii) by division of both sides by (1 + r)t1 .

Next, assuming that (ii) holds we have
Vet = Ve0 +
t
X
(Vej Vej1 )
j=1
= Ve0 +
t
X
j j1 X
j1
j X
j=1
= Ve0 +
t
X
j j X
j1
j X
j=1
= Ve0 +
t
X
j X
j1 ),
j (X
t = 1, 2, . . . , N.
j=1
Finally, assuming that (iii) holds we get

t X
t1 ),
Vet Vet1 = t (X
hence
t t1 X
t1 = t (X
t X
t1 ),
t X
t1 = t X
t1 ,
t1 X
t = 1, 2, . . . , N.
t X
t1 ) represents the profit and loss
In Relation (3.4), the term t (X
t X
t1 ),
Vet Vet1 = t (X
48
"

of the self-financing portfolio strategy (j )j=1,...,N over the time period
[t 1, t], computed by multiplication of the portfolio allocation t with the
t X
t1 , t = 1, 2, . . . , N .
change of price X
Relation (3.4) admits a natural interpretation by saying that when a portfolio is self-financing the value Vet of the (discounted) portfolio at time t is
given by summing up the (discounted) profits and losses registered over all
time periods from time 0 to time t.
The sum (3.4) is also referred to as a discrete time stochastic integral
of the portfolio process (t )t=1,...,N with respect to the random process
t )t=0,1,...,N . In particular, it can be shown from (3.4) that (Vet )t=0,1,...,N
(X
is a martingale under P by the martingale transform argument of Proposition 2.1, as in the proof of Theorem 3.1 below.
As a consequence of the above Lemma 3.1, if a contingent claim C with
discounted payoff
C
e :=
C
(1 + r)N
is attainable by a self-financing portfolio strategy (t )t=1,...,N then we have
e = N X
N = VeN = Ve0 +
C
N
X
t X
t1 ).
t (X
(3.5)
t=1
t that
Note that in the above formula it is the use of discounted asset price X
t X
t1 ) since they are
allows us to add up the profits and losses t (X
expressed in units of currency at time 0. In general, $1 at time t = 0 and
$1 at time t = 1 cannot be added without proper discounting.
Theorem 3.1. The arbitrage price t (C) of a contingent claim C is given
by
1
t (C) =
IE [C | Ft ],
t = 0, 1, . . . , N,
(3.6)
(1 + r)N t
where P denotes any risk-neutral probability measure.
e = C/(1 + r)N denote the discounted payoff of the claim C. We
Proof. Let C
will show that under any risk-neutral measure P the discounted value of any
self-financing portfolio hedging C is given by
h
i
e | Ft ,
Vet = IE C
t = 0, 1, . . . , N,
(3.7)
which shows that
Vt =
"
1
IE [C | Ft ]
(1 + r)N t
49

N. Privault
after multiplication of both sides by (1 + r)t . To conclude, we note that the
arbitrage price t (C) of the claim at any time t is by definition equal to the
value Vt of the corresponding self-financing portfolio, and that (3.7) follows
from the martingale transform result of Proposition 2.1.
In other words, since the portfolio strategy (t )t=1,...,N is self-financing,
from Lemma 3.1 we have
h
i
h
i
e | Ft = IE VeN | Ft
IE C
N

X

e
j (Xj Xj1 )Ft
= IE V0 +
j=1
N
i X

j X
j1 ) | Ft
= IE Ve0 | Ft +
IE j (X
j=1
= Ve0 +
t
X
N
X

j X
j1 ) | Ft
j X
j1 ) | Ft +
IE j (X
IE j (X
j=t+1
j=1
= Ve0 +
t
X
j X
j1 ) +
j (X
j=1
= Vet +
N
X
N
X

j X
j1 ) | Ft
IE j (X
j=t+1

j X
j1 ) | Ft ,
IE j (X
j=t+1
where we used Relation (3.4) of Lemma 3.1. In order to obtain (3.7) we need
to show that
N
X

j X
j1 ) | Ft = 0.
IE j (X
j=t+1
Let us show that

j X
j1 ) | Ft = 0,
IE j (X
for all j = t + 1, . . . , N . We have 0 t j 1 hence Ft Fj1 , and by the
tower property of conditional expectations we get

j X
j1 ) | Ft = IE IE j (X
j X
j1 ) | Fj1 | Ft ,
IE j (X
therefore it suffices to show that

j X
j1 ) | Fj1 = 0.
IE j (X
We note that the porfolio allocation j over the time period [j 1, j] is
predictable, i.e. it is decided at time j 1 and it thus depends only on the
information Fj1 known up to time j 1, hence
50
"

j X
j1 ) | Fj1 = j IE X
j X
j1 | Fj1 .
IE j (X
Finally we note that

j X
j1 | Fj1 = IE X
j | Fj1 IE X
j1 | Fj1
IE X

j1
= IE Xj | Fj1 X
= 0,
j = 1, 2, . . . , N,
t )t=0,1,...,N is a martingale under the risk-neutral measure P , and

because (X
this concludes the proof.

Note that (3.6) admits an interpretation in an insurance framework, in which
t (C) represents an insurance premium and C represents the random value
of an insurance claim made by a subscriber. In this context, the premium
of the insurance contract reads as the average of the values (3.6) of the random claims after time discounting. In addition, the discounted price process
((1 + r)t t (C))t=0,1,...,N is a martingale under P .
As a consequence of Theorem 3.1, the discounted portfolio process (Vet )t=0,1,...,N
is a martingale under P , since
h
i
h h
i
i
e | Ft+1 | Ft
IE Vet+1 | Ft = IE IE C
h
i
e | Ft
= IE C
= Vet ,
t = 0, . . . , N 1,
from the tower property of conditional expectations. In particular, for t = 0

we obtain the price of the contingent claim C at time 0:
h
i
h i
e | F0 = IE C
e =
0 (C) = IE C
1
IE [C].
(1 + r)N
3.2 Hedging of Contingent Claims - Backward Induction

The basic idea of hedging is to allocate assets in a portfolio in order to protect oneself from a given risk. For example, a risk of increasing oil prices
can be hedged by buying oil-related stocks, whose value should be positively
correlated with the oil price. In this way, a loss connected to increasing oil
prices could be compensated by an increase in the value of the corresponding
portfolio.
In the setting of this chapter, hedging an attainable contingent claim C
means computing a self-financing portfolio strategy (t )t=1,...,N such that
"
51
N. Privault
N SN = C, i.e.
N = C,
e
N X
(3.8)
by first solving Equation (3.8) for N . The idea is then to work by backward
induction and to compute successively N 1 , N 2 , . . ., 4 , down to 3 , 2 ,
and finally 1 .
In order to implement this algorithm we may use the self-financing condition which yields N 1 equations written as
t = t+1 X
t,
t X
t = 1, 2, . . . , N 1,
(3.9)
or
t t1 X
t1 = t (X
t X
t1 ),
t X
(3.10)
t = 2, . . . , N 1, allowing us in principle to compute the portfolio strategy

(t )t=1,...,N .
Based on the predictability of (k )k=1,2,...,N we start by solving (3.8) for
N which depends only on information up to time N 1. Then we then use
the values of N to solve the next self-financing condition
N 1 SN 1 = N SN 1
for N 1 , then
N 2 SN 2 = N 1 SN 2
for N 2 , and successively 2 down to 1 , see Exercise 3.4 for an application

in a two-step model. In Section 3.4 the backward induction (3.10) will be
implemented in the CRR model.
Then the discounted value Vet at time t of the portfolio claim can be obtained from
0
Ve0 = 1 X
and
t,
Vet = t X
t = 1, . . . , N.
(3.11)
In the proof of Theorem 3.1 we actually showed that the price t (C) of the
claim at time t coincides with the value Vt of any self-financing portfolio
hedging the claim C, i.e.
t (C) = Vt ,
t = 0, 1, . . . , N,
as given by (3.11). In addition, (3.6) shows that

Vt =
1
IE [C | Ft ],
(1 + r)N t
t = 0, 1, . . . , N,
52
(3.12)
"

hence the price of the claim can be computed either algebraically by solving (3.8) and (3.9) and then using (3.11), or by a probabilistic method by
evaluating the expectation (3.12).
3.3 Pricing of Vanilla Options in the CRR Model

In this section we consider the pricing of contingent claims in the discrete
time Cox-Ross-Rubinstein model, with d = 1. More precisely we are concerned with vanilla options whose payoffs depend on the terminal value of
the underlying asset, as opposed to exotic options whose payoffs may depend
on the whole path of the underlying asset price until expiration time.
Recall that the portfolio value process (Vt )t=0,1,...N and the discounted
portfolio value process respectively satisfy
Vt = t St
and
Vet =
1
t,
Vt = t X
(1 + r)t
t = 1, 2, . . . , N.
Here we will be concerned with the pricing of vanilla options with payoffs of
the form
C = f (SN ),
e.g. f (x) = (x K)+ in the case of a European call. Equivalently, the discounted claim
C
e=
C
(1 + r)N
e = fe(SN ) with fe(x) = f (x)/(1 + r)N , i.e.
satisfies C
fe(x) =
1
+
(x K)
(1 + r)N
in the case of a European call with strike K.

From Theorem 3.1, the discounted value of a portfolio hedging the attaine is given by
able (discounted) claim C
h
i
Vet = IE fe(SN ) | Ft ,
t = 0, 1, . . . , N,
under the risk-neutral measure P . Equivalently, the arbitrage price t (C) of
the contingent claim C = f (SN ) is given by
t (C) =
"
1
IE [f (SN ) | Ft ],
(1 + r)N t
t = 0, 1, . . . , N.
(3.13)
53
N. Privault
In the next proposition we implement the calculation of (3.13).
Proposition 3.1. The price t (C) of the contingent claim C = f (SN ) satisfies
t (C) = v(t, St ),
t = 0, 1, . . . , N,
where the function v(t, x) is given by

N
Y
1

IE f x
(1 + Rj )
v(t, x) =
(1 + r)N t
j=t+1
=
(3.14)

N
t

X
N t
1
j
N tj
(p )j (1 p )N tj f x (1 + b) (1 + a)
.
N
t
(1 + r)
j
j=0
Proof. From the relations

SN = St
N
Y
(1 + Rj ),
j=t+1
and (3.13) we have, using Property (v) of the conditional expectation and
the independence of the returns {R1 , . . . , Rt } and {Rt+1 , . . . , RN },
1
IE [f (SN ) | Ft ]
(1 + r)N t

N

Y
1

=
IE f St
(1 + Rj ) Ft
(1 + r)N t
j=t+1

N
Y
1

(1 + Rj )
IE f x
.
=
(1 + r)N t
j=t+1
t (C) =
x=St
Next we note that the number of times Rj is equal to b for j {t + 1, . . . , N },

has a binomial distribution with parameter (N t, p ), where
p =
ra
ba
and
1 p =
br
,
ba
(3.15)
since the set of paths

from time t + 1 to time N containing j times (1 + b)

has cardinal Njt and each such path has the probability (p )j (1p )N tj ,
j = 0, . . . , N t. Hence we have
1
IE [f (SN ) | Ft ]
(1 + r)N t

N
t

X
1
N t
j
N tj
=
(p )j (1 p )N tj f St (1 + b) (1 + a)
.
N
t
(1 + r)
j
j=0
t (C) =
54
"

In the above proof we have also shown that t (C) is given by the conditional
expectation
1
t (C) =
IE [f (SN ) | St ]
(1 + r)N t
given the value of St at time t = 0, 1, . . . , N , i.e. the price of the claim C is
written as the average (path integral) of the values of the contingent claim
over all possible paths starting from St .
In the CRR model, the discounted portfolio price Vet can be computed by
backward induction as in Section 3.2. Namely, by the tower property of
conditional expectations we have
Vet = ve(t, St )
h
i
= IE fe(SN ) | Ft
h h
i
i
= IE IE fe(SN ) | Ft+1 | Ft
h
i
= IE Vet+1 | Ft
= IE [e
v (t + 1, St+1 ) | Ft ]
= ve (t + 1, (1 + a)St ) P (Rt+1 = a) + ve (t + 1, (1 + b)St ) P (Rt+1 = b)
= (1 p )e
v (t + 1, (1 + a)St ) + p ve (t + 1, (1 + b)St ) ,
which shows that ve(t, x) satisfies the induction relation

ve(t, x) = (1 p )e
v (t + 1, x(1 + a)) + p ve (t + 1, x(1 + b)) ,
(3.16)
while the terminal condition VeN = f(SN ) implies

ve(N, x) = fe(x).
3.4 Hedging of Vanilla Options in the CRR model

In this section we implement the backward induction (3.10) of Section 3.2 for
the hedging of contingent claims in the discrete time Cox-Ross-Rubinstein
model. Our aim is to compute a self-financing portfolio strategy hedging a
vanilla option with payoff of the form
C = f (SN ).
(0)
(1)
Proposition 3.2. The replicating portfolio (t , t )t=1,...,N hedging the

contingent claim C = f (SN ) is given by
"
55
N. Privault
(1)
ve (t, (1 + b)St1 ) ve (t, (1 + a)St1 )

,
Xt1 (b a)/(1 + r)
and
t = 1, 2, . . . , N,
(3.17)
(1)
(0)
ve(t 1, St1 ) t Xt1

(0)
0
t = 1, 2, . . . , N,
(3.18)
where the function ve(t, x) = (1 + r)t v(t, x) is given by (3.14).

Proof. Recall that by Lemma 3.1 the following statements are equivalent:
(i) The portfolio strategy (t )t=1,...,N is self-financing.
(ii) We have
Vet = Ve0 +
t
X
j X
j1 ),
j (X
t = 1, 2, . . . , N.
j=1
As a consequence, any self-financing hedging strategy (t )t=1,...,N should satisfy

t X
t1 ).
ve(t, St ) ve(t 1, St1 ) = Vet Vet1 = t (X
(0)
Note that since the discounted price Xt

(0)
Xt
of the riskless asset satisfies

(0)
= (1 + r)t St
(0)
= 0 ,
we have
t X
t1 ) = t(0) (Xt(0) X (0) ) + t(1) (Xt(1) X (1) )
t (X
t1
t1
(0)
(0)
(0)
(1)
(1)
= t (0 0 ) + t (Xt
=
=
(1)
(1)
(1)
t (Xt Xt1 )
(1)
t (Xt Xt1 ),
(1)
Xt1 )
t = 1, 2, . . . , N.
Hence we have
(1)
ve(t, St ) ve(t 1, St1 ) = t (Xt Xt1 ),
t = 1, 2, . . . , N,
and from this we deduce the two equations

ve (t, (1 + a)St1 ) ve(t 1, St1 ) = t(1) 1 + a Xt1 Xt1 ,
1+r

1+b
(1)
Xt1 Xt1 ,
ve (t, (1 + b)St1 ) ve(t 1, St1 ) = t
1+r
t = 1, 2, . . . , N , i.e.
56
"
(1) a r
ve (t, (1 + a)St1 ) ve(t 1, St1 ) = t 1 + r Xt1 ,
br
ve (t, (1 + b)St1 ) ve(t 1, St1 ) = t(1)

Xt1 ,
1+r
t = 1, 2, . . . , N , hence
(1)
ve (t, (1 + a)St1 ) ve (t 1, St1 )

,
Xt1 (a r)/(1 + r)
t = 1, 2, . . . , N,
(1)
ve (t, (1 + b)St1 ) ve (t 1, St1 )

,
Xt1 (b r)/(1 + r)
t = 1, 2, . . . , N.
t
and
From the obvious relation

(1)
b r (1) a r (1)

,
ba t
ba t
we get
(1)
ve (t, (1 + b)St1 ) ve (t, (1 + a)St1 )

,
Xt1 (b a)/(1 + r)
t = 1, 2, . . . , N,
which only depends on St1 as expected. This is consistent with the fact
(1)
that t represents the (possibly fractional) quantity of the risky asset to be
present in the portfolio over the time period [t 1, t] in order to hedge the
claim C at time N , and is decided at time t 1.
(0)
Concerning the quantity t

t, recall that we have
of the riskless asset in the portfolio at time
t = t(0) Xt(0) + t(1) Xt(1) ,

Vet = t X
t = 1, 2, . . . , N,
hence
(0)
(1) (1)
Vet t Xt
(0)
Xt
=
=
(1) (1)
Vet t Xt
(0)
0
(1) (1)
ve(t, St ) t Xt
,
(0)
0
t = 1, 2, . . . , N.
"
57
N. Privault
Note that we have
(1)
(0)
(1)
ve(t 1, St1 ) + (e
v (t, St ) ve(t 1, St1 )) t Xt
ve(t 1, St1 ) + t (Xt Xt1 ) t Xt
ve(t 1, St1 )
(0)
0
(1)
(1)
(0)
0
(1)
t Xt1
(0)
(1)
t = 1, 2, . . . , N,
which coincides with the discounted self-financing condition, written as

(0)
(1)
ve(t 1, St1 ) = t (0)0 + t Xt1 ,
t = 1, 2, . . . , N.
(1)
We also note, by (3.14) and (3.17), that the quantity t invested in the
risky asset is non-negative (no short selling) when the payoff function f (x)
is nondecreasing.
(0) (0)
In addition, as a consequence of (3.18), the discounted amount t 0

invested on the riskless asset is given by
(0) (0)
t 0 = ve(t 1, St1 )
ve (t, (1 + b)St1 ) ve (t, (1 + a)St1 )

,
(b a)/(1 + r)
(0)
t = 1, 2, . . . , N , and we recover the fact that t

not on St .
(3.19)
depends only on St1 and
Using the relation

v(t 1, St1 ) = (1 + r)t1 ve(t 1, St1 )
the amount (3.19) can be rewritten without discount as
(0)
(0)
t St
(0) (0)
= (1 + r)t t 0
ve (t, (1 + b)St1 ) ve (t, (1 + a)St1 )

(b a)/(1 + r)
v (t, (1 + b)St1 ) v (t, (1 + a)St1 )
= (1 + r)v(t 1, St1 ) (1 + r)
ba
1+r
=
((b a)v(t 1, St1 ) v (t, (1 + b)St1 ) + v (t, (1 + a)St1 )) ,
ba
= (1 + r)t ve(t 1, St1 ) (1 + r)t
(0)
(1)
t = 1, 2, . . . , N . Recall that this portfolio strategy (t , t )t=1,...,N hedges

the claim C = f (SN ), i.e. at time N we have
VN = f (SN ),
58
"

and it is self-financing by Lemma 3.1.
3.5 Hedging of Exotic Options in the CRR Model

In this section we take p = p given by (3.15) and we consider the hedging of
path dependent options. Here we choose to use the finite difference gradient
and the discrete Clark-Ocone formula of stochastic analysis, see also [34],
[70], [88], Chapter 1 of [89], [102], or 15-1 of [116]. See [82] and Section 8.2
of [89] for a similar approach in continuous time. Given
= (1 , . . . , N ) = {1, 1}N ,
and k {1, 2, . . . , N }, let
k
+
= (1 , . . . , k1 , +1, k+1 , . . . , N )
and
k
= (1 , . . . , k1 , 1, k+1 , . . . , N ).
We also assume that the return Rt () is constructed as

t
Rt (+
)=b
and
t
Rt (
) = a,
t = 1, 2, . . . , N,
Definition 3.1. The operator Dt is defined on any random variable F and

t 1 by
t
t
Dt F () = F (+
) F (
),
t = 1, 2, . . . , N.
(3.20)
Recall the following predictable representation formula for the functionals
of the binomial process.
Definition 3.2. Let the centered and normalized return Yt be defined by
br
= q,
t = +1,
Rt r b a
Yt :=
=
t = 1, . . . , N.
ba
ar
= p, t = 1,
ba
Note that under the risk-neutral measure P we have

Rt r
IE [Yt ] = IE
ba
ar
br
=
P (Rt = a) +
P (Rt = b)
ba
ba
ar br
br ra
=
+
baba baba
"
59
N. Privault
= 0,
and
Var [Yt ] = pq 2 + qp2 = pq,
t = 1, 2, . . . , N.
In addition the discounted asset price increment reads

Xt Xt1 = Xt1
1 + Rt
Xt1
1+r
1
Xt1 (Rt r)
1+r
ba
=
Yt Xt1 ,
t = 1, . . . , N.
1+r
We also have
Dt Yt =
br
ra
+
= 1,
ba ba
t = 1, 2, . . . , N,
and
Dk SN = S0 (1 + b)
N
Y
(1 + Rt ) S0 (1 + a)
t=1
t6=k
= S0 (b a)
N
Y
N
Y
(1 + Rt )
t=1
t6=k
(1 + Rt )
t=1
t6=k
= S0
=
N
ba Y
(1 + Rt )
1 + Rk t=1
ba
SN ,
1 + Rk
k = 1, . . . , N.
The next proposition is the Clark-Ocone predictable representation formula

in discrete time, cf. e.g. [89], Proposition 1.7.1.
Proposition 3.3. For any square-integrable random variables F on we
have
X
F = IE [F ] +
IE [Dk F |Fk1 ]Yk .
(3.21)
k=1
The Clark-Ocone formula has the following consequence.

Corollary 3.1. Assume that (Mk )kN is a square-integrable Ft -martingale.
Then we have
MN = IE [MN ] +
N
X
Yk Dk Mk ,
N 0.
k=1
60
"

Proof. We have
MN = IE [MN ] +
= IE [MN ] +
= IE [MN ] +
X
k=1
X
k=1
IE [Dk MN |Fk1 ]Yk

Dk IE [MN |Fk ]Yk
Yk Dk Mk
k=1
= IE [MN ] +
N
X
Yk Dk Mk .
k=1

In addition to the Clark-Ocone formula we also state a discrete-time analog of
Itos change of variable formula, which can be useful for option hedging. The
next result extends Proposition 1.13.1 of [89] by removing the unnecessary
martingale requirement on (Mt )nN .
Proposition 3.4. Let (Zn )nN be an Fn -adapted process and let f : R
N R be a given function. We have
f (Zt , t) = f (Z0 , 0) +
t
X
Dk f (Zk , k)Yk
k=1
t
X
(IE [f (Zk , k)|Fk1 ] f (Zk1 , k 1)) .
(3.22)
k=1
Proof. First, we note that the process

t 7 f (Zt , t)
t
X
(IE [f (Zk , k)|Fk1 ] f (Zk1 , k 1))
k=1
is a martingale under P . Indeed we have

"
#
t

X

IE f (Zt , t)
(IE [f (Zk , k)|Fk1 ] f (Zk1 , k 1)) Ft1
k=1
= IE [f (Zt , t)|Ft1 ]
t
X
(IE [IE [f (Zk , k)|Fk1 ]|Ft1 ] IE [IE [f (Zk1 , k 1)|Fk1 ]|Ft1 ])

k=1
= IE [f (Zt , t)|Ft1 ]
t
X
(IE [f (Zk , k)|Fk1 ] f (Zk1 , k 1))
k=1
"
61
N. Privault
= f (Zt1 , t 1)
t1
X
(IE [f (Zk , k)|Fk1 ] f (Zk1 , k 1)) ,
t 1.
k=1

2
Note that if (Zt )tN is a martingale in L () with respect to (Ft )tN and
written as
t
X
Zt = Z0 +
uk Yk ,
t N,
k=1
where (ut )tN is a predictable process locally in L2 (N), (i.e. u()1[0,N ] ()

L2 ( N) for all N > 0), then we have
Dt f (Zt , t) = f (Zt1 + qut , t) f (Zt1 put , t) ,
(3.23)
t = 1, 2, . . . , N . On the other hand, the term

IE[f (Zt , t) f (Zt1 , t 1)|Ft1 ]
is analog to the finite variation part in the continuous time It
o formula, and
can be written as
pf (Zt1 + qut , t) + qf (Zt1 put , t) f (Zt1 , t 1) .
Naturally, if (f (Zt , t))tN is a martingale we recover the decomposition
f (Zt , t) = f (Z0 , 0)
t
X
(f (Zk1 + quk , k) f (Zk1 puk , k)) Yk
+
k=1
= f (Z0 , 0) +
t
X
Yk Dk f (Zk , k).
(3.24)
k=1
This identity follows from Corollary 3.1 as well as from Proposition 3.3. In
this case the Clark-Ocone formula (3.21) and the change of variable formula
(3.24) both coincide and we have in particular
Dk f (Zk , k) = IE[Dk f (ZN , N )|Fk1 ],
k = 1, 2, . . . , N . For example this recovers the martingale representation
Xt = S0 +
t
X
Yk Dk Xk
k=1
= S0 +
t
ba X
Xk1 Yk
1+r
k=1
62
"
= S0 +
= S0 +
t
X
Xk1
k=1
t
X
Rk r
1+r
(Xk Xk1 ),
k=1
of the discounted asset price.

Our goal is to hedge an arbitrary claim C on , i.e. given an FN measurable random variable C we search for a portfolio (t , t )t=1,...,N such
that the equality
C = VN = N AN + N SN
(3.25)
holds, where AN = A0 (1 + r)N denotes the value of the riskless asset at time
N N.
The next proposition is the main result of this section, and provides a
solution to the hedging problem under the constraint (3.25).
Proposition 3.5. Given C a contingent claim, let
(1 + r)(N t)
IE [Dt C|Ft1 ],
St1 (b a)
(3.26)

1
(1 + r)(N t) IE [C|Ft ] t St ,
At
(3.27)
t =
t = 1 . . . , N , and
t =
t = 1 . . . , N . Then the portfolio (t , t )t=1...,N is self financing and satisfies

Vt = t At + t St = (1 + r)(N t) IE [C|Ft ],
t = 1 . . . , N,
in particular we have VN = C, hence (t , t )t=1...,N is a hedging strategy

leading to C.
Proof. Let (t )t=1...,N be defined by (3.26), and consider the process (t )t=0,1...,N
defined by
0 = (1 + r)N
IE [C]
S0
and t+1 = t
(t+1 t )St
,
At
t = 0, . . . , N 1. Then (t , t )t=1,...,N satisfies the self-financing condition

At (t+1 t ) + St (t+1 t ) = 0,
t = 1, 2, . . . , N 1.
Let now
"
63
N. Privault
V0 :=
1(1 + r)N
[C],
IE
and
Vet =
Vt := t At + t St ,
Vt
(1 + r)t
t = 1, 2, . . . , N,
t = 0, . . . , N.
Since (t , t )t=1,...,N is self-financing, by Lemma 3.1 we have

Vet = Ve0 + (b a)
t
X
k=1
1
Yk k Sk1 ,
(1 + r)k
(3.28)
t = 1, 2, . . . , N . On the other hand, from the Clark-Ocone formula (3.21) and

the definition of (t )t=1,...,N we have
1
IE [C|Ft ]
(1 + r)N
=
"
#
N

X
1

IE
IE
[C]
+
Y
I
E
[D
C|F
]
F
i
i
i1
t
(1 + r)N
i=0
t
X
1
1
IE
[C]
+
IE [Di C|Fi1 ]Yi
(1 + r)N
(1 + r)N i=0
t
X
1
1
IE [C] + (b a)
i Si1 Yi
N
(1 + r)
(1
+
r)i
i=0
= Vet
from (3.28). Hence
Vet =
1
IE [C|Ft ],
(1 + r)N
t = 0, 1, . . . , N,
and
Vt = (1 + r)(N t) IE [C|Ft ],
t = 0, 1, . . . , N.
(3.29)
In particular, (3.29) shows that we have VN = C. To conclude the proof we

note that from the relation Vt = t At + t St , t = 1, 2, . . . , N , the process
(t )t=1,...,N coincides with (t )t=1,...,N defined by (3.27).

From Proposition 3.1, when C = f (SN ), the price t (C) of the contingent
claim C = f (SN ) is given by
t (C) = v(t, St ),
where the function v(t, x) is given by
64
"

N
Y
1
1

v(t, St ) =
IE [C|Ft ] =
IE f x
(1 + Rj )
(1 + r)N t
(1 + r)N t
j=t+1
x=St
Note that in this case we have C = v(N, SN ), IE[C] = v(0, M0 ), and the
e = C/(1 + r)N = ve(N, SN ) satisfies
discounted claim payoff C
n
h i X
e = IE C
e +
C
Yt IE [Dt ve(N, SN )|Ft1 ]
t=1
n
h i X
e +
= IE C
Yt Dt ve(t, St )
t=1
n
h i X
e +
= IE C
h i
e +
= IE C
t=1
n
X
1
Yt Dt v(t, St )
(1 + r)t
Yt Dt IE [e
v (N, SN )|Ft ]
t=1
h i
e +
= IE C
n
X
1
Yt Dt IE[C|Ft ],
(1 + r)N t=1
hence we have
IE[Dt v(N, SN )|Ft1 ] = (1 + r)N t Dt v(t, St ),
t = 1, 2, . . . , N,
and by Proposition 3.5 the hedging strategy for C = f (SN ) is given by

(1 + r)(N t)
IE[Dt v(N, SN )|Ft1 ]
St1 (b a)
1
=
Dt v(t, St )
St1 (b a)
1
=
(v (t, St1 (1 + b)) v (t, St1 (1 + a)))
St1 (b a)
1
=
(e
v (t, St1 (1 + b)) ve (t, St1 (1 + a))) ,
Xt1 (b a)/(1 + r)
t =
t = 1, 2, . . . , N , which recovers Proposition 3.2 as a particular case. Note that

t is non-negative (i.e. there is no short-selling) when f is a non decreasing
function, because a < b. This is in particular true in the case of a European
call option for which we have f (x) = (x K)+ .
"
65
N. Privault
3.6 Convergence of the CRR Model

In this section we consider the convergence of the discrete time model to the
continuous-time Black Scholes model.
Continuous compounding - riskless asset

Consider the subdivision

(N 1)T
T 2T
,...,
,T
0, ,
N N
N
of the time interval [0, T ] into N time steps. Note that
lim (1 + r)N = ,
thus we need to renormalize r so that the interest rate on each time interval
becomes rN , with limN rN = 0. It turns out that the correct renormalization is
T
rN = r ,
N
so that

N
T
lim (1 + rN )N = lim 1 + r
= erT ,
T R+ .
(3.30)
N
N
N
Hence the price At of the riskless asset is given by
At = A0 ert ,
t R+ ,
(3.31)
which solves the differential equation

dAt
= rAt ,
dt
A0 = 1,
t R+ ,
(3.32)
also written as
dAt = rAt dt,
or
dAt
= rdt,
At
which means that the return of the riskless asset is rdt on the small time
interval [t, t + dt]. Equivalently, one says that r is the instantaneous interest
rate per unit of time.
The same equation rewrites in integral form as
66
"

AT A0 =
wT
0
dAt = r
wT
0
At dt.
Continuous compounding - risky asset

We need to apply a similar renormalization to the coefficients a and b of the
CRR model. Let > 0 denote a positive parameter called the volatility and
let aN , bN be defined from
1 + aN
= e T /N
1 + rN
and
1 + bN
= e T /N ,
1 + rN
and
bN = (1 + rN )e
i.e.
aN = (1 + rN )e
T /N
T /N
1.
(N )
Consider the random return Rk {aN , bN } and the price process defined
as
t
Y
(N )
(t)
St = S0
(1 + Rk ),
t = 1, 2, . . . , N.
k=1
Note that the risk-neutral probabilities are given by
bN rN
e T /N 1
p
P (Rt = aN ) =
=
,
bN a N
2 sinh 2 T /N
t = 1, . . . , N,
and
rN aN
1 e T /N
p
P (Rt = bN ) =
=
,
bN aN
2 sinh 2 T /N
t = 1, . . . , N,
which both converge to 1/2 as N goes to infinity.
Continuous-time limit
We have the following convergence result.
Proposition 3.6. Let f be a continuous and bounded function

on R. The
(N )
price at time t = 0 of a contingent claim with payoff C = f SN
converges
as follows:
h
i
h
i
2
1
(N )
IE f (SN ) = erT IE f (S0 e T X+rT T /2 )
N
N (1 + rT /N )
(3.33)
where X ' N (0, 1) is a standard Gaussian random variable.
lim
"
67
N. Privault
Proof. This result is consequence of the weak convergence in distribution of
(N )
the sequence (SN )N 1 to a lognormal distribution, cf. Theorem 5.53 of [34].
The convergence of the discount factor follows directly from (3.30).

Note that the expectation (3.33) can be written as a Gaussian integral:
2
h
w
i
ex /2
2
2
erT IE f S0 e T X+rT T /2 = erT
f S0 e T x+rT T /2
dx,
hence we have
2
h
i
w
ex /2
2
1
(N )
IE f SN
= erT
f S0 e T x+rT T /2
dx.
N
N (1 + rT /N )
2
lim
It is a remarkable fact that in case f (x) = (xK)+ , i.e. when C = (ST K)+
is the payoff of a European call option with strike K, the above integral can
be computed according to the Black-Scholes formula:
h
i
2
erT IE (S0 e T X+rT T /2 K)+ = S0 (d+ ) KerT (d ),
where
d =
and
(r 2 /2)T + log(S0 /K)
,
T
d+ = d + T ,
1 w x y2 /2
(x) =
e
dy,
2
x R,
is the Gaussian cumulative distribution function.

The Black-Scholes formula will be derived explicitly in the subsequent
chapters using both the PDE and probabilistic method, cf. Propositions 1.12nd
6.4. It can be considered as a building block for the pricing of financial derivatives, and its importance is not restricted to the pricing of options on stocks.
Indeed, the complexity of the interest rate models makes it in general difficult
to obtain closed form expressions, and in many situations one has to rely on
the Black-Scholes framework in order to find pricing formulas, for example
in the case of interest rate derivatives as in the Black caplet formula of the
BGM model, cf. Proposition 12.3 in Section 12.3.
Our aim later on will be to price and hedge options directly in continuous
time using stochastic calculus, instead of applying the limit procedure described in the previous section. In addition to the construction of the riskless
asset price (At )tR+ via (3.31) and (3.32) we now need to construct a mathematical model for the price of the risly asset in continuous time.
68
"

The return of the risky asset St over the time period [t, d + dt] will be
defined as
dSt
= dt + dBt ,
St
where dBt is a small Gaussian random component, also called Brownian
increment, parametrized by the volatility parameter > 0. In the next Chapter 4 we will turn to the formal definition of the stochastic process (Bt )tR+
which will be used for the modeling of risky assets in continuous time.
Exercises
Exercise 3.1 (Exercise 2.1 continued)
a) We consider a forward contract on SN with strike K and payoff
C := SN K.
Find a portfolio allocation (N , N ) with price VN = N N + N SN at
time N , such that
VN = C,
(3.34)
by writing Condition (3.34) as a 2 2 system of equations.
b) Find a portfolio allocation (N 1 , N 1 ) with price VN 1 = N 1 N 1 +
N 1 SN 1 at time N 1, and verifying the self-financing condition
VN 1 = N N 1 + N SN 1 .
Next, at all times t = 1, . . . , N 1, find a portfolio allocation (t , t ) with
price Vt = t t + t St verifying (3.34) and the self-financing condition
Vt = t+1 t + t+1 St ,
where t , resp. t , represents the quantity of the riskless, resp. risky, asset
in the portfolio over the time period [t 1, t], t = 1, . . . , N .
c) Compute the arbitrage price t (C) = Vt of the forward contract C, at
time t = 0, 1, . . . , N .
d) Check that the arbitrage price t (C) satisfies the relation
t (C) =
1
IE [C | Ft ],
(1 + r)N t
t = 0, 1, . . . , N.
Exercise 3.2 Consider the discrete-time Cox-Ross-Rubinstein model with

N + 1 time instants t = 0, 1, . . . , N . The price St0 of the riskless asset evolves
as St0 = 0 (1 + r)t , t = 0, 1, . . . , N . The return of the risky asset, defined as
"
69
N. Privault
Rt :=
St St1
,
St1
t = 1, . . . , N,
is random and allowed to take only two values a and b, with 1 < a < r < b.
The discounted asset price is Xt = St /(1 + r)t , t = 0, 1, . . . , N .
a) Show that this model admits a unique risk-neutral measure P and explicitly compute P (Rt = a) and P (Rt = b) for all t = 1, . . . , N .
b) Does there exist arbitrage opportunities in this model ? Explain why.
c) Is this market model complete ? Explain why.
d) Consider a contingent claim with payoff
C = (SN )2 .
Compute the discounted arbitrage price Vt , t = 0, . . . , N , of a selffinancing portfolio hedging the claim C, i.e. such that
(SN )2
VN = C =
.
(1 + r)N
e) Compute the portfolio strategy
(t )t=1,...,N = (t0 , t1 )t=1,...,N
associated to Vt , i.e. such that
t = 0X 0 + 1X 1,
Vt = t X
t t
t t
t = 1, . . . , N.
f) Check that the above portfolio strategy is self-financing, i.e.

t+1 St = t St ,
t = 1, . . . , N 1.
Exercise 3.3 We consider the discrete-time Cox-Ross-Rubinstein model with

N + 1 time instants t = 0, 1, . . . , N .
The price t of the riskless asset evolves as t = 0 (1 + r)t , t = 0, 1, . . . , N .
The evolution of St1 to St is given by
(1 + b)St1
St =
(1 + a)St1
with 1 < a < r < b. The return of the risky asset is defined as
This is the payoff of a power call option with strike K = 0.
70
"
Rt :=
St St1
,
St1
t = 1, . . . , N.
Let t , resp. t , denote the (possibly fractional) quantities of the risky, resp.
riskless, asset held over the time period [t 1, t] in the portfolio with value
Vt = t St + t t ,
t = 0, . . . , N.
(3.35)
a) Show that
Vt = (1 + Rt )t St1 + (1 + r)t t1 ,
t = 1, . . . , N.
(3.36)
b) Show that under the probability P defined by

P (Rt = a | Ft1 ) =
br
,
ba
P (Rt = b | Ft1 ) =
ra
,
ba
where Ft1 represents the information generated by {R1 , . . . , Rt1 }, we

have
IE [Rt | Ft1 ] = r.
c) Under the self-financing condition
Vt1 = t St1 + t t1
show that
Vt1 =
t = 1, . . . , N,
(3.37)
1
IE [Vt | Ft1 ],
1+r
using the result of Question (a).

d) Let a = 5%, b = 25% and r = 15%. Assume that the price Vt at time t of
the portfolio is $3 if Rt = a and $8 if Rt = b, and compute the price Vt1
of the portfolio at time t 1.
Exercise 3.4 Consider a two-step binomial random asset model (Sk )k=0,1,2
with possible returns a = 0 and b = 200%, and a risk-free asset Ak = A0 (1 +
r)k , k = 0, 1, 2 with interest rate r = 100%, and S0 = A0 = 1, under the
risk-neutral measure p = (r a)/(b a) = 1/2.
a) Draw a binomial tree for the possible values of (Sk )k=0,1,2 and compute
the values Vk of the hedging portfolio at times k = 0, 1, 2 of a European
call option with strike K = 8 on ST , with maturity T = 2.
Hint: You need to consider three cases for k = 2 and two cases for k = 1.
b) Compute the self-financing hedging portfolio (k , k )k=1,2 with price
Vk = k Sk + k Ak = k Sk1 + k Ak1 ,
"
k = 1, 2,
71
N. Privault
hedging the European call with strike K = 8 and maturity T = 2.
Hint: Consider two separate cases for k = 2 and one case for k = 1.
72
"
Chapter 4
Brownian Motion and Stochastic Calculus
The modeling of random assets in finance is based on stochastic processes,

which are families (Xt )tI of random variables indexed by a time interval I. In
this chapter we present a description of Brownian motion and a construction
of the associated Ito stochastic integral.
4.1 Brownian Motion

We start by recalling the definition of Brownian motion, which is a fundamental example of a stochastic process. The underlying probability space
(, F, P) of Brownian motion can be constructed on the space = C0 (R+ )
of continuous real-valued functions on R+ started at 0.
Definition 4.1. The standard Brownian motion is a stochastic process
(Bt )tR+ such that
(i) B0 = 0 almost surely,
(ii) The sample trajectories t 7 Bt are continuous, with probability 1.
(iii) For any finite sequence of times t0 < t1 < < tn , the increments
Bt1 Bt0 , Bt2 Bt1 , . . . , Btn Btn1
are independent.
(iv) For any given times 0 s < t, Bt Bs has the Gaussian distribution
N (0, t s) with mean zero and variance t s.
"
N. Privault
We refer to Theorem 10.28 of [33] and to Chapter 1 of [98] for the proof of
the existence of Brownian motion as a stochastic process (Bt )tR+ satisfying
the above properties (i)-(iv).
In particular, Condition (iv) above implies
IE[Bt Bs ] = 0
and
Var[Bt Bs ] = t s,
0 s t,
and we have
Cov(Bs , Bt ) = IE[Bs Bt ] = IE[Bs (Bt Bs )] + IE[(Bs )2 ] = s,
0 s t,
hence
Cov(Bs , Bt ) = IE[Bs Bt ] = min(s, t),
s, t R+ ,
cf. also Exercise 4.1-(d).

In the sequel the filtration (Ft )tR+ will be generated by the Brownian
paths up to time t, in other words we write
Ft = (Bs : 0 s t),
t 0.
(4.1)
A random variable F is said to be Ft -measurable if the knowledge of F

depends only on the information known up to time t. As an example, if
t =today,
the date of the past course exam is Ft -measurable, because it belongs to
the past.
the date of the next Chinese new year, although it refers to a future event,
is also Ft -measurable because it is known at time t.
the date of the next typhoon is not Ft -measurable since it is not known
at time t.
the maturity date T of a European option is Ft -measurable for all t R+ ,
because it has been determined at time 0.
the exercise date of an American option after time t (see Section 9.4) is
not Ft -measurable because it refers to a future random event.
Property (iii) above shows that Bt Bs is independent of all Brownian increments taken before time s, i.e.
(Bt Bs )
(Bt1 Bt0 , Bt2 Bt1 , . . . , Btn Btn1 ),
74
"

0 t0 t1 tn s t, hence Bt Bs is also independent of the
whole Brownian history up to time s, hence Bt Bs is in fact independent
of Fs , s 0.
As a consequence, Brownian motion is a continuous-time martingale as we
have
IE[Bt | Fs ] = IE[(Bt Bs ) | Fs ] + IE[Bs | Fs ]
= IE[Bt Bs ] + Bs
0 s t,
= Bs ,
because it has centered and independent increments, cf. Section 6.1.

For convenience we will informally regard Brownian motion as a random
walk over infinitesimal time intervals of length t, with increments
Bt := Bt+t Bt
over the time interval [t, t + t] given by
Bt = t
(4.2)
with equal probabilities (1/2, 1/2).

The choice of the square root in (4.2) is in fact not fortuitous. Indeed, any
choice of (t) with a power > 1/2 would lead to explosion of the process
as dt tends to zero, whereas a power (0, 1/2) would lead to a vanishing
process.
Note that we have
IE[Bt ] =
1
1
t
t = 0,
2
2
and
1
1
t + t = t.
2
2
According to this representation, the paths of Brownian motion are not differentiable, although they are continuous by Property (ii), as we have
dt
1
dBt
'
= ' .
(4.3)
dt
dt
dt
Var[Bt ] = IE[(Bt )2 ] =
After splitting the interval [0, T ] into N intervals

k1
k
T, T ,
k = 1, . . . , N,
N
N
"
75
N. Privault
of length t = T /N with N large, and letting
Xk = T = N t = N Bt
with probabilities (1/2, 1/2) we have V ar(Xk ) = T and
Xk
Bt = = t
N
is the increment of Bt over ((k 1)t, kt], and we get
BT '
Bt '
0<t<T
X1 + + XN
.
N
Hence by the central limit theorem we recover the fact that BT has a centered
Gaussian distribution with variance T , cf. point (iv) of the above definition
of Brownian motion. Indeed, the central limit theorem states that given any
sequence (Xk )k1 of independent identically distributed centered random
variables with variance 2 = Var[Xk ] = T , the normalized sum
X1 + + Xn
n
converges (in distribution) to a centered Gaussian random variable N (0, 2 )
with variance 2 as n goes to infinity. As a consequence, Bt could in fact
be replaced by any centered random variable with variance t in the above
description.
2
1.5
Bt
0.5
-0.5
-1
-1.5
-2
0
0.2
0.4
0.6
0.8
Fig. 4.1: Sample paths of a one-dimensional Brownian motion.
In Figure 4.1 we draw three sample paths of a standard Brownian motion

obtained by computer simulation using (4.2). Note that there is no point
in computing the value of Bt as it is a random variable for all t > 0,
76
"

however we can generate samples of Bt , which are distributed according to
the centered Gaussian distribution with variance t.
2
1.5
0.5
-0.5
-1
-1.5
-2
-2
-1.5
-1
-0.5
0.5
1.5
2.5
Fig. 4.2: Two sample paths of a two-dimensional Brownian motion.

The n-dimensional Brownian motion can be constructed as (Bt1 , . . . , Btn )tR+
where (Bt1 )tR+ , . . .,(Btn )tR+ are independent copies of (Bt )tR+ . Next, we
turn to simulations of 2 dimensional and 3 dimensional Brownian motions in
Figures 4.2 and 4.3. Recall that the movement of pollen particles originally
observed by R. Brown in 1827 was indeed 2-dimensional.
2
1.5
1
0.5
0
-0.5
-2
-1.5
-1
-1
-0.5
-1.5
-2 -2
-1.5
-1
0
0.5
-0.5
0.5
1.5
1
1.5
2
Fig. 4.3: Sample paths of a three-dimensional Brownian motion.
"
77
N. Privault
4.2 Wiener Stochastic Integral

In this section we construct the Ito stochastic integral of square-integrable
deterministic function with respect to Brownian motion.
Recall that Bachelier originally modeled the price St of a risky asset by
St = Bt where is a volatility parameter. The stochastic integral
wT
0
f (t)dSt =
wT
0
f (t)dBt
can be used to represent the value of a portfolio as a sum of profits and

losses f (t)dSt where dSt represents the stock price variation and f (t) is the
quantity invested in the asset St over the short time interval [t, t + dt].
A naive definition of the stochastic integral with respect to Brownian motion would consist in writing
w
0
f (t)dBt =
w
0
f (t)
dBt
dt,
dt
and evaluating the above integral with respect to dt. However this definition
fails because the paths of Brownian motion are not differentiable, cf. (4.3).
Next we present Itos construction of the stochastic integral with respect to
Brownian motion. Stochastic integrals will be first constructed as integrals
of simple step functions of the form
f (t) =
n
X
t R+ ,
ai 1(ti1 ,ti ] (t),
(4.4)
i=1
i.e. the function f takes the value ai on the interval (ti1 , ti ], i = 1, . . . , n,

with 0 t0 < < tn , as illustrated in Figure 4.4.
f
6
a2
a1
a4
t0
t1
t2
t3
t4
Fig. 4.4: Step function.

In the sequel we will make a repeated use of the space L2 (R+ ) of measurable
functions f : R+ R, called square-integrable functions, endowed with the
norm
78
"
kf kL2 (R+ ) :=
rw
|f (t)|2 dt < ,
f L2 (R+ ),
(4.5)
which induces the distance

kf gkL2 (R+ ) :=
rw
|f (t) g(t)|2 dt < ,
between f and g in L2 (R+ ), cf. e.g. Chapter 3 of [101] for details.

Note that the set of simple step functions f of the form (4.4) is a linear
space which is dense in L2 (R+ ) for the norm (4.5), cf. e.g. Theorem 3.13 in
[101], namely, given f a function satisfying (4.7) and (fn )nN a sequence of
simple functions converging to f for the norm
kf fn kL2 (R+ ) :=
w
0
1/2
|f (t) fn (t)|2 dt
Recall that the classical integral of f given in (4.4) is interpreted as the area
under the curve f and computed as
w
f (t)dt =
n
X
ai (ti ti1 ).
i=1
In the next definition we adapt this construction to the setting of stochastic

integration with respect to Brownian motion.
Definition 4.2. The stochastic integral with respect to Brownian motion
(Bt )tR+ of the simple step function f of the form (4.4) is defined by
w
0
f (t)dBt :=
n
X
ai (Bti Bti1 ).
(4.6)
i=1
In the next Proposition 4.1 we determine the probability distribution of

w
f (t)dBt and we show that it is independent of the particular representa0

tion (4.4) chosen for f (t).
w
Proposition 4.1. The stochastic integral
f (t)dBt defined in (4.6) has a
0
centered Gaussian distribution
w

w
|f (t)|2 dt
f (t)dBt ' N 0,
0
with mean IE
Var
"
hw
0
hw
0
f (t)dBt = 0 and variance given by the Ito isometry

w
i
2 w
|f (t)|2 dt.
f (t)dBt = IE
f (t)dBt
=
0
79
N. Privault
Proof. Recall that if X1 , . . . , Xn are independent Gaussian random variables
with probability laws N (m1 , 12 ), . . . , N (mn , n2 ) then then sum X1 + +Xn
is a Gaussian random variable with distribution
N (m1 + + mn , 12 + + n2 ).
As a consequence, when f is the simple function
f (t) =
n
X
t R+ ,
ai 1(ti1 ,ti ] (t),
i=1
the sum
w
0
f (t)dBt =
n
X
ak (Btk Btk1 )
k=1
has a centered Gaussian distribution with variance

n
X
|ak |2 (tk tk1 ),
k=1
since
Var [ak (Btk Btk1 )] = a2k Var [Btk Btk1 ] = a2k (tk tk1 ),
hence the stochastic integral
w
0
f (t)dBt =
n
X
ak (Btk Btk1 )
k=1
of the step function

f (t) =
n
X
ak 1(tk1 ,tk ] (t)
k=1
has a centered Gaussian distribution with variance

Var
hw
0
n
i X
f (t)dBt =
|ak |2 (tk tk1 )
k=1
n
X
|ak |2
k=1
=
=
n
wX
0
w
0
w tk
tk1
dt
|ak |2 1(tk1 ,tk ] (t)dt
k=1
|f (t)|2 dt.
Finally we note that

80
"
Var
hw
0
w
i
2 hw
i2
f (t)dBt = IE
f (t)dBt
IE
f (t)dBt
0
0
w
2
= IE
f (t)dBt
.
0
In order to exted the definition (4.6) of the stochastic integral
w
0

f (t)dBt
to any function f L2 (R+ ), i.e. to f : R R measurable such that

w
|f (t)|2 dt < ,
(4.7)
0
we will make use of the space L2 () of random variables F : R called

square-integrable random variables, endowed with the norm
p
kF kL2 (R+ ) := IE[F 2 ] < ,
which induces the distance
kF GkL2 (R+ ) :=
IE[(F G)2 ] < ,
between the square-integrable random variables F and g in L2 ().

w
Proposition 4.2. The definition (4.6) of the stochastic integral
f (t)dBt
0
w
2
can be extended to any function f L (R+ ). In this case,
f (t)dBt has a
0
centered Gaussian distribution
w

w
f (t)dBt ' N 0,
|f (t)|2 dt
0
with mean IE
Var
hw
0
hw
0
f (t)dBt = 0 and variance given by the It

o isometry
w
i
2 w
f (t)dBt = IE
f (t)dBt
=
|f (t)|2 dt.
0
(4.8)
Proof. The extension of the stochastic integral to all functions satisfying (4.7)
is obtained by density and a Cauchy sequence argument, based on the isometry relation (4.8). Namely, given f a function satisfying (4.7) and (fn )nN a
sequence of simple functions converging to f for the norm
kf fn kL2 (R+ ) :=
w
0
1/2
|f (t) fn (t)|2 dt
2
i.e.
shows that
r in L (R+), cf. Theorem 3.13 in [101], the isometry (4.8)
fn (t)dBt nN is a Cauchy sequence in the space L2 () of square0
integrable random variables
"
81
N. Privault
F : R
such that
kF k2L2 (R+ ) := IE[F 2 ] < .
Indeed, by the triangle inequality for the L2 ()-norm
kf gkL2 () kf hkL2 () + kh gGkL2 () ,
f, g, h L2 (R+ ),
we have
w

w

fk (t)dBt
fn (t)dBt

0

=
IE
w
L2 ()
fk (t)dBt
w
0
fn (t)dBt
2 1/2
= kfk fn kL2 (R+ )

kf fk kL2 (R+ ) + kf fn kL2 (R+ ) ,

r
which tends to 0 as k, n tend to infinity. Since the sequence 0 fn (t)dBt nN
2
is Cauchy and the space L () is complete,
3.11 in [101] or
w cf. e.g. Theorem

Chapter 4 of [25], we conclude that
fn (t)dBt
converges for the
nN
L2 -norm to a limit in L2 (). In this case we let

w
w
f (t)dBt := lim
fn (t)dBt
n
and the uniqueness of this limit, which also satisfies (4.8), can be shown from
(4.8).

w
t
For example,
e dBt has a centered Gaussian distribution with variance
0

1
1
= .
e2t dt = e2t
0
2
2
0
w
Again, the Wiener stochastic integral
f (s)dBs is nothing but a Gaussian
0
random variable and it cannot be computed in the way standard integral
are computed via the use of primitives. However, when f L2 (R+ ) is C 1 , i.e.
when f is continuously differentiable on R+ , we have the following formula
w
w
f (t)dBt =
f 0 (t)Bt dt,
(4.9)
0
provided limt t|f (t)| = 0 and f L (R+ ), cf. e.g. Remark 2.5.9 in [89].
This triangle inequality follows from the Minkowski inequality.
82
"
4.3 It
o Stochastic Integral
In this section we extend the Wiener stochastic integral to square-integrable
adapted processes. Recall that a process (Xt )tR+ is said to be Ft -adapted if
Xt is Ft -measurable for all t R+ , where the information flow (Ft )tR+ has
been defined in (4.1).
Recall, as examples, that
- the process (Bt )tR+ is adapted,
- the process (Bt+1 )tR+ is not adapted,
- the process (Bt/2 )tR+ is adapted,
- the process (Bt )tR+ is not adapted.
In other words, a process (Xt )tR+ is Ft -adapted if the value of Xt at time t
depends only on information known up to time t. Note that the value of Xt
may still depend on known future data, for example a fixed future date in
the calendar, such as a maturity time T > t, as long as its value is known at
time t.
The extension of the stochastic integral to adapted random processes is
actually necessary in order to compute a portfolio value when the portfolio
process is no longer deterministic. This happens in particular when one needs
to update the portfolio allocation based on random events occurring on the
market.
Stochastic integrals of adapted processes will be first constructed as integrals
of simple predictable processes (ut )tR+ of the form
ut :=
n
X
Fi 1(ti1 ,ti ] (t),
t R+ ,
(4.10)
i=1
where Fi is an Fti1 -measurable random variable for i = 1, . . . , n. For example, a natural approximation of (Bt )tR+ by a simple predictable process can
be constructed as
ut :=
n
X
Bti1 1(ti1 ,ti ] (t),
t R+ ,
i=1
since Bti1 is Fti1 -measurable for i = 1, . . . , n.

The notion of simple predictable process is natural in the context of portfolio investment, in which Fi will represent an investment allocation decided
"
83
N. Privault
at time ti1 and to remain unchanged over the time period (ti1 , ti ].
By convention, u : R+ R is denoted in the sequel by ut (),
t R+ , , and the random outcome is often dropped for convenience
of notation.
Definition 4.3. The stochastic integral with respect to Brownian motion
(Bt )tR+ of any simple predictable process (ut )tR+ of the form (4.10) is
defined by
n
w
X
ut dBt :=
Fi (Bti Bti1 ).
(4.11)
0
i=1
The use of predictability in the definition (4.11) is essential from a financial

point of view, as Fi will represent a portfolio allocation made at time ti1
and kept constant over the trading interval [ti1 , ti ], while Bti Bti1 represents a change in the underlying asset price over [ti1 , ti ]. See also the related
discussion on self-financing portfolios in Section 5.2 and Lemma 5.1 on the
use of stochastic integrals to represent the value of a portfolio.
The next proposition gives the extension of the stochastic integral from simple predictable processes to square-integrable Ft -adapted processes (Xt )tR+
for which the value of Xt at time t only depends on information contained
in the Brownian path up to time t. This also means that knowing the future
is not permitted in the definition of the Ito integral, for example a portfolio
strategy that would allow the trader to buy at the lowest and sell at the
highest is not possible as it would require knowledge of future market data.
Note that the difference between Relation (4.12) below and Relation (4.8)
is the expectation on the right hand side.
Proposition 4.3. The stochastic integral with respect to Brownian motion
(Bt )tR+ extends to all adapted processes (ut )tR+ such that
hw
i
IE
|ut |2 dt < ,
0
with the It
o isometry
IE
w
ut dBt
2
= IE
hw
0
i
|ut |2 dt .
(4.12)
In addition, the It
o integral of an adapted process (ut )tR+ is always a centered random variable:
hw
i
IE
us dBs = 0.
(4.13)
0
Proof. We start by showing that the Ito isometry (4.12) holds for the simple
predictable process u of the form (4.10). We have
84
"
IE
w
ut dBt
2
= IE
n
X
!2
Fi (Bti Bti1 )
i=1
= IE
n
X
Fi Fj (Bti Bti1 )(Btj Btj1 )
i,j=1
"
= IE
n
X
|Fi | (Bti Bti1 )
i=1
+2 IE
Fi Fj (Bti Bti1 )(Btj Btj1 )
1i<jn
n
X
IE[IE[|Fi |2 (Bti Bti1 )2 |Fti1 ]]
i=1
+2
IE[IE[Fi Fj (Bti Bti1 )(Btj Btj1 )|Ftj1 ]]
1i<jn
n
X
IE[|Fi |2 IE[(Bti Bti1 )2 |Fti1 ]]
i=1
+2
IE[Fi Fj (Bti Bti1 ) IE[(Btj Btj1 )|Ftj1 ]]
1i<jn
n
X
IE[|Fi |2 IE[(Bti Bti1 )2 ]]
i=1
+2
IE[Fi Fj (Bti Bti1 ) IE[(Btj Btj1 )]]
1i<jn
n
X
IE[|Fi |2 (ti ti1 )]
i=1
"
= IE
n
X
#
2
|Fi | (ti ti1 )
i=1
= IE
hw
0
i
|ut |2 dt ,
where we applied the tower property (16.25) of conditional expectations

and the facts that Bti Bti1 is independent of Fti1 and

IE[Bti Bti1 ] = 0, IE (Bti Bti1 )2 = ti ti1 , i = 1, . . . , n.
The extension of the stochastic integral to square-integrable adapted processes (ut )tR+ is obtained as in Proposition 4.2 by density and a Cauchy sequence argument using the isometry (4.12), in the same way as in the proof
of Proposition 4.2. Let L2 ( R+ ) denote the space of square-integrable
"
85
N. Privault
stochastic processes u : R+ R such that
hw
i
|ut |2 dt < .
kuk2L2 (R+ ) := IE
0
By Lemma 1.1 of [54], p. 22 and p. 46, or Proposition 2.5.3 of [89], the

set of simple predictable processes forms a linear space which is dense in
the subspace L2ad ( R+ ) made of square-integrable adapted processes in
L2 ( R+ ). In other words, given u a square-integrable adapted process
there exists a sequence (un )nN of simple predictable processes

r n converging
to u in L2 ( R+ ), and the isometry (4.12) shows that
ut dBt nN is a
Cauchy sequence in L2 (), hence it converges in the complete space L2 ().
In this case we let
w
w
ut dBt := lim
unt dBt
n
and the limit is

unique from (4.12) and satisfies (4.12). The fact that the ranw
dom variable
us dBs is centered can be proved first on simple predictable
0
process u of the form (4.10) as
" n
#
hw
i
X
IE
ut dBt = IE
Fi (Bti Bti1 )
0
i=1
"
= IE
n
X
#
Fi (Bti Bti1 )
i=1
=
=
=
n
X
i=1
n
X
i=1
n
X
IE[IE[Fi (Bti Bti1 )|Fti1 ]]

IE[Fi IE[Bti Bti1 |Fti1 ]]
IE[Fi IE[Bti Bti1 ]]
i=1
= 0,
and this identity extends as above from simple predictable processes to
adapted processes u in L2 ( R+ ).

Note also that by bilinearity, the Ito isometry (4.12) can also be written as
hw
i
hw
i
w
IE
ut dBt
vt dBt = IE
ut vt dt ,
0
for all square-integrable adapted processes u, v.
86
"

In addition, whenwthe integrand (ut )tR+ is not a deterministic function,
the random variable

us dBs no longer has a Gaussian distribution, except
0
in some exceptional cases.
Definite stochastic integral

The definite stochastic integral of u over the interval [a, b] is defined as
wb
a
ut dBt :=
w
0
1[a,b] (t)ut dBt ,
with in particular
w
0
1[a,b] (t)dBt = Bb Ba ,
and
wb
a
dBt = Bb Ba ,
We also have the Chasles relation

wc
wb
wc
ut dBt =
ut dBt +
ut dBt ,
a
0 a b,
0 a b c,
and the stochastic integral has the following linearity property:

w
w
w
(ut + vt )dBt =
ut dBt +
vt dBt ,
u, v L2 (R+ ).
0
As an application of the Ito isometry (4.12) we note in particular that

"
2 #
w
w
wT
wT
T
T

T2
.
IE
Bt dBt
= IE
|Bt |2 dt =
IE |Bt |2 dt =
tdt =
0
0
0
0
2
Stochastic modeling of asset returns

In the sequel we will define the return at time t R+ of the risky asset
(St )tR+ as
dSt
= dt + dBt ,
St
with R and > 0. This equation can be formally rewritten in integral
form as
wT
wT
ST = S0 +
St dt +
St dBt ,
0
hence the need to define an integral with respect to dBt , in addition to the
usual integral with respect to dt. Note that in view of the definition (4.11),
this is a continuous-time extension of the notion portfolio value based on a
"
87
N. Privault
predictable portfolio strategy.
In Proposition 4.3 we have defined the stochastic integral of squareintegrable processes with respect to Brownian motion, thus we have made
sense of the equation
ST = S0 +
wT
0
St dt +
wT
0
St dBt ,
for (St )tR+ an Ft -adapted process, which can be rewritten in differential

notation as
dSt = St dt + St dBt ,
or
dSt
= dt + dBt .
St
(4.14)
This model will be used to represent the random price St of a risky asset
at time t. Here the return dSt /St of the asset is made of two components: a
constant return dt and a random return dBt parametrized by the coefficient
, called the volatility.
4.4 Stochastic Calculus

Our goal is now to solve Equation (4.14) and for this we will need to introduce
It
os calculus in Section 4.4 after reviewing classical deterministic calculus at
the beginning of Section 4.4.
Deterministic calculus
The fundamental theorem of calculus states that for any continuously differentiable (deterministic) function f we have
wx
f (x) = f (0) +
f 0 (y)dy.
0
In differential notation this relation is written as the first order expansion

df (x) = f 0 (x)dx,
(4.15)
where dx is small. Higher order expansions can be obtained from Taylors

formula, which, letting
df (x) = f (x + dx) f (x),
states that
88
"

1
1
1
df (x) = f 0 (x)dx + f 00 (x)(dx)2 + f 000 (x)(dx)3 + f (4) (x)(dx)4 + .
2
3!
4!
Note that Relation (4.15) can be obtained by neglecting the terms of order
larger than one in Taylors formula, since (dx)n << dx when n 2 and dx
is small.
Stochastic calculus
Let us now apply Taylors formula to Brownian motion, taking
dBt = Bt+dt Bt ,
and letting
df (Bt ) = f (Bt+dt ) f (Bt ),
we have
1
1
1
df (Bt ) = f 0 (Bt )dBt + f 00 (Bt )(dBt )2 + f 000 (Bt )(dBt )3 + f (4) (Bt )(dBt )4 + .
2
3!
4!
From
the construction of Brownian motion by its small increments dBt =
dt, it turns out that the terms in (dt)2 and dtdBt = (dt)3/2 can be neglected in Taylors formula at the first order of approximation in dt. However,
the term of order two
(dBt )2 = ( dt)2 = dt
can no longer be neglected in front of dt.
Hence Taylors formula written at the second order for Brownian motion
reads
1
df (Bt ) = f 0 (Bt )dBt + f 00 (Bt )dt,
(4.16)
2
for small dt. Note that writing this formula as
df (Bt )
dBt
1
= f 0 (Bt )
+ f 00 (Bt )
dt
dt
2
does not make sense because the derivative
dBt
dt
1
'
' '
dt
dt
dt
does not exist.
Integrating (4.16) on both sides and using the relation
f (Bt ) f (B0 ) =
"
wt
0
df (Bs )
89

N. Privault
we get the integral form of Itos formula for Brownian motion, i.e.
f (Bt ) = f (B0 ) +
wt
0
f 0 (Bs )dBs +
1 w t 00
f (Bs )ds.
2 0
We now turn to the general expression of It

os formula which applies to It
o
processes of the form
wt
wt
vs ds +
us dBs ,
t R+ ,
(4.17)
Xt = X0 +
0
or in differential notation
dXt = vt dt + ut dBt ,
where (ut )tR+ and (vt )tR+ are square-integrable adapted processes.
f
x
denote partial differentiation with respect to the second variable in f (t, x),
f
while
denote partial differentiation with respect to the first (time) variable
s
in f (t, x).
Given f (t, x) a smooth function of two variables, from now on we let
Theorem 4.1. (It

o formula for It
o processes). For any It
o process (Xt )tR+
of the form (4.17) and any f C 1,2 (R+ R) we have
w t f
w t f
f (t, Xt ) = f (0, X0 ) +
vs (s, Xs )ds +
us (s, Xs )dBs
0
0
x
x
w t f
2
1wt
f
2
(s, Xs )ds +
|us |
(s, Xs )ds.
(4.18)
+
0 s
2 0
x2
Proof. cf. [96], Theorem II-32.
Using the relation

wt
0
df (s, Xs ) = f (t, Xt ) f (0, X0 ),
we get
wt
0
df (s, Xs ) =
wt
w t f
f
vs (s, Xs )ds +
us (s, Xs )dBs
0
x
x
w t f
2f
1wt
|us |2 2 (s, Xs )ds,
+
(s, Xs )ds +
0 s
2 0
x
0
which allows us to rewrite the integrated version (4.18) of It

os formula in
differential notation, as
90
"
df (t, Xt ) =
f
f
f
1
2f
(t, Xt )dt+ut (t, Xt )dBt +vt (t, Xt )dt+ |ut |2 2 (t, Xt )dt,
t
x
x
2
x
(4.19)
or
f
1
2f
f
(t, Xt )dt +
(t, Xt )dXt + |ut |2 2 (t, Xt )dt.
t
x
2
x
Next, consider two Ito processes (Xt )tR+ and (Yt )tR+ written in integral
form as
w
w
df (t, Xt ) =
Xt = X0 +
and
Yt = Y0 +
wt
0
vs ds +
bs ds +
us dBs ,
t R+ ,
as dBs ,
t R+ ,
wt
0
or in differential notation as
dXt = vt dt + ut dBt ,
and dYt = bt dt + at dBt ,
t R+ .
The Ito formula also shows that

d(Xt Yt ) = Xt dYt + Yt dXt + dXt dYt
where the product dXt dYt is computed according to the It
o rule
(dt)2 = 0,
dtdBt = 0,
(dBt )2 = dt,
(4.20)
i.e.
dXt dYt = (vt dt + ut dBt )(bt dt + at dBt )
= bt vt (dt)2 + bt ut dtdBt + at vt dtdBt + at ut (dBt )2
= ut at dt.
Hence we have
(dXt )2 = (vt dt + ut dBt )2
= (vt )2 (dt)2 + (ut )2 (dBt )2 + 2ut vt dt dBt
= (ut )2 dt,
according to the Ito multiplication table
dt
dBt
dt
0
0
dBt
0
dt
(4.21)
and (4.19) can also be rewriten as
"
91
N. Privault
f
f
1 2f
(t, Xt )(dXt )2
(t, Xt )dt +
(t, Xt )dXt +
t
x
2 x2
f
f
f
1
2f
=
(t, Xt )dt + vt (t, Xt )dt + ut (t, Xt )dBt + (ut )2 2 (t, Xt )dt.
t
x
x
2
x
df (t, Xt ) =
Taking ut = 1 and vt = 0 in (4.17) yields Xt = Bt , in which case the It

o
formula (4.18) reads
f (t, Bt ) = f (0, B0 ) +
w t f
w t f
1 w t 2f
(s, Bs )ds +
(s, Bs )dBs +
(s, Bs )ds,
0 s
0 x
2 0 x2
i.e. in differential notation:

df (t, Bt ) =
f
1 2f
f
(t, Bt )dt +
(t, Bt )dBt +
(t, Bt )dt.
t
x
2 x2
(4.22)
As another example, applying Itos formula (4.22) to Bt2 with

Bt2 = f (t, Bt )
and f (t, x) = x2 ,
we get
d(Bt2 ) = df (Bt )
f
1 2f
f
(t, Bt )dt +
(t, Bt )dBt +
=
(t, Bt )dt
t
x
2 x2
= 2Bt dBt + dt,
since
f
(t, x) = 0,
t
f
(t, x) = 2x,
x
1 2f
(t, x) = 1,
2 x2
and
hence by integration we find

wT
Bs dBs +
wT
Bs dBs =
BT2 = B0 + 2
and
wT
0
dt = 2
wT
0
Bs dBs + T,

1
BT2 T .
2
We close this section with some comments on the practice of It
os calculus.
In some finance textbooks, Itos formula for e.g. geometric Brownian motion
can be found written in the notation
wT
wT
f
f
St
f (T, ST ) = f (0, X0 ) +
St
(t, St )dBt +
(t, St )dt
0
0
St
St
w
w T f
2
T
1
f
(t, St )dt + 2
St2 2 (t, St )dt,
+
0 t
0
2
St
0
92
"

or
df (St ) = St
f
f
1
2f
(St )dBt + St
(St )dt + 2 St2 2 (St )dt.
St
St
2
St
f
(St ) can in fact be easily misused in combination with the
St
fundamental theorem of classical calculus, and lead to the wrong identity
The notation
df (St ) =
f
(St )dSt .
St
Similarly, writing
df (Bt ) =
1 d2 f
df
(Bt )dBt +
(Bt )dt
dx
2 dx2
is consistent, while writing

df (Bt ) =
df (Bt )
1 d2 f (Bt )
dt
dBt +
dBt
2 dBt2
is potentially a source of confusion.
4.5 Geometric Brownian Motion

Our aim in this section is to solve the stochastic differential equation
dSt = St dt + St dBt
(4.23)
that will defined the price St of a risky asset at time t, where R and
> 0. This equation is rewritten in integral form as
wt
wt
t R+ .
(4.24)
St = S0 + Ss ds + Ss dBs ,
0
It can be solved by applying Itos formula to f (St ) = log St with f (x) = log x,
which shows that
1
d log St = St f 0 (St )dt + St f 0 (St )dBt + 2 St2 f 00 (St )dt
2
1
= dt + dBt 2 dt,
2
hence
log St log S0 =
"
wt
0
d log Sr
93
N. Privault
wt

wt
1
2 dr +
dBr
0
0
2

1
= 2 t + Bt ,
t R+ ,
2
=
and

St = S0 exp

1
2 t + Bt ,
2
t R+ .
The above provides a proof of the next proposition.

Proposition 4.4. The solution of (4.23) is given by
St = S0 et+Bt
t/2
t R+ .
Proof. Let us provide an alternative proof by searching for a solution of the

form
St = f (t, Bt )
where f (t, x) is a function to be determined. By Itos formula (4.22) we have
dSt = df (t, Bt ) =
f
1 2f
f
(t, Bt )dt +
(t, Bt )dBt +
(t, Bt )dt.
t
x
2 x2
Comparing this expression to (4.23) and identifying the terms in dBt we get
(t, Bt ) = St ,
f
1 2f
(t, Bt ) +
(t, Bt ) = St .
t
2 x2
Using the relation St = f (t, Bt ), these two equations rewrite as
f (t, B ) = f (t, B ),
t
t
f
1 2f
(t, Bt ) +
(t, Bt ) = f (t, Bt ).
t
2 x2
Since Bt is a Gaussian random variable taking all possible values in R, the
equations should hold for all x R, as follows:
(4.27a)
(t, x) = f (t, x),
f
1 2f
(t, x) +
(t, x) = f (t, x).
t
2 x2
94
(4.27b)
"

Letting g(t, x) = log f (t, x), the first equation (4.27a) shows that
g
log f
1 f
(t, x) =
(t, x) =
(t, x) = ,
x
x
f (t, x) x
i.e.
g
(t, x) = ,
x
which is solved as
g(t, x) = g(t, 0) + x,
hence
f (t, x) = eg(t,0) ex = f (t, 0)ex .
Plugging back this expression into the second equation (4.27b) yields
ex
f
1
(t, 0) + 2 ex f (t, 0) = f (t, 0)ex ,
t
2
i.e. after division by ex :

f
(t, 0) = 2 /2 f (t, 0),
t
or
g
(t, 0) = 2 /2,
t
i.e.

g(t, 0) = g(0, 0) + 2 /2 t,
and
f (t, x) = eg(t,x)
= eg(t,0)+x
2
= eg(0,0)+x+( /2)t
x+( 2 /2)t
= f (0, 0)e
,
hence
St = f (t, Bt ) = f (0, 0)eBt +(
/2)t
and the solution to (4.23) is given by

St = S0 eBt +(
/2)t
t R+ .
"
95
N. Privault
4
St
ert
3.5
3
St
2.5
2
1.5
1
0.5
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
Fig. 4.5: Geometric Brownian motion started at 1, with r = 1 and 2 = 0.5.

Conversely, taking St = f (t, Bt ) with f (t, x) = S0 ex
It
os formula to check that
t/2+t
we may apply
dSt = df (t, Bt )
f
f
1 2f
=
(t, Bt )dt +
(t, Bt )dBt +
(t, Bt )dt
t
x
2 x2

2
Bt +( 2 /2)t
2
= /2 S0 e
dt + S0 eBt +( /2)t dBt
2
1
+ 2 S0 eBt +( /2)t dt
2
2
2
= S eBt +( /2)t dt + S eBt +( /2)t dB
0
= St dt + St dBt .
4.6 Stochastic Differential Equations

In addition to geometric Brownian motion there exists a large family of
stochastic differential equations that can be studied, although most of the
time they cannot be explicitly solved. Let now
: R+ Rn Rd Rn
where Rd Rn denotes the space of d n matrices, and
b : R+ Rn R
The animation works in Acrobat reader on the entire pdf file.
96
"

satisfy the global Lipschitz condition
k(t, x) (t, y)k2 + kb(t, x) b(t, y)k2 K 2 kx yk2 ,
t R+ , x, y Rn . Then there exists a unique strong solution to the stochastic
differential equation
wt
wt
Xt = X0 +
(s, Xs )dBs +
b(s, Xs )ds,
t R+ ,
0
where (Bt )tR+ is a d-dimensional Brownian motion, see e.g. [96], Theorem V7.
Next, we consider a few examples of stochastic differential equations that
can be solved explicitly using Ito calculus, in addition to geometric Brownian
motion.
Examples
1. Consider the stochastic differential equation
dXt = Xt dt + dBt ,
X0 = x0 ,
with > 0 and > 0.

Looking for a solution of the form

wt
Xt = a(t) x0 +
b(s)dBs
0
where a() and b() are deterministic functions, yields

Xt = x0 et +
wt
0
e(ts) dBs ,
t R+ ,
(4.28)
rt
after applying Theorem 4.1 to the Ito process x0 + 0 b(s)dBs of the form
(4.17) with ut = b(t) and v(t) = 0, and to the function f (t, x) = a(t)x.
Remark: the solution of this equation cannot be written as a function
f (t, Bt ) of t and Bt as in the proof of Proposition 4.4.
dXt = tXt dt + et
/2
dBt ,
X0 = x0 .

rt
Looking for a solution of the form Xt = a(t) X0 + 0 b(s)dBs , where
a() and b() are deterministic functions we get a0 (t)/a(t) = t and
"
97
N. Privault
2
a(t)b(t) = et /2 , hence a(t) = et

2
et /2 (X0 + Bt ), t R+ .
/2
and b(t) = 1, which yields Xt =

dYt = (2Yt + 2 )dt + 2
Yt dBt ,
where , > 0.
Letting Xt =
Yt we have dXt = Xt dt + dBt , hence

Yt =
p
2
wt
et Y0 + e(ts) dBs .
0
Exercises
Exercise 4.1 Let (Bt )tR+ denote a standard Brownian motion.

a) Let c > 0. Among the following processes, tell which is a standard Brownian motion and which is not. Justify your answer.
(i) (Bc+t Bc )tR+ .
(ii) (cBt/c2 )tR+ .
(iii) (Bct2 )tR+ .
b) Compute the stochastic integrals
wT
0
2dBt
wT
and
(2 1[0,T /2] (t) + 1(T /2,T ] (t))dBt
and determine their probability laws (including mean and variance).

c) Determine the probability law (including mean and variance) of the
stochastic integral
w
2
sin(t) dBt .
d) Compute IE[Bt Bs ] in terms of s, t 0.

e) Let T > 0. Show that if f is a differentiable function with f (0) = f (T ) = 0
we have
wT
wT
f (t)dBt =
f 0 (t)Bt dt.
0
Hint: Apply Itos calculus to t 7 f (t)Bt .

Exercise 4.2 Given T > 0, find an expansion of BT3 of the form
98
"

BT3 =
wT
0
t dBt ,
where (t )t[0,T ] is an adapted process to be determined.

Exercise 4.3 Let f L2 ([0, T ]). Compute the conditional expectation
i
h rT

0 t T,
IE e 0 f (s)dBs Ft ,
where (Ft )t[0,T ] denotes the filtration generated by (Bt )t[0,T ] .
Exercise 4.4 Compute the expectation

w

T
IE exp
Bt dBt
0
for all < 1/T . Hint: expand (BT ) using Itos formula.
Exercise 4.5
a) Solve the ordinary differential equation df (t) = cf (t)dt and the stochastic
differential equation dSt = rSt dt + St dBt , t R+ , where r, R are
constants and (Bt )tR+ is a standard Brownian motion.
b) Show that
IE[St ] = S0 ert
and
Var[St ] = S02 e2rt (e t 1),
t R+ .
c) Compute d log St .
d) Assume that (Wt )tR+ is another standard Brownian motion, correlated
to (Bt )tR+ according to the Ito rule dWt dBt = dt, for [1, 2],
and consider the solution (Yt )tR+ of the stochastic differential equation
dYt = Yt dt + Yt dWt , t R+ , where , R are constants. Compute
f (St , Yt ), for f a C 2 function of R2 .
Exercise 4.6
a) Solve the stochastic differential equation
dXt = bXt dt + ebt dBt ,
t R+ ,
where (Bt )tR+ is a standard Brownian motion and , b > 0.

b) Solve the stochastic differential equation
dXt = bXt dt + eat dBt ,
t R+ ,
where (Bt )tR+ is a standard Brownian motion and , a, b > 0 are positive
constants.
"
99
N. Privault
Exercise 4.7 Given T > 0, let (XtT )t[0,T ) denote the solution of the stochastic
dXtT = dBt
XtT
dt,
T t
t [0, T ),
(4.29)
under the initial condition X0T = 0 and > 0.

a) Show that
XtT = (T t)
wt
0
1
dBs ,
T s
t [0, T ).
Hint: start by computing d(XtT /(T t)) using Itos calculus.

b) Show that IE[XtT ] = 0 for all t [0, T ).
c) Show that Var[XtT ] = 2 t(T t)/T for all t [0, T ).
d) Show that limtT XtT = 0 in L2 (). The process (XtT )t[0,T ] is called a
Brownian bridge.
Exercise 4.8 Exponential Vasicek model. Consider a short term rate interest
rate proces (rt )tR+ in the exponential Vasicek model:
drt = rt ( a log rt )dt + rt dBt ,
(4.30)
where , a, are positive parameters.

a) Find the solution (zt )tR+ of the stochastic differential equation
dzt = azt dt + dBt
as a function of the initial condition z0 , where a and are positive parameters.
b) Find the solution (Yt )tR+ of the stochastic differential equation
dYt = ( aYt )dt + dBt
(4.31)
as a function of the initial condition Y0 . Hint: let zt = Yt /a.

c) Let xt = eYt , t R+ . Determine the stochastic differential equation satisfied by (xt )tR+ .
d) Find the solution (rt )tR+ of (4.30) in terms of the initial condition r0 .
e) Compute the mean IE[rt ] of rt , t 0.
f) Compute the asymptotic mean limt IE[rt ].
Exercise 4.9 Cox-Ingerson-Ross model. Consider the equation
drt = ( rt )dt + rt dBt
One may use the generating function IE[e ] = e
2 /2
100
(4.32)
2
for X ' N (0, ).
"

modeling the variations of a short term interest rate process rt , where , ,
and r0 are positive parameters.
a) Write down the equation (4.32) in integral form.
b) Let u(t) = IE[rt ]. Show, using the integral form of (4.32), that u(t) satisfies
the differential equation
u0 (t) = u(t).
c) By an application of Itos formula to rt2 , show that
3/2
drt2 = rt (2 + 2 2rt )dt + 2rt dBt .
(4.33)
d) Using the integral form of (4.33), find a differential equation satisfied by

v(t) = IE[rt2 ].
Exercise 4.10 Let (Bt )tR+ denote a standard Brownian motion generating
the filtration (Ft )tR+ .
a) Consider the Ito formula
w t f
1 w t 2 2f
f
(Xs )dBs + vs (Xs )ds+
u
(Xs )ds,
0
x
x
2 0 s x2
(4.34)
wt
wt
where Xt = X0 +
us dBs +
vs ds.
f (Xt ) = f (X0 )+
wt
0
us
Xt
Compute St := e by the Ito formula (4.34) applied to f (x) = ex and

Xt = Bt + t, > 0, R.
b) Let r > 0. For which value of does (St )tR+ satisfy the stochastic differential equation
dSt = rSt dt + St dBt ?
c) Let the process (St )tR+ be defined by St = S0 eBt +t , t R+ . Using the
result of Exercise 16.2, show that the conditional probability P (ST >
K | St = x) is given by

log(x/K) + (T t)
,
P(ST > K | St = x) =
T t
Hint: use the decomposition ST = St e(BT Bt )+(T t) .
d) Given 0 t T and > 0, let
X = (BT Bt )
and
2 = Var[X],
> 0.
What is equal to ?
"
101
N. Privault
Problem 4.11 Consider (Bt )tR+ a standard Brownian motion, and for any
n 1 and T > 0, define the discretized quadratic variation
(n)
QT :=
n
X
(BkT /n B(k1)T /n )2 ,
n 1.
k=1
h
i
(n)
a) Compute IE QT , n 1.
(n)
b) Compute Var[QT ], n 1.
c) Show that
(n)
lim QT = T,
(n)
where the limit is taken in L2 (), that is, show that lim kQT
n
T kL2 () = 0, where
s

(n)

QT T
L2 ()
IE
:=

(n)
QT T
2
,
n 1.
d) By the result of Question (c), show that the limit

wT
0
Bt dBt := lim
n
X
(BkT /n B(k1)T /n )B(k1)T /n
k=1
exists in L2 (), and compute it.

Hint: Use the identity
(x y)y =
1 2
(x y 2 (x y)2 ),
2
x, y R.
e) Consider the modified quadratic variation defined by

(n) :=
Q
T
n
X
(B(k1/2)T /n B(k1)T /n )2 ,
n 1.
k=1
(n)
Compute the limit limn Q

T in L () by repeating the steps of Questions (a)-(c).
f) By the result of Question (e), show that the limit
wT
0
Bt dBt := lim
n
X
(BkT /n B(k1)T /n )B(k1/2)T /n
k=1
102
"

Hint: Use the identities
(x y)y =
1 2
(x y 2 (x y)2 ),
2
and
1 2
(x y 2 + (x y)2 ),
x, y R.
2
g) More generally, by repeating the steps of Questions (e) and (f), show that
for any [0, 1] the limit
(x y)x =
wT
0
Bt d Bt := lim
n
X
(BkT /n B(k1)T /n )B(k)T /n
k=1

h) Comparison with deterministic calculus. Compute the limit
lim
n
X
k=1
(k )
T
n

k
T
T
(k 1)
n
n
for all values of in [0, 1].

Exercise 4.12 Let (Bt )tR+ be a standard Brownian motion generating the
information flow (Ft )tR+ .
a) Let 0 t T . What is the probability law of BT Bt ?
b) From the answer to Exercise 16.5, show that
r

T t Bt2 /(2(T t))
Bt
IE[(BT )+ | Ft ] =
,
e
+ Bt
2
T t
0 t T . Hint: write BT = BT Bt + Bt .
c) Let > 0, R, and Xt = Bt + t. Compute eXt using the It
o formula
w t f
f
1 w t 2 2f
(Xs )dBs + vs (Xs )ds +
u
(Xs )ds
0
x
x
2 0 s x2
wt
wt
stated here for a process Xt = X0 +
us dBs +
vs ds, t R+ , and
0
0
applied to f (x) = ex .
d) Let St = eXt , t R+ , and r > 0. For which value of does (St )tR+
satisfy the stochastic differential equation
f (Xt ) = f (X0 ) +
wt
0
us
dSt = rSt dt + St dBt
Exercise 4.13 From the answer to Exercise 16.4-(b), show that

"
103
N. Privault
r
IE[(BT )+ | Ft ] =
T t (Bt )2 /(2(T t))

e
+(Bt )
2
Bt
T t

,
0 t T.
Hint: write BT = BT Bt + Bt .
104
"
Chapter 5
The Black-Scholes PDE
In this chapter we review the notions of assets, self-financing portfolios, riskneutral measures, and arbitrage in continuous time. We also derive the BlackScholes PDE for self-financing portfolios, and we solve this equation using the
heat kernel method.
5.1 Continuous-Time Market Model

Let (At )tR+ be the riskless asset given by
dAt
= rdt,
At
t R+ ,
i.e. At = A0 ert ,
t R+ .
For t > 0, let (St )tR+ be the price process defined as

t R+ .
By Proposition 4.4 we have

1
St = S0 exp Bt + 2 t ,
2
t R+ .
5.2 Self-Financing Portfolio Strategies

Let t and t denote the (possibly fractional) quantities invested at time t,
respectively in the assets St and At , and let
t = (t , t ),
"
St = (At , St ),
t R+ ,
N. Privault
denote the associated portfolio and asset price processes. The value of the
portfolio Vt at time t is given by
Vt = t St = t At + t St ,
t R+ .
(5.1)
Our description of portfolio strategies proceeds in four steps which correspond

to different interpretations of the self-financing condition.
Portfolio update
The portfolio strategy (t , t )tR+ is self-financing if the portfolio value remains constant after updating the portfolio from (t , t ) to (t+dt , t+dt ), i.e.
t St+dt = At+dt t + St+dt t = At+dt t+dt + St+dt t+dt = t+dt St+dt , (5.2)
which is the continuous-time equivalent of the self-financing condition already
encountered in the discrete setting of Chapter 2, see Definition 2.1. A major
difference with the discrete-time case of Definition 2.1, however, is that the
continuous-time differentials dSt and dt do not make pathwise sense as the
stochastic integral is defined by an L2 limit, cf. Proposition 4.3, or by convergence in probability.
Portfolio value
t St
- t St+dt = t+dt St+dt
- t+dt St+2dt
Asset value
St
St+dt
St+dt
St+2dt
Time scale
t
t
t + dt t + dt
t t+dt
t + 2dt
t+2dt
Portfolio allocation
Fig. 5.1: Illustration of the self-financing condition (5.2).
Portfolio re-allocation
Equivalently, Condition (5.2) can be rewritten as
At+dt dt + St+dt dt = 0,
(5.3)
or
At+dt (t+dt t ) = St+dt (t+dt t ),
i.e. when one sells a quantity dt > 0 of the risky asset St+dt between the
time periods [t, t + dt] and [t + dt, t + 2dt] for a total amount St+dt dt , one
106
"

should entirely use this income to buy a quantity dt > 0 of the riskless asset
for an amount At+dt dt > 0.
Similarly, if one sells a (possibly fractional) quantity dt > 0 of the
riskless asset At+dt between the time periods [t, t + dt] and [t + dt, t + 2dt]
for a total amount At+dt dt , one should entirely use this income to buy a
quantity dt > 0 of the risky asset for an amount St+dt dt > 0, i.e.
St+dt dt = At+dt dt ,
(5.4)
It
o calculus version
Condition (5.4) can be rewritten as
At+dt (t+dt t ) + St (t+dt t ) + (St+dt St )(t+dt t ) = 0,
which rewrites as
At dt + St dt + dSt dt = 0
(5.5)
in differential notation, since

dAt dt = (At+dt At )(t+dt t ) = dAt dt = rAt dt dt ' 0
in the sense of the Ito calculus, by the Ito table (4.21).
Portfolio differential
In practice we will use the following definition for the self-financing portfolio
property.
Proposition 5.1. The portfolio Vt is said to be self-financing if
dVt = t dAt + t dSt .
(5.6)
Proof. We check that by Itos calculus we have

dVt = t dAt + t dSt + At dt + St dt + dt dAt + dt dSt
= t dAt + t dSt + At dt + St dt + dt dSt ,
since dt dAt = rAt dt dt = 0, hence Condition (5.6) rewrites as (5.5),
which is equivalent to (5.2) and (5.3).

Let
Vt = ert Vt
and
Xt = ert St
respectively denote the discounted portfolio value and discounted risky asset
prices at time t 0. We have
"
107
N. Privault
dXt = d(ert St )
= rert St dt + ert dSt
= rert St dt + ert St dt + ert St dBt
= Xt (( r)dt + dBt ).
In the next lemma we show that when a portfolio is self-financing, its discounted value is a gain process given by the sum of discounted profits and
losses (number of risky assets t times discounted price variation dXt ) over
time.
The following lemma is the continuous-time analog of Lemma 3.1.
Lemma 5.1. Let (t , t )tR+ be a portfolio strategy with value
Vt = t At + t St ,
t R+ .
The following statements are equivalent:

i) the portfolio strategy (t , t )tR+ is self-financing,
ii) we have
Vt = V0 +
wt
u dXu ,
t R+ .
(5.7)
Proof. Assuming that (i) holds, the self-financing condition shows that
dVt = t dAt + t dSt
= rt At dt + t St dt + t St dBt
= rVt dt + ( r)t St dt + t St dBt
t R+ ,
hence
ert dVt = rert Vt dt + ( r)ert t St dt + ert t St dBt ,
t R+ ,
and
dVt = d ert Vt
= rert Vt dt + ert dVt

= ( r)t ert St dt + t ert St dBt
= ( r)t Xt dt + t Xt dBt
= t dXt ,
t R+ ,
i.e. (5.7) holds by integrating on both sides as

Vt V0 =
wt
0
dVu =
wt
0
u dXu ,
108
t R+ .
"

Conversely, if (5.7) is satisfied we have
dVt = d(ert Vt )
= rert Vt dt + ert dVt
= rert Vt dt + ert t dXt
= rVt dt + ert t dXt
= rVt dt + ert t Xt (( r)dt + dBt )
= rVt dt + t St (( r)dt + dBt )
= t dAt + t dSt ,
hence the portfolio is self-financing according to Definition 5.1.
As a consequence of (5.7), the hedging problem of a claim C with maturity

T is reduced to that of finding the representation of the discounted claim
C = erT C as a stochastic integral:
C = V0 +
wT
0
u dXu .
Note also that (5.7) shows that the value of a self-financing portfolio can also
be written as
wt
wt
Vt = ert V0 + ( r) er(tu) u Su du + er(tu) u Su dBu ,
t R+ .
0
0
(5.8)
5.3 Arbitrage and Risk-Neutral Measures

In continuous-time, the definition of arbitrage follows the lines of its analogs
in the discrete and two-step models. In the sequel we will only consider admissible portfolio strategies whose total value Vt remains non-negative for
all times t [0, T ].
Definition 5.1. A portfolio strategy (t , t )t[0,T ] constitutes an arbitrage
opportunity if all three following conditions are satisfied:
i) V0 0,
ii) VT 0,
iii) P(VT > 0) > 0.
Roughly speaking, (ii) means that the investor wants no loss, (iii) means
that he wishes to sometimes make a strictly positive gain, and (i) means that
he starts with zero capital or even with a debt.
"
109
N. Privault
Next we turn to the definition of risk-neutral measures in continuous time.
Recall that the filtration (Ft )tR+ is generated by Brownian motion (Bt )tR+ ,
i.e.
Ft = (Bu : 0 u t),
t R+ .
Definition 5.2. A probability measure P on is called a risk-neutral measure if it satisfies
IE [St |Fu ] = er(tu) Su ,
0 u t,
(5.9)
where IE denotes the expectation under P .

From the relation
At = er(tu) Au ,
0 u t,
we interpret (5.9) by saying that the expected return of the risky asset St
under P equals the return of the riskless asset At . The discounted price Xt
of the risky asset is defined by
Xt = ert St =
St
,
At /A0
t R+ ,
i.e. At /A0 plays the role of a numeraire in the sense of Chapter 10.
Definition 5.3. A continuous time process (Zt )tR+ of integrable random
variables is a martingale with respect to the filtration (Ft )tR+ if
IE[Zt |Fs ] = Zs ,
0 s t.
Note that when (Zt )tR+ is a martingale, Zt is in particular Ft -measurable

for all t R+ .
As in the discrete case, the notion of martingale can be used to characterize
risk-neutral measures.
Proposition 5.2. The measure P is risk-neutral if and only if the discounted
price process (Xt )tR+ is a martingale under P .
Proof. If P is a risk-neutral measure we have
IE [Xt |Fu ] = IE [ert St |Fu ]
= ert IE [St |Fu ]
= ert er(tu) Su
= eru Su
= Xu ,
0 u t,
hence (Xt )tR+ is a martingale. Conversely, if (Xt )tR+ is a martingale then

IE [St |Fu ] = ert IE [Xt |Fu ]
110
"

= ert Xs
= er(ts) Ss ,
0 s t,
hence the measure P is risk-neutral according to Definition 5.2.
As in the discrete time case, P would be called a risk-premium measure if it

satisfied
IE [St |Fu ] > er(tu) Su ,
0 u t,
meaning that by taking risks in buying St , one could make an expected return
higher than that of
At = er(tu) Au ,
0 u t.
Next we note that the first fundamental theorem of asset pricing also holds
in continuous time, and can be used to check for the existence of arbitrage
opportunities.
admits at least one risk-neutral measure.
Proof. cf. [48] and Chapter VII-4a of [108].
5.4 Market Completeness

Definition 5.4. A contingent claim with payoff C is said to be attainable if
there exists a (self-financing) portfolio strategy (t , t )t[0,T ] such that
C = VT .
In this case the price of the claim at time t will be equal to the value Vt of
any self-financing portfolio hedging C.
claim C is attainable.
The next result is a continuous-time restatement of the second fundamental
theorem of asset pricing.
it admits only one risk-neutral measure.
Proof. cf. [48] and Chapter VII-4a of [108].
In the Black-Scholes model one can show the existence of a unique risk-neutral
measure, hence the model is without arbitrage and complete.
"
111
N. Privault
5.5 The Black-Scholes PDE

We start by deriving the Black-Scholes partial differential equation (PDE)
for the price of a self-financing portfolio.
Proposition 5.3. Let (t , t )tR+ be a portfolio strategy such that
(i) (t , t )tR+ is self-financing,
(ii) the value Vt := t At + t St , t R+ , takes the form
Vt = g(t, St ),
t R+ ,
for some g C 1,2 ((0, ) (0, )).

Then the function g(t, x) satisfies the Black-Scholes PDE
rg(t, x) =
g
g
1
2g
(t, x)+rx (t, x)+ 2 x2 2 (t, x),
t
x
2
x
x > 0,
t [0, T ],
and t is given by
t =
g
(t, St ),
x
t R+ .
(5.10)
Proof. First, note that the self-financing condition implies

dVt = t dAt + t dSt
(5.11)
= rVt dt + ( r)t St dt + t St dBt ,

t R+ . We now rewrite (4.24) under the form of an Ito process
St = S0 +
wt
0
vs ds +
wt
0
us dBs ,
t R+ ,
as in (4.17), by taking
ut = St ,
and vt = St ,
t R+ .
The application of Itos formula Theorem 4.1 to g(t, x) leads to

g
g
(t, St )dt + ut (t, St )dBt
x
x
g
1
2g
+ (t, St )dt + |ut |2 2 (t, St )dt
t
2
x
g
g
1
2g
g
=
(t, St )dt + St (t, St )dt + 2 St2 2 (t, St )dt + St (t, St )dBt .
t
x
2
x
x
112
"
dg(t, St ) = g(0, S0 ) + vt


(5.12)
By respective identification of the terms in dBt and dt in (5.11) and (5.12)
we get
g
g
1 2 2 2g
rt At dt + t St dt = t (t, St )dt + St x (t, St )dt + 2 St x2 (t, St )dt,
t St dBt = St g (t, St )dBt ,

x
hence
g
1
2g
(t, St ) + 2 St2 2 (t, St ),

rVt rt St =
t
2
x
t = g (t, St ),
x
i.e.
g
g
1 2 2 2g
rg(t, St ) = t (t, St ) + rSt x (t, St ) + 2 St x2 (t, St ),
t = g (t, St ).
x
(5.13)

The derivative giving t in (5.10) is called the Delta of the option price.
The amount invested on the riskless asset is
t At = Vt t St = g(t, St ) St
g
(t, St ),
x
and t is given by
t =
Vt t St
At
g(t, St ) St
At
g(t, St ) St
g
(t, St )
x
g
(t, St )
x
.
A0 ert
In the next proposition we add a terminal condition g(T, x) = f (x) to the

Black-Scholes PDE in order to hedge claim C of the form C = f (ST ).
"
113
N. Privault
Proposition 5.4. The price of any self-financing portfolio of the form Vt =
g(t, St ) hedging an option with payoff C = f (ST ) satisfies the Black-Scholes
PDE
g
g
1
2g
rg(t, x) =
(t, x) + rx (t, x) + 2 x2 2 (t, x),
t
x
2
x
g(T, x) = f (x).
When C = ST K is the (linear) payoff of a forward contract, i.e. f (x) =

x K, the Black-Scholes PDE admits the easy solution
g(t, x) = x Ker(T t) ,
x > 0,
t [0, T ],
and the Delta of the option price is given by

t =
g
(t, St ) = 1,
x
t [0, T ],
cf. Exercise 5.3. The forward contract can be realized by the option issuer as
follows:
a) At time t, receive the option premium St er(T t) K from the option
buyer.
b) Borrow er(T t) K from the bank, to be refunded at maturity.
c) Buy the risky asset using the amount St er(T t) K + er(T t) K = St .
d) Hold the risky asset until maturity (do nothing, constant portfolio strategy).
e) At maturity T , hand in the asset to the option holder, who gives the price
K in exchange.
f) Use the amount K = er(T t) er(T t) K to refund the bank of the sum
er(T t) K borrowed at time t.
Recall that in the case of a European call option with strike K the payoff
function is given by f (x) = (x K)+ and the Black-Scholes PDE reads
g
g
1
2g
rgc (t, x) = c (t, x) + rx c (t, x) + 2 x2 2c (t, x)

t
x
2
x
+
gc (T, x) = (x K) .
In Sections 5.6 and 5.7 we will prove that the solution of this PDE is given
by the Black-Scholes formula
gc (t, x) = BS(K, x, , r, T t) = x(d+ ) Ker(T t) (d ),
114
(5.14)
"

cf. Proposition 5.8 below, where
1 w x y2 /2
(x) =
e
dy,
2
x R,
denotes the standard Gaussian distribution function and

d+ =
log(x/K) + (r + 2 /2)(T t)
,
T t
with
d =
log(x/K) + (r 2 /2)(T t)
,
T t
(5.15)
d+ = d + T t.
One can easily check that

lim d+ = lim d =
t%T
t%T
+,
x > K,
x < K,
which allows one to recover the boundary condition
x>K
x(+) K(+) = x K,
gc (T, x) =
= (x K)+
x() K() = 0,
x<K
at t = T . Similarly we can check that
+,
lim d =
T
r > 2 /2,
r < 2 /2,
and limT d+ = +, hence

lim BS(K, St , , r, T t) = St ,
t R+ .
Figure 5.2 presents an interactive graph of the Black call price function, i.e.
the solution
(t, x) 7 gc (t, x) = x(d+ ) Ker(T t) (d )
of the Black-Scholes PDE for a call option.
"
115
N. Privault
Fig. 5.2: Graph of the Black-Scholes call price function with strike K = 100.
In Figure 5.3 we consider the stock price of HSBC Holdings (0005.HK) over
one year:
Fig. 5.3: Graph of the stock price of HSBC Holdings.

Consider a call option issued by Societe Generale on 31 December 2008 with
strike K=$63.704, maturity T = October 05, 2009, and an entitlement ratio of
100, meaning that one option contract is divided into 100 warrants, cf. page
6. The next graph gives the time evolution of the Black-Scholes portfolio
price
t 7 gc (t, St )
driven by the market price t 7 St of the underlying risky asset as given in
Figure 5.3, in which the number of days is counted from the origin and not
from maturity.
The next proposition is proved by direct differentiation of the Black-Scholes
function, and will be recovered later using a probabilistic argument in Proposition 6.7 below.
Right click on the figure for interaction and full screen view (works in Acrobat
reader on the entire pdf file).
116
"
40
35
30
25
20
15
10
5
0
100
90
80
underlying HK$
70
60
50
40
200
150
100
50
time in days
Fig. 5.4: Path of the Black-Scholes price for a call option on HSBC.
Proposition 5.5. The Black-Scholes Delta of a European call option is given

by
t = (d+ ) [0, 1],
where d+ is given by (5.15).
Proof. By (5.14) we have
gc
(t, x) =
(5.16)
x

2
log(x/K) + (r + /2)(T t)
x
x
T t

log(x/K) + (r 2 /2)(T t)
Ker(T t)
x
T t

log(x/K) + (r 2 /2)(T t)
=
T t

log(x/K) + (r + 2 /2)(T t)
+x
x
T t

log(x/K) + (r 2 /2)(T t)
r(T t)
Ke
x
T t

log(x/K) + (r 2 /2)(T t)
=
T t

2 !
1
1 log(x/K) + (r + 2 /2)(T t)
exp
+
2
T t
2 T t

2 !
Ker(T t)
1 log(x/K) + (r 2 /2)(T t)
exp
2
T t
2x T t

log(x/K) + (r 2 /2)(T t)
=
.
T t
"
117
N. Privault

As a consequence of Proposition 5.5, in the Black-Scholes model the amount
invested in the risky asset is

log(St /K) + (r + 2 /2)(T t)
St t = St (d+ ) = St
0,
T t
which is always positive, i.e. there is no short-selling, and the amount invested
on the riskless asset is

log(St /K) + (r 2 /2)(T t)
t At = KA0 er(T t)
0,
T t
which is always negative, i.e. we are constantly borrowing money, as noted
in Figure 5.5.
Black-Scholes price
Risky investment
Riskless investment
Underlying
100
80
K
60
HK$
40
20
-20
-40
-60
0
50
100
150
200
Fig. 5.5: Time evolution of the hedging portfolio for a call option on HSBC.
Similarly, in the case of a European put option with strike K the payoff
function is given by f (x) = (K x)+ and the Black-Scholes PDE reads
g
1
2 gp
g
rgp (t, x) = p (t, x) + rx p (t, x) + 2 x2

(t, x),
t
x
2
x2
gp (T, x) = (K x)+ ,
with explicit solution
gp (t, x) = Ker(T t) (d ) x(d+ ),
as illustrated in Figure 5.6.
Note that the call-put parity relation
g(t, St ) = x Ker(T t) = gc (t, St ) gp (t, St ),
118
0 t T,
"

14
12
10
8
6
4
2
0
7
90
95
100 105
underlying HK$
110
115
120
10
6
5
4 time to maturity T-t
Fig. 5.6: Graph of the Black-Scholes put price function with strike K = 100.
is satisfied here.
For one more example, we consider a put option issued by BNP Paribas on
04 November 2008 with strike K=$77.667, maturity T = October 05, 2009,
and entitlement ratio 92.593, cf. page 6. In the next Figure 5.7 the number
of days is counted from the origin and not from maturity.
45
40
35
30
25
20
15
10
5
0
0
50
100
time in days
150
200
100
90
80
70
40
50
60
underlying HK$
Fig. 5.7: Path of the Black-Scholes price for a put option on HSBC.
In the case of a Black-Scholes put option the Delta is given by
t = (d+ ) [1, 0],
and the amount invested on the risky asset is

log(St /K) + (r + 2 /2)(T t)
St (d+ ) = St
0,
T t
i.e. there is always short-selling, and the amount invested on the riskless asset
is
"
119
N. Privault

log(St /K) + (r 2 /2)(T t)
Ker(T t)
0,
T t
which is always positive, i.e. we are constantly investing on the riskless asset.
Black-Scholes price
Risky investment
Riskless investment
Underlying
100
80
60
HK$
40
20
-20
-40
-60
0
50
100
150
200
Fig. 5.8: Time evolution of the hedging portfolio for a put option on HSBC.
5.6 The Heat Equation

In this section we study the heat equation
g
1 2g
(t, y) =
(t, y)
t
2 y 2
which is used to model the diffusion of heat over time through solids. Here,
the data of g(x, t) represents the temperature measured at time t and point
x. We refer the reader to [115] for a complete treatment of this topic.
120
"

4
t=0.0209
3.5
3
g(x,t)
2.5
2
1.5
1
0.5
0
-2
-1.5
-1
-0.5
0.5
1.5
Fig. 5.9: Time-dependent solution of the heat equation.

In Section 5.7 this equation will be shown to be equivalent to the BlackScholes PDE after a change of variables. In particular this will lead to the
explicit solution of the Black-Scholes PDE.
Proposition 5.6. The heat equation
g
1 2g
(t, y) =
(t, y)
t
2 y 2
g(0, y) = (y)
(5.17)
with initial condition

g(0, y) = (y)
has the solution
g(t, y) =
(z)e(yz)
/(2t)
dz
,
2t
t > 0.
(5.18)
Proof. We have
2
g
w
dz
(t, y) =
(z)e(yz) /(2t)
t
t
2t
!
2
w
e(yz) /(2t)
=
(z)
dz
t
2t

1w
(y z)2
1 (yz)2 /(2t) dz
=
(z)
2
t2
t
2t
1w
2 (yz)2 /(2t) dz
(z) 2 e
=
2
z
2t
"
121
N. Privault
2
1w
dz
2
(z) 2 e(yz) /(2t)
2
y
2t
1 2 w
(yz)2 /(2t) dz
=
(z)e
2 y 2
2t
1 2g
=
(t, y).
2 y 2
On the other hand it can be checked that at time t = 0,

lim
t0
(z)e(yz)
/(2t)
w
2
dz
dz
= lim
(y + z)ez /(2t)
= (y),
2t t0
2t
y R.
Let us provide a second proof of Proposition 5.6 using stochastic calculus and
Brownian motion. Note that under the change of variable x = z y we have
w
dz
2t
w
2
dx
=
(y + x)ex /(2t)
2t
= IE[(y + Bt )],
g(t, y) =
(z)e(yz)
/(2t)
where (Bt )tR+ is a standard Brownian motion. Applying It

os formula we
have
w

w

t
t
1
IE[(y + Bt )] = (y) + IE
0 (y + Bs )dBs + IE
00 (y + Bs )ds
0
0
2
1wt
00
= (y) +
IE [ (y + Bs )] ds
2 0
w
1 t 2
= (y) +
IE [(y + Bs )] ds,
2 0 y 2
since the expectation of the stochastic integral is zero. Hence
g
(t, y) =
IE[(y + Bt )]
t
t
1 2
=
IE [(y + Bt )]
2 y 2
2
1 g
=
(t, y).
2 y 2
Concerning the initial condition we check that
g(0, y) = IE[(y + B0 )] = IE[(y)] = (y).
122
"

The expression g(t, y) = IE[(y + Bt )] provides a probabilistic interpretation
of the heat diffusion phenomenon based on Brownian motion.
5.7 Solution of the Black-Scholes PDE

In this section we will solve the Black-Scholes PDE by the kernel method
of Section 5.6 and a change of variables. This solution method uses a transformation of variables (5.20) which involves the time inversion t 7 T t
on the interval [0, T ], so that the terminal condition at time T in the BlackScholes equation (5.19) becomes an initial condition at time t = 0 in the heat
equation (5.22).
Proposition 5.7. Assume that f (t, x) solves the Black-Scholes PDE
f
f
1
2f
rf (t, x) =
(t, x) + rx (t, x) + 2 x2 2 (t, x),
t
x
2
x
+
f (T, x) = (x K) ,
(5.19)
with terminal condition h(x) = (x K)+ . Then the function g(t, y) defined
by

2
g(t, y) = ert f T t, ey+( /2r)t
(5.20)
solves the heat equation (5.17) with initial condition
g(0, y) = h(ey ),
i.e.
y R,
g (t, y) = 1 g (t, y)
t
2 y 2
g(0, y) = h(ey ).
Proof. Letting s = T t and x = ey+(
/2r)t
(5.21)
(5.22)
we have

2
2
f
g
(t, y) = rert f (T t, ey+( /2r)t ) ert
T t, ey+( /2r)t
t
s
2

2
2
f
+
r ert ey+( /2r)t
T t, ey+( /2r)t
2
x
2

f
f
= rert f (T t, x) ert (T t, x) +
r ert x (T t, x)
s
2
x
1 rt 2 2 2 f
2 rt f
= e x
(T t, x) +
e x (T t, x),
(5.23)
2
x2
2
x
"
123
N. Privault
where on the last step we used the Black-Scholes PDE. On the other hand
we have

2
2
g
f
(t, y) = ert ey+( /2r)t
T t, ey+( /2r)t
y
x
and

2
2 rt y+(2 /2r)t f
1 g 2
(t, y) =
e e
T t, ey+( /2r)t
2 y 2
2
x

2
2
2
2f
+ ert e2y+2( /2r)t 2 T t, ey+( /2r)t
2
x
2 rt f
2 rt 2 2 f
=
e x (T t, x) +
e x
(T t, x).
(5.24)
2
x
2
x2
We conclude by comparing (5.23) with (5.24), which shows that g(t, x) satisfies the heat equation (5.17) with initial condition
g(0, y) = f (T, ey ) = h(ey ).

In the next proposition we recover the Black-Scholes formula by solving the
PDE (5.19). The Black-Scholes will also be recovered by probabilistic arguments and the computation of an expectation in Proposition 6.4.
Proposition 5.8. When h(x) = (x K)+ , the solution of the Black-Scholes
PDE (5.19) is given by
f (t, x) = x(d+ ) Ker(T t) (d ),
where
1 w x y2 /2
(x) =
e
dy,
2
x R,
and
d+ =
log(x/K) + (r + 2 /2)(T t)
,
T t
d =
log(x/K) + (r 2 /2)(T t)
.
T t
2
Proof. By inversion of (5.20) with s = T t and x = ey+( /2r)t we get

( 2 /2 r)(T s) + log x
.
f (s, x) = er(T s) g T s,
Hence using the solution (5.18) and Relation (5.21) we get

( 2 /2 r)(T t) + log x
f (t, x) = er(T t) g T t,
124
"

w

2
( 2 /2 r)(T t) + log x
dz
+ z ez /(2(T t)) p
2(T t)

w
2
2
dz
= er(T t)
h xez( /2r)(T t) ez /(2(T t)) p
2(T t)
+
w
2
2
dz
= er(T t)
xez( /2r)(T t) K ez /(2(T t)) p
2(T t)

w
2
2
dz
= er(T t) (r+2 /2)(T t)+log(K/x) xez( /2r)(T t) K ez /(2(T t)) p
2(T t)
w
2
2
dz
= xer(T t)
ez( /2r)(T t) ez /(2(T t)) p
d T t
2(T t)
w
2
dz
ez /(2(T t)) p
Ker(T t)
d T t
2(T t)
w
2
2
dz
ez (T t)/2z /(2(T t)) p
=x
d T t
2(T t)
w
2
dz
Ker(T t)
ez /(2(T t)) p
d T t
2(T t)
w
2
dz
=x
e(z(T t)) /(2(T t)) p
d T t
2(T t)
w
2
dz
Ker(T t)
ez /(2(T t)) p
d T t
2(T t)
w
2
dz
ez /(2(T t)) p
=x
d T t(T t)
2(T t)
w
2
dz
Ker(T t)
ez /(2(T t)) p
d T t
2(T t)
w
w
2
2
dz
dz
=x
ez /2 Ker(T t)
ez /2
d T t
d
2
2
= x (1 (d+ )) Ker(T t) (1 (d ))
= er(T t)
= x (d+ ) Ker(T t) (d ) ,
where we used the relation
1 (a) = (a),
a R.
"
125
N. Privault
Exercises
Exercise 5.1
dSt = St dt + dBt
in terms of , > 0, and the initial condition S0 .
b) Write down the Black-Scholes PDE satisfied by the function C(t, x), where
C(t, St ) is the price at time t [0, T ] of the contingent claim with payoff
(ST ) = exp(ST ).
c) Solve the Black-Scholes PDE of Question (b).
Hint: Seach for a solution of the form

2 2
C(t, x) = exp r(T t) + xh(t) +
(h (t) 1) ,
4r
where h(t) is a function to be determined, with h(T ) = 1.
d) Compute the strategy (t , t )t[0,T ] that hedges the contingent claim with
payoff exp(ST ).
Exercise 5.2 On December 18, 2007, a call warrant has been issued by
Fortis Bank on the stock price S of the MTR Corporation with maturity
T = 23/12/2008, Strike K = HK$ 36.08 and Entitlement ratio=10. Recall
that in the Black-Scholes model, the price at time t of a European claim on
the underlying asset St , with strike price K, maturity T , interest rate r and
volatility is given by the Black-Scholes formula as
f (t, St ) = St (d+ ) Ker(T t) (d ),
where
d =
(r 2 /2)(T t) + log(St /K)
T t
Recall that
and d+ = d + T t.
f
(t, St ) = (d+ ),
x
cf. Proposition 5.5.

a) Using the values of the Gaussian cumulative distribution function, compute the Black-Scholes price of the corresponding call option at time
t =November 07, 2008 with St = HK$ 17.200, assuming a volatility =
90% = 0.90 and an annual risk-free interest rate r = 4.377% = 0.04377,
126
"

b) Still using the values of the Gaussian cumulative distribution function,
compute the quantity of the risky asset required in your portfolio at time
t =November 07, 2008 in order to hedge one such option at maturity
T = 23/12/2008.
c) Figure 1 represents the Black-Scholes price of the call option as a function
of [0.5, 1.5] = [50%, 150%].
0.6
Black-Scholes price
0.5
HK$
0.4
0.3
0.2
0.1
0
0.5
0.6
0.7
0.8
0.9
1
sigma
1.1
1.2
1.3
1.4
1.5
Fig. 5.10: Option price as a function of the volatility .

Knowing that the closing price of the warrant on November 07, 2008 was
HK$ 0.023, which value can you infer for the implied volatility at this
date ?
Exercise 5.3 Forward contracts. Recall that the price t (C) of a claim C =
h(ST ) of maturity T can be written as t (C) = g(t, St ), where the function
g(t, x) satisfies the Black-Scholes PDE
g
g
1
2g
rg(t, x) =
(t, x) + rx (t, x) + x2 2 2 (t, x),
t
x
2
x
g(T, x) = h(x),
(1)
with terminal condition g(T, x) = h(x).
a) Assume that C is a forward contract with payoff
C = ST K,
at time T . Find the function h(x) in (1).
b) Find the solution g(t, x) of the above PDE and compute the price t (C)
at time t [0, T ].
Hint: search for a solution of the form g(t, x) = x (t) where (t) is a
function of t to be determined.
"
127
N. Privault
c) Compute the quantity
g
(t, St )
x
of risky assets in a self-financing portfolio hedging C.
t =
Exercise 5.4
a) Solve the Black-Scholes PDE
rg(t, x) =
g
g
2 2 2 g
(t, x) + rx (t, x) +
x
(t, x)
t
x
2
x2
(5.25)
with terminal condition g(T, x) = 1.

Hint: Try a solution of the form g(t, x) = f (t) and find f (t).
b) Find the respective quantities t and t of the risky asset St and riskless
asset At = ert in the portfolio with value
Vt = g(t, St ) = t St + t At
hedging the contract with payoff $1 at maturity.
Similar exercise: Repeat the above questions with the terminal condition
g(T, x) = x.
Exercise 5.5 Forward contracts revisited. Consider a risky asset whose price
2
St is given by St = S0 eBt +rt t/2 , t R+ , where (Bt )tR+ is a standard
Brownian motion. Consider a forward contract with maturity T and payoff
ST .
a) Compute the price Ct of this claim at any time t [0, T ].
b) Compute a hedging strategy for the option with payoff ST .
Exercise 5.6 Computation of Greeks. Consider an underlying asset whose
price (St )tR+ is given by a stochastic differential equation of the form
dSt = rSt dt + (St )dWt ,
where (x) is a Lipschitz coefficient, and an option with payoff function
and price

h
i

C(x, T ) = erT IE (ST )S0 = x ,
where (x) is a twice continuously differentiable (C 2 ) function, with S0 = x.
Using the Ito formula, show that the sensitivity
128
"
ThetaT =

i
rT h

e
IE (ST )S0 = x
T
of the option price with respect to maturity T can be expressed as

i
i
h
h

ThetaT = rerT IE (ST )S0 = x + rerT IE St 0 (ST )S0 = x

i
h
1

+ erT IE 00 (ST ) 2 (ST )S0 = x .
2
Exercise 5.7 Black-Scholes PDE with dividends. Consider an underlying asset
price process (St )tR+ modeled as
dSt = ( D)St dt + St dBt ,
where (Bt )tR+ is a standard Brownian motion and D > 0 is a continuoustime dividend rate. By absence of arbitrage, the payment of a dividend entails
a drop in the stock price by the same amount.
a) Write down the corresponding Black-Scholes PDE for the price g(t, St ) of
a European call option.
b) Compute the price at time t [0, T ] of the European call option in a
market with dividend rate D.
Exercise 5.8 Show that the Black-Scholes PDE of Proposition 5.3 can be
recovered from the induction relation (3.16) when the number N of time
steps tends to infinity, using the renormalizations rN := rT /N and
aN := (1 + rN )e T /N 1, bN := (1 + rN )e T /N 1, N 1,
of Section 3.6.
"
129
Chapter 6
Martingale Approach to Pricing and

Hedging
In this chapter we present the probabilistic martingale approach method to

the pricing and hedging of options. In particular, this allows one to compute
option prices as the expectations of the discounted option payoffs, and to
determine the associated hedging portfolios.
6.1 Martingale Property of the It

o Integral
Recall (Definition 5.3) that an integrable process (Xt )tR+ is said to be a
martingale with respect to the filtration (Ft )tR+ if
IE[Xt | Fs ] = Xs ,
0 s t.
The following result shows that the indefinite Ito integral is a martingale with
respect to the Brownian filtration (Ft )tR+ . It is the continuous-time analog
of the discrete-time Proposition 2.1.
r

t
Proposition 6.1. The indefinite stochastic integral
u dBs
of a
0 s
tR+
square-integrable adapted process u L2ad ( R+ ) is a martingale, i.e.:

w
ws
t

u dB Fs =
u dB ,
0 s t.
IE
0
Proposition 6.1 is a consequence of Proposition 6.2 below, which shows

that
w

i
hw
t

IE
u dB Fs = IE
1[0,t] ( )u dB Fs
0
0
i
hw s

= IE
1[0,t] ( )u dB Fs
0
"
N. Privault
=
ws
0
0 s t.
u dB ,
Proposition 6.2. For any u L2ad ( R+ ) we have

i wt
hw

us dBs ,
t R+ .
IE
us dBs Ft =
0
In particular,
rt
0
us dBs is Ft -measurable, t R+ .
Proof. The statement is first proved in case u is a simple predictable process,

and then extended to the general case, cf. e.g. Proposition 2.5.7 in [89]. For
example, for u of the form us := F 1[a,b] (s) with F and Fa -measurable random
variable and t [a, b] we have
i
i
hw
hw

IE
us dBs Ft = IE
F 1[a,b] (s)dBs Ft
0
i
h 0

= IE F (Bb Ba )Ft
i
h

= F IE (Bb Ba )Ft
= F (Bt Ba )
wt
=
us dBs ,
a t b.
On the other hand, when t [0, a] we have

i
i
hw
hw

IE
us dBs Ft = IE
F 1[a,b] (s)dBs Ft
0
0
i
h

= IE F (Bb Ba )Ft
i i
h h

= IE IE F (Bb Ba )Fa Ft
i i
h
h

= IE F IE Bb Ba Fa Ft
=0
wt
=
us dBs ,
0 t a.
The extension from simple processes to square-integrable processes in L2ad (

R+ ) as in Proposition 4.3. Indeed, given (un )nN be a sequence of simple predictable processes converging to u in L2 ( [0, T ]), by Fatous lemma and
the continuity of the conditional expectation on L2 we have:
"
#
i2
hw
wt

IE
us dMs IE
us dMs Ft
0
lim inf IE
n
"
w
uns dMs IE
hw
0
i2

us dMs Ft
132
"
Martingale Approach to Pricing and Hedging

i2
hw
w

IE
uns dMs
us dMs Ft
n
0
0
w
2
w

lim IE IE
uns dMs
us dMs Ft
n
0
0
w
2
= lim IE
(uns us )dMs
n
0
i
hw
|uns us |2 ds
= lim IE
= lim IE

= 0,
where we used the Ito isometry (4.12).
In particular, since F0 = {, }, this recover the fact that the It

o integral is
a centered random variable:
i w0
hw
i
hw

IE
us dBs = IE
us dBs F0 =
us dBs = 0.
0
Examples
1. Given any square-integrable random variable F L2 (), the process
(Xt )tR+ defined by Xt := IE[F | Ft ], t R+ , is a martingale under P,
as follows from the tower property
IE[Xt | Fs ] = IE[IE[F | Ft ] | Fs ] = IE[F | Fs ] = Xs ,
0 s t, (6.1)
cf. (16.25) in appendix.

2. Any integrable stochastic process (Xt )tR+ with centered and independent increments is a martingale:
IE[Xt |Fs ] = IE[Xt Xs + Xs |Fs ]
= IE[Xt Xs |Fs ] + IE[Xs |Fs ]
= IE[Xt Xs ] + Xs
= Xs ,
0 s t.
(6.2)
In particular, the standard Brownian motion (Bt )tR+ is a martingale

because it has centered and independent increments. This fact can also
be recovered from Proposition 6.1 since Bt can be written as
wt
Bt =
dBs ,
t R+ .
0
3. The discounted asset price

Xt = X0 e(r)t+Bt
"
t/2
133
N. Privault
is a martingale when = r. Indeed we have
IE[Xt |Fs ] = IE[X0 eBt
t/2
|Fs ]
= X0 e
t/2
IE[eBt |Fs ]
= X0 e
t/2
IE[e(Bt Bs )+Bs |Fs ]
= X0 e
= X0 e
2 t/2+Bs
IE[e(Bt Bs ) |Fs ]
2 t/2+Bs
IE[e(Bt Bs ) ]
= X0 e
t/2+Bs 2 (ts)/2
= X0 eBs
s/2
0 s t.
= Xs ,
This fact can also be recovered from Proposition 6.1 since Xt satisfies the
equation
dXt = Xt dBt ,
i.e. it can be written as the Brownian stochastic integral
wt
Xt = X0 + Xu dBu ,
t R+ .
0
4. The discounted value

Vet = ert Vt
of a self-financing portfolio is given by
wt
Vet = Ve0 +
u dXu ,
0
t R+ ,
cf. Lemma 5.1 is a martingale when = r by Proposition 6.1 because

wt
t R+ ,
Vet = Ve0 + u Xu dBu ,
0
since
dXt = Xt (( r)dt + dBt ) = Xt dBt .
Since the Black-Scholes theory is in fact valid for any value of the parameter
we will look forward to including the case 6= r in the sequel.
6.2 Risk-neutral Measures

Recall that by definition, a risk-neutral measure is a probability measure P
under which the discounted asset price (Xt )tR+ = (ert St )tR+ is a martingale. From the analysis of Section 6.1 it appears that when = r, (Xt )tR+
134
"

is a martingale and P = P is risk-neutral.
In this section we address the construction of a risk-neutral measure in
the general case 6= r and for this we will use the Girsanov theorem.
Note that the relation
dXt = Xt (( r)dt + dBt )
can be rewritten as
t ,
dXt = Xt dB
where
t := r t + Bt ,
B
t R+ .
Therefore the search for a risk-neutral measure can be replaced by the search
t )tR is a standard Brownian
for a probability measure P under which (B
+
motion.
Let us come back to the informal interpretation of Brownian motion via
its infinitesimal increments:
Bt = dt,
with
1
P(Bt = + dt) = P(Bt = dt) = .
2
2
Drifted Brownian motion
Drift
1.6
1.2
0.8
0.4
0.2
0.4
0.6
0.8
Fig. 6.1: Drifted Brownian path.

Clearly, given R, the drifted process Bt := t+Bt is no longer a standard
Brownian motion because it is not centered:
"
135
N. Privault
IE[t + Bt ] = t + IE[Bt ] = t 6= 0,
cf. Figure 6.1. This identity can be formulated in terms of infinitesimal increments as
IE[dt + dBt ] =
1
1
(dt + dt) + (dt dt) = dt 6= 0.
2
2
In order to make t+Bt a centered process (i.e. a standard Brownian motion,

since t + Bt conserves all the other properties (i)-(iii) in the definition of
Brownian motion, one may change the probabilities of ups and downs, which
have been fixed so far equal to 1/2.
That is, the problem is now to find two numbers p, q [0, 1] such that
p(dt + dt) + q(dt dt) = 0
p + q = 1.
The solution to this problem is given by

p=
1
(1 dt)
2
and q =
1
(1 + dt).
2
Coming back to Brownian

motion considered as a discrete random walk with
independent increments dt, we try to construct a new probability measure
denoted P , under which the drifted process Bt := t + Bt will be a standard
Brownian motion. This probability measure will be defined through its density dP /dP with respect to the historical probability measure P, obtained
by taking the product of the above probabilities divided by the reference
probability 1/2N corresponding to the symmetric random walk, that is:
Y 1 1
dP
1
'
dt
dP
(1/2)N
2 2
0<t<T
where 2N is a normalization factor and N = T /dt is the (infinitely large)

number of discrete time steps. Using elementary calculus, this density can
be informally shown to converge as follows as N tends to infinity, i.e. as the
time step dt = T /N tends to zero:
Y 1 1
Y

2N
dt =
1 dt
2 2
0<t<T
0<t<T
!
Y

= exp log
1 dt
0<t<T
136
"
= exp

log 1 dt
0<t<T
1 X
dt
( dt)2
2
0<t<T
0<t<T
!
X
1 2 X
= exp
dt
dt
2
0<t<T
0<t<T

1
= exp BT 2 T ,
2
based on the approximations
X
BT '
dt
' exp
and T '
0<t<T
dt.
0<t<T
6.3 Girsanov Theorem and Change of Measure

In this section we restate the Girsanov theorem in a more rigorous way, using
changes of probability measures. Recall that, given Q a probability measure
on , the notation
dQ
=F
dP
means that the probability measure Q has a density F with respect to P,
where F is a non-negative random variable such that IE[F ] = 1. We also
write
dQ = F dP,
which is equivalent to stating that
w
w
IEQ [] =
()dQ() =
()F ()dP() = IE [F ] ,
where is an integrable random variable. In addition we say that Q is equivalent to P when F > 0 with P-probability one.
Recall that here, = C0 ([0, T ]) is the Wiener space and is a
continuous function on [0, T ] starting at 0 in t = 0. Consider the probability
Q defined by

1
dQ() = exp BT 2 T dP().
2
Then the process t + Bt is a standard (centered) Brownian motion under
Q.
"
137
N. Privault
For example, the fact that T + BT has a standard (centered) Gaussian law
under Q can be recovered as follows:
w
IEQ [f (T + BT )] =
f (T + BT )dQ

w
1
=
f (T + BT ) exp BT 2 T dP
2

w
2
dx
1 2
=
f (T + x) exp x T ex /(2T )
2
2T
w
y 2 /(2T ) dy
=
f (y)e
2T
w
=
f (BT )dP
= IEP [f (BT )].

The Girsanov theorem can actually be extended to shifts by adapted processes as follows, cf. e.g. [96], Theorem III-42. Section 14.6 will cover the
extension of the Girsanov theorem to jump processes.
Theorem 6.1. Let (t )t[0,T ] be an adapted process satisfying the Novikov
integrability condition

w

1 T
IE exp
|t |2 dt
< ,
(6.3)
2 0
and let Q denote the probability measure defined by
w

T
dQ
1wT 2
= exp
s dBs
s ds .
0
dP
2 0
Then
t := Bt +
B
wt
0
s ds,
t [0, T ],
is a standard Brownian motion under Q.

When applied to
t :=
r
,
the Girsanov theorem shows that

t := r t + Bt ,
B
t [0, T ],
(6.4)
is a standard Brownian motion under the probability measure P defined by

dP
r
( r)2
= exp
BT
T .
(6.5)
2
dP
138
"

Hence the discounted price process given by
dXt
t ,
= ( r)dt + dBt = dB
Xt
t R+ ,
is a martingale under P , hence P is a risk-neutral measure. We obviously

have P = P when = r.
6.4 Pricing by the Martingale Method

In this section we give the expression of the Black-Scholes price using expectations of discounted payoffs.
Recall that from the first fundamental theorem of mathematical finance,
a continuous market is without arbitrage opportunities if there exists (at
least) a risk-neutral probability measure P under which the discounted price
process
Xt := ert St ,
t R+ ,
is a martingale under P . In addition, when the risk-neutral measure is
unique, the market is said to be complete.
In case the price process (St )s[t,) satisfies the equation
dSt
= dt + dBt ,
St
t R+ ,
S0 > 0
we have
St = S0 eBt
t/2+t
and Xt = S0 e(r)t+Bt
t/2
t R+ ,
hence from Section 6.2 the discounted price process is a martingale under the
probability measure P defined by (6.5), and P is a martingale measure.
We have
t ,
dXt = ( r)Xt dt + Xt dBt = Xt dB
t R+ ,
(6.6)
hence the discounted value Vet of a self-financing portfolio is written as

wt
Vet = Ve0 +
u dXu
0
wt
eu ,
t R+ ,
= Ve0 + u Xu dB
0
"
139
N. Privault
by Lemma 5.1, and becomes a martingale under P .
As in Chapter 3, the value Vt at time t of a self-financing portfolio strategy
(t )t[0,T ] hedging an attainable claim C will be called an arbitrage price of
the claim C at time t and denoted by t (C), t [0, T ].
Proposition 6.3. Let (t , t )t[0,T ] be a portfolio strategy with price
Vt = t At + t St ,
t [0, T ],
and let C be a contingent claim, such that

(i) (t , t )t[0,T ] is a self-financing portfolio, and
(ii) (t , t )t[0,T ] hedges the claim C, i.e. we have VT = C.
Then the arbitrage price of the claim C is given by
Vt = er(T t) IE [C|Ft ],
0 t T,
(6.7)
where IE denotes expectation under the risk-neutral measure P .

Proof. Since the portfolio strategy (t , t )tR+ is self-financing, by Lemma 5.1
and (6.6) we have
Vet = Ve0 +
wt
0
u ,
u Xu dB
t R+ ,
which is a martingale under P from Proposition 6.1, hence

h
i
Vet = IE VeT | Ft
= erT IE [VT | Ft ]
= erT IE [C | Ft ],
which implies
Vt = ert Vet = er(T t) IE [C | Ft ].

When the process (St )tR+ has the Markov property, the value
Vt = er(T t) IE [(ST )|Ft ] = C(t, St ),
0 t T,
of the portfolio at time t [0, T ] can be written from (6.7) as a function

C(t, St ) of t and St , and by Proposition 5.3 the function C(t, x) solves the
Black-Scholes PDE
140
"
C
C
1
2C
rC(t, x) =
(t, x) + x2 2 2 (t, x) + rx
(t, x)
t
2
x
x
C(T, x) = (x).
In the case of European options with payoff function (x) = (x K)+ we recover the Black-Scholes formula (5.14), cf. Proposition 5.8, by a probabilistic
argument.
Proposition 6.4. The price at time t of a European call option with strike
K and maturity T is given by
C(t, St ) = St (d+ ) Ker(T t) (d ),
t [0, T ].
Proof. The proof of Proposition 6.4 is a consequence of (6.7) and Lemma 6.1
below. Using the relation
ST = St er(T t)+(BT Bt )
(T t)/2
t [0, T ],
by Proposition 6.3 the price of the portfolio hedging C is given by

Vt = er(T t) IE [C|Ft ]
= er(T t) IE [(ST K)+ |Ft ]
= er(T t) IE [(St er(T t)+(BT Bt )

= er(T t) IE [(xe
(T t)/2
T B
t ) 2 (T t)/2
r(T t)+(B
= er(T t) IE [(em(x)+X K)+ ]x=St ,
K)+ |Ft ]
K)+ ]x=St
0 t T,
where
m(x) = r(T t) 2 (T t)/2 + log x
T B
t ) is a centered Gaussian random variable with variance
and X = (B
T B
t )] = 2 Var [B
T B
t ] = 2 (T t)
Var [X] = Var [(B
under P . Hence by Lemma 6.1 below we have
Vt = er(T t) IE [(em(x)+X K)+ ]x=St
= er(T t) em(St )+
Ke
r(T t)
(T t)/2
(v + (m(St ) log K)/v)
((m(St ) log K)/v)
= St (v + (m(St ) log K)/v) Ker(T t) ((m(St ) log K)/v)

= St (d+ ) Ker(T t) (d ),
0 t T.
Lemma 6.1. Let X be a centered Gaussian random variable with variance

v 2 . We have
"
141
N. Privault
IE[(em+X K)+ ] = em+v
/2
(v + (m log K)/v) K((m log K)/v).
Proof. We have
IE[(em+X K)+ ] =
=
=
2v 2 m+log K
em w
2v 2
2
m+v 2 /2
=e
2v 2
(em+x K)+ ex
(em+x K)ex
/(2v 2 )
/(2v 2 )
dx
dx
2
2
K w
ex /(2v ) dx
2v 2 m+log K
w
2
2
2
2
K w
e(v x) /(2v ) dx
ex /2 dx
m+log K
2 (m+log K)/v
w
2
2
ex /(2v ) dx K((m log K)/v)
2
m+log K
em+v /2
=
2v 2
2
em+v /2
=
2v 2
exx
/(2v 2 )
dx
v m+log K
(v + (m log K)/v) K((m log K)/v).

Denoting by
P (t, St ) = er(T t) IE [(K ST )+ |Ft ]
the price of the put option with strike K and maturity T , we check from
Proposition 6.3 that
C(t, St ) P (t, St )
= er(T t) IE [(ST K)+ |Ft ] er(T t) IE [(K ST )+ |Ft ]
= er(T t) IE [(ST K)+ (K ST )+ |Ft ]
= er(T t) IE [ST K|Ft ]
= St er(T t) K.
This relation is called the put-call parity, and it shows that
P (t, St ) = C(t, St ) St + er(T t) K
= St (d+ ) + er(T t) K St er(T t) K(d )
= St (1 (d+ )) + er(T t) K(1 (d ))
= St (d+ ) + er(T t) K(d ).
6.5 Hedging Strategies

In the next proposition we compute a self-financing hedging strategy leading
to an arbitrary square-integrable random variable C admitting a stochastic
integral representation formula of the form
142
"

C = IE [C] +
wT
0
t ,
t dB
(6.8)
where (t )t[0,t] is a square-integrable adapted process. Consequently, the

mathematical problem of finding the predictable representation (6.8) of a
given random variable has important applications in finance. For example we
have
wT
BT2 = T + 2
Bt dBt ,
0
and
BT3 = 3
wT
0
(T t + Bt2 )dBt ,
cf. Exercise 4.2.

Recall that the risky asset follows the equation
dSt
= dt + dBt ,
St
t R+ ,
S0 > 0,
and the discounted asset price satisfies

t ,
dXt = Xt dB
t R+ ,
X0 = S0 > 0,
t )tR is a standard Brownian motion under the risk-neutral probwhere (B

+
ability measure P .
The following proposition applies to arbitrary square-integrable payoff
functions, i.e. it covers exotic and path-dependent options.
Proposition 6.5. Consider a random payoff C L2 () such that (6.8)
holds, and let
er(T t)
t ,
St
r(T t)
e
IE [C|Ft ] t St
t =
,
At
(6.9)
t =
t [0, T ].
(6.10)
Then the portfolio (t , t )t[0,T ] is self-financing, and letting

Vt = t At + t St ,
t [0, T ],
(6.11)
we have
Vt = er(T t) IE [C|Ft ],
t [0, T ].
(6.12)
In particular we have
VT = C,
(6.13)
i.e. the portfolio (t , t )t[0,T ] yields a hedging strategy leading to C, starting

from the initial value
"
143
N. Privault
V0 = erT IE [C].
Proof. Relation (6.12) follows from (6.10) and (6.11), and it implies
V0 = erT IE [C] = 0 A0 + 0 S0
at t = 0, and (6.13) at t = T . It remains to show that the portfolio strategy
(t , t )t[0,T ] is self-financing. By (6.8) and Proposition 6.1 we have
Vt = t At + t St = er(T t) IE [C|Ft ]

wT
u Ft
= er(T t) IE IE [C] +
u dB
0

wt
r(T t)
u
=e
IE [C] +
u dB
0
rt
= e V0 + e
r(T t)
= ert V0 +
= ert V0 +
wt
0
wt
0
= ert V0 + ert
wt
0
u
u dB
u
u Su er(tu) dB
u
u Xu ert dB
wt
0
u dXu ,
t [0, T ],
which shows that the discounted portfolio value Vet = ert Vt satisfies
wt
Vet = V0 +
u dXu ,
t [0, T ],
0
and this implies that (t , t )t[0,T ] is self-financing by Lemma 5.1.
The above proposition shows that there always exists a hedging strategy
starting from
V0 = IE [C]erT .
In addition, since there exists a hedging strategy leading to
VeT = erT C,
then (Vet )t[0,T ] is necessarily a martingale with
h i

Vet = IE VeT Ft = erT IE [C|Ft ],
and initial value
t [0, T ],
h i
Ve0 = IE VeT = erT IE [C].
In practice, the hedging problem can now be reduced to the computation of

the process (t )t[0,T ] appearing in (6.8). This computation, called the Delta
hedging, can be performed by application of the Ito formula and the Markov
144
"

property, see e.g. [95]. Consider the Markov semi-group (Pt )0tT associated
to (St )t[0,T ] , and defined by
Pt f (Su ) = IE [f (St+u ) | Fu ] = IE [f (St+u ) | Su ],
t, u R+ ,
which acts on functions f Cb2 (R), with

Pt Ps = Ps+t ,
s, t R+ .
Note that (PT t f (St ))t[0,T ] is an Ft -martingale, i.e.:

IE [PT t f (St ) | Fu ] = IE [IE [f (ST ) | Ft ] | Fu ]
= IE [f (ST ) | Fu ]
= PT u f (Su ),
(6.14)
0 u t T , and we have
Ptu f (x) = IE [f (St ) | Su = x] = IE [f (xSt /Su )],
0 u t.
(6.15)
The next lemma allows us to compute the process (t )t[0,T ] in case the payoff
C is of the form C = (ST ) for some function . In case C L2 () is the
payoff of an exotic option, the process (t )t[0,T ] can be computed using the
Malliavin gradient on the Wiener space, cf. [82], [89].
Lemma 6.2. Let Cb2 (Rn ). The predictable representation
(ST ) = IE [(ST )] +
wT
0
t
t dB
(6.16)
t [0, T ].
(6.17)
is given by
t = St
(PT t )(St ),
x
Proof. Since PT t is in C (R), we can apply the Ito formula to the process
t 7 PT t (St ) = IE [(ST ) | Ft ],
which is a martingale from the tower property (6.1) of conditional expectations as in (6.14). From the fact that the finite variation term in the It
o
formula vanishes when (PT t (St ))t[0,T ] is a martingale, (see e.g. Corollary II-6-1 page 72 of [96]), we obtain:
PT t (St ) = PT (S0 ) +
wt
0
Su
u ,
(PT u )(Su )dB
x
t [0, T ],
(6.18)
with PT (S0 ) = IE [(ST )]. Letting t = T , we obtain (6.17) by uniqueness

of the predictable representation (6.16) of C = (ST ).

By (6.15) we also have
"
145
N. Privault

IE [(ST ) | St = x]x=St
x

= St
IE [(xST /St )]x=St ,
t [0, T ],
x
t = St
hence
1 r(T t)
e
t
St
= er(T t)
IE [(xST /St )]x=St ,
x
(6.19)
t =
t [0, T ],
which recovers the formula (5.10) for the Delta of a vanilla option. As a consequence we have t 0 and there is no short selling when the payoff function
is nondecreasing.
In the case of European options, the process can be computed via the next
proposition.
Proposition 6.6. Assume that C = (ST K)+ . Then for 0 t T we
have

ST
ST
t = St IE
1[K,) x
.
St
St
x=St
Proof. This result follows from Lemma 6.2 and the relation PT t f (x) =
x
IE [f (St,T
)], after approximation of x 7 (x) = (xK)+ with C 2 functions.

From Proposition 6.6 we can recover the formula for the Delta of a European
call option in the Black-Scholes model, cf. Proposition 5.5. Proposition 6.7
shows that the Black-Scholes self-financing hedging strategy is to hold a (possibly fractional) quantity

log(St /K) + (r + 2 /2)(T t)
t = (d+ ) =
0
(6.20)
T t
of the risky asset, and to borrow a quantity

log(St /K) + (r t2 /2)(T t)
t = KerT
0
T t
(6.21)
of the riskless (savings) account, cf. also Corollary 10.2 in Chapter 10.
In the next proposition we provide another proof of the result of Proposition 5.5.
Proposition 6.7. The Delta of a European call option with payoff function
f (x) = (x K)+ is given by
146
"

t = (d+ ) =
log(St /K) + (r + 2 /2)(T t)
T t

0 t T.
Proof. By Propositions 6.5 and 6.6 we have

1 r(T t)
e
t
St

ST
ST
1[K,) x
= er(T t) IE
St
St
x=St
t =
t)
= er(T
h
i
2
2
(B
IE e T Bt ) (T t)/2+r(T t) 1[K,) (xe(BT Bt ) (T t)/2+r(T t) )
x=St
w
1
y 2 (T t)/2y 2 /(2(T t))
= p
e
dy
2(T t) (T t)/2r(T t)/+1 log(K/St )
w
2
1
= p
e(y(T t)) /(2(T t)) dy
2(T t) d / T t
1 w (y(T t))2 /2
=
e
dy
2 d
w
2
1
=
ey /2 dy
2 d+
1 w d+ y2 /2
e
dy
=
2
= (d+ ).

As noted above, the result of Proposition 6.7 also follows from (5.10) or
(6.19) and direct differentiation of the Black-Scholes function, cf. (5.16). In
Figure 6.2 we plot the value of the Delta of a European as a function of the
underlying and of time to maturity.
2
1.5
1
0.5
020
15
200
10
150
Time to maturity T-t

100
5
50
0 0
underlying
Fig. 6.2: Delta of a European option with strike K = 100.

"
0198
147
N. Privault
The gamma of the European call option is defined as the second derivative
of the option price with respect to the underlying, i.e.

1
1 log(St /K) + (r + 2 /2)(T t)
p
t =
exp
2
T t
St 2(T t)
In Figure 6.3 we plot the (truncated) value of the Gamma of a European as
a function of the underlying and of time to maturity.
4
3.5
3
2.5
2
1.5
1
0.5
0
101
100.5
0
0.005
100
0.01
underlying
0.015
99.5
0.02
0.025
99 0.03
Fig. 6.3: Gamma of a European option with strike K = 100.
0198
Since Gamma is always nonnegative, the Black-Scholes hedging strategy is

to keep buying the underlying risky asset when its price increases, and to sell
it when its price decreases, as can be checked from Figure 6.3.
Exercises
Exercise 6.1 Consider an asset price (St )tR+ which is a martingale under
the risk-neutral measure P in a market with interest rate r = 0, and let
(x) = (x K)+ be the (convex) European call payoff function.
Show that, for any two maturities T1 < T2 and p, q [0, 1] such that
p + q = 1, the price of the average option with payoff (pST1 + qST2 ) is upper
bounded by the price of the European call option with maturity T2 , i.e. show
that
IE [(pST1 + qST2 )] IE [(ST2 )].
Hint 1: For a convex function we have (px + qy) p(x) + q(y) for any
x, y R and p, q [0, 1] such that p + q = 1.
148
"

Hint 2: Any convex function (St ) of a martingale St is a submartingale.
Exercise 6.2 Consider an underlying asset price process (St )tR+ .

a) Show that the price at time t of a European call option with strike price
K and maturity T is lower bounded by (St Ker(T t) )+ , i.e.
er(T t) IE [(ST K)+ | Ft ] (St Ker(T t) )+ ,
t [0, T ].
b) Show that the price at time t of a European putoption with strike price
K and maturity T is lower bounded by Ker(T t) St , i.e.
er(T t) IE [(K ST )+ | Ft ] Ker(T t) )+ St ,
t [0, T ].
Exercise 6.3 Forward start options [100]. Given two maturity dates T1 < T2 ,
compute the price

er(T1 t) IE er(T2 T1 ) IE (ST2 ST1 )+ | FT1 | Ft
at time t [0, T1 ], of a forward start European call option, i.e. an option
whose holder receives at time T1 the value of a standard European call option at the money, with maturity T2 .
Exercise 6.4 Consider the price process (St )t[0,T ] given by

dSt
= rdt + dBt
St
and a riskless asset of value At = A0 ert , t [0, T ], with r > 0. In this
problem, (t , t )t[0,T ] denotes a portfolio strategy with value
Vt = t At + t St ,
0 t T.
a) Compute the arbitrage price

C(t, St ) = er(T t) IE [|ST |2 | Ft ],
at time t [0, T ], of the power option with payoff |ST |2 .
b) Compute a self-financing portfolio strategy (t , t )t[0,T ] hedging the claim
|ST |2 .
Exercise 6.5 Let again (t , t )t[0,T ] denote a portfolio strategy with value
Vt = t At + t St ,
"
0 t T,
149

N. Privault
where St , resp. At , denotes the price at time t of a risky, resp. riskless, asset.
dSt = St dt + dBt
in terms of , > 0, and the initial condition S0 .
b) For which value M of is the discounted price process St = ert St ,
t [0, T ], a martingale under P ?
c) For each value of , build a probability measure P under which the
discounted price process St = ert St , t [0, T ], is a martingale.
d) Compute the arbitrage price
C(t, St ) = er(T t) IE [exp(ST ) | Ft ]
at time t [0, T ] of the contingent claim with payoff exp(ST ), and recover
the result of Exercise 5.1.
e) Explicitly compute the portfolio strategy (t , t )t[0,T ] that hedges the
contingent claim exp(ST ).
f) Check that this strategy is self-financing.
Exercise 6.6 Let (Bt )tR+ be a standard Brownian motion generating a
filtration (Ft )tR+ . Recall that for f C 2 (R+ R), Itos formula for Brownian
motion reads
w t f
f (t, Bt ) = f (0, B0 ) +
(s, Bs )ds
0 s
w t f
1 w t 2f
+
(s, Bs )dBs +
(s, Bs )ds.
0 x
2 0 x2
2
a) Let r R, > 0, f (x, t) = ert+x t/2 , and St = f (t, Bt ). Compute

df (t, Bt ) by Itos formula, and show that St solves the stochastic differential equation
dSt = rSt dt + St dBt ,
where r > 0 and > 0.
b) Show that

2
IE eBT | Ft = eBt + (T t)/2 ,
0 t T.
Hint: Use the independence of increments in the decomposition

BT = (BT Bt ) + (Bt B0 )
2 2
and the Laplace transform IE[eX ] = e

c) Show that the process (St )tR+ satisfies
/2
when X ' N (0, 2 ).
150
"

IE[ST | Ft ] = er(T t) St ,
0 t T.
d) Let C = ST K denote the payoff of a forward contract with exercise

price K and maturity T . Compute the discounted expected payoff
Vt := er(T t) IE[C | Ft ].
e) Find a self-financing portfolio strategy (t , t )tR+ such that
Vt = t St + t At ,
0 t T,
where At = A0 ert is the price of a riskless asset with interest rate r > 0.
Show that it recovers the result of Exercise 5.3-(c).
f) Show that the portfolio (t , t )t[0,T ] found in Question (e) hedges the
payoff C = ST K at time T , i.e. show that VT = C.
Exercise 6.7 Digital options. Consider a price process (St )tR+ given by
dSt
= rdt + dBt ,
St
S0 = 1,
under the risk-neutral measure P. A digital (or binary) call, resp. put, option
is a contract with maturity T , strike K, and payoff
$1 if ST K,
$1 if ST K,
Cd :=
resp. Pd :=
0 if ST < K,
0 if ST > K.
Recall that the prices t (Cd ) and t (Pd ) at time t of the digital call and put
options are given by the discounted expected payoffs
t (Cd ) = er(T t) IE[Cd | Ft ]
and t (Pd ) = er(T t) IE[Pd | Ft ]. (6.22)
a) Show that the payoffs Cd and Pd can be rewritten as

Cd = 1[K,) (ST )
and
Pd = 1[0,K] (ST ).
b) Using Relation (6.22), Question (a), and the relation

IE[1[K,) (ST ) | St = x] = P(ST K | St = x),
show that the price t (Cd ) is given by
t (Cd ) = Cd (t, St ),
where Cd (t, x) is the function defined by
Cd (t, x) := er(T t) P(ST K | St = x).
"
151
N. Privault
c) Using the results of Exercise 4.10-(c) and of Question (b), show that the
price t (Cd ) of the digital call option is given by

(r 2 /2)(T t) + log(x/K)
Cd (t, x) = er(T t)
T t
= er(T t) (d ),
where
d =
(r 2 /2)(T t) + log(St /K)
.
T t
d) Assume that the binary option holder is entitled to receive a return

amount [0, 1] in case the underlying ends out of the money at maturity. Compute price at time t [0, T ] of this modified contract.
e) Using Relation (6.22) and Question (a), prove the call-put parity relation
t (Cd ) + t (Pd ) = er(T t) ,
0 t T.
(6.23)
If needed, you may use the fact that P(ST = K) = 0.

f) Using the results of Questions (e) and (c), show that the price t (Pd ) of
the digital put is given by
t (Pd ) = er(T t) (d ).
g) Using the result of Question (c), compute the Delta
t :=
Cd
(t, St )
x
of the digital call option. Does the Black-Scholes hedging strategy of such
a call option involve short-selling ? Why ?
h) Using the result of Question (f), compute the Delta
t :=
Pd
(t, St )
x
of the digital put option. Does the Black-Scholes hedging strategy of such
a put option involve short-selling ? Why ?
Exercise 6.8 Option pricing with dividends. (Exercise 5.7 continued) Consider
an underlying asset price process (St )tR+ modeled under the risk-neutral
measure as
dSt = (r D)St dt + St dBt ,
where (Bt )tR+ is a standard Brownian motion and D > 0 is a continuoustime dividend rate. Compute the price at time t [0, T ] of the European call
option in a market with dividend rate D by the martingale method.
152
"

Exercise 6.9 Log options.
a) Consider a market model made of a risky asset with price (St )tR+ as in
Exercise 4.12-(d) and a riskless asset with price At = $1 ert and riskless
interest rate r = 2 /2. From the answer to Exercise 4.12-(b), show that
the arbitrage price
Vt = er(T t) IE[(log ST )+ | Ft ]
at time t [0, T ] of a log call option with payoff (log ST )+ is equal to
r

Bt
T t Bt2 /(2(T t))
e
+ er(T t) Bt
.
Vt = er(T t)
2
T t
b) Show that Vt can be written as
Vt = g(T t, St ),
where g(, x) = er f (, log x), and
r

y2 /(22 )
y
f (, y) =
e
+ y
.
2

c) Figure 6.4 represents the graph of (, x) 7 g(, x), with r = 0.05 = 5%
per year and = 0.1. Assume that the current underlying price is $1 and
there remains 700 days to maturity. What is the price of the option ?
Price
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
700
600
500
400
T-t
300
200
100
0
0.5
1.5
St
Fig. 6.4: Option price as a function of the underlying and of time to maturity
g
(T t, St ) of St at
x
time t in a portfolio hedging the payoff (log ST )+ is equal to
d) Show that the (possibly fractional) quantity t =
Recall the chain rule of derivation
"
1 f
f (, log x) =
(, y)|y=log x .
x
x y
153
N. Privault
t = er(T t)
St
log S
t
T t

0 t T.
g
e) Figure 6.5 represents the graph of (, x) 7 x
(, x). Assuming that the
current underlying price is $1 and that there remains 700 days to maturity,
how much of the risky asset should you hold in your portfolio in order to
hedge one log option ?
Delta
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0 2
1.8
1.6
1.4
1.2
St
0.8
0.6
0.4
0.2
700
600
500
400
300
200
100
T-t
Fig. 6.5: Delta as a function of the underlying and of time to maturity
f) Based on the framework and answers of Questions (c) and (e), should you
borrow or lend the riskless asset At = $1 ert , and for what amount ?
2g
g) Show that the Gamma of the portfolio, defined as t =
(T t, St ),
x2
equals

!
1
log St
(log St )2 /(2 2 (T t))
r(T t) 1
p
e
,
t = e
St2 2(T t)
T t
0 t T.
h) Figure 6.6 represents the graph of Gamma. Assume that there remains
60 days to maturity and that St , currently at $1, is expected to increase.
Should you buy or (short) sell the underlying asset in order to hedge the
option ?
154
"

Gamma
1
0.8
0.6
0.4
0.2
0
-0.20.2
0.4
0.6
0.8
St
1.2
1.4
1.6
1.8
180200
140160
T-t
100120
80
60
Fig. 6.6: Gamma as a function of the underlying and of time to maturity
i) Let now = 1. Show that the function f (, y) of Question (b) solves the
heat equation
f
1 2f
(, y) =
(, y)
2 y 2
f (0, y) = (y)+ .
Exercise 6.10 Log options with given strike.
a) Consider a market model made of a risky asset with price (St )tR+ as in
Exercise 4.10, a riskless asset with price At = $1 ert , riskless interest
rate r = 2 /2 and S0 = 1. From the answer to Exercise 16.4-(b), show
that the arbitrage price
Vt = er(T t) IE [(K log ST )+ | Ft ]
at time t [0, T ] of a log call option with strike K and payoff (Klog ST )+
is equal to
r

K/ Bt
T t (Bt K/)2 /(2(T t)) r(T t)
e
+e
(KBt )
.
Vt = er(T t)
2
T t
b) Show that Vt can be written as
Vt = g(T t, St ),
where g(, x) = er f (, log x), and
r

(Ky)2 /(22 )
K y
e
+ (K y)
.
f (, y) =
2

"
155
N. Privault
c) Figure 6.7 represents the graph of (, x) 7 g(, x), with r = 0.125 per
year and = 0.5. Assume that the current underlying price is $3 and
there remains 700 days to maturity. What is the price of the option ?
Price
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
2.2
2.4
2.6 2.8
St
3.2
3.4
3.6
3.8
100
200
300
400
500
600
700
T-t
Fig. 6.7: Option price as a function of the underlying and of time to maturity
g
(T t, St ) of St at time t in a portfolio
x
+
hedging the payoff (K log ST ) is equal to

1
K log St
t = er(T t)
,
0 t T.
St
T t
d) Show that the quantity t =
g
e) Figure 6.8 represents the graph of (, x) 7 x
(, x). Assuming that the
current underlying price is $3 and that there remains 700 days to maturity,
how much of the risky asset should you hold in your portfolio in order to
hedge one log option ?
Recall the chain rule of derivation
1 f
f (, log x) =
(, y)|y=log x .
x
x y
156
"

Delta
0
-0.05
-0.1
-0.15
-0.2
-0.25
-0.3
-0.35
-0.4
-0.45
-0.5
4
3.8
3.6
3.4
3.2
St
2.8
2.6
2.4
2.2
700
600
500
400
300
200
100
T-t
Fig. 6.8: Delta as a function of the underlying and of time to maturity
f) Based on the framework and answers of Questions (c) and (e), should you
borrow or lend the riskless asset At = $1 ert , and for what amount ?
2g
g) Show that the Gamma of the portfolio, defined as t =
(T t, St ),
x2
equals
t = er(T t)
1
St2
2(T t)
e(Klog St )
/(2 2 (T t))

+
K log St
T t
!
h) Figure 6.9 represents the graph of Gamma. Assume that there remains
10 days to maturity and that St , currently at $3, is expected to increase.
Should you buy or (short) sell the underlying asset in order to hedge the
option ?
Gamma
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
010
20
30
T-t
40
50
60
70
80
90
100
3.8
3.6
3.4
3.2
2.8 2.6
St
2.4
2.2
Fig. 6.9: Gamma as a function of the underlying and of time to maturity
i) Show that the function f (, y) of Question (b) solves the heat equation
"
157
0t
N. Privault
1 2f
f
(, y) = 2 2 (, y)
2 y
f (0, y) = (K y)+ .
158
"
Chapter 7
Estimation of Volatility
The values of the parameters r, t, St , T , and K used to price a call option via the Black-Scholes formula can be easily obtained from market data.
Estimating the volatility coefficient can be a more difficult task, and several estimation methods are considered in this section with some examples
of how the Black-Scholes formula can be fitted to market data. We cover the
historical, implied, and local volatility models, and refer to [35] for stochastic
volatility models.
7.1 Historical Volatility

We consider the problem of estimating the parameters and from market
data in the stock price model
dSt
= dt + dBt .
St
A natural estimator for the trend parameter can be written as
N :=
N 1
Stk+1 Stk
1 X
1
,
N
tk+1 tk
Stk
(7.1)
k=0
where (Stk+1 Stk )/Stk , k = 0, . . . , N 1 is a family of returns observed at

different times t0 , . . . , tN on the market.
Observe that by replacing (7.1) by actual log-returns with tk+1 tk = T /N ,
k = 0, 1, . . . , N 1, one can get the simplified estimate
N 1

1 X
1
1
ST
log Stk+1 log Stk = log
.
N
tk+1 tk
T
S0
k=0
"
N. Privault
Similarly the parameter can be estimated as by the estimator
N built as
2
N
:=

2
N 1
Stk+1 Stk
1 X
1
N (tk+1 tk ) .
N 1
tk+1 tk
Stk
k=0
Parameter estimation based on historical data requires a lot of samples and

it can only be valid on a given time interval, or as a moving average.
Fig. 7.1: [107] The fugazi: its a wazy, its a woozie. Its fairy dust.
7.2 Implied Volatility

In contrast with the historical volatility, the computation of the implied
volatility can be done at a fixed time and requires much less data.
Recall that when h(x) = (x K)+ , the solution of the Black-Scholes PDE
is given by
f (t, x, K, , r, T ) = x(d+ ) Ke(T t)r (d ),
where
1 w x y2 /2
e
dy,
(x) =
2
x R,
and
Click on the figure to play the video (works in Acrobat reader on the entire pdf
file).
160
"
d+ =
log(x/K) + (r + 2 /2)(T t)
,
T t
d =
log(x/K) + (r 2 /2)(T t)
.
T t
Equating
f (t, St , K, , r, T ) = M
to the observed value M of a given market price, when t, St , r, T are known,
allows one to infer a value for , as in e.g. Figure 5.10.
This value is called the implied volatility and denoted here by imp (K, T ).
The implied volatility value can be used as an alternative way to quote the
option price, based on the knowledge of the remaining parameters (such as
underlying asset price, time to maturity, interest rate, and strike price). For
example, option price data provided by the Hong Kong stock exchange includes implied volatility computed by inverting the Black-Scholes formula,
cf. Figure S.1.
Given two European call options with strikes K1 , resp. K2 and maturities
T1 , resp. T2 , on the same stock S, this procedure should yield two estimates
imp (K1 , T1 ) and imp (K2 , T2 ) of implied volatilites.
Clearly, there is no reason a priori for the implied volatilites imp (K1 , T1 )
and imp (K2 , T2 ) to coincide. However, in the standard Black-Scholes model
the value of the parameter should be unique for a given stock S. This contradiction between a model and market data is a reason for the development
of more sophisticated stochastic volatility models.
Plotting the different values of the implied volatility as a function of K
and T will yield a planar curve called the volatility surface.
Figure 7.2 presents an estimation of implied volatility for Asian options
whose underlying asset is the price of light sweet crude oil futures traded on
the New York Mercantile Exchange (NYMEX), based on contract specifications and market data obtained from the Chicago Mercantile Exchange.
"
161
N. Privault
Implied volatility surface
0.6
0.55
0.5
0.45
Vol.
0.4
0.35
0.3
0.25
8000
8500
9000
9500
10000
Strike
10500
11000
11500
35
30
25
20
15
10
Time to maturity
Fig. 7.2: Implied volatility of Asian options on light sweet crude oil futures.
As observed in Figure 7.2, the volatility surface can exhibit a smile phenomenon, in which implied volatility is higher at a given end (or at both
ends) of the range of strike values.
7.3 The Black-Scholes Formula vs Market Data

On July 28, 2009 a call warrant has been issued by Merrill Lynch on the
stock price S of Cheung Kong Holdings (0001.HK) with Strike K=$109.99
and Maturity T = December 13, 2010.
Fig. 7.3: Graph of the (market) stock price of Cheung Kong Holdings.
This graph is courtesy of Tan Yu Jia.
162
"
The market price of the option (17838.HK) on September 28 was $12.30, as
obtained from http://www.hkex.com.hk/dwrc/search/listsearch.asp
The next graph in Figure 7.4 shows the evolution of the market price of
the option over time. One sees that the option price is much more volatile
than the underlying stock price.
Fig. 7.4: Graph of the (market) call option price on Cheung Kong Holdings.
In Figure 7.5 we have fitted the path
t 7 gc (t, St )
of the Black-Scholes price to the data of Figure 7.4 using the stock price data
of Figure 7.3, by varying the values of the volatility .
0.2
Black-Scholes price
0.18
HK$
0.16
0.14
0.12
0.1
Jul17
Aug06
Aug26
Sep15
Fig. 7.5: Graph of the Black-Scholes call option price on Cheung Kong Holdings.
"
163
N. Privault
Another example
Let us consider the stock price of HSBC Holdings (0005.HK) over one year:
Fig. 7.6: Graph of the (market) stock price of HSBC Holdings.
Next we consider the graph of the price of a call option issued by Societe
Generale on 31 December 2008 with strike K=$63.704, maturity T = October
05, 2009, and entitlement ratio 100, cf. page 6.
Fig. 7.7: Graph of the (market) call option price on HSBC Holdings.
As above, in Figure 7.8 we have fitted the path t 7 gc (t, St ) of the BlackScholes price to the data of Figure 7.7 using the stock price data of Figure 7.6.
In this case we are in the money at maturity, and we also check that the
option is worth 100 0.2650 = $26.650 at that time which, by absence of
arbitrage, is very close to the value $90 - $63.703 = $26.296 of its payoff.
164
"
0.3
Black-Scholes price
HK$
0.2
0.1
0
Nov 08
Jan 09
Mar 09
May 09
Jul 09
Sep 09
Fig. 7.8: Graph of the Black-Scholes call option price on HSBC Holdings.
For one more example, consider the graph of the price of a put option issued
by BNP Paribas on 04 November 2008 with strike K=$77.667, maturity T =
October 05, 2009, and entitlement ratio 92.593.
Fig. 7.9: Graph of the (market) put option price on HSBC Holdings.
One checks easily that at maturity, the price of the put option is worth $0.01
(a market price cannot be lower), which almost equals the option payoff $0,
by absence of arbitrage opportunities. Figure 7.10 is a fit of the Black-Scholes
put price graph
t 7 gp (t, St )
to Figure 7.9 as a function of the stock price data of Figure 7.8. Note that the
Black-Scholes price at maturity is strictly equal to 0 while the corresponding
market price cannot be lower than one cent.
The normalized market data graph in Figure 7.11 shows how the option
price can track the values of the underlying. Note that the range of values
[26.55, 26.90] for the underlying corresponds to [0.675, 0.715] for the option
"
165
N. Privault
0.5
Black-Scholes price
0.4
HK$
0.3
0.2
0.1
0
Nov 08
Jan 09
Mar 09
May 09
Jul 09
Sep 09
Fig. 7.10: Graph of the Black-Scholes put option price on HSBC Holdings.
price, meaning 1.36% vs 5.9% in percentage. This is a European call option

on the ALSTOM underlying with strike price K =e 20, maturity March 20,
2015, and entitlement ratio 10.
Fig. 7.11: Call option price vs ALSTOM underlying.
7.4 Local Volatility

Since the constant volatility assumption in the Black-Scholes model appears
to be not satisfying due to the existence of volatility smiles, it makes sense
to consider models of the form
dSt
= rdt + t dBt
St
166
"
where t is a random process. Such models are called stochastic volatility
models.
A particular class of stochastic volatility models can be written as
dSt
= rdt + (t, St )dBt
St
(7.2)
where (t, x) is a deterministic function of time and the stock price. Such
models are called local volatility models. The corresponding Black-Scholes
PDE can be written as
g
g
1
2g
rg(t, x, K) =
(t, x, K) + rx (t, x, K) + x2 2 (t, x) 2 (t, x, K),
t
x
2
x
g(T, x, K) = (x K)+ ,
(7.3)
with terminal condition g(T, x, K) = (x K)+ , i.e. we consider European
call options.
Note that the Black-Scholes PDE would allow one to recover the value of
(t, x) as a function of the option price g(t, x, K), as
v
u
u 2rg(t, x, K) 2 g (t, x, K) 2rx g (t, x, K)
u
t
x
(t, x) = u
,
x, t > 0,
t
2
2 g
x
(t, x, K)
2
x
however this formula requires the knowledge of the option price for different
values of the underlying x, in addition to the knowledge of the strike price K.
The Dupire formula brings a solution to the local volatility calibration
problem by providing an estimator of (t, x) as a function (t, K) based on
the values of the strike price K.
Proposition 7.1. Assume that a family (C(T, K))T,K>0 of market call option prices with maturities T and strikes K is given at time t with St = x,
while the values of r and x are fixed.
The Dupire formula states that, defining the volatility function (t, y) by
v
u C
C
u
(t, y) + 2ry
(t, y)
u2
y
u t
(t, y) := u
,
(7.4)
2
t
C
y 2 2 (t, y)
y
"
167
N. Privault
the prices g(t, x, K) computed from the Black-Scholes PDE (7.3) will match
the option prices C(T, K) in the sense that
g(t, x, K) = C(T, K),
T, K > 0.
(7.5)
Proof. We use the probabilistic approach that allows us to write g(t, x, K)

as
g(t, x, K) = er(T t) IE[(ST K)+ | St = x],
(7.6)
where (St )tR+ is defined by (7.2), and use stochastic calculus. Hence the
condition (7.5) can be written at t = 0 as
w
C(T, K) = erT
(y K)+ T (y)dy
w
= erT
(y K)T (y)dy
wK
w
= erT
yT (y)dy KerT
T (y)dy
(7.7)
K
wK
rT
rT
=e
yT (y)dy Ke
P(ST K),
K
where T (y) is the probability density of ST . By differentiation of (7.7) with

respect to K, one gets
w
C
(T, K) = erT KT (K) erT
T (y)dy + erT KT (K)
K
K
w
= erT
T (y)dy,
hence twice differentiation of C(T, K) with respect to K shows that

2C
(T, K) = erT T (K),
K 2
(7.8)
cf. Relation (1) in [8]. On the other hand, for any sufficiently smooth function
f , using the Ito formula we have
w
T (y)f (y)dy = IE[f (ST )]

wT
1 w T 00
f (St ) 2 (t, St )dt
= IE f (S0 ) +
f 0 (St )dSt +
0
2 0

wT
wT
1 w T 00
0
f (St )St dt +
f 0 (St )St dBt +
f (St ) 2 (t, St )dt
= IE f (S0 ) + r
0
0
2 0
w

wT
T
1
= f (S0 ) + IE r
f 0 (St )St dt +
f 00 (St )St2 2 (t, St )dt
0
2 0
w wT
1 w w T 2 00
0
yf (y)t (y)dtdy +
y f (y) 2 (t, y)t (y)dtdy,
= f (S0 ) + r
0
2 0
hence after differentiating both sides of the equality with respect to T ,
168
"
w
w
1 w 2 00
T
(y)f (y)dy = r
yf 0 (y)T (y)dy+
y f (y) 2 (T, y)T (y)dy.
T
2
Integrating by parts in the above relation yields
w
T
(y)f (y)dy
T
w
1w
2
f (y) 2 (y 2 2 (T, y)T (y))dy,

= r
f (y) (yT (y))dy +
y
2
y
for all smooth functions f (y) with compact support, hence
1 2 2 2
T
(y) = r (yT (y)) +
(y (T, y)T (y)),
T
y
2 y 2
y R.
Making use of (7.8) we get

r
2C
2C
(T, y)
(T, y)
y 2
T y 2

2

1 2
C
2C
y 2 (T, y)
y 2 2 (T, y) 2 (T, y) ,
=r
2
y
y
2 y
y
y R.
After a first integration with respect to y under the limiting condition

limK+ C(T, K) = 0, we obtain

C
C
2C
1
2C
r
(T, y)
(T, y) = ry 2 (T, y)
y 2 2 (T, y) 2 (T, y) ,
y
T y
y
2 y
y
i.e.
C
C
(T, y)
(T, y)
y
T y

C
C
1
2C
y
(T, y) r
(T, y)
y 2 2 (T, y) 2 (T, y) ,
=r
y
y
y
2 y
y
or
(T, y) = r
y T
y

C
1
2C
y
(T, y)
y 2 2 (T, y) 2 (T, y) .
y
2 y
y
Integrating one more time with respect to y yields
C
C
1
2C
(T, y) = ry
(T, y) y 2 2 (T, y) 2 (T, y),
T
y
2
y
which conducts to (7.4) and is called the Dupire [28] PDE.
y R,
From (7.4) the local volatility (t, y) can be estimated by computing

C(T, y) by the Black-Scholes formula, based on a value of the implied volatil"
169
N. Privault
ity . See [1] and in particular Figure 8.1 therein for numerical methods
applied to volatility estimation in this framework.
7.5 Stochastic Volatility

Time-Dependent Stochastic Volatility
The next Figure 7.12 refers to the EURO/SGD exchange rate, and shows
some spikes that cannot be generated by Gaussian returns with constant
variance.
Fig. 7.12: Euro / SGD exchange rate.
This type data shows that, in addition to jump models that are commonly
used to take into account the slow decrease of probability tails observed in
market data, other tools should be implemented in order to model a possibly
random and time varying volatility.
We consider an asset price driven by the stochastic differential equation
dSt = rSt dt + St vt dBt
(7.9)
under the risk-neutral measure P , where (vt )tR+ is a (possibly random)

(1)
squared volatility process adapted to the filtration Ft generated by (Bt )tR+ .
Time-dependent deterministic volatility

When (v(t))tR+ is a deterministic function of time, the solution
170
"

wT p
1wT
ST = St exp r(T t) +
v(s)dBs
v(s)ds
t
2 t
of (7.9) is a lognormal random variable at time T with conditional logvariance
wT
v(s)ds
t
given Ft . In particular, a European call option on ST can be priced by the

Black-Scholes formula
!
rT
v(s)ds
er(T t) IE [(ST K)+ | Ft ] = BS St , r, K, T t, t
,
T t
with integrated squared volatility parameter
v(t) :=
rT
t
v(s)ds
,
T t
t [0, T ).
Independent volatility
(2)
When (vt )tR+ is a random process generating a filtration Ft independent

(1)
of the driving Brownian motion (Bt )tR+ under P , the equation (7.9) can
still be solved as

wT
1wT
ST = St exp r(T t) +
vs dBs
vs ds ,
t
2 t
and ST is a lognormal random variable with random variance
wT
t
vs ds
(2)
given FT . In this case we can still price options with payoff (ST ) on the
underlying ST using the tower property

h
h
i
i
(1)
(2) (1)
(2)
IE [(ST ) | Ft ] = IE IE (ST ) Ft FT Ft Ft .
For example, a European call option on ST can be priced by averaging the
Black-Scholes formula as follows:

h
h
ii
(1)
(2)
er(T t) IE [(ST K)+ | Ft ] = er(T t) IE IE (ST K)+ Ft FT
.
#
!
"
rT

vs ds
= er(T t) IE BS St , r, K, T t, t
Ft
T t
"
171
N. Privault
"
= er(T t) IE BS St , r, K, T t,
rT
vs ds
T t
#

(2)
,
Ft
with the random integrated volatility

vt :=
1 wT
vs ds.
T t t
On the other hand, when (vt )tR)+ is a geometric Brownian motion, the probwT
(2)
ability distribution of the time integral
vs ds given Ft can be computed
t
using integral expressions.
Two-factor Stochastic Volatility Models

Evidence based on financial market data shows that the variations in volatility tend to be negatively correlated with the variations of underlying asset
prices, Figure 1 of [84] and cf. 2.3.1 of [36]. For this reason we need to
consider an asset price process (St )tR+ and a stochastic volatility process
(vt )tR+ driven by
(1)
dSt = rSt + vt St dBt
dv = (t, v )dt + (t, v )dB (2) ,

t
t
t
t
(1)
(2)
Here, (Bt )tR+ and (Bt )tR+ are two Brownian motions such that
(1)
dBt
(2)
dBt
= dt,
where the correlation parameter satisfies 1 1, and the coefficients

(t, x) and (t, x) can be chosen e.g. from mean-reverting models (CIR) or
geometric Brownian models, as follows.
Heston model
In the Heston model [50], the stochastic volatility (vt )tR+ is chosen to be a
CIR process, i.e. we have
(1)
dSt = rSt dt + St vt dBt
dv = (v m)dt + v dB (2) ,
t
t
t
t
172
"
and (t, v) = (vt v) and (t, v) = v, where , m, > 0.

Option pricing formulas can be derived in the Heston model using complex
integrals.
SABR model
In the SABR model ([46], here with = 1), we have
(1)
dSt = t St dBt
d = dB (2) ,
t
t
t
with > 0 and (0, 1], which is not mean-reverting, i.e. it is preferably
used in short time. This model is typically used for the modeling of LIBOR
rates and it allows for short-time asymptotics of Black implied volatilities that
can be used for pricing by inputting them into the Black pricing formula, cf.
3.3 of [97].
Pricing PDE with Stochastic Volatility

Consider a portfolio priced as
i
h

Vt = f (t, vt , St ) = er(T t) IE h(ST ) Ft ,
0 t T , for an option with payoff h(ST ) on ST .
(1)
In the sequel we will assume that (Bt )tR+ is a standard Brownian motion under P , i.e. the discounted price process (ert St )tR+ is a martingale
under P . For simplicity of exposition we will make the assumption that
(2)
(Bt )tR+ is also a standard Brownian motion under P .
(1)
By Ito calculus with respect to the correlated Brownian motions (Bt )tR+
(2)
and (Bt )tR+ , the portfolio value f (t, vt , St ) can be differentiated as follows:
df (t, vt , St )
(7.10)
f
f
f
(1)
(t, vt , St )dt + rSt (t, vt , St )dt + vt St (t, vt , St )dBt
=
t
x
x
2
1
f
2 f
+ vt St 2 (t, vt , St )dt + (t, vt ) (t, vt , St )dt
2
x
v
f
1 2
2f
(2)
+(t, vt ) (t, vt , St )dBt + (t, vt ) 2 (t, vt , St )dt
v
2
v
"
173
N. Privault
2f
+(t, vt ) vt St
(t, vt , St )dt.
vx
(2)
Assuming that (Bt )tR+ is also a standard Brownian motion under the riskneutral measure P and knowing that the discounted portfolio price process
(ert f (t, vt , St ))tR+ is also a martingale under P , from the relation
d(ert f (t, vt , St )) = rert f (t, vt , St )dt + ert df (t, vt , St ),
we obtain
f
1
2f
f
(t, vt , St )dt + rSt (t, vt , St )dt + vt St2 2 (t, vt , St )dt
t
x
2
x
f
1
2f
+ (t, vt ) (t, vt , St )dt + 2 (t, vt ) 2 (t, vt , St )dt
v
2
v
2f
+ (t, vt )St vt
(t, vt , St )dt
vx
= 0,
rf (t, vt , St )dt +
and the pricing PDE

f
1
f
2f
(t, v, x) + rx (t, v, x) + vx2 2 (t, v, x)
(7.11)
t
x
2
x
2f
f
1 2
2f
+ (t, v) (t, v, x) + (t, v) 2 (t, v, x) + (t, v)x v
(t, v, x)
v
2
v
vx
= rf (t, v, x),
under the terminal condition f (T, v, x) = h(x).
Hedging
Consider a portfolio of the form
Vt = t ert + t St
based on the riskless asset At = ert and on the risky asset St . When this
portfolio is self-financing we have
dVt = df (t, vt , St )
= rt ert dt + t dSt
(1)
= rt ert dt + t (rSt dt + St vt dBt )
(1)
= rVt dt + t St vt dBt
When this condition is not satisfied we need to introduce a drift that yields a market
price of volatility.
174
"
(1)
= rf (t, vt , St )dt + t St vt dBt .
(7.12)
However, trying to match (7.12) to (7.10) yields
vt St
f
f
(1)
(1)
(2)
(t, vt , St )dBt + (t, vt ) (t, vt , St )dBt = t St vt dBt , (7.13)
x
v
which admits no solution unless (t, v) = 0, i.e. when volatility is deterministic. A solution to that problem is to consider instead a portfolio
Vt = f (t, vt , St ) = t ert + t St + t P (t, vt , St )
that includes an additional asset with price P (t, vt , St ), which can be an
option depending on the volatility vt . In that case, (7.12) is replaced with
dVt = df (t, vt , St )
= rt ert dt + t dSt + t dP (t, vt , St )
P
P
(1)
= rt ert dt + t (rSt dt + St vt dBt ) + rt St
(t, vt , St )dt + t (t, vt )
(t, vt , St )dt
x
v
2
2
P
1
P
1
P
+t
(t, vt , St )dt + t St2 vt 2 (t, vt , St )dt + t 2 (t, vt ) 2 (t, vt , St )dt
t
2
x
2
v
2P
P
(1)
+t (t, vt )St vt
(t, vt , St )dt + t St vt
(t, vt , St )dBt
xv
x
P
(2)
+t (t, vt )
(t, vt , St )dBt ,
v
P
(1)
= r(Vt t P (t, vt , St ))dt + t St vt dBt + rt St
(t, vt , St )dt
x
P
+t (t, vt )
(t, vt , St )dt
v
1
2P
1
2P
P
+t
t
2
x
2
v
P
2P
(1)
(t, vt , St )dBt
(7.14)
+t (t, vt )St vt
xv
x
P
(2)
+t (t, vt )
(t, vt , St )dBt ,
v
P
P
(1)
= rf (t, vt , St )dt + t St vt dBt + rt St
(t, vt , St )dt + t (t, vt )
(t, vt , St )dt
x
v
2
2
1
P
P
P
1
+t
t
2
x
2
v
P
2P
(1)
+t (t, vt )St vt
(t, vt , St )dBt
xv
x
P
(2)
+t (t, vt )
(t, vt , St )dBt ,
v
and by matching (7.14) to (7.10), the equation (7.13) now becomes
"
175
N. Privault
vt St
f
f
(1)
(2)
(t, vt , St )dBt + (t, vt ) (t, vt , St )dBt
x
v
P
P
(1)
(1)
(2)
= t St vt dBt + t St vt
(t, vt , St )dBt + t (t, vt )
(t, vt , St )dBt .
x
v
This leads to the equations
vt St f (t, vt , St ) = t St vt + t St vt P (t, vt , St ),
x
x
f
P
(t, vt ) (t, vt , St ) = t (t, vt )

(t, vt , St ),
v
v
hence
f
(t, vt , St )
,
t = v
P
(t, vt , St )
v
(7.15)
and

P
f
(t, vt , St ) t St vt
(t, vt , St ),
St vt
x
x
f
P
=
(t, vt , St ) t
(t, vt , St )
x
x
P
(t, vt , St )
f
f
=
(t, vt , St )
(t, vt , St ) v
.
P
x
x
(t, vt , St )
v
t =
vt St
(7.16)
In addition, indentifying the dt terms when equating (7.14) to (7.10) would

now lead to the more complicated PDE
P
P
(t, vt , St ) + t (t, vt )
(t, vt , St )
x
v
2
2
1
P
P
P
1
(t, vt , St ) + t St2 vt 2 (t, vt , St ) + t 2 (t, vt ) 2 (t, vt , St )
+t
t
2
x
2
v
2P
(t, vt , St )
+t (t, vt )St vt
xv
f
f
1
2f
=
(t, vt , St ) + rSt (t, vt , St ) + vt St2 2 (t, vt , St )
t
x
2
x
2f
f
1 2
2f
+(t, vt ) (t, vt , St ) + (t, vt ) 2 (t, vt , St ) + (t, vt )St vt
(t, vt , St ),
v
2
v
vx
r(f (t, vt , St ) t P (t, vt , St )) + rt St
which can be rewritten using (7.15) as

f
P
P
P
(t, v, x) rP (t, v, x) + rx
(t, v, x) + (t, v)
(t, v, x) +
(t, v, x)
v
x
v
t
176
"

2P
f
1
1 2 2P
2P
(t, v, x)
x v 2 (t, v, x) + 2 (t, v) 2 (t, v, x) + (t, v)x v
(t, v, x)
v
2
x
2
v
xv

P
f
f
1 2 2f
(t,
v,
x)
=
(t, v, x) rf (t, v, x) +
(t, v, x) + rx (t, v, x) + vx
v
t
x
2
x2

2f
P
f
1 2
2f
+
(t, v, x) (t, v) (t, v, x) + (t, v) 2 (t, v, x) + (t, v)x v
(t, v, x) ,
v
v
2
v
vx
+
or

f
1
2f
f
(t, v, x) + rx (t, v, x) + vx2 2 (t, v, x)
t
x
2
x

2f
1
2f
1 2
+ f
(t, v) 2 (t, v, x) + (t, v)x v
(t, v, x)
2
v
vx
v (t, v, x)

1
P
P
= P
(t, v, x) +
(t, v, x)
rP (t, v, x) + rx
x
t
v (t, v, x)

2P
1
1 2
2P
1 2 2P
+ P
x v 2 (t, v, x) + (t, v) 2 (t, v, x) + (t, v)x v
(t, v, x)
2
x
2
v
xv
v (t, v, x)
= (t, v, x),
1
f
v (t, v, x)
rf (t, v, x) +
where (t, v, x) is a function that does not depend of P , without requir(2)

ing (Bt )tR+ to be a standard Brownian motion under P . The function
(t, v, x) is linked to the market price of volatility risk, cf. Chapter 1 of [38]
and 2.4.1 of [36] for details.
The pricing PDE rewrites as
f
f
1
2f
(t, v, x) + rx (t, v, x) + vx2 2 (t, v, x)
t
x
2
x
2f
1
2f
f
+ 2 (t, v) 2 (t, v, x) + (t, v)x v
(t, v, x) = rf (t, v, x) + (t, v, x) (t, v, x),
2
v
vx
v
and (7.11) corresponds to the choice (t, v, x) = (t, v), i.e. a vanishing
market price of volatility risk.
Heston model
In the Heston model with (t, v) = (vt v) and (t, v) = v, we find

the PDE
f
f
1
2f
(t, v, x) + rx (t, v, x) + vx2 2 (t, v, x)
t
x
2
x
f
1 2 2f
2f
(v m) (t, v, x) + v 2 (t, v, x) + xv
(t, v, x) = rf (t, v, x).
v
2
v
vx
"
177
N. Privault
The solution of this PDE has been expressed in [50] as a complex integral by
inversion of a characteristic function.
Using the change of variable y = log x with g(t, v, y) = f (t, v, ey ) we find
g
1 g
g
1 2g
(t, v, y) + r (t, v, y) + v 2 (t, v, y) v (t, v, x)
t
y
2 y
2 y
g
2 2 g
2g
(v m) (t, v, y) + v
(t, v, y) + v
(t, v, y) = rg(t, v, y).
2
v
2 v
vy
Using the Fourier transform
g(t, v, z) :=
eiyz g(t, v, y)dy
iz
g (t, v, z) =
eiyz
and the relation
g
(t, v, y)dy,
y
we find, using the rule i2 = 1, that g(t, v, z) satisfies the equation

1
1
g
(t, v, z) irz
g (t, v, z) vz 2 g(t, v, z) + iz v
g (t, v, z)
t
2
2
2 2
g
g
g
(v m) (t, v, z) + v
(t, v, z) izv (t, v, z) = 0,
v
2 v 2
v
r
g (t, v, z) +
which is an affine PDE with respect to the variable v with z a constant

parameter. This equation can be solved in closed form, and the final solution
g(t, v, y) can then be obtained by the Fourier inversion
g(t, v, y) =
1 w izy
e g(t, v, z)dz,
2
cf. [50].
Perturbation Analysis
We refer to Chapter 4 of [36] for the contents of this section.
time-rescaled model
(1)
dSt = rSt dt + St vt/ dBt
Consider the
(7.17)
dv = (v )dt + (v )dB (2) .

t
t
t
t
We note that vt/ satisfies the SDE
178
"
dvt/ ' v(t+dt)/ vt/
= vt/+dt/ vt/
1
(2)
= (vt/ )dt + (vt/ )dBt/ ,
with
1
dt
(2)
' (dBt )2 '
(2)
(dBt/ )2 '
1
dBt(2)
2
,
hence the SDE can be rewritten as

dvt =
1
1
(2)
(vt )dt + (vt )dBt .
In other words, 0 corresponds to fast mean-reversion and (7.17) can be

rewritten as
p
(1)
dSt = rSt dt + vt St dBt

1
1
(2)
dvt = (vt )dt + (vt )dBt ,
> 0.
The perturbed PDE

f
f
1
2 f
(t, v, x) + rx
(t, v, x) + vx2
(t, v, x)
t
x
2
x2
2
2 f
f
1
f
1
(t, v, x) + 2 (v) 2 (t, v, x) + (v)x v
(t, v, x)
+ (v)
v
2
v
vx
= rf (t, v, x)
with terminal condition f (T, v, x) = (x K)+ , rewrites as
1
1
L0 f (t, v, x) + L1 f (t, v, x) + L2 f (t, v, x) = rf (t, v, x),
(7.18)
where
2 f
1
f
(t, v, x),
L0 f (t, v, x) = 2 (v) 2 (t, v, x) + (v)
2
v
v
2 f
L1 f (t, v, x) = x(v) v
(t, v, x),
vx
L2 f (t, v, x) = f (t, v, x) + rx f (t, v, x) + 1 vx2 f (t, v, x).

t
x
2
x2
Note that
"
179
N. Privault

L0 is the infinitesimal generator of the process vs1 sR+ , see (7.22) below,
and
L2 is the Black-Scholes operator, i.e. L2 f = rf is the Black-Scholes PDE.
The solution f (t, v, x) will be expanded as
f (t, v, x) = f (0) (t, v, x) + f (1) (t, v, x) + f (2) (t, v, x) +
(7.19)
with f (T, v, x) = (x K)+ , f (1) (T, v, x) = 0, and f (2) (T, v, x) = 0.

Since L0 contains only differentials with respect to v we will choose
f (0) (t, v, x) of the form f (0) (t, v, x) = f (0) (t, x), cf. 4.2 of 4.1.1 of [36]
for details, with
L0 f (0) (t, x) = L1 f (0) (t, x) = 0.
(7.20)
By identifying the terms of order 1/ when plugging (7.19) in (7.18) we

also find
L0 f (1) (t, v, x) + L1 f (0) (t, x) = 0,
hence L0 f (1) (t, v, x) = 0. Similarly, by identifying the terms that do not
depend on in (7.18) and taking f (1) (t, v, x) = f (1) (t, x), we have L1 f (1) = 0
and
L0 f (2) (t, v, x) + L2 f (0) (t, x) = 0.
(7.21)
Using the Ito formula we have
w

h
s f (2)
i

IE f (2) t, vs1 , x = f (2) t, v01 , x +
t, v1 , x dB(2)
0
x
w

s
(2)
1 2 1 2 f (2)

1 f
1
+ IE
v
t, v , x + v
t, v1 , x d
2
0
v
2
v
h
i w s h
i
= IE f (2) t, v01 , x +
IE L0 f (2) t, v1 , x d.
(7.22)
0
vt1 tR+
When the process

is started under its stationary probability distribution with density function (v) we have
w

f (2) (t, v, x)(v)dv,
R+ ,
IE[f (2) t, v1 , x ] =
0
hence (7.22) rewrites as

w
w
wsw
f (2) (t, v, x)(v)dv =
f (2) (t, v, x)(v)dv+
L0 f (2) (t, v, x)(v)dvd.
0
By differentiation with respect to s > 0 this yields

w
L0 f (2) (t, v, x)(v)dv = 0,
0
hence by (7.21) we find

180
"
w
0
L2 f (0) (t, x)(v)dv = 0,
cf. 3.2 of [36], i.e.

f (0)
f (0)
1 w
2 f (0)
(t, x) = rf (0) (t, x),
(t, x) + rx
(t, x) + 2
v(v)dv
0
t
x
2
x2
with the terminal condition f (0) (T, x) = (x K)+ .
Consequently the first expansion term f (0) (t, x) in (7.18) is the BlackScholes function

w
f (0) (t, x) = BS St , r, K, T t,
v(v)dv ,
0
with the averaged squared volatility

w

v(v)dv = IE v1 ,
0
R+ ,
under the stationary distribution of the process with infinitesimal generator

L0 , i.e. the stationary distribution of the solution to
(2)
dvt1 = (vt1 )dt + (vt1 )dBt .
Heston model
We have
p
(1)
dS = rSt dt + St vt dBt
t
r
dvt = (vt m)dt + vt dBt(2) ,
under the modified short mean-reversion time scale, and the SDE can be
rewritten as
r
vt (2)
dvt = (vt m)dt +

dBt .
In other words, 0 corresponds to fast mean-reversion.

The CIR process vt1 tR+ has a gamma stationary distribution with shape
parameter 2m/ 2 , scale parameter 2 /(2), probability density function
(v) =
2
2
1
v 1+2m/ e2x/ 1[0,) (v),
(2m/ 2 )( 2 /(2))2m/2
and mean
"
181
N. Privault
w
v(v)dv = m.
Hence the first expansion term f
(0)
(t, x) in (7.18) reads
f (0) (t, x) = BS (St , r, K, T t, m) ,

with the averaged squared volatility
w

v(v)dv = m = IE v1 ,
R+ ,
under the stationary distribution of the process with infinitesimal generator

L0 , i.e. the stationary distribution of the solution to
(2)
dvt1 = (vt1 )dt + (vt1 )dBt .
7.6 Volatility Derivatives

Another look at historical volatility
When tk = kT /N , k = 0, 1, . . . , N , a natural estimator for the trend parameter can be written as
N :=
N
Stk Stk1
1 X
1
N
tk tk1
Stk1
k=1
'
N
1
1 X
Stk
log
N
tk tk1
Stk1
k=1
N

1 X
=
log(Stk ) log(Stk1 )
T
k=1
1
ST
=
log
.
T
S0
Similarly we can use
N
N
:=
1 X
1
N 1
tk tk1
k=1
Stk Stk1
Stk1
2
(
N )2
N
2
1 X
1
log(Stk ) log(Stk1 )
N
tk tk1
k=1
2

2
N
Stk
ST
1 X
1
=
log
2 log
.
T
Stk1
T
S0
'
N
Stk Stk1
1 X
1
N
tk tk1
Stk1
k=1
(7.23)
k=1
182
!2
"
Volatility swaps are forward contracts that allow for the exchange of the
estimated volatility (7.23) against a fixed value , with the payoff

2
2
N
ST
1 X
Stk
1
log
2 log
.
T
Stk1
T
S0
k=1
Note that the above payoff has to be multiplied by the vega notional, which
is part of the contract, in order to convert it into currency units.
Exercise. ([38], Ch. 11) Compute the expected total realized variance in the
Heston model with
dvt = (vt m)dt + vt dBt .

Answer; We need to compute
w

T
1
1 wT
1 wT
IE
vt dt =
IE[vt ]dt =
u(t)dt,
0
T
T 0
T 0
where u(t) := IE[vt ] satisfies the ordinary differential equation
u0 (t) = m u(t),
i.e.
(et u(t))0 = et u(t) + et u0 (t) = met ,
hence

wt
u(t) = et u(0) + m es ds = IE[v0 ]et + m(1 et ),
0
t R+ .
Exercises
Exercise 7.1 Consider an index whose level St is given in the Heston stochastic
volatility model
p
(1)
dSt = (r vt )St dt + St + vt dBt
dv = (v m)dt + v dB (2) ,
t
t
t
t
(1)
(2)
where (Bt )tR+ and (Bt )tR+ are standard Brownian motions with correlation [1, 1] and 0, 0, > 0, m > 0, r > 0, > 0. Compute
the variance swap rate
"
183
N. Privault
VST :=
1
IE
T
"
lim

n
X
SkT /n S(k1)T /n 2
k=1
S(k1)T /n
184
w

T 1
1
2
IE
(dS
)
.
t
0 St2
T
"
Chapter 8
Exotic Options
In this chapter we work in a continuous geometric Brownian model in which

the asset price (St )t[0,T ] has the dynamics
dSt = rSt dt + St dBt ,
t R+ ,
where (Bt )tR+ is a standard Brownian motion under the risk-neutral probability measure P . In particular the value Vt of a self-financing portfolio
satisfies
wT
VT erT = V0 +
t St ert dBt , t [0, T ].
0
8.1 Generalities
An exotic option is an option whose payoff may depend on the whole path
{St : t [0, T ]} of the price process via a complex operation such as
averaging or computing a maximum. They are opposed to vanilla options
whose payoff
C = (ST ),
where is called a payoff function, depends only on the terminal value ST of
the price process.
An option with payoff C = (ST ) can be priced as
w
erT IE[(ST )] = erT
(y)fST (y)dy
0
where fST (y) is the (one parameter) probability density function of ST , which
satisfies
wy
P(ST y) =
fST (v)dv,
y R.
0
"
N. Privault
Recall that typically we have
+
(x) = (x K) =
x K
if x K,
if x < K,
for a European call option with strike K, and
$1 if x K,
(x) = 1[K,) (x) =
0 if x < K,
for a binary call option with strike K.
Exotic Options
Exotic options, also called path-dependent options, are options whose payoff
C may depend on the whole path
{St : 0 t T }
of the underlying price process instead of its terminal value ST . Next we
review some examples of exotic options.
Options on Extrema
We take
C := (M0T ),
where
M0T = max St
t[0,T ]
is the maximum of (St )tR+ over the time interval [0, T ].

Figure 5.9 represents the running maximum process (Mt )tR+ of Brownian
motion (Bt )tR+ .
186
"
Exotic Options
3
Xt
Bt
2.5
2
Bt , X t
1.5
1
0.5
0
-0.5
-1
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
2.0
Fig. 8.1: Brownian motion Bt and its supremum Xt .

Note that the running maximum of Brownian motion admits (almost surely)
no last point of increase before it switches to a flat behavior. Examples
of deterministic functions having a similar property can be built as
f (t) := (1 r)
rn1 1[1rn ,1) (t) + 1[1,) (t),
t R+ ,
n=1
where r (0, 1), which admits no last point of increase before t = 1, as

illustrated in Figure 8.2 with r = 3/4.
1.2
f(t)
0.8
0.6
0.4
0.2
0.2
0.4
0.6
0.8
1.2
Fig. 8.2: A function with no last point of increase before t = 1.
"
187
N. Privault
Barrier Options
The payoff of an up-and-out barrier put option on the underlying asset St
with exercise date T , strike K and barrier B is
C = (K ST ) 1(
max St < B
0tT
(K ST )+
if max St < B,
if max St B.
0tT
0tT
This option is also called a Callable Bear Contract with no residual value, in
which the call price B usually satisfies B K.
The payoff of a down-and-out barrier call option on the underlying asset
St with exercise date T , strike K and barrier B is
C = (ST K) 1(
min St > B
0tT
(S K)+
if min St > B,
if min St B.
0tT
0tT
This option is also called a Callable Bull Contract with no residual value, in
which B denotes the call price B K. It is also called a turbo warrant with
no rebate.
Lookback Options
The payoff of a floating strike lookback call option on the underlying asset
St with exercise date T is
C = ST min St .
0tT
The payoff of a floating strike lookback put option on the underlying asset
St with exercise date T is

C = max St ST .
0tT
Options on Average
In this case we can take

C=

1 wT
St dt
T 0
188
"
Exotic Options
where
1 wT
St dt
T 0
represents the average of (St )tR+ over the time interval [0, T ] and : R R
is a payoff function.
3
Xt
Bt
2.5
2
Bt , X t
1.5
1
0.5
0
-0.5
-1
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
2.0
Fig. 8.3: Brownian motion Bt and its moving average.

Figure 8.3 shows a graph of Brownian motion and its moving average process
Xt .
Asian Options
Asian options are particular cases of options on average, and they were first
traded in Tokyo in 1987. The payoff of the Asian call option on the underlying
asset St with exercise date T and strike K is given by

C=
+
1 wT
St dt K
.
T 0
Similarly, the payoff of the Asian put option on the underlying asset St with
exercise date T and strike K is

+
1 wT
C= K
St dt
.
T 0
Due to the fact that their dependence on averaged asset prices, Asian options are less volatile than plain vanilla options whose payoffs depend only
"
189
N. Privault
on the terminal value of the underlying asset. Asian options have become
particularly popular in commodities trading.
8.2 The Reflexion Principle

In order to price barrier options we will have to derive the probability density
of the maximum
M0T = max St
t[0,T ]
of geometric Brownian motion (St )tR+ over a given time interval [0, T ].
In such situations the option price at time t = 0 can be expressed as
w w
erT IE[(M0T , ST )] = erT
(x, y)f(MT ,ST ) (x, y)dxdy
where f(MT ,ST ) is the joint probability density function of (MT , ST ), which
satisfies
wx wy
P(M0T x, ST y) =
f(MT ,ST ) (u, v)dudv,
x, y R.
In order to price such options by the above probabilistic method, we will

compute f(MT ,ST ) (u, v) by the reflection principle.
Maximum of Standard Brownian Motion

Let (Bt )tR+ denote a standard Brownian motion started at B0 = 0. While
it is well-known that BT ' N (0, T ), computing the law of the maximum
XT = max Bt
t[0,T ]
might seem a difficult problem. However this is not the case, due to the
reflection principle. Note that since B0 = 0 we have
XT 0,
almost surely.
Given a > B0 = 0, let
a = inf{t R+ : Bt = a}
190
"
Exotic Options
denote the first time (Bt )tR+ hits the level a > 0. Due to the spatial symmetry of Brownian motion we note the identity
P(BT > a | a < T ) = P(BT < a | a < T ) =
1
.
2
In addition, due to the relation

{XT a} = {a < T },
(8.1)
we have
P(a < T ) = P(BT > a & a < T ) + P(BT < a & a < T )
= 2P(BT > a & a < T )
= 2P(BT > a & XT a)
= 2P(BT > a)
= P(BT > a) + P(BT < a)
= P(|BT | > a),
where we used the fact that
{BT > a} {BT > a & XT a} {BT > a}.
Figure 8.4 shows a graph of Brownian motion and its reflected path.
3
2.5
2
Bt
1.5
1
0.5
0
-0.5
-1
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
2.0
Fig. 8.4: Reflected Brownian motion with a = 1.

As a consequence of the equality
P(a < T ) = 2P(BT > a),
a R,
"
191
N. Privault
the maximum XT of Brownian motion has same distribution as the absolute
value |BT | of BT . In other words, XT is a non-negative random variable with
distribution function
P(XT a) = P(|BT | a)
1 w a x2 /(2T )
=
e
dx
2T a
w
a
2
2
ex /(2T ) dx,
=
2T 0
a R+ ,
and probability density
dP(XT a)
=
fXT (a) =
da
2 a2 /(2T )
e
1[0,) (a),
T
a R.
(8.2)
density
0.8
0.6
0.4
0.2
-4
-3
-2
-1
Fig. 8.5: Probability density of the maximum of Brownian motion over [0, 1].
Using the density of XT we can price an option with payoff (XT ), as

w
erT IE [(XT )] = erT
(x)dP(XT = x)
r
2
2 w
= erT
(x)e|x| /(2T ) dx.
T 0
We have
M0T = max St
t[0,T ]
= S0 max eBt
t[0,T ]
192
"
Exotic Options
= S0 e maxt[0,T ] Bt
= S0 eXT ,
since > 0, hence the probability density function of the maximum MT of
(St )t[0,T ] = (S0 eBt )t[0,T ] is given by
x 7
1
x

2
1
2
exp
(log (x/S0 )) ,
T
2T
x > 0,
cf. Figure 8.6.

1.4
1.2
density
1
0.8
0.6
0.4
0.2
0
0.5
1.5
2.5
Fig. 8.6: Density of the supremum of geometric Brownian motion.
When the payoff takes the form

C = (M0T ),
where
ST = S0 eBT ,
we have
C = (M0T ) = (S0 eXT ),
hence

erT IE [C] = erT IE (S0 eXT )
w
(S0 ex )dP(XT = x)
= erT
r
2
2 rT w
=
e
(S0 ex )ex /(2T ) dx.
0
T
This is however not sufficient since this imposes the condition r = 2 /2. In
order to do away with this condition we need to consider the maximum of
"
193
N. Privault
drifted Brownian motion, and for this we have to compute the joint density
of XT and BT .
Joint Density
The reflection principle also allows us to compute the joint density of Brownian motion BT and its maximum XT . Indeed, for b [0, a] we also have
P(BT > a + (a b) | a < T ) = P(BT 2a b & a < T ) = P(BT 2a b & XT a) = P(BT 2a b & XT a) = P(BT 2a b & XT 2a b}
{BT > 2a b & XT a} {BT > a},
which shows that {BT 2a b} = {BT > 2a b & XT a}. Consequently,
by (8.3) we find
P(BT < b & XT a) = P(BT 2a b) =
1 w x2 /(2T )
e
dx,
2T 2ab
0 b a, which yields the joint probability density

fXT ,BT (a, b) =
dP(XT a & BT b)
dP(XT a & BT b)
=
,
dadb
dadb
a, b R, by (16.16), i.e., letting a b := max(a, b),
194
"
Exotic Options
r
fXT ,BT (a, b) =
2 (2a b) (2ab)2 /(2T )

e
1{ab0}
T
T
2 (2a b) (2ab)2 /(2T )
e
,
T
T
=
0,
(8.4)
a > b 0,
a < b 0.
Density function
0.5
0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0-1
-0.5
b
0.5
1.5
2.5
2.5
1.5
0.5
a
-0.5 -1
Fig. 8.7: Joint probability density of B1 and its maximum over [0,1].
Figure 8.8 presents the corresponding heat map of the same graph seen from
above.
"
195
N. Privault
3
2.5
Density function
1.5
1
0.5
0
-0.5
-1
-1
-0.5
0.5
1.5
2.5
Fig. 8.8: Heat map of the joint probability density of B1 and its maximum over [0,1].
Maximum of Drifted Brownian Motion

Using the Girsanov theorem, it is even possible to compute the probability
density function of the maximum
T = max B
t = max (Bt + t)
X
t[0,T ]
t[0,T ]
t = Bt + t, R. The arguments previof the drifted Brownian motion B

t because drifted Brownian
ously applied to Bt cannot be directly applied to B
motion is no longer symmetric in space when 6= 0.
t is a standard Brownian motion under the probaOn the other hand, B
defined from
bility measure P
2
dP
= eBT T /2 ,
dP
(8.5)
is given by (8.4).
T under P
hence the density of X
Now, using the density (8.5) we get
h
T a & B
T b) = IE 1
P(X
{XT a &
h
i
BT +2 T /2
= IE e
1{X T a & BT b}
h
i
eBT 2 T /2 1
= IE
T b}
{XT a & B
i
T b}
B
196
"
Exotic Options
r
=
2
2 wawb
(2x y) (2xy)2 /(2T )
1(,x] (y)ey T /2
e
dxdy,
T 0
T
0 b a, which yields the joint probability density

fX T ,BT (a, b) =
T a & B
T b)
dP(X
,
dadb
i.e.
fX T ,BT (a, b) = 1{ab0}
1
T
2
2
2
(2a b)eb(2ab) /(2T ) T /2
T
2
2
1
2
(2a b)e T /2+b(2ab) /(2T ) ,

T
T
=
0,
(8.6)
a > b 0,
a < b 0.
By the completion of squares

y
(2a y)2
1
= 2a
(y (T + 2a))2
2T
2T
and standard changes of variables, we also find

r
2
(2x y) (2xy)2 /(2T )
2 waw
T a) =
1(,x] (y)ey T /2
e
dydx
P(X
T 0
T
r
2 2 T /2 w a y w a (2x y) (2xy)2 /(2T )
=
e
e
e
dxdy
y0
T
T

wa
2
2
2
1
e T /2
ey(2(y0)y) /(2T ) ey(2ay) /(2T ) dy
=
2T

2
2
2
1 w a yy2 /(2T )2 T /2
=
e
ey2a /T +2ay/T y /(2T ) T /2 dy
2T

2
1 w a (yT )2 /(2T )
=
e
e(y(T +2a)) /(2T )+2a dy
2T
1 w a (yT )2 /(2T )
1 w a (y(T +2a))2 /(2T )
=
e
dy e2a
e
dy
2T
2T
w
w
aT
aT
2
2
1
1
=
ey /(2T ) dy e2a
ey /(2T ) dy
2T
2T
"
197
N. Privault

=
a T
e2a
a T

,
(8.7)
cf. Corollary 7.2.2 and pages 297-299 of [109] for another derivation. This
yields the density
r

T a)
dP(X
2 (aT )2 /(2T )
a T
,
=
e
2e2a
da
T
T
of the supremum of drifted Brownian motion, and recovers (8.2) for = 0,
and will be used for the pricing of lookback options in Section 8.4.
=0
=-0.5
=0.5
1.4
1.2
density
0.8
0.6
0.4
0.2
0
-1
Fig. 8.9: Probability density of the maximum of drifted Brownian motion.
Note from Figure 8.2 that small values of the maximum are more likely to
occur when takes large negative values.
t = max (B
t ), the joint density f
Based on the relation min B
RT ,BT
t[0,T ]
t[0,T ]
of the minimum
T = min B
t = min (Bt + t)
R
t[0,T ]
t[0,T ]
t := Bt + t and its value B

T at time T
of the drifted Brownian motion B
can similarly be computed as follows, letting a b := min(a, b):
198
"
Exotic Options
fR T ,BT (a, b) = 1{ab0}
1
T
2
2
2
(b 2a)eb(2ab) /(2T ) T /2
T
2
2
2
1
(b 2a)e T /2+b(2ab) /(2T ) ,

T
T
=
0,
(8.8)
a b 0.
8.3 Barrier Options

General Case
T and X
T we are able to price any exotic option
Using the joint density of B
T , X
T ), as
with payoff (B
#
"

T , B
T ) Ft ,
er(T t) IE (X

with in particular
h
i
w w
T , B
T ) = erT
T = x, B
T = y).
erT IE (X
(x, y)dP(X
y0
When the payoff takes the form

C = (M0T , ST ),
where
ST = S0 eBT
T /2+rT
= S0 eBT ,
T = BT + T , and
with = /2 + r/ and B
M0T = max St
t[0,T ]
= S0 max eBt
t/2+rt
t[0,T ]
= S0 max eBt
t[0,T ]
= S0 e maxt[0,T ] Bt
= S0 e
"
T
X
,
199

N. Privault
we have
C = (ST , M0T )
= (S0 eBT
= (S0 e
T
B
T /2+rT
, S0 e
T
X
, M0T )
),
hence
h
i
erT IE[C] = erT IE (S0 eBT , S0 eXT )

w w
T = x, B
T = y)
= erT
(S0 ey , S0 ex )dP(X
y0
r
2
2
1
2 rT w w
=
e
(S0 ey , S0 ex )(2x y)e T /2+y(2xy) /(2T ) dxdy
y0
T T
r
2
2
1
2 ww
= erT
(S0 ey , S0 ex )(2x y)e T /2+y(2xy) /(2T ) dxdy
T
T 0 y
r
2
2
1
2 w0 w
(S0 ey , S0 ex )(2x y)e T /2+y(2xy) /(2T ) dxdy.
+ erT
T
T 0
We can distinguish 8 different versions of barrier options according to the
following table.
200
"
Exotic Options
option type
behavior
down-and-out
payoff
+
(ST K) 1(
min St > B
0tT
down-and-in
(ST K) 1(
min St B
0tT
down-and-out
(K ST ) 1(
min St > B
0tT
down-and-in
(K ST ) 1(
min St B
0tT
We have the following obvious relations between the prices of barrier and
vanilla call and put options:
Cupin (t) + Cupout (t) = C(t) = er(T t) IE [(ST K)+ ],
Cdownin (t) + Cdownout (t) = C(t) = er(T t) IE [(ST K)+ ],
"
201
N. Privault
Pupin (t) + Pupout (t) = P (t) = er(T t) IE [(K ST )+ ],
Pdownin (t) + Pdownout (t) = P (t) = er(T t) IE [(K ST )+ ],
where C(t), resp. P (t) denotes the price of a European call, resp. put option
with strike K as obtained from the Black-Scholes formula. Consequently, in
the sequel we will only compute the prices of the up-and-out call and put,
and down-and-out barrier call and put options.
Up-and-Out Barrier Call Option

Let us consider an up-and-out call option with maturity T , strike K, barrier
(or call price) B, and payoff
C = (ST K) 1(
max St B
0tT
S K
if max St B,
if max St > B,
0tT
0tT
with B > K. Our goal is to prove the following result.

Proposition 8.1. When K < B, the price

+
ST t
r(T t) n
e
1 M t B o IE x
K
1(
S0
0
x
max
0rT t
Sr /S0 B
x=St
of the up-and-out call option with maturity T , strike K and barrier B is given
by
er(T t) IE [C | Ft ]
(8.9)

S
S
t
t
T t
T t
= St 1nM t B o +
+
K
B
0
2

!
1+2r/2
B
B
B
T t
T t
+
+
St
KSt
St

St
St
T t
T t
er(T t) K1nM t B o

K
B
0
12r/2
2

!
St
B
B
T t
T t

,
B
KSt
St
where
202
"
Exotic Options
(s) =

1
log s + r 2 ,
2

1
s > 0.
(8.10)
Note that taking B = + in the above identity (8.9) recovers the BlackScholes formula for the price of a European call option, and that the price of
the up-and-out barrier call option is 0 when B < K.
The graph of Figure 8.10 represents the up-and-out call option price given
the value St of the underlying and the time t [0, T ] with T = 220 days.
18
16
14
12
10
8
6
4
2
0 50
55
60
65
70
underlying
75
80
85
220
200
180
160
Time in days
140
120
90 100
Fig. 8.10: Graph of the up-and-out call option price with B = 80 > K = 65.
Proof of Proposition 8.1. We have C = (ST , M0T ) with
(x K)+ if y B,
+
(x, y) = (x K) 1{yB} =
0
if y > B,
hence
h
i
+
er(T t) IE [C | Ft ] = er(T t) IE (ST K) 1{M0T B} Ft
+ n
r(T t)
(
)
=e
IE (ST K) 1 M t B o 1
Ft
0
max Sr B
trT

+ (
r(T t) n
)
o
=e
1 M t B IE (ST K) 1
Ft
0
max Sr B
trT

+
ST
)
= er(T t) 1nM t B o IE x
K
1(
St
0
x max Sr /St B
trT
x=St
Right click on the figure for interaction and full screen view (works in Acrobat
reader on the entire pdf file).
"
203
N. Privault
er(T t) 1n

+
S
o IE x T t K
1(
t
M0 B
S0
x
max
0rT t
Sr /S0 B
x=St
It suffices to compute
h
i
+
erT IE [C] = erT IE (ST K) 1{M0T B}

+
= erT IE S0 eBT K 1{S0 eX T B}

w w
+
T = x, B
T = y)
= erT
(S0 ey K) 1{S0 ex B} dP(X
y0
r
1
w
2 rT log(B/S0 )
1
=
e
T T
w
2
2
(S0 ey K) 1{S0 ex B} (2x y)e T /2+y(2xy) /(2T ) dxdy
y0
r
2 w 1 log(B/S0 )
1
= erT
T
T 0
w
2
2
y
r
1
2 w0
+ erT
T
T
w
2
2
0
r
1
2 w 1 log(B/S0 )
= erT
T
T 0
w
2
2
y
(S0 e K) 1{x1 log(B/S0 )} (2x y)e T /2+y(2xy) /(2T ) dxdy
y
r
2 w0
1
+ erT
T
T
w
2
2
y
(S0 e K) 1{x1 log(B/S0 )} (2x y)e T /2+y(2xy) /(2T ) dxdy
0
r
1
2 w 1 log(B/S0 )
= erT
T
T 0
w 1 log(B/S0 )
2
2
(S0 ey K) (2x y)e T /2+y(2xy) /(2T ) dxdy
y
r
1
2 w0
+ erT
T
T
w 1 log(B/S0 )
2
2
0
r
1
2 w 1 log(B/S0 )
= erT
T
T 1 log(K/S0 )
204
"
Exotic Options
w 1 log(B/S0 )

r
2
2
1
2 w 1 log(B/S0 )
= erT T /2
(S0 ey K) eyy /(2T )
T
T 1 log(K/S0 )
w 1 log(B/S0 )
(2x y)e2x(yx)/T dxdy,
y0
y0
if B S0 (otherwise the option price is 0), with = r/ /2 and

y 0 = max(y, 0).
Letting a = y 0 and b = 1 log(B/S0 ), we have
wb
a
(2x y)e2x(yx)/T dx =
wb
(2x y)e2x(yx)/T dx
T h 2x(yx)/T ix=b
=
e
2
x=a
T
= (e2a(ya)/T e2b(yb)/T )
2
T 2(y0)(yy0)/T
= (e
e2b(yb)/T )
2
T
= (1 e2b(yb)/T ),
2
a
hence, letting c = 1 log(K/S0 ), we have

2
1 wb
(S0 ey K) eyy /(2T ) (1 e2b(yb)/T )dy
2T c
2
1 w b y(+)y2 /(2T )
e
(1 e2b(yb)/T )dy
= S0 eT (r+ /2)
2T c
2
1 w b yy2 /(2T )
KeT (r+ /2)
e
(1 e2b(yb)/T )dy
2T c
2
1 w b y(+)y2 /(2T )
= S0 eT (r+ /2)
e
dy
2T c
w
b
2
2
2
1
S0 eT (r+ /2)2b /T
ey(++2b/T )y /(2T ) dy
2T c
2
1 w b yy2 /(2T )
KeT (r+ /2)
e
dy
2T c
2
2
1 w b y(+2b/T )y2 /(2T )
+KeT (r+ /2)2b /T
e
dy.
2T c
2
erT IE [C] = eT (r+
/2)
Using the relation
"

2
1 w b yy2 /(2T )
c + T
b + T
e
dy = e T /2
,
c
2T
T
T
205
N. Privault
we find
h
i
+

2
2
c + ( + )T
b + ( + )T
= S0 eT (r+ /2)+(+) T /2
T
T
2
S0 eT (r+ /2)2b /T +(++2b/T ) T /2

c + ( + + 2b/T )T
b + ( + + 2b/T )T
T
T

c + T
b + T
rT
Ke
T
T
2
+KeT (r+ /2)2b /T +(+2b/T ) T /2

b + ( + 2b/T )T
c + ( + 2b/T )T

T
T

S0
S0
T
T
= S0 +
+
K
B
2

B
B
T (r+2 /2)2b2 /T +(++2b/T )2 T /2
T
T
S0 e
+
+
KS0
S0

S0
S0
T
T
KerT

K
B
2

B
B
T (r+2 /2)2b2 T +(+2b/T )2 T /2
+Ke

,
KS0
S0
T
0 x B, where
(s) is defined in (8.10). Given the relations

2

r
2
T
2b

2r
r+
2b2 /T +
++
= 2b
+
= 1 + 2 log(B/S0 ),
2
2
T
and
T

2

2
b2 T
2b
2r
r+
2 +
+
= rT +2b = rT + 1 + 2 log(B/S0 ),
2
T 2
T
this yields
h
i
+

S0
S0
T
T
+
= S0 +
K
B

S0
S0
rT
T
T
e
K

K
B
206
(8.11)
"
Exotic Options
2r/2 2

B
B
T
T
+
+
KS0
S0
12r/2 2

S0
B
B
T
T
+erT K

B
KS0
S0

S0
S0
T
T
= S0 +
+
K
B

1+2r/2 2
B
B
B
T
T
S0
+
+
S0
KS0
S0

S0
S0
T
T
erT K

K
B
12r/2 2

S
B
B
0
T
T
erT K

,

B
KS0
S0

B
S0
and this yields the result of Proposition 8.1, cf. 7.3.3 pages 304-307 of [109]
for a different calculation. This concludes the proof of Proposition 8.1.
Up-and-Out Barrier Put Option

The price
er(T t) 1n
+

S
o IE K x T t
1(
t
M0 B
S0
x
max
0rT t
Sr /S0 B
x=St
of the up-and-out put option with maturity T , strike K and barrier B is

given by
er(T t) IE [P | Ft ]

St
T t
1
= St 1nM t B o +
K
0
2
1+2r/
2
!
B
B
T t
+
1
St
KSt

St
T t
er(T t) K1nM t B o
1
K
0
!
12r/2
2

St
B
T t

1
B
KSt

1+2r/2
2 !
St
B
B
T t
T t
= St 1nM t B o +
+
+
K
St
KSt
0
"
207
N. Privault
Ker(T t)
1nM t B o
0

12r/2
2 !
St
St
B
T t
T t

+
,

K
B
KSt
(8.12)
if B > K, and

St
T t
er(T t) IE [P | Ft ] = St 1nM t B o +
1
B
0
1+2r/2
!

B
B
T t
1
+
St
St

St
T t
er(T t) K1nM t B o
1
B
0
12r/2
!

St
B
T t

1
B
St

1+2r/2
!
B
St
B
T t
T t
n
o
+
= St 1 M t B
+
+
B
St
St
0
Ker(T t)
1nM t B o
0
!

12r/2
St
St
B
T t
T t
+

,

B
B
St
(8.13)
if B < K.
12
10
8
6
4
2
0
-2 50
55
60
65
70
underlying
75
80
85
220
200
180
160
time in days
140
120
90 100
Fig. 8.11: Graph of the up-and-out put option price (8.12) with B = 80 > K = 60.
208
"
Exotic Options
50
40
30
20
10
0 50
55
60
65
70
underlying
75
80
85
220
200
180
160
time in days
140
120
90 100
Fig. 8.12: Graph of the up-and-out put option price (8.13) with K = 100 > B = 80.
Down-and-Out Barrier Call Option

Let us now consider a down-and-out barrier call option on the underlying
asset St with exercise date T , strike K, barrier B, and payoff
C = (ST K)
1(
min St > B
0tT
S K
if min St > B,
if min St B,
0tT
0tT
with 0 B K. This option is also called a Callable Bull Contract with no

residual value, in which B denotes the call price, or a turbo warrant with no
rebate.
When B < K we have
er(T t) IE [C | Ft ]

St
St
T t
T t
= g(t, St ) = St +
er(T t) K
K
K
2
2r/2
B
B
T t
B
+
St
Kx
2
12r/2
St
B
T t
+er(T t) K

B
KSt
= BSc (St , r, T t, K)
2r/2
2
B
B
T t
B
+
St
KSt
"
(8.14)
209
N. Privault
12r/2

2
B
T t

KSt
12r/2
1 St
= BSc (St , r, T t, K)
BSc (B/St , r, T t, K/B),
B B
+er(T t) K
St
B
St > B, 0 t T , and
+
IE (ST K) 1(
min St > B
0tT

Ft = 1{mint[0,T ] St >B} g(t, St ),

t [0, T ]. When B > K we find

er(T t) IE [C | Ft ] = g(t, St )

St
St
T t
T t
= St +
er(T t) K
B
B
2r/2

B
B
T t
B
+
St
St
12r/2

St
B
T t
+er(T t) K

,
B
St
(8.15)
St > B, 0 t T , cf. Exercise 8.3 below.
16
14
12
10
8
6
4
100
2
090
120
140
85
80
160
75
underlying
70
180
65
60
time in days
200
55
50
220
Fig. 8.13: Graph of the down-and-out call option price (8.14) with B = 60 < K = 80.
210
"
Exotic Options
60
50
40
30
20
10
0 50
55
60
65
70
underlying
75
80
85
220
200
180
160
time in days
140
120
90 100
Fig. 8.14: Graph of the down-and-out call option price (8.15) with K = 40 B, the price

+
ST t
er(T t) 1nmt B o IE K x
1(
S0
0
x
min
0rT t
Sr /S0 B
x=St
of the down-and-out put option with maturity T , strike K and barrier B is

given by
er(T t) IE [P | Ft ]
(8.16)

S
S
t
t
T t
T t
+
= St 1nmt B o +
K
B
0
2

!
1+2r/2
B
B
B
T t
T t
+
+
St
KSt
St

St
St
T t
T t
er(T t) K1nmt B o

K
B
0

2
!
12r/2
St
B
B
T t
T t
+

,
B
KSt
St
while the corresponding price vanishes when K < B.
"
211
N. Privault
16
14
12
10
8
6
220
200
180
160
time in days
140
120
90 100
4
2
0 50
55
60
65
70
underlying
75
80
85
Fig. 8.15: Graph of the down-and-out put option price (8.16) with K = 80 > B = 65.
Note that although Figures 8.11 and 8.13, resp. 8.12 and 8.14, appear to share
some symmetry property, the functions themselves are not exactly symmetric.
Concerning Figures 8.10 and 8.15, the pricing function is actually the same,
but the conditions B < K and B > K play opposite roles.
PDE Method
Having computed the up-and-out call option price by probabilistic arguments,
we are now interested in deriving a PDE for this price.
The option price can be written as
#
"

+
er(T t) IE (ST K) 1{M0T B} Ft

er(T t) 1(
) IE (S K)+ 1(
T
max Sr B
0rt
max Sr B
trT

Ft

= g(t, St , Mt ),
where the function g(t, x) of t and St is given by
+
g(t, x, y) = 1{yB} er(T t) IE (ST K) 1(
max Sr B
trT

St = x .

(8.17)
Next, by the same argument as in the proof of Proposition 5.3 we derive the
Black-Scholes partial differential equation (PDE) satisfied by g(t, x, y), for
the price of a self-financing portfolio.
212
"
Exotic Options
t R+ .
Vt = g(t, St , Mt ),
Then the function g(t, x, y) satisfies the Black-Scholes PDE

rg(t, x, y) =
g
g
1
2g
(t, x, y) + rx (t, x, y) + x2 2 2 (t, x, y),
t
x
2
x
(8.18)
t > 0, x > 0, 0 < y < B, and t is given by

t =
g
(t, St , Mt ),
x
t [0, T ],
(8.19)
provided M0t < B.

Proof. By (8.17) the price at time t of the down-and-out call barrier option
discounted to time 0 is given by
+
) S = x
ert g(t, St , Mt ) = 1{M t B } erT IE (ST K) 1(
t
0
max Sr B
trT
+
= erT IE (ST K) 1{M t B } 1(
0
max Sr B
trT
+
= erT IE (ST K) 1(
max Sr B
0rT
St = x

St = x ,

which is a martingale. We conclude by applying the Ito formula to t 7

ert g(t, St , Mt ) on {M0t y, 0 t T } and noting that the sum of
components in factor of dt vanishes. Using the fact that dM0t dt = dM0t dBt =
0 because (Mt )tR+ is a finite variation process as it is non-decreasing, we
have
d(ert g(t, St , Mt )) = rert g(t, St , Mt )dt + ert dg(t, St , Mt )
g
g
= rert g(t, St , Mt )dt + ert (t, St , Mt )dt + rert St (t, St , Mt )dt
t
x
2g
1
+ ert 2 St2 2 (t, St , Mt )dt
2
x
g
g
+ert (t, St , Mt )dM0t + ert St (t, St , Mt )dBt
y
x
"
213
N. Privault
g
g
(t, St , Mt )dt + rert St (t, St , Mt )dt
t
x
1
g
2g
+ ert 2 St2 2 (t, St , Mt )dt + ert St (t, St , Mt )dBt ,
2
x
x
= rert g(t, St , Mt )dt + ert
hence
rg(t, St , Mt ) +
g
1
2g
g
(t, St , Mt ) + rSt (t, St , Mt ) + 2 St2 2 (t, St , Mt ) = 0.
t
x
2
x
In the sequel we will drop the variable y in g(t, x, y) and simply write g(t, x)
since
g
(t, x, y) = 0,
0 < y < B,
y
and the function g(t, x, y) is constant in y (0, B).
In the next proposition we add a boundary condition to the Black-Scholes
PDE (8.18) in order to hedge the up-and-out call option with maturity T ,
strike K, barrier (or call price) B, and payoff
C = (ST K) 1(
max St B
0tT
S K
if max St B,
if max St > B,
0tT
0tT
with B > K.
Proposition 8.3. The price of any self-financing portfolio of the form Vt =
1{M0t B} g(t, St ) hedging the up-and-out barrier call option satisfies the BlackScholes PDE
g
g
1
2g
rg(t, x) =
(t, x) + rx (t, x) + x2 2 2 (t, x),
(8.20a)
t
x
2
x
g(t, x) = 0,
x B, t [0, T ],
g(T, x) = (x K)+ 1{x<B} ,

on the time-space domain [0, T ] [0, B] with terminal condition
g(T, x) = (x K)+ 1{x<B}
and additional boundary condition
g(t, x) = 0,
x B.
214
(8.21)
"
Exotic Options
Condition (8.21) holds since the price of the claim at time t is 0 whenever
St = B, cf. e.g. [31].
The closed-form solution for the PDE (8.20a) is given by (8.11), as

x

x
T t
T t
g(t, x) = x +
+
(8.22)
K
B

2
x 12r/2
B
B
T t
T t
x
+
+
B
Kx
x

x

x
T t
T t
r(T t)
Ke

K
B
2

x 12r/2
B
B
T t
T t
r(T t)
+Ke

,
B
Kx
x
0 < x B, 0 t T .
We note that the expression (8.22) can be rewritten using the standard BlackScholes formula

S
S
T
T
BSc (S, K, r, , T ) = S +
KerT
K
K
for the price of a European call option, as
x
x

T t
T t
g(t, x) = BSc (x, K, r, , T t) x +
+ er(T t) K
B
B
2r/2
2

B
B
B
T t
T t
B
+
+
x
Kx
x

x 12r/2
B2
B
T t
T t
r(T t)
+e
K

,
B
Kx
x
0 < x B, 0 t T .
Figure 8.16 represents the value of Delta obtained from (8.19) for the
up-and-out call option, cf. Exercise 8.3-(a).
"
215
N. Privault
90
85
80
75
70
St
65
100
120
140
160
180
200
220
60
55
50
Fig. 8.16: Delta for the up-and-out option.
Checking the Boundary Conditions

For x = B we check that

B
T t
T t
g(t, B) = B +
+
(1)
K

B
T t
T t
er(T t) K

(1)
K

B
T t
T t
+
(1)
B +
K

B
T t
T t
+er(T t) K

(1)
K
= 0,
and the function g(t, x) is extended to x > B by letting
g(t, x) = 0,
x > B.
For x = K and t = T we find
216
"
Exotic Options
(s) = 1{s<1} + 1{s>1}
+ if s > 1,
if s = 1,
= 0
if s < 1,
hence when x < K < B we have

g(T, K) = x ( () ())
K ( () ())
2r/2
B
( (+) (+))
B
x
2r/2
B
+K
( (+) (+))
K
= 0,
when K < x B we obtain
g(T, K) = x ( (+) (+))
K ( (+) (+))
2r/2
B
( () ())
B
x
2r/2
B
( () ())
+K
K
= 0.
Down-and-Out Barrier Call Option

Similarly the price g(t, St ) at time t of the down-and-out barrier call option
satisfies the Black-Scholes PDE
"
217
N. Privault
g
1
2g
g
(t, x) + rx (t, x) + x2 2 2 (t, x),

rg(t, x) =
t
x
2
x
g(t, B) = 0, t [0, T ],
g(T, x) = (x K)+ 1{xB} ,

on the time-space domain [0, T ] [0, B] with terminal condition g(T, x) =
(x K)+ 1{xB} and the additional boundary condition g(t, B) = 0 since the
price of the claim at time t is 0 whenever St = B.
8.4 Lookback Options

Let
mts = inf Su
u[s,t]
and Mst = sup Su ,

u[s,t]
0 s t T , and let Mts be either mts or Mst . In the lookback option case
the payoff (ST , MT0 ) depends not only on the price of the underlying asset
at maturity but it also depends on all price values of the underlying asset
over the period which starts from the initial time and ends at maturity.
The payoff of such of an option is of the form (ST , MT0 ) with (x, y) =
x y in the case of lookback call options, and (x, y) = y x in the case of
lookback put options. We let

er(T t) IE (ST , MT0 )|Ft
denote the price at time t [0, T ] of such an option.
Maximum selling price over [0, T ]

The standard lookback put option gives its holder the right to sell the underlying asset at its historically highest price. In this case the strike is M0T
and the payoff is
C = M0T ST .
In the next proposition we start by computing the average of the maximum
selling price M0T over the time interval [0, T ]. As in (8.10), we denote

1
1
(s) =
log s + r 2 ,
s > 0.
2

218
"
Exotic Options
Proposition 8.4. The average maximum value of (St )t[0,T ] over [0, T ] is
given by

IE M0T | Ft
(8.23)

2
St
St
T t
T t
r(T t)
t
+ St e
1+
+
= M0
M0t
2r
M0t
2

t 2r/
2
t
M0
M0
T t
St

.
2r St
St
When t = 0 we have S0 = M00 , and given that
T
(1) =
r 2 /2
T,
(8.24)
the formula (8.23) simplifies to

IE M0T
2

2

/2 r
2
/2 + r
2
= S0 1
T + S0 erT 1 +
T ,
2r
2r

with IE M0T = 2S0 when r = 0.
In general, when T tends to infinity we find that
2

1 +
if r > 0,
IE M0T | Ft
2r
lim
=
T IE[ST | Ft ]
2
if r = 0,
see Exercise 8.1 in the case r = 2 /2.
Proof of Proposition 8.4. We have

IE M0T | Ft = IE M0t MtT | Ft

= IE M0t 1{M0t >MtT } | Ft + IE MtT 1{MtT >M0t } | Ft

= M0t IE 1{M0t >MtT } | Ft + IE MtT 1{MtT >M0t } | Ft

= M0t P M0t > MtT | Ft + IE MtT 1{MtT >M0t } | Ft .
Next, we have
P
"
M0t
>
MtT
| Ft = P
MT
M0t
> t
St
St
!

Ft

219

N. Privault
=P x>
=P
MtT
St
!

Ft

x=M0t /St
!
M0T t
<x
S0
.
x=M0t /St
On the other hand, letting := r/ /2, from (8.7) we have

T
M0
T < 1 log x)
P
< x = P(X
S0

1
T + 1 log x
T 1 log x
=
e2 log x
T
T

2
T
T
=
(1/x) x1+2r/
(x) .
Hence

P M0t > MtT = P
!
M0T t
<x
S0
x=M0t /St

t 1+2r/2
t
M0
St
M0
T
T
=

.
t
M0
St
St
Next, we have
"

MtT
t
IE MtT 1{MtT >M0t } | Ft = St IE
1 T
St {Mt /St >M0 /St }
#
"

Sr

1{maxr[t,T ] Sr /St >x} Ft
= St IE max

r[t,T ] St
x=M0t /St

Sr
= St IE
max
1{maxr[0,T t] Sr /S0 >x}
,
r[0,T t] S0
x=M t /St
#

Ft

and

Sr
IE max
1{maxr[0,T ] Sr /S0 >x}
r[0,] S0

= IE max eBr 1{maxr[0,T ] eBr >x}

r[0,T ]
h
i
= IE e maxr[0,T ] Br 1{maxr[0,T ] Br >1 log x}

h
i
= IE eXT 1{X T >1 log x}

w
= 1
ex fX T (z)dz
log x
220
"
Exotic Options
1 log x
r
=
ez
2 (zT )2 /(2T )
e
2e2z
T
z T
!
dz

w
2
2 w
z T
dz.
ez(zT ) /(2T ) dz 2 1
ez(+2)
1
log x
T log x
T
By standard arguments we have

2
1 w
ez(zT ) /(2T ) dz
2T 1 log x
2
2 2
1 w
=
e(z + T 2(+)T z)/(2T ) dz
2T 1 log x
w
2
2
1
e T /2+T 1
e(z(+)T ) /(2T ) dz
=
log x
2T
w
2
1
=
erT
ez /(2T ) dz
(+)T + 1 log x
2T

1
T
,
= erT +
x
since + 2 /2 = r. The second integral

w
z T
z(+2)
dz
1 log x
T
can be computed by integration by parts using the identity
w
w
v 0 (z)u(z)dz = u(+)v(+) u(a)v(a)
v(z)u0 (z)dz,
a
with a =
log x. We let

z T
u(z) =
T
and v 0 (z) = ez(+2)
which satisfy
u0 (z) =
and
w
a
"
2
1
e(z+T ) /(2T )
2T
z T
and v(z) =
1
ez(+2) ,
+ 2
v 0 (z)u(z)dz
w
= u(+)v(+) u(a)v(a)
v(z)u0 (z)dz
a

1
a T
=
ea(+2)
+ 2
T
ez(+2)
dz =
221
N. Privault
w
2
1
ez(+2) e(z+T ) /(2T ) dz

( + 2) 2T a

1
a T
=
ea(+2)
+ 2
T
w
2
1
(T (+)2 2 T )/2
+
e
e(zT (+)) /(2T ) dz
a
( + ) 2T

1
a T
ea(+2)
=
+ 2
T
w
1
(T (+)2 2 T )/2
z 2 /2
e
dz
+
e
(aT (+))/ T
( + 2) 2

a T
1
=
ea(+2)
+ 2
T

1
a + T ( + )
(T (+)2 2 T )/2
+
e
+ 2
T

2r
(r/ /2)T 1 log x
2r/ 2
= (x)
T

2r T (+2)/2
T (r/ + /2) 1 log x
+ e
T

2
rT
1
T
T
=
e +
x2r/
(x) ,
2r
x
2r
+
cf. pages 317-319 of [109] for a different derivation using double integrals.
Hence we have
#
"

Sr

T
max
1{maxr[0,T t] Sr /S0 >x}
IE Mt 1{MtT >M0t } Ft = St IE

r[0,T t] S0
x=M0t /St

St
St
T t
T t
= 2St er(T t) +
St er(T t) +
M0t
r
M0t

2r/2
t
M0t
M0
T t
+St

,
r
St
St
and consequently this yields, since /r = 1 2 /(2r),
IE [M0T | Ft ] = IE [M0T | M0t ]
= M0t P(M0t > MtT | M0t ) + IE [MtT 1{MtT >M0t } | M0t ]

t 2r/2

t
M0
St
M0
T t
T t
= M0t
t
M0t
St
St

S
t
T t
+2St er(T t) +
M0t
222
"
Exotic Options

2
St
T t
St 1
er(T t) +
t
2r
M0

t 2r/2
t
2
M0
M0
T t
+St 1

2r
St
St

2
St
St
T t
T t
r(T t)
t
+ St e
1+
+
= M0
t
t
M0
2r
M0

2r/2
t
2 M0t
M
0
T t
St
.

2r St
St
This concludes the proof of Proposition 8.4.
The Lookback Put Option

As a corollary of Proposition 8.4 we immediately obtain the following pricing
formula for lookback put options.
Proposition 8.5. The price at time t [0, T ] of the lookback put option with
payoff M0T ST is given by
er(T t) IE [M0T ST | Ft ]

St
2
St
T t
T t
= M0t er(T t)
+ St 1 +
+
t
t
M0
2r
M0
t 2r/2
t
2
M
M
0
0
T t

St er(T t)
St .
2r St
St
Figure 8.17 represents the lookback put price as a function of St and M0t , for
different values of the time to maturity T t.
"
223
N. Privault
Lookback put option price
T = 7.0
100
80
60
40
20
Mt
0
80
60
40
20
0
20
40
60
St
80
Fig. 8.17: Graph of the lookback put option price.

IE [M0T ST | Ft ] = IE [M0T | Ft ] IE [ST | Ft ]
= IE [M0T | Ft ] er(T t) St ,
hence Proposition 8.4 shows that
er(T t) IE [M0T ST | Ft ] = er(T t) IE [M0T | Ft ] er(T t) IE [ST | Ft ]
= er(T t) IE [M0T | M0t ] St

St
St
T t
T t
= M0t er(T t)
St +
t
t
M0
M0

t 2r/2
t
2
2
S
M
M0
t
0
T t
T t
St er(T t)

.
+St +
t
2r
M0
2r
St
St
2
PDE Method
Since the couple (St , Mt ) is a Markov process, the price can be written as a
function
f (t, St , M0t ) = erT IE [(ST , M0T ) | Ft ],
0 t T,
and in this case the function f (t, x, y) can solve a PDE.
224
"
Exotic Options
Black-Scholes PDE for Lookback Put Options
In the next proposition we derive the partial differential equation (PDE) for
the pricing function f (t, x, y) of a self-financing portfolio hedging a lookback
option.
(ii) the portfolio value Vt := t At + t St , t R+ , takes the form
Vt = f (t, St , M0t ),
t R+ ,
for some f C 2 ((0, ) (0, )2 ).

Then the function f (t, x, y) satisfies the Black-Scholes PDE
rf (t, x, y) =
f
f
1
2f
(t, x, y) + rx (t, x, y) + x2 2 2 (t, x, y),
t
x
2
x
t, x, y > 0,
(8.25)
under the boundary conditions
r(T t)
y,
0 t T, y R+ ,
f (t, 0, y) = e
f
(t, x, y)x=y = 0,
0 t T, y > 0,
f (T, x, y) = y x,
0 x y.
(8.26a)
(8.26b)
(8.26c)
The replicating portfolio of the lookback put option is given by

f
(t, St , M0t ),
t [0, T ].
(8.27)
x
Proof. The existence of f (t, x, y) follows from the Markov property, more
precisely, the function f (t, x, y) satisfies
t =
f (t, x, y) = er(T t) IE [(ST , M0T ) | St = x, M0t = y]

ST
= er(T t) IE x , y MtT
St

ST t
r(T t)
IE x
=e
, y M0T t , t [0, T ],
S0
from the time homogeneity of the asset price process (St )tR+ . Applying the
change of variable formula to the discounted portfolio value
f(t, x, y) = ert f (t, x, y) = erT IE [(ST , M0T ) | St = x, M0t = y]
"
225
N. Privault
which is a martingale for t [0, T ], we have
df(t, St , M0t ) = rert f (t, St , M0t )dt + ert df (t, St , M0t )
f
f
= rert f (t, St , M0t )dt + ert (t, St , M0t )dt + rert St (t, St , M0t )dt
t
x
1 rt 2 2 2 f
t
+ e St 2 (t, St , M0 )dt
2
x
f
rt f
+e
(t, St , M0t )dM0t + ert St (t, St , M0t )dBt .
y
x
T
Since IE [(ST , M0 ) | Ft ] t[0,T ] is a P-martingale and (M0t )t[0,T ] has finite
variation (it is in fact a nondecreasing process), we have:
df (t, St , M0t ) = St
f
(t, St , M0t )dBt ,
x
t [0, T ],
(8.28)
and the function f (t, x, y) satisfies the equation

f
f
(t, St , M0t )dt + rSt f (t, St , M0t )dt
t
x
1
f
2f
+ 2 St2 2 (t, St , M0t )dt +
(t, St , M0t )dM0t = rf (t, St , M0t )dt,
2
x
y
which implies
f
f
1
2f
(t, St , M0t ) + rSt (t, St , M0t ) + 2 St2 2 (t, St , M0t ) = rf (t, St , M0t ),
t
x
2
x
which is (8.25), and
f
(t, St , M0t )dM0t = 0,
y
because M0t increases only on a set of zero measure (which has no isolated
points). This implies
f
(t, St , St ) = 0,
y
which shows the boundary condition (8.26b), since M0t hits St when M0t
increases. On the other hand, (8.28) shows that
(ST , M0T ) = IE [(ST , M0T )] +
wT
0
St
f
(t, x, M0t )|x=St dBt ,
x
0 t T , which implies (8.27) as in the proof of Proposition 5.3.
In other words, the price of the lookback put option takes the form
226
"
Exotic Options
#
"

f (t, St , Mt ) = er(T t) IE M0T ST Ft ,

where the function f (t, x, y) is given by

2
T t
T t
f (t, x, y) = yer(T t)
(x/y) + x 1 +
(x/y)
+
2r

2
2r/ 2
T t
x er(T t) (y/x)

(y/x) x.
2r
(8.29)

The boundary condition (8.26a) is explained by the fact that
f (t, 0, y) = er(T t) IE [M0T ST | St = 0, M0t = y]
= er(T t) IE [M0t ST | St = 0, M0t = y]
= er(T t) IE [M0t | M0t = y] er(T t) IE [ST | St = 0]
= yer(T t) ,
since IE [ST | St = 0] = 0 as St = 0 implies ST = 0. On the other hand,
(8.26c) follows from the fact that
f (T, x, y) = IE [M0T ST | ST = x, M0T = y] = y x.
Note that we have
f (t, x, x) = xC(T t),
with

2 r

2
+
(1)
e

(1) 1,
C( ) = er
(1) + 1 +
2r
2r
> 0, hence
f
(t, x, x) = C(T t),
x
t [0, T ],
while we also have

f
(t, x, y)y=x = 0,
y
"
0 x y.
227
N. Privault
Scaling Property of Lookback Put Prices
From (8.29) and the following argument we note the scaling property

#
"

f (t, x, y) = er(T t) IE M0T ST St = x, M0t = y

"
#

= er(T t) IE M0t MtT ST St = x, M0t = y

"
#
t
T
Mt
ST
M0
r(T t)
t
=e
x IE
St = x, M0 = y
St
St
St

"
#
y MtT
ST
= er(T t) x IE
St = x, M0t = y
x
x
x

"
#

y

t
T
r(T t)
t
=e
x IE M0 Mt ST St = 1, M0 =

x

#
"

y

= er(T t) x IE M0T ST St = 1, M0t =

x
= xf (t, 1, y/x)
= xg(T t, x/y),
where we let
g(, z) :=
2

2 r 1 2r/
1 r
2
e

(z) + 1 +
+
(z)
e
(
(1/z)) 1,
z
2r
2r
z
with the boundary condition
(, 1) = 0,
g(0, z) = 1 1,
z
> 0,
z (0, 1].
(8.30a)
(8.30b)
The next Figure 8.18 shows a graph of the function g(, z).
228
"
Exotic Options
normalized lookback put price
1.2
1
0.8
0.6
0.4
0.2
0
0.6
0.7
z
0.8
0.9
0.2
0.4
0.6
0.8
Fig. 8.18: Graph of the normalized lookback put option price.
Black-Scholes Approximation of Lookback Put Prices

Letting

S
S
BSp (S, K, r, , ) = Ker

S +
K
K
denote the standard Black-Scholes formula for the price of a European put
option, we observe that the lookback put option price satisfies

er(T t) IE M0T ST | Ft = BSp (St , M0t , r, , T t)

t 2r/2
t !
2
St
M0
M0
T t
T t
r(T t)
+St
+
e

,
2r
M0t
St
St
i.e.

St
er(T t) IE M0T ST Ft = BSp (St , M0t , r, , T t) + St hp T t, t
M0
where the function
hp (, z) =

2
2
+
(z) er z 2r/
(1/z) ,
2r
(8.31)
depends only on time and z = St /M0t . In other words, due to the relation

x
x
BSp (x, y, r, , ) = yer

x +
y
y
= xBSp (1, y/x, r, , )
for the standard Black-Scholes put formula, we observe that f (t, x, y) satisfies
"
229
N. Privault
f (t, x, y) = xBSp (1, y/x, r, , T t) + xh(T t, x/y),
i.e.
f (t, x, y) = xg(T t, x/y),
with
g(, z) = BSp (1, 1/z, r, , ) + hp (, z),
(8.32)
where hp (, z) is the function given by (8.31), and (x, y) 7 xhp (T t, x/y)

also satisfies the Black-Scholes PDE (8.25), i.e. (, z) 7 BSp (1, 1/z, r, , )
and hp (, z) both satisfy the PDE
hp
hp
1
2 hp
(, z) = z r + 2
(, z) + 2 z 2
(, z),
z
2
z 2
(8.33)
R+ , z [0, 1], under the boundary condition

0 z 1.
hp (0, z) = 0,
The next Figures 8.19 and 8.20 show the decompositions (8.32) of the normalized lookback put option price g(, z) in Figure 8.18 into the Black-Scholes
put function BSp (1, 1/z, r, , ) and hp (, z).
normalized Black-Scholes put price BSp(1,1/z,r,,)
1.2
1
0.8
0.6
0.4
0.2
0
0.6
0.7
z
0.8
0.9
0.2
0.4
0.6
0.8
Fig. 8.19: Black-Scholes put price in the decomposition (8.32).
230
"
Exotic Options
h(,x)
1.2
1
0.8
0.6
0.4
0.2
0
0.6
0.7
z
0.8
0.9
0.2
0.4
0.6
0.8
Fig. 8.20: Function hp (, z) in the decomposition (8.32).
Note that in Figures 8.19-8.20 the condition hp (0, z) = 0 is not fully respected
as z 1 due to numerical error in the approximation of the function .
The Lookback Call Option

The following result gives the value of the average minimum IE [mT0 | Ft ] of
(St )t[0,T ] over the interval [0, T ].
Proposition 8.7. The average minimum value of (St )t[0,T ] over [0, T ] is
given by

2r/2
t

St
2 mt0
m0
T t
T t
IE mT0 | Ft = mt0
mt0
2r St
St

2
S
t
T t
+St er(T t) 1 +
+
.
2r
mt0
(8.34)
We note a certain symmetry between the expressions of (8.34) and (8.34).
When t = 0 we have S0 = m00 , and given (8.24) the formula (8.34) simplifies
to

2r/2

r 2 /2
2 mt0
r 2 /2
T S0
2r St

2
2
/2
+
r
+S0 erT 1 +

T ,
2r

IE mT0 = S0
"
231
N. Privault

with IE mT0 = 2S0 when r = 0.
In general, when T tends to infinity we find that

IE mT0 | Ft
lim
= 0.
T IE[ST | Ft ]

IE mT0 | Ft = IE mt0 mTt | Ft

= IE mt0 1{mt0 <mTt } | Ft + IE mTt 1{mt0 >mTt } | Ft

= mt0 IE 1{mt0 <mTt } | Ft + IE mTt 1{mt0 >mTt } | Ft

= mt0 P mt0 < mTt | Ft + IE mTt 1{mt0 >mTt } | Ft .
By computations similar to those of the lookback put option case we find
!

mTt
mt0
t
T
P m0 < mt | Ft = P
<
Ft
St
St
!
mT
= P x < t Ft
St
x=mt0 /St
!
T t
m0
=P
>x
S0
t
x=m0 /St
t 1+2r/2

t
m0
m0
St
T t
T t

,
=
t
m0
St
St
and

IE mTt 1{mt0 >mTt } | Ft = St IE
= St IE

min
r[t,T ]
Sr
1{mt0 /St >mTt /St }
r[t,T ] St

min
Sr
1{minr[t,T ] Sr /St <x}
St
x=mt0 ,y=St
x=mt0 /St

Sr
1{minr[0,T t] Sr /S0 <x}
r[0,T t] S0
x=mt0 /St

St
r(T t)
St
T t
T t
= 2St er(T t) +
S
e
t
+
mt0
r
mt0
t 2r/2
t
m0
m0
T t
+St

.
r
St
St
= St IE
min
Given the relation /r = 1 2 /(2r), this yields

232
"
Exotic Options

IE mT0 | Ft = mt0 P
!
m0T t
>x
S0
x=mt /St
0

Sr
+St IE
min
1{minr[0,T t] Sr /S0 <x}
r[0,T t] S0
x=mt0 /St

t 1+2r/2
t
St
m0
m0
T t
T t
= mt0
mt0

t
m0
St
St

St
St
T t
T t
r(T t)
r(T t)
+2St e
+
St e
+
t
t
m0
r
m0
t 2r/2
t
m0
m0
T t
+St

r
St
St

2
S
St
t
T t
T t
= mt0
+ St er(T t) 1 +
+
t
t
m0
2r
m0

2r/2
t
2 mt0
m
0
T t
St

.
2r St
St
2
The standard Lookback call option gives the right to buy the underlying asset
at its historically lowest price. In this case the strike is mT0 and the payoff is
C = ST mT0 .
The following result gives the price of the lookback call option, cf. e.g. Proposition 9.5.1, page 270 of [18].
Proposition 8.8. The price at time t [0, T ] of the lookback call option
with payoff ST mT0 is given by

er(T t) IE ST mT0 | Ft

St
St
T t
T t
t r(T t)
m
e
= St +
0
mt0
mt0
t 2r/2
t

2
m0
m0
2
St
T t
T t
+er(T t) St

St +
.
2r St
St
2r
mt0
Figure 8.21 represents the price of the lookback call option as a function of
mt0 and St for different values of the time to maturity T t.
"
233
N. Privault
T = 7.0
90
80
70
60
50
40
30
20
10
0
Lookback call option price
80
60
80
60
mt
40
40
20
20
0
St
Fig. 8.21: Graph of the lookback call option price.

er(T t) IE ST mT0 | Ft = St er(T t) IE mT0 | Ft

St
St
T t
T t
= St +
er(T t) mt0
mt0
mt0
t 2r/2
t

!
m0
m0
St
St 2
T t
T t

+er(T t)
er(T t) +
.
2r
St
St
mt0
2
Black-Scholes Approximation of Lookback Call Prices
Letting

S
S
Ker
BSc (S, K, r, , ) = S +
K
K
denote the standard Black-Scholes formula for the price of a European call
option, we observe that the lookback call option price satisfies
er(T t) IE [ST mT0 | Ft ] = BSc (St , mt0 , r, , T t)

t 2r/2
t !
2
St
m0
m0
T t
T t
r(T t)
St
+
e
,

2r
mt0
St
St
i.e.
234
"
Exotic Options

St
er(T t) IE [ST mT0 | Ft ] = BSc (St , mt0 , r, , T t) + St hc T t, t
m0
where the function
hc (, z) =

2
2
+
(z) er z 2r/
(1/z) ,
2r
(8.35)
depends only on z = St /mt0 and satisfies

hc (, z) = hp (, z)

2
2
1 er z 2r/ ,
2r
R+ ,
z R+ ,
where (z, ) 7 er z 2r/ also solves the PDE (8.33).
Black-Scholes PDE for Lookback Call Options

By the same argument as in the proof of Proposition 8.6, the function
f (t, x, y) satisfies the Black-Scholes PDE
rf (t, x, y) =
f
f
1
2f
(t, x, y) + rx (t, x, y) + x2 2 2 (t, x, y),
t
x
2
x
t, x > 0,
under the boundary conditions
lim f (t, x, y) = x,
y&0
f
y (t, x, y)x=y = 0,
f (T, x, y) = x y,
0 t T,
0 t T,
x > 0,
(8.36a)
y > 0,
(8.36b)
0 y x,
(8.36c)
and the corresponding self-financing hedging strategy is given by

t =
f
(t, St , mt0 ),
x
t [0, T ],
(8.37)
which represents the quantity of the risky asset St to be held at time t in the
hedging portfolio.
In other words, the price of the lookback call option takes the form
f (t, St , mt ) = er(T t) IE [ST mT0 | Ft ],
where the function f (t, x, y) is given by
"
235
N. Privault

x
x
T t
T t
f (t, x, y) = x +
er(T t) y
(8.38)
y
y

2
x
2 y 2r/ T t y
T t
er(T t) +

+er(T t) x
2r
x
x
y

2
x
x
T t
T t
= x yer(T t)
x 1+
+
y
2r
y

2 2r/ 2
y
T t y
r(T t)
.
+xe

2r x
x

The boundary condition (8.36a) is explained by the fact that
f (t, x, 0) = er(T t) IE [ST mT0 | St = x, mt0 = 0]
= er(T t) IE [ST | St = x, mt0 = 0]
= er(T t) IE [ST | St = x]
= er(T t) x.
On the other hand, (8.36b) follows from the fact that
f (T, x, y) = IE [ST mT0 | ST = x, mT0 = y] = x y.
We have
f (t, x, x) = xC(T t),
with

2
2
+
(1) + er
(1) ,
C( ) = 1 er
(1) 1 +
2r
2r
> 0, hence
f
(t, x, x) = C(T t),
x
t [0, T ],
while we also have

f
(t, x, y)y=x = 0,
y
0 x y.
236
"
Exotic Options
Scaling Property of Lookback Call Prices
We note the scaling property

#
"

f (t, x, y) = er(T t) IE ST mT0 St = x, mt0 = y

"
#

r(T t)
t
t
T
=e
IE ST m0 mt St = x, m0 = y

"
#
t
ST
m
mT
= er(T t) x IE
0 t St = x, mt0 = y
St
St
St

#
"
T
y m
ST
t St = x, mt0 = y
= er(T t) x IE
x
x
x

"
#

r(T t)
t
T
t
=e
x IE ST m0 mt St = 1, m0 = y/x

"
#

= er(T t) x IE ST mT0 St = 1, mt0 = y/x

= xf (t, 1, y/x),
hence letting
g(, z) :=

2 r 2r/2
1
2
1 er
(z) 1 +
+
(z) +
e
z
(
(1/z)),
z
2r
2r
we have g(, 1) = C(T t), and
f (t, x, y) = xg(T t, x/y)
and the boundary condition
(, 1) = 0,
g(0, z) = 1 1 ,
z
> 0,
z 1.
(8.39a)
(8.39b)
The next Figure 8.22 shows a graph of the function g(, z).
"
237
N. Privault
normalized lookback call price
option price path
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
03
2.5
z
1.5
200
150
100
50
Fig. 8.22: Normalized lookback call option price.
The next Figure 8.23 represents the path of the underlying asset price used
in Figure 8.22.
Fig. 8.23: Graph of the underlying asset price.
Next we represent the option price as a function of time.
238
"
Exotic Options
option price path
St-mt
60
50
40
30
20
10
0
50
100
150
200
Fig. 8.24: Graph of the lookback call option price.
The next Figure 8.25 represents the corresponding underlying asset price and
its running minimum.
100
St
mt
90
80
70
60
50
40
30
20
50
100
150
200
Fig. 8.25: Running minimum of the underlying asset price.
Due to the relation

x
x
BSc (x, y, r, , ) = x +
yer
y
y
= xBSc (1, y/x, r, , )
for the standard Black-Scholes call formula, we observe that f (t, x, y) satisfies
f (t, x, y) = xBSc (1, y/x, r, , T t) + xhc (T t, x/y),
i.e.
f (t, x, y) = xg(T t, x/y),
with
"
239
N. Privault
g(, z) = BSc (1, 1/z, r, , ) + hc (, z),
(8.40)
where hc (, z) is the function given by (8.35), and (x, y) 7 xhc (T t, x/y)

also satisfies the Black-Scholes PDE (8.25), i.e. (, z) 7 BSc (1, 1/z, r, , )
and hc (, z) both satisfy the PDE (8.33) under the boundary condition
hc (0, z) = 0,
z 1.
The next Figures 8.26 and 8.27 show the decomposition of g(t, z) in (8.40) and
Figures 8.22-8.23 into the sum of the Black-Scholes call function BSc (1, 1/z, r, , )
and h(t, z).
normalized Black-Scholes put price BSc(1,1/z,r,,T-t)
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
03
2.5
z
1.5
200
150
100
50
Fig. 8.26: Black-Scholes call price in the decomposition (8.40) of the normalized
lookback call option price g(, z).
h(T-t,x)
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
03
2.5
z
1.5
200
150
100
50
Fig. 8.27: Function hc (, z) in the decomposition (8.40) of the normalized lookback

call option price g(, z).
We also note that

240
"
Exotic Options

T t
IE [M0T mT0 | S0 = x] = x xer(T t)
(1)

2
2
T t
T t
x 1 +
+
(1) + xer(T t)
(1)
2r
2r

2
T t
T t
+
(1)
+xer(T t)
(1) + x 1 +
2r

2
T t
x er(T t)
(1) x
2r

2
T t
T t
+
(1) +
(1)
= x 1+
2r
2

T t
T t
(1)
(1) .
+xer(T t)
1
2r
Hedging Lookback call Options

In this section we compute hedging strategies for lookback options by application of the Delta hedging formula (8.37). See [3], 2.6.1, page 29, for
another approach to the following result using the Clark-Ocone formula.
Here we use (8.37) instead, cf. Proposition 4.6 of [67].
Proposition 8.9. The hedging strategy of the lookback call option is given
by

2
St
St
T t
T t
+
(8.41)
t = +
t
t
m0
2r
m0
t 2r/2 2

t
m0
m0
T t
+er(T t)
1
,
t [0, T ].
St
2r
St
Proof. We need to differentiate

x
f (t, x, y) = BSc (x, y, r, , T t) + xhc T t,
y
with respect to the variable x, where
hc (, z) =

2
2
+
(z) er z 2r/
(1/z)
2r
is given by (8.35) First we note that the relation

x
BSc (x, y, r, , ) = +
x
y
is known, cf. Propositions 5.5 and 6.7. Next, we have
"
241
N. Privault

x
x
x hc
x
xhc ,
= hc ,
+
,
,
y
y
y z
y
and

2
hc
2
(, z) =
+
(z) er z 2r/
(
(1/z))
z
2r x
z

2 2r r 12r/2
e
z
(
(1/z))
2r 2

2
(z)
=
exp +
2
2rz 2

2
2
1
2r
er z 2r/
exp (
(1/z)) + 2 er z 12r/ (
(1/z)) .
2
2rz 2
Next, we note that

2 1 4r2
4r
1
(z)
(z)
= exp +
+
2
2 2

2

2
1
1
4r
4r
1
= e 2 (+ (z)) exp
2 log z + (r + 2 )
2
2
2

2
2r2
2r
2r2
21 (+
(z))
=e
exp
+ 2 log z + 2 + r
2
e( (1/z))
/2
= er z 2r/ e(+ (z))
/2
(8.42)
as in the proof of Proposition 5.5, hence

2
x
hc
,
= er z 12r/ (
(1/z)),
z
y
and

y 2r/2 y
x
x
xhc ,
= hc ,
er

,
y
y
x
x
which concludes the proof.
In Figure 8.28 we represent the Delta of the lookback call option, as given
by (8.41).
242
"
Exotic Options
T = 200.0000
2.5
2
1.5
1
0.5
00
10 20
mt
30 40
50 60
70 80
90 90
80
70
60
50
40
St
30
20
10
Fig. 8.28: Delta of the lookback call option.
The above scaling procedure can be applied to the Delta as well, by noting
that t can be written as

St
t = t, t ,
m0
where the function (t, z) is given by
2

2
T t
T t
(t, z) = +
(z) +
(z) +er(T t) z 2r/
2r

1
2
T t
1
,
2r
z
(8.43)
t [0, T ], z [0, 1]. The graph of the function (t, x) is given in Figure 8.29.

1.4
1.2
1
0.8
0.6
0.4
0.2
0
2.4
2.2
z
2
1.8
1.6
1.4
1.2
1 70
60
50
20
30
40
time to maturity
10
Fig. 8.29: Rescaled portfolio strategy for the lookback call option.
"
243
N. Privault
Similar calculations using (8.27) can be carried out for other types of lookback options, such as options on extrema and partial lookback options, cf.
[66].
As a consequence of (8.43) we have

e
IE ST mT0 | Ft

St
St
T t
T t
mt0 er(T t)
= St +
t
t
m0
m0

2r/2
t

2
S
St
m
2
t
0
T t
T t
+er(T t) St

St +
t
t
2r m0
St
2r
m0

12r/2
t !
St
St
m0
T t
T t
= t St mt0 er(T t)
+

,
mt0
mt0
St
r(T t)
and the quantity of the riskless asset ert in the portfolio is given by
t = mt0 erT

12r/2
t !

St
m0
St
T t
T t
+
mt0
mt0
St
0,
so that the portfolio value Vt at time t satisfies
Vt = t St + t ert ,
t R+ ,
and one has to constantly borrow from the riskless account in order to hedge
the lookback option.
8.5 Asian Options

An option on average is an option whose payoff has the form
C = (T , ST ),
where
T = S0
wT
0
eBu +ru
u/2
du =
wT
0
Su du,
T R+ .
For example when (y, x) = (y/T K) this yields the Asian call option
with payoff
+
+
w
T
1 T
Su du K
=
K
,
(8.44)
T 0
T
244
"
Exotic Options
which is a path-dependent option whose price at time t [0, T ] is given
by
"
+ #
1 wT

(8.45)
er(T t) IE
Su du K
Ft .

T 0
As another example, when (y, x) = ey this yields the price
i
h rT

P (0, T ) = IE e 0 Su du = IE eT
at time 0 of a bond with underlying short term rate process St .
The option with payoff C = (T , ST ) can be priced as
"

wT
er(T t) IE [(T , ST ) | Ft ] = er(T t) IE t +
Su du, ST
t

wT S
ST
u
= er(T t) IE y + x
du, x
t St
St
y=t ,x=St

w T t S
ST t
u
r(T t)
=e
IE y + x
du, x
.
0
S0
S0
y=t ,x=St
#

Ft

(8.46)
Using the Markov property of the process (St , t )tR+ , we can write down
the option price as a function
f (t, St , t ) = er(T t) IE [(T , ST ) | Ft ],
of (t, St , t ), where the function f (t, x, y) is given by

w T t S
ST t
u
f (t, x, y) = er(T t) IE y + x
du, x
.
0
S0
S0
As we will see below there exists no easily tractable closed form solution for
the price of an arithmetically averaged Asian option.
First we note that the prices of option on avarages can be estimated numerically using the joint probability density T t ,BT t of (T t , BT t ), as
follows:
f (t, x, y) =

ww
2
er(T t)
y + xz, xeu+r(T t) (T t)/2 T t ,BT t (z, u)dzdu.
0
In [119], Proposition 2, the joint probability density of

w

t
2
(t , Bt ) =
S0 eBs p s/2 ds, Bt pt/2 ,
0
"
t > 0,
245
N. Privault
has been computed in the case = 2, cf. also [77]. In the next proposition
we restate this result for an arbitrary variance parameter after rescaling.
Let (v, ) denote the function defined as
2
ve /(2 ) w 2 /(2 ) v cosh
e
e
sinh() sin (/ ) d,
(v, ) =
2 3 0
v, > 0.
(8.47)
Proposition 8.10. For all t > 0 we have

w

t
2
P
eBs p s/2 ds dy, Bt pt/2 dz
0

z/2 2
pz/2p2 2 t/8
1 + ez
t dy
4e
e
exp 2
,
dz,
2
2 y
2 y
4
y
y > 0, z R.
The expression of this probability density can then been used for the pricing
of options on average such as (8.46), as

w T t S
ST t
v
f (t, x, y) = er(T t) IE y + x
dv, x
0
S0
S0
= er(T t)

w T t S
w
2
v
y + xz, xeu+r(T t) (T t)/2 P

dv dz, BT t du
0
0
S0

r(T t)+p2 2 (T t)/8 w w
u+r(T t) 2 (1+p)(T t)/2
= e
y + xz, xe
0
2
!
!
2
2
p
4eu/2p (T t)/4 2 (T t)
dz
1 + eup (T t)/2
,
du
exp 2
2 z
2
2 z
4
z

ww
2
2 2
= er(T t)p (T t)/8
y + x/z, xv 2 er(T t) (T t)/2
0

0

2
1+v
4vz 2 (T t)
dz
v 1p exp 2z
,
dv ,
2
2
4
z
which actually stands as a triple integral due to the definition (8.47) of (v, ).
Note that here the order of integration between du and dz cannot be exchanged without particular precautions, at the risk of wrong computations.
Asian call options

We have
e
r(T t)
IE
"
+
1 wT
Su du K
T 0
#

Ft

246
"
Exotic Options
= er(T t) IE
=e
r(T t)
"
"
1
T
"
1
T
IE
= er(T t) IE
1
T
wT
#

Ft
t

#

+

wT S

u
y+x
du K
Ft
t St

x=St ,
#

+
w T t S
u
y+x
du K
0
S0

t +

+
Su du K
y=t
x=St , y=t
Hence the option can be priced as

f (t, St , t ) = e
r(T t)
IE
"
+
1 wT
Su du K
T 0
#

Ft ,

where the function f (t, x, y) is defined by

"

+ #
w T t S
1
u
f (t, x, y) = er(T t) IE
y+x
du K
0
T
S0
"

+ #
1
x
= er(T t) IE
T t K
y+
.
T
S0
Bounds on Asian option prices

We note (see Lemma 1 of [65] for the discrete-time version of that result), that
the Asian call option price can be bounded by the corresponding European
call price using convexity arguments, as follows:
"
+ #
w

T
du
1 wT
Su du K
erT IE
(Su K)+
erT IE
0
T 0
T
w

T
du
(IE [ST | Fu ] K)+
= erT IE
0
T
w

T
rT
+ du
e
IE IE
(ST K)
Fu
0
T

rT
+
=e
IE (ST K)

erT IE (ST K)+ ,
see also Proposition 3.2-(ii) of [40] for lower bounds when r takes negative
values.
We also note the bound
"
247
N. Privault
"
+ #
1 wT
1 wT

Su du K
Su du
Ft er(T t) IE

T 0
T 0
#
#
"
"

1 wt
1 wT

er(T t) IE
Su du Ft + er(T t) IE
Su du Ft

T 0
T t
wt
wT
1
1
er(T t)
Su du + er(T t)
St er(tu) du
T 0
T t
w
w
St T r(T u)
1 t
er(T t)
Su du +
e
du
T 0
T t
w
r(T
t)
1 t
1e
er(T t)
Su du + St
,
T 0
rT
0 er(T t) IE
=
=
=
=
"
#

Ft

hence the Asian option price tends to zero for infinity maturity time:
"
+ #
1 wT

Su du K
lim er(T t) IE
Ft = 0.
T

T 0
Probabilistic Approach
First we note that the numerical computation of Asian option prices can be
done using the probability density of
T =
wT
0
St dt.
From Proposition 8.10 we deduce the marginal density of

wt
2
t =
eBs p s/2 ds,
0
as follows:

w
t
2
P
eBs p s/2 ds du
0
!
!
2
2
p2 2 t/8 w
1 + evp t/2
p
4ev/2p t/4 2 t
e
exp 2
,
dvdu
2u
2 u
2
2 u
4

w
2 2
1 + v2
4v 2 t
du
= ep t/8
v 1p exp 2 2
,
dv ,
2
0
u
u 4
u
=
u > 0. From this we get

P(t /S0 du) = P
w

St dt du
248
(8.48)
"
Exotic Options
2
= ep
2 t/8
w
0

1 + v2
4v 2 t
du
v 1p exp 2 2
,
dv ,
2
u
u 4
u
where St = S0 eBt p t/2 and p = 1 2r/ 2 . This probability density can

then be used for the pricing of Asian options, as
"

+ #
x
1
r(T t)
(8.49)
y+
T t K
f (t, x, y) = e
IE
T
S0

+
w y + xz
= er(T t)
K
P(T t /S0 dz)
0
T
+
w w y + xz
2 2
= er(T t) ep (T t)/8
K
0
0
2
T

2
2
1
+
v
4v
(T t)
dz
v 1p exp 2 2
,
dv
2
z
z
4
z
w
1 r(T t)p2 2 (T t)/8 w
= e
(xz + y KT )
0(KT y)/x 0
T

2
2
4v (T t)
dz
1+v
exp 2 2
,
dv
z
2 z
4
z
+

4x r(T t)p2 2 (T t)/8 w w 1 2 (KT y)
= 2 e
0
0
T
z
4x

1 + v2
2 (T t)
dz
1p
v
exp z
vz,
dv ,
2
4
z
cf. the Theorem in 5 of [10], which is actually a triple integral due to the
definition (8.47) of (v, t). Note that since the integrals are not absolutely
convergent, here the order of integration between dv and dz cannot be exchanged without particular precautions, at the risk of wrong computations.
Time Laplace transform

The time Laplace transform of the rescaled price
"
+ #
1wt
Su du
,
C(t) := IE
t 0
as
w
0
"
C(t)dt =
r K/2
0
2
(1 2Kx)2+ 2+p p /2 dx
p
,

( 2 + 2p) 1 + p + 2 + p2 /2
ex x2+
p+
2+p2 /2
249
N. Privault
with here = 2 and (z) denotes the gamma function, cf. Relation (3.10)
in [40]. This expression can be used for pricing by numerical inversion of the
Laplace transform. The following Figure 8.30 represents Asian option prices
computed by the Geman-Yor [40] method.
Asian option price
30
25
20
15
10
5
100
95
underlying
90
85
80
100
50
200
150
250
300
350
Time in days
Fig. 8.30: Graph of the Asian option price with = 1, r = 0.1 and K = 90.
We refer to e.g. [2], [10], [27], and references therein for more on Asian
option pricing using the probability density of the averaged geometric Brownian motion.
Figure 7.2 presents a graph of implied volatility surface for Asian options on
light sweet crude oil futures.
Lognormal approximation
Other numerical approaches to the pricing of Asian options include [73],
[111] which relies on approximations of the average price probability based
on the Lognormal distribution. The lognormal distribution with mean and
variance 2 has the probability density function
2
2 dx
1
g(x) = e(log x) /(2 ) ,
x
2
where x > 0, R, > 0, and moments

IE[X] = e+
/2
and
IE[X 2 ] = e2+2 .
(8.50)
Under the lognormal approximation, Asian options on the time integral

T :=
wT
0
St dt
250
"
Exotic Options
of geometric Brownian motion
St = eBt +(r
/2)t
t [0, T ],
are computed by approximating T by a lognormal random variable, as

"
+ #
2
1 wT
1
erT IE
St dt K
' e(+ /2)T (d1 ) K(d2 ),
(8.51)
T 0
T
where
d1 =
and
log(IE[T ]/(KT ))
T
T +
2 T log(KT )
=
2
log(IE[T ]/(KT ))
T
d2 = d1
T =
,
2
and
,
are estimated as

1
E[2T ]
2 = log
T
(IE[T ])2
and
=
1 2
1
log IE[T ]
,
T
2
based on the first two moments of the lognormal distribution as in (8.50). The
next Figure 8.31 compares the lognormal approximation to a Monte Carlo
estimate of Asian option prices with = 0.5, r = 0.05 and K/St = 1.1
1
lognormal approximation
Monte Carlo estimate
stratified lognormal approximation
0.9
0.8
asian option price
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
6
time t
10
12
Fig. 8.31: Lognormal approximation to the Asian option price.
In Figure 8.31 we also include the stratified approximation

"
+ #
1 wT
rT
e
E
St dt K
T 0
"
(8.52)
251

N. Privault

erT w p(z/x)2 (z/x)T /2+2 (z/x)T /2
e
(d+ (K, z, x)) KT (d (K, z, x))
0
T
dP (ST = z, S0 = x),
'
of [94], where
d (K, z, x) :=
1
log
2(z/x) T

2x(bT (z/x) (1 + z/x)aT (z/x))
T
(z/x)
2 K 2 T 2
2
and

1
2
bT (z)
1z
,
2 (z) := log
T
aT (z) aT (z)

log z
1
1 2
log z
1 2

aT (z) := 2
+
T
T
,
p(z)
2
2
2 T
2 T

1
log z
log z

+ 2 T
2 T
,
bT (z) := 2
q(z)
2 T
2 T
and
p(z) :=
1
2 2 T
e(
T /2+log z ) /(2 2 T )
q(z) :=
1
2 2 T
e(
T +log z ) /(2 2 T )
For reference, in the next proposition we compute the unconditional mean

and variance of T , which have been used in (8.51), cf. also (7) and (8) page
480 of [73].
Proposition 8.11. We have
IE[T ] = S0
and
IE[(T )2 ] = 2S02
re(2r+
erT 1
,
r
)T
(2r + 2 )erT + (r + 2 )
.
r(r + 2 )(2r + 2 )
Proof. The computation of the first moment is straightforward. For the

second moment we have, letting p = 1 2r/ 2 ,
wT wT
2
2
ep a/2p b/2 IE[eBa eBb ]dbda
IE[(T )2 ] = S02
0
0
wT wa
2
2
2
2
= 2S02
ep a/2p b/2 e (a+b)/2 eb dbda
0
0
wT
wa
2
2
e(p1) a/2
e(p3) b/2 dbda
= 2S02
0
0
wT
2
2
4S02
=
e(p1) a/2 (1 e(p3) a/2 )da
(p 3) 2 0
252
"
Exotic Options
wT
wT
2
2
2
4S02
4S02
e(p1) a/2 da
e(p1) a/2 e(p3) a/2 da
2
2
(p 3) 0
(p 3) 0
wT
2
2
8S02
4S02
=
(1 e(p1) T /2 )
e(2p4) a/2 da
(p 3)(p 1) 4
(p 3) 2 0
2
2
4S02
8S02
(1 e(p1) T /2 )
(1 e(p2) T )
=
(p 3)(p 1) 4
(p 3)(p 2) 4
= 2S02
re(2r+
)T
(2r + 2 )erT + (r + 2 )
,
r(r + 2 )(2r + 2 )
since r 2 /2 = p 2 /2.
PDE Method - Two Variables

The price at time t of the Asian option with payoff
"
+
1 wT
r(T t)
f (t, St , t ) = e
IE
Su du K
T 0
(8.44) can be written as

#

t [0, T ].
Ft ,

(8.53)
Next, we derive the Black-Scholes partial differential equation (PDE) for the
price of a self-financing portfolio. Until the end of this chapter we model the
asset price (St )t[0,T ] as
dSt = St dt + St dWt ,
t R+ ,
where (Wt )tR+ is a standard Brownian motion under the historical probability measure P.
Vt = f (t, St , t ),
2
t R+ ,
for some f C ((0, ) (0, ) ).

Then the function f (t, x, y) in (8.53) satisfies the PDE
"
253
N. Privault
rf (t, x, y) =
f
f
f
1
2f
(t, x, y) + x (t, x, y) + rx (t, x, y) + x2 2 2 (t, x, y),
t
y
x
2
x
t, x > 0, under the boundary conditions
y
+
lim f (t, x, y) = er(T t)

K , 0 t T,
x&0
T
lim f (t, x, y) = 0,
0 t T, x R+ ,
y
+

f (T, x, y) = y K ,
x, y R+ ,
T
y R+ , (8.54a)
(8.54b)
(8.54c)
and t is given by
f
(t, St , t ),
t R+ .
x
Proof. We note that the self-financing condition implies
t =
dVt = t dAt + t dSt

= rt At dt + t St dt + t St dWt
(8.55)
= rVt dt + ( r)t St dt + t St dWt

= rt At dt + t St dt + t St dWt ,
(8.56)
t R+ . Noting that dt = St dt, the application of Itos formula to f (t, x, y)

leads to
f
f
(t, St , t )dt + St (t, St , t )dt
t
y
f
1 2 2 2f
f
+St (t, St , t )dt + St
(t, St , t )dt + St (t, St , t )dWt .
x
2
x2
x
(8.57)
df (t, St , t ) =
By respective identification of the terms in dWt and dt in (8.55) and (8.57)

we get
254
"
Exotic Options
f
f
f
(t, St , t )dt + St (t, St , t )dt + St (t, St , t )dt

rt At dt + t St dt =
t
y
x
1
2f
+ St2 2 2 (t, St , t )dt,

2
x
S dW = S f (t, S , )dW ,
t t
t
t
t
t
t
x
hence
f
f
1 2 2 2f
rVt rt St = t (t, St , t ) + St y (t, St , t )dt + 2 St x2 (t, St , t ),
t = f (t, St , t ),
x
i.e.
f
f
f
rf (t, St , t ) =
(t, St , t ) + St (t, St , t ) + rSt (t, St , t )
t
y
x
2f
1
+ St2 2 2 (t, St , t ),
2
x
= f (t, S , ).
t
t
t
x

Next, we examine two methods which allow one to reduce the Asian option
pricing PDE from three variables (t, x, y) to two variables (t, z). Reduction
of dimensionality can be of crucial importance when applying discretization
scheme whose complexity are of the form N d where N is the number of
discretization steps and d is the dimension of the problem.
PDE Method - One Variable (1) - Time Independent Coefficients

Following [70], page 91, we define the auxiliary process

1 t
1 1 wt
Su du K =
Zt =
K ,
St T 0
St T
t [0, T ].
With this notation, the price of the Asian option at time t becomes
"
"
+ #
1 wT

er(T t) IE
Su du K
Ft = er(T t) IE ST (ZT )+

T 0
"
#

Ft .

255

N. Privault
Lemma 8.1. The price (8.45) at time t of the Asian option with payoff
(8.44) can be written as
t [0, T ],
St g(t, Zt ),
where
+ #
1 w T t Su
du
T 0
S0
"
+ #
T t
z+
= er(T t) IE
.
S0 T
g(t, z) = er(T t) IE
"
z+
(8.58)
Proof. For 0 s t T , we have

w

t
1
St
d (St Zt ) = d
Su du K = dt,
0
T
T
hence
St Zt
1 w t Su
= Zs +
du,
Ss
T s Ss
t s.
Since for any t [0, T ], St is positive and Ft -measurable, and Su /St is independent of Ft , u t, we have:
#
"
"
+ #

ST

r(T t)
+
r(T t)
e
IE ST (ZT ) Ft = e
St IE
ZT
Ft

St
#
"

+
w
1 T Su

Zt +
= er(T t) St IE
du
Ft

T t St

"
#

+

1 w T Su

r(T t)
=e
St IE
z+
du
Ft

T t St
z=Zt
"
#

+
w T t S
1
u
du
= er(T t) St IE
z+
T 0
S0
z=Zt
"
#
+
T t
= er(T t) St IE
z+
S0 T
z=Zt
= St g(t, Zt ),
which proves (8.58).
Note that as in (8.49), g(t, z) can be computed from the density (8.48) of
T t , as
"
+ #
T t
g(t, z) = IE
z+
S0 T
256
"
Exotic Options
w
0
z+
u +
P(t /S0 du)
T
= ep t/8

w
u + w 1p
1 + v2
4v 2 (T t)
du
z+
v
exp 2
,
dv
0
0
T
2
2 u
4
u
= ep t/8
w

1 + v2
4v 2 (T t)
du
u w 1p
v
exp 2
,
dv
2
2
0
(zT )0
T
u
4
u

w
w
4v 2 (T t)
du
1 + v2
p2 2 t/8
1p
= ze
,
dv
v
exp 2
(zT )0 0
2
2 u
4
u

w
w
2
2
1 p2 2 t/8
4v
(T
t)
1
+
v
+ e
,
dvdu.
v 1p exp 2
(zT )0 0
T
2
2 u
4

z+
The next proposition gives a replicating hedging strategy for Asian options.
Vt = St g(t, Zt ),
t R+ ,
for some f C 2 ((0, ) (0, )2 ).

Then the function g(t, x) satisfies the PDE
g
(t, z) +
t
1
rz
T
g
1
2g
(t, z) + 2 z 2 2 (t, z) = 0,
z
2
z
(8.59)
under the terminal condition

g(T, z) = z + ,
and the corresponding replicating portfolio is given by
g
(t, Zt ),
t [0, T ].
z
Proof. We proceed as in [99]. From the expression of 1/St we have

1
1
d
=
+ 2 dt dWt ,
St
St
t = g(t, Zt ) Zt
hence

dZt = d
"
1
St
t
K
T

257
N. Privault

t
K
T St
St

1
t
1
d
Kd
T
St
St

1 dt
t
1
+
K d
T St
T
St

dt
1
+ St Zt d
T
St

dt
+ Zt + 2 dt Zt dWt .
T
=d
=
=
=
=
By self-financing we have
dVt = t dAt + t dSt

= rt At dt + t St dt + t St dWt ,
(8.60)
t R+ . The application of Itos formula to f (t, x, y) leads to

d(St g(t, Zt )) = g(t, Zt )dSt + St dg(t, Zt ) + dSt dg(t, Zt )
g
g
(t, Zt )dt +
(t, Zt )dZt
=
t
z
1 2g
+
(t, Zt )(dZt )2 + dSt dg(t, Zt )
2 z 2
g
= St (t, Zt )dt + St g(t, Zt )dt + St g(t, Zt )dWt
t
g
1 g
g
+St Zt + 2
(t, Zt )dt + St (t, Zt )dt St Zt (t, Zt )dWt
z
T z
z
1 2 2 2g
g
2
+ Zt St 2 (t, Zt )dt St Zt (t, Zt )dt
2
z
z
g
g
1 g
= St g(t, Zt )dt + St (t, Zt )dt + St Zt + 2
(t, Zt )dt + St (t, Zt )dt
t
z
T z
2g
g
1
+ 2 Zt2 St 2 (t, Zt )dt 2 St Zt (t, Zt )dt
2
z
z
g
+St g(t, Zt )dWt St Zt (t, Zt )dWt .
z
we get
258
"
Exotic Options
g
g
rt At + t St = St g(t, Zt ) + St t (t, Zt ) St Zt z (t, Zt )
1 g
1
2g
+ St (t, Zt ) + 2 Zt2 St 2 (t, Zt ),

T z
2
z
S = S g(t, Z ) S Z g (t, Z ),
t t
t
t
t t
t
z
hence
1 g
1 2 2 2g
g
rVt rt St = St t (t, Zt ) + T St z (t, Zt ) + 2 Zt St z 2 (t, Zt ),
t = g(t, Zt ) Zt g (t, Zt ),
z
i.e.

g
1
g
1
2g
(t,
z)
+
rz
(t, z) + 2 z 2 2 (t, z) = 0,
t
T
z
2
z
= g(t, Z ) Z g (t, Z ),
t
t
t
t
z
g(T, z) = z + .

We check that
f
f
t = er(T t) St f (t, St , Zt ) Zt f (t, St , Zt )
z
x

g
r(T t)
=e
Zt (t, Zt ) + g(t, Zt )
z
!

g
1 1 wt
= er(T t) St
Su du K
+ g(t, Zt )
t,
x
x T 0
|x=St

1 1 wt
r(T t)
=
xe
g t,
Su du K
, t [0, T ].
x
x T 0
|x=St
We also find that the amount invested on the riskless asset is given by
t At = Zt St
g
(t, Zt ).
z
Next we note that a PDE with no first order derivative term can be obtained
using time-dependent coefficients.
"
259
N. Privault
PDE Method - One Variable (2) - Time Dependent Coefficients
Define now the auxiliary process

1
1 1 wt
(1 er(T t) ) + er(T t)
Su du K
rT
St T 0
1
r(T t)
r(T t)
(1 e
=
)+e
Zt ,
t [0, T ],
rT
Ut :=
i.e.
Zt = er(T t) Ut +
er(T t) 1
,
rT
t [0, T ].
We have
1
dUt = er(T t) dt + rer(T t) Zt dt + er(T t) dZt
T
= er(T t) 2 Zt dt er(T t) Zt dWt ( r)er(T t) Zt dt
t,
= er(T t) Zt dW
t R+ ,
where
t = dWt dt +
dW
r
t dt
dt = dW
is a standard Brownian motion under

= eWT
dP
t/2
dP = erT
ST
dP .
S0
Lemma 8.2. The Asian option price can be written as

"
+
1 wT
Su du K
St h(t, Ut ) = er(T t) IE
T 0
where the function h(t, y) is given by

"
#

(UT )+ Ut = y ,
h(t, y) = IE

#

Ft ,

0 t T.
Proof. We have
UT =
and
1
ST

1 wT
Su du K = ZT ,
T 0
|F
dP
2
erT ST
t
= e(WT Wt ) (T t)/2 = rt ,
dP|Ft
e St
hence the price of the Asian option is

260
"
Exotic Options
er(T t) IE [ST (ZT )+ | Ft ] = er(T t) IE [ST (UT )+ | Ft ]
#
"

erT ST
+
= St IE
(U
)
Ft
T
rt

e St

"
#

|F
d
P
t
+
= St IE
F
(U
)

t
T

dP|Ft
+
= St IE[(U
T ) | Ft ].

The next proposition gives a replicating hedging strategy for Asian options.
See 7.5.3 of [109] and references therein for a different derivation of the
PDE (8.61).
t R+ ,
Vt = St h(t, Ut )
for some f C 2 ((0, ) (0, )2 ).
Then the function h(t, z) satisfies the PDE
h
1
(t, y) + 2
t
2
1 er(T t)
y
rT
2
2h
(t, y) = 0,
y 2
(8.61)

h(T, z) = z + ,
and the corresponding replicating portfolio is given by
t = h(t, Ut ) Zt
h
(t, Ut ),
y
t [0, T ].
Proof. By the self-financing condition (8.56) we have

dVt = rVt dt + ( r)t St dt + t St dWt ,
(8.62)
t R+ . By Itos formula we get

d(St h(t, Ut )) = h(t, Ut )dSt + St dh(t, Ut ) + dSt dh(t, Ut )
= St h(t, Ut )dt + St h(t, Ut )dWt
"
261
N. Privault

+St
+
h
h
1 2h
(t, Ut )(dUt )2
(t, Ut )dt +
(t, Ut )dUt +
t
y
2 y 2
h
(t, Ut )dSt dUt
y
h
= St h(t, Ut )dt + St h(t, Ut )dWt St ( r) (t, Ut )Zt dt
y

2
h
h
1
t + 2 h (t, Ut )Z 2 dt
+St
(t, Ut )dt (t, Ut )Zt dW
t
2
t
y
2 y
h
2
St (t, Ut )Zt dt
y
h
= St h(t, Ut )dt + St h(t, Ut )dWt St ( r) (t, Ut )Zt dt
y

h
1 2h
h
+St
(t, Ut )dt (t, Ut )Zt (dWt dt) + 2 2 (t, Ut )Zt2 dt
t
y
2 y
h
2 St (t, Ut )Zt dt.
y
we get
h
h
rt At + t St = St h(t, Ut ) ( r)St Zt y (t, Ut )dt + St t (t, Ut )
1
2h
+ St 2 Zt2 2 (t, Ut ),
2
y
t = h(t, Ut ) Zt (t, Ut ),
y
hence
1
2h
h
r A = rSt (t h(t, Ut )) + St (t, Ut ) + St 2 2 (t, Ut )Zt2 ,
t t
t
2
y
t = h(t, Ut ) Zt h (t, Ut ),
y
and

2 2
h
1
1 er(T t)
h
(t, y) + 2
y
(t, y) = 0,
t
2
rT
y 2

1 er(T t)
h
t = h(t, Ut ) +
Ut
(t, Ut ),
rT
y

262
"
Exotic Options
h(T, z) = z + .

We also find

1 er(T t) h
h
t At = er(T t) St Ut
(t, Ut ) = St Zt (t, Ut ).
rT
y
y
Exercises
Exercise 8.1 Consider a risky asset whose price St is given by

dSt = St dBt + 2 St dt/2,
(8.63)
where (Bt )tR+ is a standard Brownian motion.

a) Solve the stochastic differential equation (8.63).
b) Compute the expected stock price value IE[ST ] at time T .
c) What is the probability distribution of the supremum sup Bt over the
t[0,T ]
interval [0, T ]?
d) Compute the expected value IE[ST ] of the maximum
!
ST := sup St = S0 sup eBt = S0 exp sup Bt
t[0,T ]
t[0,T ]
t[0,T ]
of the stock price over the interval [0, T ].

e) What is the probability distribution of the infimum
inf Bt over the
t[0,T ]
interval [0, T ] ?
f) Compute the expected value IE[ST ] of the minimum

ST := inf St = S0 inf eBt = S0 exp inf Bt .
t[0,T ]
t[0,T ]
t[0,T ]
of the stock price over the interval [0, T ].

Exercise 8.2 Recall that the maximum Xt := sups[0,t] Bs over [0, t] of standard Brownian motion (Bs )s[0,t] has the probability density
r
Xt (x) =
"
2 x2 /(2t)
e
1[0,) (x),
t
x R.
263
N. Privault
a) Let a = inf{s R+ : Bs = a} denote the first hitting time of a > 0 by
(Bs )sR+ . Using the relation between {a t} and {Xt a}, write down
the probability P (a t) as an integral from a to .
b) Using integration by parts on [a, ), compute the probability density of
a .
2
Hint: the derivative of ex /(2t) with respect to x is xex

c) Compute the mean value IE[(a )2 ] of 1/a2 .
/(2t)
/t.
Exercise 8.3 Barrier options.

a) Compute the hedging strategy of the up-and-out barrier call option on
the underlying asset St with exercise date T , strike K and barrier B, with
B > K.
b) Compute the joint probability density
fYT ,BT (a, b) =
dP(YT a & BT b)
,
dadb
a, b R,
of standard Brownian motion BT and its minimum

YT = min Bt .
t[0,T ]
c) Compute the joint probability density

fYT ,BT (a, b) =
T b)
dP(YT a & B
,
dadb
a, b R,
T = BT + T and its minimum

of drifted Brownian motion B
t = min (Bt + t).
YT = min B
t[0,T ]
t[0,T ]
d) Compute the price at time t [0, T ] of the down-and-out barrier call

option on the underlying asset St with exercise date T , strike K, barrier
B, and payoff
C = (ST K)
1(
min St > B
0tT
S K
if min St > B,
if min St B,
0tT
0tT
in cases 0 K.

Exercise 8.4 Barrier forward contracts. Compute the price at time t of the
following barrier forward contracts on the underlying asset St with exercise
264
"
Exotic Options
date T , strike K, barrier B, and the following payoffs. In addition, compute
the corresponding hedging strategies.
a) Up-and-in barrier long forward contract. Take
C = (ST K) 1(
max St > B
S K if max St > B,
T
0tT
0tT
if max St B.
0tT
b) Up-and-out barrier long forward contract. Take
C = (ST K) 1(
max St < B
S K if max St < B,
T
0tT
0tT
if max St B.
0tT
c) Down-and-in barrier long forward contract. Take
C = (ST K) 1(
min St B
S K if min St > B,
T
0tT
0tT
if min St B.
0tT
e) Up-and-in barrier short forward contract. Take
C = (K ST ) 1(
max St > B
K ST if max St > B,
0tT
0tT
if max St B.
0tT
f) Up-and-out barrier short forward contract. Take
C = (K ST ) 1(
max St < B
0tT
K ST if max St < B,
0tT
if max St B.
0tT
g) Down-and-in barrier short forward contract. Take
"
265
N. Privault
C = (K ST ) 1(
min St B
0tT
K ST if min St > B,
0tT
if min St B.
0tT
Exercise 8.5 Consider a risky asset whose price St is given by

dSt = St dBt + 2 St dt/2,
where (Bt )tR+ is a standard Brownian motion.
a) What is the probability distribution (distribution function and probability
density function) of the minimum min Bt over the interval [0, T ]?
t[0,T ]
b) Compute the price value

e
T /2

IE ST min St
t[0,T ]
of a lookback call option on ST with maturity T .

Exercise 8.6 Lookback options. Compute the hedging strategy of the lookback put option priced in Proposition 8.4.
Exercise 8.7 Consider the short rate process rt = Bt , where (Bt )tR+ is a
standard Brownian motion.
rT
a) Find the probability distribution of the time integral 0 rs ds.
b) Compute the price
"
+ #
wT
ru du
erT IE
0
of a caplet on the forward rate
rT
0
rs ds.
Exercise 8.8 Asian call options with negative strike. Consider the asset price
process
2
St = S0 ert+Bt t/2 ,
t R+ ,
266
"
Exotic Options
where (Bt )tR+ is a standard Brownian motion. Assuming that 0, compute the price
"
+ #
1 wT

r(T t)
e
IE
Su du
Ft

T 0
of the Asian option at time t [0, T ].
Exercise 8.9 Consider an asset price (St )tR+ which is a submartingale under
the risk-neutral measure P , in a market with risk-free interest rate r > 0,
and let (x) = (x K)+ be the (convex) payoff function of the European
call option.
Show that, for any sequence 0 < T1 < < Tn , the price of the average
option with payoff

ST1 + + STn
n
can be upper bounded by the price of the European call option with maturity
Tn , i.e. show that

ST1 + + STn
IE
IE [(STn )].
n
Exercise 8.10 Let (St )tR+ denote a risky asset whose price St is given by
where (Bt )tR+ is a standard Brownian motion under the risk-neutral measure P . Compute the price at time t [, T ] of the Asian option with payoff

+
1 wT
Ss ds K
,
T
under the condition that

At :=
1 wt
Ss ds K.
T
Exercise 8.11 Pricing of Asian options by PDEs. Show that the functions
g(t, z) and h(t, y) are linked by the relation

1 er(T t)
g(t, z) = h t,
+ er(T t) z ,
rT
"
t [0, T ],
z > 0,
267
N. Privault
and that the PDE (1.35) for h(t, y) can be derived from the PDE (1.33) for
g(t, z) and the above relation.
268
"
Chapter 9
American Options
In contrast with European option which have fixed maturities, the holder of
an American option is allowed to exercise at any given (random) time. This
transforms the valuation problem into an optimization problem in which one
has to find the optimal time to exercise in order to maximize the payoff of
the option. As will be seen in the first section below, not all random times
can be considered in this process, and we restrict ourselves to stopping times
whose value at time t be can decided based on the historical data available.
9.1 Filtrations and Information Flow

Let (Ft )tR+ denote the filtration generated by a stochastic process (Xt )tR+ .
In other words, Ft denotes the collection of all events possibly generated by
{Xs : 0 s t} up to time t. Examples of such events include the event
{Xt0 a0 , Xt1 a1 , . . . , Xtn an }
for a0 , a1 , . . . , an a given fixed sequence of real numbers and 0 t1 < <
tn < t, and Ft is said to represent the information generated by (Xs )s[0,t]
up to time t.
By construction, (Ft )tR+ is an increasing family of -algebras in the sense
that we have Fs Ft (information known at time s is contained in the information known at time t) when 0 < s < t.
One refers sometimes to (Ft )tR+ as the increasing flow of information
generated by (Xt )tR+ .
"
N. Privault
9.2 Martingales, Submartingales, and Supermartingales

Let us recall the definition of martingale (cf. Definition 5.3) and introduce in
addition the definitions of supermartingale and submartingale.
Definition 9.1. An integrable stochastic process (Zt )tR+ is a martingale
(resp. a supermartingale, resp. a submartingale) with respect to (Ft )tR+ if
it satisfies the property
Zs = IE[Zt | Fs ],
0 s t,
Zs IE[Zt | Fs ],
0 s t,
Zs IE[Zt | Fs ],
0 s t.
resp.
resp.
Clearly, a process (Zt )tR+ is a martingale if and only if it is both a supermartingale and a submartingale.
A particular property of martingales is that their expectation is constant.
Proposition 9.1. Let (Zt )tR+ be a martingale. We have
IE[Zt ] = IE[Zs ],
0 s t.
The above proposition follows from the tower property (16.25) of conditional expectations, which shows that
IE[Zt ] = IE[IE[Zt | Fs ]] = IE[Zs ],
0 s t.
(9.1)
Similarly, a supermartingale has a decreasing expectation, while a submartingale

has a increasing expectation.
Proposition 9.2. Let (Zt )tR+ be a supermartingale, resp. a submartingale.
Then we have
IE[Zt ] IE[Zs ],
0 s t,
resp.
IE[Zt ] IE[Zs ],
0 s t.
Proof. As in (9.1) above we have

IE[Zt ] = IE[IE[Zt | Fs ]] IE[Zs ],
The proof is similar in the submartingale case.
0 s t.
This obviously inappropriate nomenclature was chosen under the malign influence
of the noise level of radios SUPERman program, a favorite supper-time program of
Doobs son during the writing of [22], cf. [23], historical notes, page 808.
270
"
American Options
Independent increments processes whose increments have negative expectation give examples of supermartingales. For example, if (Xt )tR+ is such a
process then we have
IE[Xt | Fs ] = IE[Xs | Fs ] + IE [Xt Xs | Fs ]
= IE[Xs | Fs ] + IE[Xt Xs ]
IE[Xs | Fs ]
0 s t.
= Xs ,
Similarly, a process with independent increments which have positive expectation will be a submartingale. Brownian motion Bt + t with positive drift
> 0 is such an example, as in Figure 9.1 below.
5
drifted Brownian motion
drift
4.5
4
3.5
3
2.5
2
1.5
1
0.5
0
-0.5
0
10
12
14
16
18
20
Fig. 9.1: Drifted Brownian path.
The following example comes from gambling.
Fig. 9.2: Evolution of the fortune of a poker player vs number of games played.
"
271
N. Privault
A natural way to construct submartingales is to take convex functions of
martingales. Indeed, if (Mt )tR+ is a martingale and is a convex function,
Jensens inequality states that
(IE[Mt | Fs ]) IE[(Mt ) | Fs ],
0 s t,
(9.2)
which shows that

(Ms ) = (IE[Mt | Fs ]) IE[(Mt ) | Fs ],
0 s t,
i.e. ((Mt ))tR+ is a submartingale. More generally, the above shows that
(Mt )tR+ remains a submartingale when is convex nondecreasing and
(Mt )R+ is a submartingale. Similarly, ((Mt ))tR+ will be supermartingale
when (Mt )R+ is a martingale and the function is concave.
Other examples of (super, sub)-martingales include geometric Brownian
motion
2
St = S0 ert+Bt t/2 ,
t R+ ,
which is a martingale for r = 0, a supermartingale for r 0, and a
submartingale for r 0.
9.3 Stopping Times

Next we turn to the definition of stopping time.
Definition 9.2. A stopping time is a random variable : R+ {+}
such that
{ > t} Ft ,
t R+ .
(9.3)
The meaning of Relation (9.3) is that the knowledge of the event { > t}
depends only on the information present in Ft up to time t, i.e. on the knowledge of (Xs )0st .
In other words, an event occurs at a stopping time if at any time t it
can be decided whether the event has already occured ( t) or not ( > t)
based on the information generated by (Xs )sR+ up to time t.
For example, the day you bought your first car is a stopping time (one
can always answer the question did I ever buy a car), whereas the day you
will buy your last car may not be a stopping time (one may not be able to
answer the question will I ever buy another car).
Note that a constant time is always a stopping time, and if and are
stopping times then the smallest of and is also a stopping time,
272
"
American Options
since
{ > t} = { > t and > t} = { > t} { > t} Ft ,
t R+ .
Hitting times provide natural examples of stopping times. The hitting time
of level x by the process (Xt )tR+ , defined as
x = inf{t R+ : Xt = x},
is a stopping time, as we have (here in discrete time)

{x > t} = {Xs 6= x, 0 s t}
= {X0 6= x} {X1 6= x} {Xt 6= x} Ft ,
t N.
In gambling, a hitting time can be used as an exit strategy from the game.
For example, letting
x,y := inf{t R+ : Xt = x or Xt = y}
(9.4)
defines a hitting time (hence a stopping time) which allows a gambler to exit
the game as soon as losses become equal to x = 10, or gains become equal
to y = +100, whichever comes first.
However, not every R+ -valued random variable is a stopping time. For
example the random time
(
)
= inf
t [0, T ] : Xt = sup Xs
s[0,T ]
which represents the first time the process (Xt )t[0,T ] reaches its maximum
over [0, T ], is not a stopping time with respect to the filtration generated by
(Xt )t[0,T ] . Indeed, the information known at time t (0, T ) is not sufficient
to determine whether { > t}.
Given (Zt )tR+ a stochastic process and : R+ {+} a stopping
time, the stopped process (Zt )tR+ is defined as
Z if t ,
Zt =
Zt if t < ,
Using indicator functions we may also write
Zt = Z 1{ t} + Zt 1{ >t} ,
t R+ .
As a convention we let = + in case there exists no t R+ such that Xt = x.
"
273
N. Privault
The following figure is an illustration of the path of a stopped process.
0.065
0.06
0.055
0.05
0.045
0.04
0.035
0.03
0.025
0
10
t
15
20
Fig. 9.3: Stopped process
Theorem 9.1 below is called the stopping time (or optional sampling, or optional stopping) theorem, it is due to the mathematician J.L. Doob (19102004). It is also used in Exercise 9.3 below.
Theorem 9.1. Assume that (Mt )tR+ is a martingale with respect to (Ft )tR+ .
Then the stopped process (Mt )tR+ is also a martingale with respect to
(Ft )tR+ .
Proof. We only give the proof in discrete time by applying the martingale
transform argument of Proposition 2.1. Writing
M n = M0 +
n
X
(Ml Ml1 ) = M0 +
l=1
1{ln} (Ml Ml1 ),
l=1
we have
M n = M0 +
X
n
(Ml Ml1 ) = M0 +
l=1
1{l n} (Ml Ml1 ),
l=1
and for k n,
IE[M n | Fk ] = M0 +
IE[1{l n} (Ml Ml1 ) | Fk ]
l=1
= M0 +
k
X
IE[1{l n} (Ml Ml1 ) | Fk ]
l=1
274
"
American Options
IE[1{l n} (Ml Ml1 ) | Fk ]
l=k+1
= M0 +
k
X
(Ml Ml1 ) IE[1{l n} | Fk ]
l=1
IE[IE[(Ml Ml1 )1{l n} | Fl1 ] | Fk ]
l=k+1
= M0 +
k
X
(Ml Ml1 )1{l n}
l=1
IE[1{l n} IE[(Ml Ml1 ) | Fl1 ] | Fk ]
l=k+1
= M0 +
nk
X
(Ml Ml1 )1{l n}
l=1
= M0 +
k
X
(Ml Ml1 )1{l n}
l=1
= M k ,
k = 0, 1, . . . , n,
because the martingale property of (Ml )lN implies

IE[(Ml Ml1 ) | Fl1 ] = IE[Ml | Fl1 ] IE[Ml1 | Fl1 ]
= IE[Ml | Fl1 ] Ml1
= 0,
l 1.
Since the stopped process (M t )tR+ is a martingale by Theorem 9.1 we

find that its expectation is constant by Proposition 9.1. More generally, if
(Mt )tR+ is a supermartingale with respect to (Ft )tR+ , then the stopped
process (Mt )tR+ remains a supermartingale with respect to (Ft )tR+ .
As a consequence, if is a stopping time bounded by T > 0, i.e. T
almost surely, we have
IE[M ] = IE[M T ] = IE[M 0 ] = IE[M0 ].
(9.5)
In case is finite with probability one but not bounded we may also write
h
i
IE[M ] = IE lim M t = lim IE[M t ] = IE[M0 ],
(9.6)
t
provided that
"
275
N. Privault
|M t | C,
a.s.,
t R+ .
(9.7)
More generally, (9.6) will hold provided that the limit and expectation signs
can be exchanged, and this can be done using e.g. the Dominated Convergence Theorem.
In case P( = +) > 0, (9.6) will hold under the above conditions,
provided that
M := lim Mt
(9.8)
t
exists with probability one.

In addition, if and are two bounded stopping times such that ,
a.s., we have
IE[M ] IE[M ]
(9.9)
if (Mt )tR+ is a supermartingale, and
IE[M ] IE[M ]
(9.10)
if (Mt )tR+ is a submartingale, cf. Exercise 9.3 below for a proof in discrete
time. As a consequence of (9.9) and (9.10) (or directly from (9.5)), if and
are two bounded stopping times such that , a.s., we have
IE[M ] = IE[M ]
(9.11)
if (Mt )tR+ is a martingale.

Relations (9.9), (9.10) and (9.11) can be extended to unbounded stopping
times along the same lines and conditions as (9.6), such as (9.7) applied to
both and . Dealing with unbounded stopping times can be necessary in
the case of hitting times.
In general, for all stopping times (bounded or unbounded) it remains
true that
IE[M ] = IE[ lim M t ] lim IE[M t ] lim IE[M0 ] = IE[M0 ],
t
(9.12)
provided that (Mt )tR+ is a nonnegative supermartingale, where we used Fatous lemma. As in (9.6), the limit (9.8) is required to exist with probability
one if P( = +) > 0.
In the case of the exit strategy x,y of (9.4) the stopping time theorem
shows that IE[Mx,y ] = 0 if M0 = 0, which shows that on average this exit
IE[limn Fn ] limn IE[Fn ] for any sequence (Fn )nN of nonnegative random
variables, provided that the limits exist.
276
"
American Options
strategy does not increase the average gain of the player. More precisely we
have
0 = M0 = IE[Mx,y ] = xP(Mx,y = x) + yP(Mx,y = y)
= 10P(Mx,y = 10) + 100P(Mx,y = 100),
which shows that
P(Mx,y = 10) =
10
11
and P(Mx,y = 100) =
1
,
11
provided that the relation P(Mx,y = x) + P(Mx,y = y) = 1 is satisfied, see

below for further applications to Brownian motion.
As a counterexample to (9.11), the random time
(
:= inf
t [0, T ] : Mt = sup Ms
)
,
s[0,T ]
which is not a stopping time, will satisfy

IE[M ] > IE[MT ],
although T almost surely.
Similarly,

:= inf t [0, T ] : Mt = inf Ms ,
s[0,T ]
is not a stopping time and satisfies

IE[M ] < IE[MT ].
When Xt is a martingale, e.g. a centered random walk with independent increments, the message of the stopping time Theorem 9.1 is that the expected
gain of the strategy (9.4) remains zero on average since
IE[Xx,y ] = IE[X0 ] = 0.
Similar arguments are used in the examples below.
In the table below we summarize some of the results of this section for
bounded stopping times.
"
277
N. Privault
Mt
bounded stopping times
supermartingale IE[M ] IE[M ]
martingale
submartingale
IE[M ] = IE[M ]
IE[M ] IE[M ]
Examples of application
In this section we note that, as an application of the stopping time theorem,
a number of expectations can be computed in a simple and elegant way.
Brownian motion hitting a barrier

Given a, b R, a < b, let the hitting time a,b : R+ be defined by
a,b = inf{t 0 : Bt = a or Bt = b},
which is the hitting time of the boundary {a, b} of Brownian motion (Bt )tR+ ,
a, b R, a < b.
Recall that Brownian motion (Bt )tR+ is a martingale since it has independent increments, and those increments are centered:
IE[Bt Bs ] = 0,
0 s t.
Consequently, (Ba,b t )tR+ is still a martingale and by (9.6) we have

IE[Ba,b | B0 = x] = IE[B0 | B0 = x] = x,
as the exchange between limit and expectation in (9.6) can be justified since
|Bta,b | max(|a|, |b|),
t R+ .
Hence we have
x = IE[Ba,b | B0 = x] = aP(Ba,b = a | B0 = x)+bP(Ba,b = b | B0 = x),
under the additional condition

P Xa,b = a | X0 = x + P(Xa,b = b | X0 = x) = 1.
A hitting time is a stopping time
278
"
American Options
which yields
P(Ba,b = b | B0 = x) =
xa
,
ba
a x b,
bx
,
ba
a x b.
which also shows that

P(Ba,b = a | B0 = x) =
Note that the above result and its proof actually apply to any continuous
martingale, and not only to Brownian motion.
Drifted Brownian motion hitting a barrier

Next, let us turn to the case of drifted Brownian motion
t R+ .
Xt = x + Bt + t,
In this case the process (Xt )tR+ is no longer a martingale and in order to
use Theorem 9.1 we need to construct a martingale of a different type. Here
we note that the process
Mt := eBt
t/2
t R+ ,
is a martingale with respect to (Ft )tR+ . Indeed, we have

i
h
2
2

IE[Mt | Fs ] = IE eBt t/2 Fs = eBs s/2 ,
0 s t.
By Theorem 9.1 we know that the stopped process (Ma,b t )tR+ is a martingale, hence its expectation is constant by Proposition 9.1, and (9.6) gives
1 = IE[M0 ] = IE[Ma,b ],
as the exchange between limit and expectation in (9.6) can be justified since
|Mta,b | max(e|a| , e|b| ),
t R+ .
Next we note that letting = 2 we have

eXt = ex+Bt +t = ex+Bt
t/2
= ex Mt ,
hence
1 = IE[Ma,b ]
= ex IE[eXa,b ]
"
279
N. Privault

= e(ax) P Xa,b = a | X0 = x + e(bx) P(Xa,b = b | X0 = x),
under the additional condition

P Xa,b = a | X0 = x + P(Xa,b = b | X0 = x) = 1.
Finally this gives
P(Xa,b = a | X0 = x) =
e2x e2b
ex eb
= 2a
,
ea eb
e
e2b
(9.13)
a x b, and
P(Xa,b = b | X0 = x) =
e2a e2x
,
e2a e2b
a x b.
Letting b tend to infinity in the above equalities shows that the probability
P(a = +) of escape to infinity of Brownian motion started from x [a, )
is equal to
1 P(Xa, = a | X0 = x) = 1 e2(xa) , > 0,

P(a = +) =
0, 0.
and similarly for x (, b] we have
1 P(X,b = b | X0 = x) = 1 e2(xb) ,
P(b = +) =
0, 0.
< 0,
Mean hitting time for Brownian motion

The martingale method also allows us to compute the expectation IE[Ba,b ],
after checking that (Bt2 t)tR+ is also a martingale. Indeed we have
IE[Bt2 t | Fs ] = IE[(Bs + (Bt Bs ))2 t | Fs ]
= IE[Bs2 + (Bt Bs )2 + 2Bs (Bt Bs ) t | Fs ]
= IE[Bs2 s | Fs ] (t s) + IE[(Bt Bs )2 | Fs ] + 2 IE[Bs (Bt Bs ) | Fs ]
= Bs2 s (t s) + IE[(Bt Bs )2 | Fs ] + 2Bs IE[Bt Bs | Fs ]
= Bs2 s (t s) + IE[(Bt Bs )2 ] + 2Bs IE[Bt Bs ]
= Bs2 s,
0 s t.
Consequently the stopped process (B2a,b t a,b t)tR+ is still a martingale

by Theorem 9.1 hence the expectation IE[B2a,b t a,b t] is constant in
280
"
American Options
t R+ , hence by (9.6) we get
x2 = IE[B02 0 | B0 = x]
= IE[B2a,b a,b | B0 = x]
= IE[B2a,b | B0 = x] IE[a,b | B0 = x]
= b2 P(Ba,b = b | B0 = x) + a2 P(Ba,b = a | B0 = x) IE[a,b | B0 = x],
i.e.
IE[a,b | B0 = x] = b2 P(Ba,b = b | B0 = x) + a2 P(Ba,b = a | B0 = x) x2
bx
xa
+ a2
x2
= b2
ba
ba
= (x a)(b x),
a x b.
Mean hitting time for drifted Brownian motion

Finally we show how to recover the value of the mean hitting time of drifted
Brownian motion Xt = Bt + t. As above, the process Xt t is a martingale
the stopped process (Xa,b t (a,b t))tR+ is still a martingale by Theorem 9.1. Hence the expectation IE[Xa,b t (a,b t)] is constant in t R+ .
Since the stopped process (Xa,b t t)tR+ is a martingale, we have
x = IE[Xa,b a,b | X0 = x],
which gives
x = IE[Xa,b a,b | X0 = x]
= IE[Xa,b | X0 = x] IE[a,b | X0 = x]
= bP(Xa,b = b | X0 = x) + aP(Xa,b = a | X0 = x) IE[a,b | X0 = x],
i.e. by (9.13),
IE[a,b | X0 = x] = bP(Xa,b = b | X0 = x) + aP(Xa,b = a | X0 = x) x
e2a e2x
e2x e2b
+ a 2a
x
e2a e2b
e
e2b
2a
2x
2x
2b
b(e
e
) + a(e
e
) x(e2a e2b )
=
,
e2a e2b
=b
hence
Here we note that it can be showed that IE[a,b ] < in order to apply (9.6).
"
281
N. Privault
IE[a,b | X0 = x] =
b(e2a e2x ) + a(e2x e2b ) x(e2a e2b )

,
(e2a e2b )
a x b.
9.4 Perpetual American Options

The price of an American put option expiring at time T > 0 with strike K
can be expressed as the expected value of its discounted payoff:
i
h

f (t, St ) =
sup
IE er( t) (K S )+ St ,
t T
stopping time
under the risk-neutral probability measure P , where the supremum is taken

over stopping times between t and a fixed maturity T . Similarly, the price of
a finite expiration American call option with strike K is expressed as
i
h

IE er( t) (S K)+ St .
f (t, St ) =
sup
t T
stopping time
In this section we take T = +, in which case we refer to these options as

perpetual options, and the corresponding put and call are respectively priced
as
i
h

f (t, St ) =
sup
t
stopping time
and
f (t, St ) =
sup
t
stopping time
i
h

Two-choice optimal stopping at a fixed price level for perpetual

put options
In this section we consider the pricing of perpetual put options. Given L
(0, K) a fixed price, consider the following choices for the exercise of a put
option with strike K:
1. If St L, then exercise at time t.
2. Otherwise if St > L, wait until the first hitting time
L := inf{u t : Su L}
(9.14)
and exercise the option at time L if L < .

282
"
American Options
Note that by definition of (9.14) we have L = t if St L.
In case St L, the payoff will be
(K St )+ = K St
since K > L St , however in this case one would buy the option at price
K St only to exercise it immediately for the same amount.
In case St > L, the price of the option will be
i
h

fL (t, St ) = IE er(L t) (K SL )+ St
i
h

= IE er(L t) (K L)+ St
i
h

= (K L) IE er(L t) St .
(9.15)
We note that the starting date t does not matter when pricing perpetual
options, hence fL (t, x) is actually independent of t R+ , and the pricing of
the perpetual put option can be performed by taking t = 0 and in the sequel
we will work under
fL (t, x) = fL (x),
x > 0.
Recall that the underlying asset price is written as
St = S0 ert+Bt
t/2
t R+ ,
(9.16)

+
ability measure P , r is the risk-free interest rate, and > 0 is the volatility
coefficient.
Proposition 9.3. Assume that r > 0. We have

h
i

fL (x) = IE er(L t) (K SL )+ St = x
K x,
0 < x L,
=
x 2r/2
, x L.
(K L)
L
Proof. The result is obvious for St = x L since in this case we have L = t
and SL = St = x, so that we only focus on the case x L. In addition we
take t = 0 without loss of generality. By the relation

i
h
i
h

(9.17)
IE erL (K SL )+ S0 = x = IE erL (K L)S0 = x ,
"
283
N. Privault

i
h

we note that it suffices to compute IE erL S0 = x . For this, we note that
()
from (9.16), for all R the process (Zt )tR+ defined as

()
Zt

:=
St
S0
et(r+
/22 2 /2)
= eBt
2 t/2
t R+ ,
is a martingale under the risk-neutral probability measure P . Choosing

such that
(r 2 /2 + 2 2 /2) = r,
i.e.
0 = 2 2 /2 + (r 2 /2) r =
2
( + 2r/ 2 )( 1),
2
(9.18)
with solutions
and = 2r/ 2 ,
=1
we have
()
Zt

=
St
S0
ert ,
t R+ .
Choosing the negative solution = 2r/ 2 < 0 we have the bound

()
0 Zt

=
St
S0
ert
L
S0

,
0 t < L .
since r > 0, hence by dominated convergence we have

"
#

L
SL
IE erL = IE
erL 1{L <}
S0
S0
h
i
= IE Z()
1{L <}
L
h
i
()
= IE lim ZL t
t
h
i
()
= lim IE ZL t
t
i
h
()
= IE Z0
= 1,
and we find
284
"
American Options

i x 2r/2
h

IE erL S0 = x =
L
x L,
(9.19)
and we conclude by (9.17), which shows that

h
i
h
i

IE erL (K SL )+ S0 = x = (K L) IE erL S0 = x
x 2r/2
,
= (K L)
L
when S0 = x > L.
We note that taking L = K would yield a payoff always equal to 0 for the
option holder, hence the value of L should be strictly lower than K. On the
other hand, if L = 0 the value of L will be infinite almost surely, hence
the option price will be 0 when r 0 from (9.15). Therefore there should
be an optimal value L , which should be strictly comprised between 0 and K.
Figure 9.4 shows for K = 100 that there exists an optimal value L = 85.71
which maximizes the option price for all values of the underlying.
35
L=75
L=L*=85.71
L=98
(K-x)+
30
Option price
25
20
15
10
5
0
70
80
90
100
110
120
Underlying x
Fig. 9.4: Graphs of the option price by exercise at L for several values of L.
In order to compute L we observe that, geometrically, the slope of fL (x) at
x = L is equal to 1, i.e.
2
fL0 (L ) =
2r
(L )2r/ 1
(K L )
= 1,
2
(L )2r/2
i.e.
"
285
N. Privault
2r
(K L ) = L ,
2
or
L =
2r
K < K.
2r + 2
The same conclusion can be reached by the vanishing of the derivative

2
x 2r/
fL (x)
2r K L x 2r/
=
+ 2
= 0,
L
L
L
L
cf. page 351 of [109]. The next Figure 9.5 is a 2-dimensional animation that
also shows the optimal value L of L.
35
Put option price

(K-x)+
30
Option price
25
20
15
10
5
0
70
80
90
100
110
120
Underlying x
Fig. 9.5: Animated graph of the option price depending on the values of L.
The next Figure 9.6 gives another view of the put option prices according to
different values of L, with the optimal value L = 85.71.
286
"
American Options
(K-x)+
fL(x)
fL*(x)
K-L
30
25
20
15
10
5
0
70
75
80
75
85
70
90
65
Underlying x 95 100 105
110 60
80
85
90
100
95
L
Fig. 9.6: Graph of the option price as a function of L and of the underlying asset
price.
In Figure 9.7 which is based on the stock price of HSBC Holdings (0005.HK)
over year 2009, the optimal exercise strategy for an American put option
with strike K=$77.67 would have been to exercise whenever the underlying
price goes above L = $62, i.e. at approximately 54 days, for a payoff of
$38. Note that waiting a longer time, e.g. until 85 days, would have yielded a
higher payoff of at least $65. This is due to the fact that, here, optimization
is done based on the past information only and makes sense in expectation
(or average) over all possible future paths.
Payoff (K-x)+
American put price
Option price path
L*
80
70
60
50
40
30
20
10
0
0
50
100
Time in days
150
200
100
90
80
70
50
60
underlying HK$
40
30
Fig. 9.7: Path of the American put option price on the HSBC stock.
PDE approach
We can check by hand that
"
287
N. Privault
fL (x) =
K x,
0 < x L =
2r
K,
2r + 2

2r/2
2r + 2 x
K 2
2r
K,
, x L =
2
2r +
2r K
2r + 2
satisfies the PDE

rfL (x) +
rxfL0 (x)
rK < 0,
1 2 2 00
+ x fL (x) =
2
0,
0 < x L < K,
x > L .
(9.20)
in addition to the condition
fL (x) = K x,
0 < x L < K,
fL (x) > (K x)+ , x > L .
This can be summarized in the following differential inequalities, or variational differential equation:
+
(9.21a)
fL (x) (K x) ,
(9.21b)
rxfL0 (x) + 2 x2 fL00 (x) rfL (x),
2

1 2 2 00
(x) rxf (x)
x
f
(fL (x) (K x)+ ) = 0, (9.21c)
rf
(x)
L
L
L
which admits an interpretation in terms of absence of arbitrage, as shown

below.
By Itos formula the discounted portfolio price fL (St ) = ert fL (St )
satisfies

1
d(fL (St )) = rfL (St ) + rSt fL0 (St ) + 2 St2 fL00 (St ) ert dt
2
t ,
+ert St fL0 (St )dB
(9.22)
hence from Equation (9.21c), fL (St ) is a martingale when
fL (St ) > (K St )+ ,
i .e.
St > L ,
288
"
American Options
and the expected rate of return of the option price fL (St ) then equals the
rate r of the risk-free asset as
d(fL (St )) = d(ert fL (St )) = rfL (St )dt + ert dfL (St ),
and the investor prefers to wait.
On the other hand if fL (St ) = (K St )+ , i.e. 0 < St < L , it is not
worth waiting as (9.21b) and (9.21c) show that the return of the option is
lower than that of the risk-free asset, i.e.:
1
rfL (St ) + rSt fL0 (St ) + 2 St2 fL00 (St ) < 0,
2
and exercise becomes immediate since the process fL (St ) becomes a (strict)
supermartingale. In this case, (9.21c) implies fL (x) = (K x)+ .
In view of the above derivation it should make sense to assert that fL (St )
is the price at time t of the perpetual American put option. The next proposition shows that this is indeed the case, and that the optimal exercise time
is = L .
Proposition 9.4. The price of the perpetual American put option is given
for all t 0 by
fL (St ) = sup
t
stopping time
i
h

IE er( t) (K S )+ St
i
h

= IE er(L t) (K SL )+ St
K St , 0 < St L ,

2r/2
=
2r + 2 St
K 2
,
St L .
2
2r +
2r K
Proof. By Itos formula (9.22) and the inequality (9.21b) one checks that the
discounted portfolio price
u 7 eru fL (Su ),
u [t, ),
is a supermartingale. As a consequence, for all stopping times we have, by

(9.12),
i
i
h
h

ert fL (St ) IE er fL (S )St IE er (K S )+ St ,
from (9.21a), which implies
"
289
N. Privault
ert fL (St )
sup
t
stopping time
i
h

IE er (K S )+ St .
(9.23)
The converse is obvious by Proposition 9.3 as

i
h

fL (St ) = IE er(L t) (K SL )+ St
i
h

sup
t
stopping time
since L is a stopping time larger than t 0.

Note that that converse statement can also be obtained from the variational PDE (9.21a)-(9.21c) itself, without relying on Proposition 9.3. For this,
taking = L we note that the process
u 7 eruL fL (SuL ),
u t,
is not only a supermartingale, it is also a martingale until exercise at time

L by (9.20) since SuL L , hence we have
i
h

ert fL (St ) = IE er(uL ) fL (SuL )St ,
u t,
hence after letting u tend to infinity we obtain
i
h

ert fL (St ) = IE erL fL (SL )St
i
h

= IE erL fL (L )St
i
h

= IE erL (K SL )+ St
i
h

sup
IE erL (K SL )+ St ,
t
stopping time
which shows that

ert fL (St )
sup
t
stopping time
i
h

IE er (K S )+ St ,
t 0.
290
"
American Options
Two-choice optimal stopping at a fixed price level for perpetual
call options
In this section we consider the pricing of perpetual call options. Given L > K
a fixed price, consider the following choices for the exercise of a call option
with strike K:
1. If St L, then exercise at time t.
2. Otherwise, wait until the first hitting time
L = inf{u t : Su = L}
and exercise the option and time L .
In case St L, the payoff will be
(St K)+ = St K
since K < L St .
In case St < L, the price of the option will be
i
h

fL (St ) = IE er(L t) (SL K)+ St

h
i

= IE er(L t) (L K)+ St
i
h

= (L K) IE er(L t) St .
fL (x) =
x K,
x L > K,
(L K) x ,
L
0 < x L.
(9.24)
Proof. We only need to consider the case x < L. Note that for all R,

Zt :=
St
S0
ert+
t/22 2 t/2
= eBt
2 t/2
t R+ ,
Hence the
is a martingale under the risk-neutral probability measure P.
stopped process (ZtL )tR+ is a martingale and it has constant expectation. Hence we have
IE [ZtL ] = IE [Z0 ] = 1,
"
291
N. Privault
and by letting t go to infinity we get
#
"

SL
(r 2 /2+2 2 /2)L
IE
e
= 1,
S0
which yields
i S
h
2
2 2
0
,
IE e(r /2+ /2)L =
L
i.e.

IE erL =
S0
L

,
(9.25)
provided that is chosen such that

(r 2 /2 + 2 2 /2) = r,
i.e.
2
( + 2r/ 2 )( 1).
2
Here we choose the positive solution = 1 since S0 = x < L and the
expectation (9.25) is lower than 1.

0 = 2 2 /2 + (r 2 /2) r =
One sees from Figure 9.8 that the situation completely differs from the perpetual put option case, as there does not exist an optimal value L that would
maximize the option price for all values of the underlying.
450
L=150
L=250
L=400
(x-K)+
x
400
350
Option price
300
250
200
150
100
50
0
0
50
100
150
200
250
Underlying
300
350
400
450
Fig. 9.8: Graphs of the option price by exercising at L for several values of L.
The intuition behind this picture is that there is no upper limit above which
one should exercise the option, and in order to price the American perpetual
call option we have to let L go to infinity, i.e. the optimal exercise strategy
is to wait indefinitely.
292
"
American Options
+
(x-K)
fL(x)
K-L
180
160
140
120
100
80
60
40
20
0
300
150
250
200
L
150
100
250
200
Underlying x
100
Fig. 9.9: Graphs of the option prices parametrized by different values of L.

We check from (9.24) that
lim fL (x) = x lim K
x
= x,
L
x > 0.
(9.26)
As a consequence we have the following proposition.

Proposition 9.6. Assume that r 0. The price of the perpetual American
call option is given by
sup
t
stopping time
i
h

IE er( t) (S K)+ St = St ,
t R+ . (9.27)
Proof. For all L > K we have

i
h

fL (St ) = IE er(L t) (SL K)+ St
i
h

sup
IE er( t) (S K)+ St ,
t 0,
t
stopping time
hence taking the limit as L yields

i
h

St
sup
IE er( t) (S K)+ St
(9.28)
t
stopping time
from (9.26). On the other hand, for all stopping times t we have, by
(9.12),
i
i
h
h

IE er( t) (S K)+ St IE er( t) S St St ,
t 0,
since u 7 er(ut) Su is a martingale, hence
"
293
N. Privault
sup
t
stopping time
i
h

IE er( t) (S K)+ St St ,
t 0,
which shows (9.27) by (9.28).

rt
We may also check that since (e St )tR+ is a martingale, the process t 7

(ert St K)+ is a submartingale since the function x 7 (xK)+ is convex,
hence for all bounded stopping times such that t we have
i
i
h
h

(St K)+ IE (er( t) S K)+ St IE er( t) (S K)+ St ,
t 0, showing that it is always better to wait than to exercise at time t,
and the optimal exercise time is = +. This argument does not apply to
American put options.
9.5 Finite Expiration American Options

In this section we consider finite expirations American put and call options
with strike K, whose prices can be expressed as
i
h

f (t, St ) =
sup
t T
stopping time
and
f (t, St ) =
sup
t T
stopping time
i
h

Two-choice optimal stopping at fixed times with finite expiration

We start by considering the optimal stopping problem in a simplified setting
where {t, T } is allowed at time t to take only two values t (which corresponds to immediate exercise) and T (wait until maturity).
Call options
Since x 7 (x K)+ is a nondecreasing convex function and the process (ert St ert K)tR+ is a submartingale under P , we know that
t 7 (ert St ert K)+ is a submartingale by the Jensen inequality (9.2),
hence by (9.12), for any stopping time t we have
(St K)+ = ert (ert St ert K)+
ert IE [(er S er K)+ | Ft ]
294
"
American Options
IE [er( t) (S K)+ | Ft ],
i.e.
(x K)+ IE [er( t) (S K)+ |St = x],
x, t > 0.
In particular for = T t a deterministic time we get

(x K)+ er(T t) IE [(ST K)+ |St = x],
x, t > 0.
as illustrated in Figure 9.10 using the Black-Scholes formula for European

call options.
In other words, taking x = St , the payoff (St K)+ of immediate exercise at
time t is always lower than the expected payoff er(T t) IE [(ST K)+ |St = x]
given by exercise at maturity T . As a consequence, the optimal strategy for
the investor is to wait until time T to exercise an American call option, rather
than exercising earlier at time t.
Black-Scholes European call price
Payo (x-K)+
80
70
60
50
40
30
20
10
0
140
120
underlying 100
HK$
80
60
7
10 9 8
2 1 0
5 4to 3maturity T-t
6 Time
Fig. 9.10: Expected Black-Scholes European call price vs (x, t) 7 (x K)+ .
More generally it can be in fact shown that the price of an American call option equals the price of the corresponding European call option with maturity
T , i.e.
i
h

f (t, St ) = er(T t) IE (ST K)+ St ,
i.e. T is the optimal exercise date, cf. e.g. 14.4 of [110] for a proof.
Put options
For put options the situation is entirely different. The Black-Scholes formula
for European put options shows that the inequality
(K x)+ er(T t) IE [(K ST )+ |St = x],
"
295
N. Privault
does not always hold, as illustrated in Figure 9.11.
Black-Scholes European put price
Payo (K-x)+
16
14
12
10
8
6
4
2
0
0
90
100
110 underlying HK$
3 4
5
Time to maturity T-t 6
10
120
Fig. 9.11: Black-Scholes put price function vs (x, t) 7 (K x)+ .

As a consequence, the optimal exercise decision for a put option depends on
whether (K St )+ er(T t) IE [(K ST )+ |St ] (in which case one chooses
to exercise at time T ) or (K St )+ > er(T t) IE [(K ST )+ |St ] (in which
case one chooses to exercise at time t).
A view from above of the graph of Figure 9.11 shows the existence of
an optimal frontier depending on time to maturity and on the value of the
underlying asset instead of being given by a constant level L as in Section 9.4,
cf. Figure 9.12:
0
1
2
3
4
5 T-t
6
7
8
9
10
120
115
110
100
105
underlying HK$
95
90
Fig. 9.12: Optimal frontier for the exercise of a put option.

At a given time t, one will choose to exercise immediately if (St , T t) belongs
to the blue area on the right, and to wait until maturity if (St , T t) belongs
to the red area on the left.
296
"
American Options
PDE characterization of the finite expiration American put price
Let us describe the PDE associated to American put options. After discretization {0 = t0 , t1 , . . . , tN = T } of the time interval [0, T ], the optimal exercise
strategy for an American put option can be described as follow at each time
step:
If f (t, St ) > (K St )+ , wait.
If f (t, St ) = (K St )+ , exercise the option at time t.
Note that we cannot have f (t, St ) < (K St )+ .
If f (t, St ) > (K St )+ the expected return of the option equals that of
the risk-free asset. This means that f (t, St ) follows the Black-Scholes PDE
rf (t, St ) =
f
f
1
2f
(t, St ) + rSt (t, St ) + 2 St2 2 (t, St ),
t
x
2
x
whereas if f (t, St ) = (K St )+ it is not worth waiting as the return of the

option is lower than that of the risk-free asset:
rf (t, St )
f
f
1
2f
(t, St ) + rSt (t, St ) + 2 St2 2 (t, St ).
t
x
2
x
As a consequence, f (t, x) should solve the following variational PDE:
f (t, x) (K x) ,
f
f
1 2 2 2f
(t, x) rf (t, x),

(t, x) + rx (t, x) + x
x
2
x2
t
(9.29a)
(9.29b)

f (t, x) + rx f (t, x) + 1 2 x2 f (t, x) rf (t, x)
(9.29c)
x
2
x
t
(f (t, x) (K x)+ ) = 0,
subject to the terminal condition f (T, x) = (K x)+ . In other words, equality holds either in (9.29a) or in (9.29b) due to the presence of the term
(f (t, x) (K x)+ ) in (9.29c).
The optimal exercise strategy consists in exercising the put option as soon
as the equality f (u, Su ) = (K Su )+ holds, i.e. at the time
"
297
N. Privault
= inf{u t : f (u, Su ) = (K Su )+ },
after which the process fL (St ) ceases to be a martingale and becomes a
(strict) supermartingale.
A simple procedure to compute numerically the price of an American put
option is to use a finite difference scheme while simply enforcing the condition
f (t, x) (K x)+ at every iteration by adding the condition
f (ti , xj ) := max(f (ti , xj ), (K xj )+ )
right after the computation of f (ti , xj ).
The next figure shows a numerical resolution of the variational PDE
(9.29a)-(9.29c) using the above simplified (implicit) finite difference scheme.
In comparison with Figure 9.7, one can check that the PDE solution becomes
time-dependent in the finite expiration case.
Finite expiration American put price
Immediate
payoff (K-x)+
L*=2r/(2r+sigma2)
16
14
12
10
8
6
4
2
0
2
4
90
6
110
10
100
underlying
120
Fig. 9.13: Numerical values of the finite expiration American put price.
In general, one will choose to exercise the put option when
f (t, St ) = (K St )+ ,
i.e. within the blue area in Figure (9.13). We check that the optimal threshold
L = 90.64 of the corresponding perpetual put option is within the exercise
region, which is consistent since the perpetual optimal strategy should allow
one to wait longer than in the finite expiration case.
The numerical computation of the put price
i
h

f (t, St ) =
sup
t T
stopping time
298
"
American Options
can also be done by dynamic programming and backward optimization using
the Longstaff-Schwartz (or Least Square Monte Carlo, LSM) algorithm [75],
as in Figure 9.14.
Longstaff-Schwartz algorithm
Immediate
payoff (K-x)+
L*=2r/(2r+sigma2)
16
14
12
10
8
6
4
2
0
2
4
90
6
110
10
100
underlying
120
Fig. 9.14: Longstaff-Schwartz algorithm for the finite expiration American put price.
In Figure 9.14 above and Figure 9.15 below the optimal threshold of the
corresponding perpetual put option is again L = 90.64 and falls within the
exercise region. Also, the optimal threshold is closer to L for large time to
maturities, which shows that the perpetual option approximates the finite
expiration option in that situation. In the next Figure 9.15 we compare the
numerical computation of the American put price by the finite difference and
Longstaff-Schwartz methods.
10
Longstaff-Schwartz algorithm
Implicit finite differences
Immediate payoff (K-x)+
0
90
100
110
120
underlying
Fig. 9.15: Comparison between Longstaff-Schwartz and finite differences.
It turns out that, although both results are very close, the Longstaff-Schwartz
method performs better in the critical area close to exercise at it yields the
"
299
N. Privault
expected continuously differentiable solution, and the simple numerical PDE
solution tends to underestimate the optimal threshold. Also, a small error
in the values of the solution translates into a large error on the value of the
optimal exercise threshold.
The finite expiration American call option

In the next proposition we compute the price of a finite expiration American
call option with an arbitrary convex payoff function .
Proposition 9.7. Let : R R be a nonnegative convex function such
that (0) = 0. The price of the finite expiration American call option with
payoff function on the underlying asset (St )tR+ is given by
i
i
h
h

f (t, St ) =
sup
IE er( t) (S )St = er(T t) IE (ST )St ,
t T
stopping time
i.e. the optimal strategy is to wait until the maturity time T to exercise the
option, and = T .
Proof. Since the function is convex and (0) = 0 we have
(px) = ((1 p) 0 + px) (1 p) (0) + p(x) p(x),
for all p [0, 1] and x 0. Hence the process s 7 ers (St+s ) is a
submartingale since taking p = er( t) we have
ers IE [ (St+s ) | Ft ] ers (IE [St+s | Ft ])

ers IE [St+s | Ft ]
= (St ),
where we used Jensens inequality (9.2) applied to the convex function and
the fact that
(px) = ((1 p) 0 + px) (1 p)(0) + p(x) = p(x),
x > 0,
by the convexity of and the fact that (0) = 0. Hence by the optional
stopping theorem for submartingales, cf (9.10), for all (bounded) stopping
times comprised between t and T we have,
IE [er( t) (S ) | Ft ] er(T t) IE [(ST ) | Ft ],
i.e. it is always better to wait until time T than to exercise at time [t, T ],
and this yields
300
"
American Options
sup
t T
stopping time
i
i
h
h

IE er( t) (S )St er(T t) IE (ST )St .
The converse inequality

i
h

er(T t) IE (ST )St
sup
t T
stopping time
i
h

IE er( t) (S )St ,
being obvious because T is a stopping time.
As a consequence of Proposition 9.7 applied to the convex function (x) =

(x K)+ , the price of the finite expiration American call option is given by
i
h

f (t, St ) =
sup
IE er( t) (S K)+ St
t T
stopping time
i
h

= er(T t) IE (ST K)+ St ,
i.e. the optimal strategy is to wait until the maturity time T to exercise the
option.
In the following table we summarize the optimal exercise strategies for the
pricing of American options.
option
type
perpetual
K St , 0 < St L ,
put

2
option
St 2r/
(K L )
, St L .
L
= L
call
option
St
= +
finite expiration
Solve the PDE (9.29a)-(9.29c) for f (t, x)

or use Longstaff-Schwartz [75]
= T inf{u t : f (u, Su ) = (K Su )+ }
er(T t) IE [(ST K)+ | St ],
= T
Exercises
Exercise 9.1 Stopping times. Let (Bt )tR+ be a standard Brownian motion
started at 0.
a) Consider the random time defined by
:= inf{t R+ : Bt = B1 },
"
301
N. Privault
which represents the first time Brownian motion Bt hits the level B1 . Is
a stopping time?
b) Consider the random time defined by

:= inf t R+ : eBt = et/2 ,
which represents the first time the exponential of Brownian motion Bt
crosses the path of t 7 et/2 , where > 1.
Is a stopping time? If is a stopping time, compute IE[e ] by the
stopping time theorem.
c) Consider the random time defined by

:= inf t R+ : Bt2 = 1 + t ,
which represents the first time the process (Bt2 )tR+ crosses the straight
line t 7 1 + t, with (, 1).
Is a stopping time? If is a stopping time, compute IE[ ] by the Doob
stopping time theorem.
Exercise 9.2 Consider a standard Brownian motion (Bt )tR+ started at B0 =
0, and let
L = inf{t R+ : Bt = L}
denote the first hitting time of level L > 0.
a) Compute the Laplace transform IE[erL ] of L for all r 0.

Hint: Use the stopping time theorem and the fact that e 2rBt rt tR+
is a martingale when r > 0.
b) Find the optimal level stopping strategy depending on the value of r > 0
for the maximization problem

sup IE erL BL .
L>0
Exercise 9.3 (Doob-Meyer decomposition in discrete time). Let (Mn )nN

be a discrete-time submartingale with respect to a filtration (Fn )nN , with
F1 = {, }.
a) Show that there exists two processes (Nn )nN and (An )nN such that
(i) (Nn )nN is a martingale with respect to (Fn )nN ,
(ii) (An )nN is nondecreasing, i.e. An An+1 a.s., n N,
302
"
American Options
(iii) (An )nN is predictable in the sense that An is Fn1 -measurable,
n N, and
(iv) Mn = Nn + An , n N.
Hint: Let A0 = 0,
An+1 = An + IE[Mn+1 Mn | Fn ],
n 0,
and define (Nn )nN in such a way that it satisfies the four required properties.
b) Show that for all bounded stopping times and such that a.s.,
we have
IE[M ] IE[M ].
Hint: Use the stopping time Theorem 9.1 for martingales and (9.11).
Exercise 9.4 American digital options. An American digital call (resp. put)
option with maturity T > 0 can be exercised at any time t [0, T ], at the
choice of the option holder.
The call (resp. put) option exercised at time t yields the payoff 1[K,) (St )
(resp. 1[0,K] (St )), and the option holder wants to find an exercise strategy
that will maximize his payoff.
a) Consider the following possible situations at time t:
(i) St K,
(ii) St < K.
In each case (i) and (ii), tell whether you would choose to exercise the
call option immediately or to wait.
b) Consider the following possible situations at time t:
(i) St > K,
(ii) St K.
In each case (i) and (ii), tell whether you would choose to exercise the
put option immediately or to wait.
c) The price CdAm (t, St ) of an American digital call option is known to satisfy
the Black-Scholes PDE
rCdAm (t, x) =
Am
1
2
C (t, x) + rx CdAm (t, x) + 2 x2 2 CdAm (t, x).
t d
x
2
x
Based on your answers to Question (a), how would you set the boundary
conditions CdAm (t, K), 0 t < T , and CdAm (T, x), 0 x < K?
d) The price PdAm (t, St ) of an American digital put option is known to satisfy
the same Black-Scholes PDE
"
303
N. Privault
Am
1
2
P (t, x) + rx PdAm (t, x) + 2 x2 2 PdAm (t, x).
t d
x
2
x
(9.30)
Based on your answers to Question (b), how would you set the boundary
conditions PdAm (t, K), 0 t < T , and PdAm (T, x), x > K?
e) Show that the optimal exercise strategy for the American digital call option with strike K is to exercise as soon as the underlying reaches the
level K, at the time
rPdAm (t, x) =
K = inf{u t : Su = K},
starting from any level St K, and that the price CdAm (t, St ) of the
American digital call option is given by
CdAm (t, x) = IE[er(K t) 1{K <T } | St = x].
f) Show that the price CdAm (t, St ) of the American digital call option is equal
to

(r + 2 /2) + log(x/K)
x
CdAm (t, x) =
K

x 2r/2 (r + 2 /2) + log(x/K)
,
0 x K,
+
K

where = T t. Show that this formula is consistent with your answer
to Question (c).
g) Show that the optimal exercise strategy for the American digital put option with strike K is to exercise as soon as the underlying reaches the
level K, at the time
K = inf{u t : Su = K},
starting from any level St K, and that the price PdAm (t, St ) of the
American digital put option is
PdAm (t, x) = IE[er(K t) 1{K <T } | St = x],
x K.
h) Show that the price PdAm (t, St ) of the American digital put option is equal
to

x
(r + 2 /2) log(x/K)
PdAm (t, x) =
K

x 2r/2 (r + 2 /2) log(x/K)
,
x K,
K

and that this formula is consistent with your answer to Question (d).
i) Does the call-put parity hold for American digital options?
304
"
American Options
Exercise 9.5 American forward contracts. Consider (St )tR+ an asset price
process given by
dSt
= rdt + dBt ,
St
where r > 0 and (Bt )tR+ is a standard Brownian motion.
a) Compute the price
f (t, St ) =
sup
t T
stopping time
i
h

IE er( t) (K S )St ,
and optimal exercise strategy of a finite expiration American type short

forward contract with strike K on the underlying asset (St )tR+ , with
payoff K ST .
b) Compute the price
i
h

IE er( t) (S K)St ,
f (t, St ) =
sup
t T
stopping time
and optimal exercise strategy of a finite expiration American type long

forward contract with strike K on the underlying asset (St )tR+ , with
payoff ST K.
c) How are the answers to Questions (a) and (b) modified in the case of
perpetual options?
Exercise 9.6 Consider an underlying asset price process written as
St = S0 ert+Bt
t/2
t R+ ,

+
ability measure P , with , r > 0.
a) Show that the processes (Yt )tR+ and (Zt )tR+ defined as
2r/ 2
Yt := ert St
and Zt := ert St ,
t R+ ,
are both martingales under P .

b) Let L denote the hitting time
L = inf{u R+ : Su = L}.
By application of the stopping time theorem to the martingales (Yt )tR+
and (Zt )tR+ , show that
rL
x/L,
IE e
| S0 = x =
2
(x/L)2r/ ,
0 < x L,
"
x L.
305

N. Privault
c) Compute the price IE [erL (K SL )] of a short forward contract under
the exercise strategy L .
d) Show that for every value of S0 = x there is an optimal value Lx of L
that maximizes L 7 IE[erL (K SL )].
e) Would you use the strategy
Lx = inf{u R+ : Su = Lx }
as an optimal exercise strategy for the short forward contract with payoff
K S ?
Exercise 9.7 Let p 1 and consider a power put option with payoff
( S )p if S ,
(( S )+ )p =
0
if S > ,
when exercised at time , on an underlying asset whose price St is written as
St = S0 ert+Bt
t/2
t R+ ,
where (Bt )tR+ is a standard Brownian motion under the risk-neutral probability measure P , r 0 is the risk-free interest rate, and > 0 is the
volatility coefficient.
Given L (0, ) a fixed price, consider the following choices for the exercise
of a put option with strike :
(i) If St L, then exercise at time t.
(ii) Otherwise, wait until the first hitting time L := inf{u t : Su = L},
and exercise the option at time L .
a) Under the above strategy, what is the option payoff equal to if St L ?
b) Show that in case St > L, the price of the option is equal to
i
h

fL (St ) = ( L)p IE er(L t) St .
c) Compute the price fL (St ) of the option at time t.
2r/ 2
Hint. Recall that by (9.19) we have IE [er(L t) | St = x] = (x/L)

,
x L.
d) Compute the optimal value L that maximizes L 7 fL (x) for all fixed
x > 0.
Hint. Observe that, geometrically, the slope of x 7 fL (x) at x = L is
equal to p( L )p1 .
306
"
American Options
e) How would you compute the American option price
i
h

f (t, St ) =
sup
IE er( t) (( S )+ )p St ?
t
stopping time
Exercise 9.8 Same questions as in Exercise 9.7 for the option with payoff
(S )p when exercised at time , with p > 0.
Exercise 9.9 (cf. Exercise 8.5 page 372 of [109]). Consider an underlying asset
price process written as
St = S0 e(ra)t+Bt
t/2
t R+ ,

+
ability measure P , r > 0 is the risk-free interest rate, > 0 is the volatility
coefficient, and a > 0 is a constant dividend rate.
a) Show that for all x L and R the process (Zt )tR+ defined as

Zt :=
St
S0
e(ra)t+
t/22 2 t/2
t R+ ,
is a martingale under P .
b) Let L denote the hitting time
L = inf{u R+ : Su L}.
By application of the stopping time theorem to the martingale (Zt )tR+ ,
show that

S0
,
(9.31)
IE erL =
L
with
=
(r a 2 /2)
p
(r a 2 /2)2 + 4r 2 /2
.
2
(9.32)
c) Show that for all L (0, K) we have

h
i

IE erL (K SL )+ S0 = x
0 < x L,
K x,
=
2 /2)2 +4r 2 /2
(ra2 /2) (ra
2
(K L) x
, x L.
L
d) Show that the value L of L that maximizes
"
307
N. Privault

i
h

fL (x) := IE erL (K SL )+ S0 = x
for all x is given by
L =
K.
1
e) Show that
fL (x) =
K x,
0 < x L =
K,
1
1

x
1
, x L =
K,
f) Show by hand computation that fL (x) satisfies the variational differential

equation
(9.33a)
fL (x) (K x) ,
1 2 2 00
(9.33b)
(r a)xfL (x) + x fL (x) rfL (x),
2

(9.33c)
rfL (x) (r a)xfL0 (x) 2 x2 fL00 (x)
(f (x) (K x)+ ) = 0.
L
g) By Itos formula, check that the discounted portfolio price

t 7 ert fL (St )
is a supermartingale.
h) Show that we have
fL (S0 )
sup
stopping time
i
h

IE er (K S )+ S0 .
i) Show that the stopped process

s 7 er(sL ) fL (SsL ),
s R+ ,
is a martingale, and that

fL (S0 )
sup
j) Fix t R+ and let

IE er (K S )+ .
stopping time
denote the hitting time
308
"
American Options
L = inf{u t : Su = L }.
Conclude that the price of the perpetual American put option with dividend is given for all t 0 by
i
h

K St ,
0 < St
K,
=
1

St
,
St
K,
1
where is given by (9.32), and
L = inf{u t : Su L}.
Exercise 9.10 This exercise is a simplified adaptation of the paper [41].
We consider two risky assets S1 and S2 modeled by
2
S1 (t) = S1 (0)e1 Wt +rt2 t/2
S2 (t) = S2 (0)e2 Wt +rt2 t/2 ,

(9.34)
t R+ , with 2 > 1 0, and the perpetual optimal stopping problem
and
IE[er (S1 ( ) S2 ( ))+ ],
sup
stopping time
where (Wt )tR+ is a standard Brownian motion under P.

a) Find > 1 such that the process
Zt := ert S1 (t) S2 (t)1 ,
t R+ ,
(9.35)
is a martingale.
b) For some fixed L 1, consider the hitting time

L = inf t R+ : S1 (t) LS2 (t) ,
and show that
IE[erL (S1 (L ) S2 (L ))+ ] = (L 1) IE[erL S2 (L )].
c) By an application of the stopping time theorem to the martingale (9.35),
show that we have
IE[erL (S1 (L ) S2 (L ))+ ] =
"
L1
S1 (0) S2 (0)1 .
L
309

N. Privault
d) Show that the price of the perpetual exchange option is given by
sup
IE[er (S1 ( ) S2 ( ))+ ] =
stopping time
where
L 1
S1 (0) S2 (0)1 ,
(L )
.
1
L =
e) As an application of Question (d), compute the perpetual American put

option price
sup
IE[er ( S2 ( ))+ ]
stopping time
when r =
22 /2.
Exercise 9.11 Consider an underlying asset whose price is written as

St = S0 ert+Bt
t/2
t R+ ,
where (Bt )tR+ is a standard Brownian motion under the risk-neutral probability measure P , > 0 denotes the volatility coefficient, and r R is the
()
risk-free rate. For any R we consider the process (Zt )tR+ defined by
()
Zt
:= ert (St ) = S0 eBt
2 t/2+(1)(+2r/ 2 ) 2 t/2
t R+ .
(9.36)
()
a) Assume that r 2 /2. Show that, under P , the process (Zt )tR+ is
a supermartingale when 2r/ 2 1, and that it is a submartingale
when (, 2r/ 2 ] [1, ).
()
b) Assume that r 2 /2. Show that, under P , the process (Zt )tR+ is
2
a supermartingale when 1 2r/ , and that it is a submartingale
when (, 1] [2r/ 2 , ).
c) From now on we assume that r < 0. Given L > 0, let L denote the hitting
time
L = inf{u R+ : Su = L}.
()
By application of the stopping time theorem to (Zt )tR+ , show that
min(1,2r/ 2 )
,
rL
(x/L)
IE e
1{L <} | S0 = x
2
(x/L)max(1,2r/ ) ,
x L,
0 < x L.
d) Deduce an upper bound on the price

IE erL (K SL )+ | S0 = x
310
"
American Options
of a European put option exercised in finite time under the strategy L
when L (0, K) and x L.
e) Show that the upper bound of Question (d) increases and tends to +
when L decreases to 0.
f) Find an upper bound on the price

IE erL (SL K)+ 1{L <} | S0 = x
of a European call option exercised in finite time under the strategy L
when L K and x L.
g) Show that the upper bound of Question (f) increases in L K and tends
to S0 as L increases to +.
"
311
Chapter 10
Change of Num
eraire and Forward
Measures
In this chapter we introduce the notion of numeraire. This allows us to consider pricing under random discount rates using forward measures, with the
pricing of exchange options (Margrabe formula) and foreign exchange options (Garman-Kohlagen formula) as main applications. A short introduction to the computation of self-financing hedging strategies under change of
numeraire is also given in Section 10.5. The change of numeraire technique
and associated forward measures will also be applied to the pricing of bonds
and interest rate derivatives such as bond options in Chapter 12.
10.1 Notion of Num

eraire
A numeraire is any strictly positive Ft -adapted stochastic process (Nt )tR+
that can be taken as a unit of reference when pricing an asset or a claim.
In general the price St of an asset in the numeraire Nt is given by
St
St =
,
Nt
t R+ .
Deterministic numeraires transformations are easy to handle as a change of

numeraire by a deterministic factor is a formal algebraic transformation that
does not involve any risk. This can be the case for example when a currency
is pegged to another currency, e.g. the exchange rate 6.55957 from Euro to
French Franc has been fixed on January 1st, 1999.
On the other hand, a random numeraire may involve risk and allow for
arbitrage opportunities.
Examples of numeraire include:
"
N. Privault
- the money market account
Nt = exp
w
t
0

rs ds ,
where (rt )tR+ is a possibly random and time-dependent risk-free interest

rate.
In this case,
rt
St
St =
= e 0 rs ds St ,
Nt
t R+ ,
represents the discounted price of the asset at time 0.

- an exchange rate Nt = Rt with respect to a foreign currency.
In this case,
St
,
St =
Rt
t R+ ,
represents the price of the asset in units of the foreign currency. For example, if Rt = 1.7 is the exchange rate from Euro to Singapore dollar and
St = S$1, then St = St /Rt =e 0.59.
- forward numeraire: the price
i
h rT

P (t, T ) = IE e t rs ds Ft ,
0 t T,
of a bond paying P (T, T ) = $1 at maturity T , in this case Rt = P (t, T ),

0 t T . We check that
i
h rT
rt

t 7 e 0 rs ds P (t, T ) = IE e 0 rs ds Ft ,
0 t T,
is a martingale.
- annuity numeraire of the form
Rt =
n
X
(Tk Tk1 )P (t, Tk )
k=1
Anyone who believes exponential growth can go on forever in a finite world is

either a madman or an economist, Kenneth E. Boulding, in: Energy Reorganization
Act of 1973: Hearings, Ninety-third Congress, First Session, on H.R. 11510, page 248,
United States Congress, U.S. Government Printing Office, 1973.
314
"
Change of Numeraire and Forward Measures

where P (t, T1 ), . . . , P (t, Tn ) are bond prices with maturities T1 < < Tn
arranged according to a tenor structure.
- rcombinations of the above, for example a foreign money market account
t f
e 0 rs ds Rt , expressed in local (or domestic) currency, where (rtf )tR+ represents a short term interest rate on the foreign market.
When the numeraire is a random process, the pricing of a claim whose value
has been transformed under change of numeraire, e.g. under a change of currency, has to take into account the risks existing on the foreign market.
In particular, in order to perform a fair pricing, one has to determine a
probability measure (for example on the foreign market), under which the
transformed process St = St /Nt will be a martingale.
rt
For example in case Nt = e 0 rs ds , the risk-neutral measure P is a measure

under which the discounted price process
rt
St
St =
= e 0 rs ds St ,
Nt
t R+ ,
is a martingale.
In the next section we will see that this property can be extended to any
kind of numeraire.
10.2 Change of Num

eraire
In this section we review the pricing of options by a change of measure associated to a numeraire Nt , cf. e.g. [39] and references therein.
Most of the results of this chapter rely on the following assumption, which
expresses absence of arbitrage. In the sequel, (rt )tR+ denotes an Ft -adapted
short term interest rate process.
Assumption (A) Under the risk-neutral measure P , the discounted
numeraire
rt
t 7 e 0 rs ds Nt
is an Ft -martingale.
Taking the process (Nt )t[0,T ] as a numeraire, we define the forward measure
by
P
"
315
N. Privault
rT
dP
NT
= e 0 rs ds
.
dP
N0
(10.1)
Recall that from Section 6.3 the above relation rewrites as

= e
dP
rT
0
rs ds NT
N0
dP ,
which is equivalent to stating that

w
()dP()
=
rT
0
rs ds NT
N0
dP ,
or, under a different notation,

r
= IE e 0T rs ds NT ,
IE[]
N0
for all integrable FT -measurable random variable .
More generally, (10.1) and the fact that
t 7 e
rt
0
rs ds
Nt
is a martingale under P under Assumption (A), imply that

"
#

dP
NT r T rs ds
Nt r t rs ds
IE
e 0
e 0
,
Ft = IE
Ft =
dP
N0
N0
(10.2)
for all integrable Ft -measurable random variables , 0 t T . The next

lemma shows that for any integrable random claim F we have

rT
NT
0 t T.
IE[F
| Ft ] = IE F e t rs ds
Ft ,
Nt
Note that (10.2), which is Ft -measurable, should not be confused with (10.3),
which is FT -measurable. In the next lemma we compute the probaorpbility
|F /dP of P
|F with respect to P .
density dP
t
t
|Ft
|Ft
Lemma 10.1. We have
|F
rT
dP
NT
t
= e t rs ds
,
dP|Ft
Nt
0 t T.
(10.3)
Proof. The proof of (10.3) relies on the abstract version of the Bayes formula.
we start by noting that for all integrable Ft -measurable random variable G
we have
t ]] = IE[
t ]]
IE[
|F
IE[G
|F
IE[G
316
"
]
= IE[G

r
0T rs ds NT
= IE Ge
N0

Nt r t rs ds r T rs ds NT
IE e t
= IE G e 0
Ft
N0
Nt

rT
N
t rs ds T Ft ,
G IE e
= IE
Nt
which shows that

r
t ] = IE e
tT rs ds NT Ft ,
|F
IE[
Nt
i.e. (10.3) holds.
We note that in case the numeraire Nt = e

= P .
account we simply have P
rt
0
rs ds
is equal to the money market
Pricing using Change of Num

eraire
The change of numeraire technique is specially useful for pricing under random interest rates, in which case an expectation of the form
i
h rT

IE e t rs ds Ft
becomes a path integral, see e.g. [19] for an account of path integral methods
in quantitative finance. The next proposition is the basic result of this section,
it provides a way to
price an option with arbitrary payoff under a random
rT
discount factor e t rs ds by use of the forward measure. It will be applied in
Chapter 12 to the pricing of bond options and caplets, cf. Propositions 12.1,
12.2 and 12.3 below.
Proposition 10.1. An option with integrable payoff L1 (P , FT ) is priced
at time t as

i

h rT

Ft ,
IE e t rs ds Ft = Nt IE
0 t T,
(10.4)
NT
FT ).
provided /NT L1 (P,
Each application of the formula (10.4) will require to
a) identify a suitable numeraire (Nt )tR+ , and to
b) make sure that the ratio /NT takes a sufficiently simple form,
in order to allow for the computation of the expectation in the right-hand
side of (10.4).
"
317
N. Privault
Proof. Proposition 10.1 relies on Relation (10.3), which shows that
"
#

|F

dP
t
Nt IE
Ft
Ft = Nt IE
NT
NT dP|Ft
i
h rT

0 t T.
= IE e t rs ds Ft ,

Next we consider further examples of numeraires and associated examples of
option prices.
Examples:
rt
a) Money market account: Nt = e 0 rs ds , where (rt )tR+ is a possibly random

and time-dependent risk-free interest rate.
= P and (10.4) simply reads
In this case we have P
i
i
h rT
h rT
rt

IE e t rs ds Ft = e 0 rs ds IE e 0 rs ds Ft ,
0 t T,
which yields no particular information.

b) Forward numeraire: Nt = P (t, T ) is the price P (t, T ) of a bond maturing
atr ttime T , 0 t T . Here, the discounted bond price process e 0 rs ds P (t, T )
is an Ft -martingale under P , i.e. Assumpt[0,T ]
tion (A) is satisfied and Nt = P (t, T ) can be taken as numeraire. In this

case, (10.4) shows that a random claim can be priced as
i
h rT
h i

Ft ,
IE e t rs ds Ft = P (t, T )IE
0 t T,
(10.5)
satisfies
since P (T, T ) = 1, where the forward measure P
rT
rT
dP
P (T, T )
e 0 rs ds
= e 0 rs ds
=
dP
P (0, T )
P (0, T )
(10.6)
by (10.1).
c) Annuity numeraire of the form
Nt =
n
X
(Tk Tk1 )P (t, Tk )
k=1
where P (t, T1 ), . . . , P (t, Tn ) are bond prices with maturities T1 < <
Tn . Here, (10.4) shows that a swaption on the cash flow P (T, Tn )
P (T, T1 ) NT can be priced as
318
"

i
h rT

IE e t rs ds (P (T, Tn ) P (T, T1 ) NT )+ Ft
"
+ #
P (T, Tn ) P (T, T1 )

= Nt IE
Ft ,
NT
0 t T , where (P (T, Tn ) P (T, T1 ))/NT becomes a swap rate, cf.
(11.31) in Proposition 11.5 and Section 12.5.
In the sequel, given (Xt )tR+ an asset price process, we define the process of
forward prices
t := Xt ,
X
0 t T,
(10.7)
Nt
which represents the values at times t of Xt , expressed in units of the
t )tR under
numeraire Nt . It will be useful to determine the dynamics of (X
+
the forward measure P.

Proposition 10.2. Let (Xt )tR+ denote a continuous Ft -adapted asset price
process such that
rt
t 7 e 0 rs ds Xt
is a martingale under P . Then under change of numeraire, the process
provided it is integrable
t )t[0,T ] of forward prices is a martingale under P,
(X
under P.
Proof. We need to show that

Xt Fs = Xs ,
IE
Nt
Ns
0 s t,
(10.8)
and we achieve this using a standard characterization of conditional expectation. Namely, for all bounded Fs -measurable random variables G we note
that under Assumption (A) we have
"
#

G Xt = IE G Xt dP
IE
Nt
Nt dP
"
"
##

Xt dP
= IE G
IE
F
t
Nt
dP

rt
Xt
= IE Ge 0 ru du
N0

rs
Xs
= IE Ge 0 ru du
N0

X
s
G
= IE
,
0 s t,
Ns
because
"
319
N. Privault
t 7 e
rt
0
rs ds
Xt
is a martingale. Finally, the identity

G Xt = IE
G Xs ,
IE
Nt
Ns
0 s t,
for all bounded Fs -measurable G, implies (10.8).
Next we will rephrase Proposition 10.2 in Proposition 10.3 using the Girsanov
theorem, which briefly recalled below.
Girsanov theorem
Recall that letting
"
t = IE
#

dP
F

t ,
dP
t [0, T ],
t )tR defined by
the Girsanov theorem, shows that the process (W
+
t = dWt
dW
1
dt dWt ,
t
t R+ ,
(10.9)
is a standard Brownian motion under P.

Next, Relation (10.2) shows that
"
#

dP
t = IE
F
t
dP

NT r T rs ds
e 0
= IE
N0
Nt r t rs ds
=
e 0
,
N0
hence
dt = t rt dt +

Ft
0 t T,
t
dNt ,
Nt
and Relation (10.9) becomes

t = dWt
dW
1
dNt dWt ,
Nt
t R+ .
(10.10)
See e.g. Theorem III-35 in [96].
320
"

The next proposition confirms the statement of Proposition 10.2, and in
See Exer t )tR under P.
addition it determines the precise dynamics of (X
+
cise 10.1 for another calculation based on geometric Brownian motion, and
Exercise 10.4 for an extension to correlated Brownian motions.
Proposition 10.3. Assume that (Xt )tR+ and (Nt )tR+ satisfy the stochastic differential equations
dXt = rt Xt dt + tX Xt dWt ,
where
have
(tX )tR+
(tN )tR+
and
and
dNt = rt Nt dt + tN Nt dWt ,
(10.11)
are Ft -adapted volatility processes. Then we
t = (tX tN )X
t dW
t.
dX
(10.12)
Proof. First we note that by (10.10) and (10.11),

t = dWt
dW
1
dNt dWt = dWt tN dt,
Nt
t R+ ,
Next, by Itos calculus we have


t = d Xt
dX
Nt
dXt
Xt
1
(dNt )2
=
2 dNt 2 dNt dXt + Xt
Nt
Nt
Nt
Nt3
1
Xt
=
(rt Xt dt + tX Xt dWt ) 2 (rt Nt dt + tN Nt dWt )
Nt
Nt
Xt Nt X N
| N |2 N 2
dt + Xt t 3 t dt
Nt2 t t
Nt
Xt N
Xt X N
| N |2
Xt X
dWt
dWt
dt + Xt t dt
=
Nt t
Nt t
Nt t t
Nt

Xt X
N
X N
N 2
=
dWt t dWt t t dt + |t | dt
Nt t
= Xt (tX tN )dWt Xt (tX tN )tN dt
t,
= Xt ( X N )dW
t
t = dWt tN dt, t R+ .
since dW
We end this section with a comment on inverse changes of measure.
Inverse Change of Measure
In the next proposition we compute conditional inverse density dP /dP.

"
321
N. Privault

w

t
dP Ft = N0 exp
IE
rs ds
0
Nt
dP
0 t T,
(10.13)
and the process

t 7
w

t
N0
exp
rs ds ,
0
Nt
0 t T,
is an Ft -martingale under P.
Proof. For all bounded and Ft -measurable random variables F we have,

F dP = IE [F ]
IE
dP

Nt
= IE F
Nt

w

T
NT
= IE F
exp
rs ds
t
Nt

w

t
N
0
F
= IE
exp
rs ds .
0
Nt

By Itos calculus and (10.11) we also have

1
1
1
d
= 2 dNt + 3 (dNt )2
Nt
Nt
Nt
1
| N |2
(rt Nt dt + tN Nt dWt ) + t dt
Nt2
Nt
N 2
1
t + tN dt)) + |t | dt
= 2 (rt Nt dt + tN Nt (dW
Nt
Nt
1
N
= (rt dt + t dWt ),
Nt
=
and

d
w

w

t
t
1
1
t,
exp
rs ds
=
exp
rs ds tN dW
0
0
Nt
Nt
which recovers the second part of Proposition 10.4, i.e. the martingale property of

w
t
1
t 7
exp
rs ds
0
Nt
under P.
322
"
10.3 Foreign Exchange

Currency exchange is a typical application of change of numeraire that illustrate the principle of absence of arbitrage.
Let Rt denote the foreign exchange rate, i.e. Rt is the (possibly fractional)
quantity of local currency that correspond to one unit of foreign currency.
Consider an investor that intends to exploit an overseas investment opportunity by
a) at time 0, changing one unit of local currency into 1/R0 units of foreign
currency,
b) investing 1/R0 on the foreign market at the rate rf to make the amount
f
etr /R0 until time t,
f
f
c) changing back etr /R0 into a quantity etr Rt /R0 of his local currency.
f
In other words, the foreign money market account etr is valued etr Rt on
the local (or domestic) market, and its discounted value on the local market
is
f
etr+tr Rt ,
t R+ .
f
The outcome of this investment will be obtained by comparing etr Rt /R0

to the amount ert that could have been obtained by investing on the local
market.
Taking
f
Nt := etr Rt ,
t R+ ,
(10.14)
as numeraire, absence of arbitrage is expressed by stating that the discounted

numeraire process
f
t 7 ert Nt = et(rr ) Rt
is a martingale under P , which is Assumption (A).
Next, we find a characterization of this arbitrage condition under the parameters of the model.
Proposition 10.5. Assume that the foreign exchange rate Rt satisfies a
stochastic differential equation of the form
dRt = Rt dt + Rt dWt ,
(10.15)
where (Wt )tR+ is a standard Brownian motion under P . Under the absence
of arbitrage Assumption (A) for the numeraire (10.14), we have
"
323
N. Privault
= r rf ,
(10.16)
hence the exchange rate process satisfies

dRt = (r rf )Rt dt + Rt dWt .
(10.17)
under P .
Proof. The equation (10.15) has solution
Rt = R0 et+Wt
t/2
t R+ ,
f
hence the discounted value of the foreign money market account etr on the
local market is
f
etr+tr Rt = R0 et(r
r+)+Wt 2 t/2
t R+ .
Under absence of arbitrage, et(rr ) Rt = etr Nt should be a martingale

under P and this holds provided rf r + = 0, which yields (10.16) and
(10.17).

As a consequence of Proposition 10.5, under absence of arbitrage a local
investor who buys a unit of foreign currency in the hope of a higher return
rf >> r will have to face a lower (or even more negative) drift
= r rf << 0
in his exchange rate Rt ,
The local money market account Xt := ert is valued ert /Rt on the foreign
market, and its discounted value on the foreign market is
t 7
et(rr
Rt
Xt
Nt
1 t(rrf )tWt +2 t/2
=
e
R0
1 t(rrf )tW
t 2 t/2
=
e
,
R0
=
(10.18)
where
1
dNt dWt
Nt
1
= dWt
dRt dWt
Rt
= dWt dt,
t R+ ,
t = dWt
dW
324
"

by (10.10). Under absence of arbitrage
is a standard Brownian motion under P
f
by
et(rr ) Rt is a martingale under P and (10.18) is a martingale under P
Proposition 10.2, which recovers (10.16).
Proposition 10.6. Under the absence of arbitrage condition (10.16), the
exchange rate 1/Rt satisfies

1
1

= (rf r) dt
d
d Wt ,
(10.19)
Rt
Rt
Rt
, where (Rt )tR is given by (10.17).
under P
+
Proof. By (10.16), the exchange rate 1/Rt is written by Itos calculus as

1
1
1
= 2 (Rt dt + Rt dWt ) + 3 2 Rt2 dt
d
Rt
Rt
Rt
1
= ( 2 ) dt
dWt
Rt
Rt

= dt
d Wt
Rt
Rt

1
dWt ,
= (rf r) dt
Rt
Rt
t is a standard Brownian motion under P.

where W
Consequently, under absence of arbitrage, a foreign investor who buys a unit

of the local currency in the hope of a higher return r >> rf will have to face
a lower (or even more negative) drift = rf r in his exchange rate 1/Rt
as written in (10.19) under P.
Foreign exchange options

We now price a foreign exchange option with payoff (RT )+ under P on
the exchange rate RT by the Black-Scholes formula as in the next proposition,
also known as the Garman-Kohlagen [37] formula.
Proposition 10.7. (Garman-Kohlagen formula). Consider an exchange rate
process (Rt )tR+ given by (10.17). The price of the foreign exchange call
option on RT with maturity T and strike is given by
e(T t)r IE [(RT )+ | Rt ] = e(T t) r Rt + (t, Rt ) e(T t)r (t, Rt ),
0 t T , where
"
325
N. Privault

+ (t, x) =
and

(t, x) =
log(x/) + (T t)(r rf + 2 /2)
T t
log(x/) + (T t)(r rf 2 /2)
T t
Proof. As a consequence of (10.17) we find the numeraire dynamics

f
dNt = d(etr Rt )
f
= rf etr Rt dt + etr dRt

f
= retr Rt dt + etr Rt dWt

= rNt dt + Nt dWt .
Hence a standard application of the Black-Scholes formula yields
f
e(T t)r IE [(RT )+ | Ft ] = e(T t)r IE [(eT r NT )+ | Ft ]

f
= e(T t)r eT r IE [(NT eT r )+ | Ft ]

f
=e
T r f
Nt
log(Nt eT r /) + (r + 2 /2)(T t)
T t
!!
f
log(Nt eT r /) + (r 2 /2)(T t)
T t

f
log(R
/)
+
(T
t)(r
rf + 2 /2)
t
= eT r Nt
T t

log(Rt /) + (T t)(r rf 2 /2)
T r f (T t)r
T t
eT r
(T t)r
= e(T t)r Rt + (t, Rt ) e(T t)r (t, Rt ).

Similarly, from (10.19) rewritten as
rt
e
ert
ert
= rf
dt dW
d
t,
Rt
Rt
Rt
a foreign exchange call option with payoff (1/RT )+ can be priced under P
in a Black-Scholes model by taking ert /Rt as underlying price, rf as risk-free
interest rate, and as volatility parameter. In this framework the BlackScholes formula (5.14) yields
"
+ #
f
1

e(T t)r IE
(10.20)
Rt
RT
326
"

+ #
erT

erT
Rt
RT

f
er(T t)
1
1
=
+ t,
e(T t)r t,
,
Rt
Rt
Rt
f
= e(T t)r erT IE
where

+ (t, x) =
and

(t, x) =
"
log(x/) + (T t)(rf r + 2 /2)
T t
log(x/) + (T t)(rf r 2 /2)
T t
(10.21)
Call/put duality for foreign exchange options

f
Let Nt = etr Rt , where Rt is an exchange rate with respect to a foreign currency and rf is the foreign market interest rate.
From Proposition 10.1 and (10.4) we have
"
"
+ #

+ #
1
1
1

RT
R
e(T t)r IE
Rt = Nt IE
Rt ,
T
f
eT r RT
and this yields the call/put duality
"
"
+ #

+ #
f
f
1
1

RT
e(T t)r IE
Rt = e(T t)r IE
Rt
RT
RT
#
"
+

f
1
R
= etr IE
Rt
T
f
eT r RT
"
+ #
1
trf (T t)r

=
e
IE
RT
Rt
Nt
"
+ #
(T t)r
1

=
e
IE
RT
(10.22)
Rt ,
Rt
between a call option with strike and a (possibly fractional) quantity /Rt
of put option(s) with strike 1/.
In the Black-Scholes case the duality (10.22) can be directly checked by
verifying that (10.20) coincides with
"
327
N. Privault
(T t)r
e
IE
Rt
"
1
RT
+ #

Rt
!+
f

(T t)r T rf eT r

T rf
=
e
e RT
e
IE
Rt
Rt
!+
f

(T t)r T rf eT r

=
NT
e
e
IE
Rt
Rt

f
e(T t)r p
(t, Rt ) e(T t)r Rt p+ (t, Rt )
=
Rt
f
e(T t)r p
(t, Rt ) e(T t)r p+ (t, Rt )
Rt

f
er(T t)
1
1
=
+ t,
e(T t)r t,
,
Rt
Rt
Rt
where
and

log(x) + (T t)(r rf t2 /2)
,
p (t, x) =
T t

log(x) + (T t)(r rf + 2 /2)
.
p+ (t, x) =
T t
10.4 Pricing of Exchange Options

t of forward prices as a
Based on Proposition 10.2 we model the process X
written as
continuous martingale under P,
t =
t,
dX
t dW
t R+ ,
(10.23)
and (
t )tR is a standard Brownian motion under P
where (W
t )tR+ is
+
an Ft -adapted process. The following lemma is a consequence of the Markov
t )tR and leads to the Margrabe formula of Propoproperty of the process (X
+
sition 10.8 below.
t )tR has the dynamics
Assuming that (X
+
t =
t )dW
t,
dX
t (X
(10.24)
where x 7
t (x) is a Lipschitz function, uniformly in t R+ , the Markov
t )tR shows that the price
property of the process (X
+
328
"

i
h rT

g(X
T ) Ft ,
T ) Ft = Nt IE
IE e t rs ds NT g(X
T ) can be written using a (meaof a vanilla option with payoff = NT g(X
x) of t and X
t on R+ R+ , cf. Theorem V-6-32 of [96]
surable) function C(t,
as
i
h r
T ) Ft .
X
t ) = IE e tT rs ds NT g(X
(10.25)
Nt C(t,
The next proposition states the Margrabe [76] formula for the pricing of
exchange options by the zero interest rate Black-Scholes formula. It will be
applied in particular in Proposition 12.2 below for the pricing of bond options.
t) =
t , i.e.
Proposition 10.8. (Margrabe formula). Assume that
t (X
(t)X
with de t )t[0,T ] is a geometric Brownian motion under P
the martingale (X
terministic volatility (
(t))t[0,T ] . Then we have
i
h rT
+
t ) Nt 0 (t, X
t ),
IE e t rs ds (XT NT ) Ft = Xt 0+ (t, X
(10.26)
t [0, T ], where

log(x/) v(t, T )
0+ (t, x) =
+
,
v(t, T )
2
and v 2 (t, T ) =
wT
t
0 (t, x) =
log(x/) v(t, T )
v(t, T )
2
2 (s)ds.
Proof. Taking g(x) = (x )+ in (10.25), the call option with payoff

T )+ ,
(XT NT )+ = NT (X
and floating strike NT is priced by (10.25) as
i
i
h rT
h rT

T )+ Ft
IE e t rs ds (XT NT )+ Ft = IE e t rs ds NT (X
i
h
(X
T )+ Ft
= Nt IE
X
t ),
= Nt C(t,
X
t ) is given by the Black-Scholes formula
where the function C(t,
x) = x0+ (t, x) 0 (t, x),
C(t,
t )t[0,T ] is a geometric Brownian motion which
with zero interest rate, since (X
and X
T is a lognormal random variable with variance
is a martingale under P,
wT
2 (s)ds. Hence we have

coefficient v 2 (t, T ) =
t
"
329
N. Privault
i
h rT
+
X
t)
IE e t rs ds (XT NT ) Ft = Nt C(t,
t 0+ (t, X
t ) Nt 0 (t, X
t ),
= Nt X
t R+ .
In particular, from Proposition 10.3 and (10.12), we can take

(t) =
when (tX )tR+ and (tN )tR+ are deterministic.
tX tN
Examples:
a) When the short rate process (r(t))t[0,T ] is a deterministic function and
rT
Nt = e t r(s)ds , 0 t T , Proposition 10.8 yields Mertons [78] zero

interest rate version of the Black-Scholes formula
i
h
rT
+
e t r(s)ds IE (XT ) Ft
rT

rT

rT
= Xt 0+ t, e t r(s)ds Xt e t r(s)ds 0 t, e t r(s)ds Xt ,
where (Xt )tR+ satisfies the equation
dXt
= r(t)dt +
(t)dWt ,
Xt
b) In the case of pricing under
t T , we get
h rT
+
IE e t rs ds (XT )
0 t T.
a forward numeraire, i.e. Nt = P (t, T ), 0

i

t ) P (t, T ) (t, X
t ),
Ft = Xt + (t, X
t R+ , since P (T, T ) = 1. In particular, when Xt = P (t, S) the above

formula allows us to price a bond call option on P (T, S) as
i
h rT
+
t )P (t, T ) (t, X
t ),
IE e t rs ds (P (T, S) ) Ft = P (t, S)+ (t, X
is given
t = P (t, S)/P (t, T ) under P
0 t T , provided the martingale X
by a geometric Brownian motion, cf. Section 12.2.
10.5 Hedging by Change of Num

eraire
In this section we reconsider and extend the Black-Scholes self-financing hedging strategies found in (6.20)-(6.21) and Proposition 6.7 of Chapter 6, and
use the stochastic integral representation of the forward payoff /NT and
change of numeraire to compute self-financing portfolio strategies. Our hedging portfolios will be built on the assets (Xt , Nt ), not on Xt and the money
330
"

rt
market account Bt = e 0 rs ds , extending the classical hedging portfolios that

are available in from the Black-Scholes formula, using a technique from [58],
cf. also [92].
Assume that the forward claim /NT L2 () has the stochastic integral
representation

w
T
+
t,
= IE
t dX
(10.27)
0
NT
NT
t )t[0,T ] is given by (10.23) and (t )t[0,T ] is a square-integrable
where (X
from which it follows that the forward claim price
adapted process under P,

Ft ,
Vt := IE
0 t T,
NT
that can be decomposed as
is a martingale under P,

w
t
+
s,
Vt = IE
s dX
0 t T.
0
NT
(10.28)
The next proposition extends the argument of [58] to the general framework of pricing using change of numeraire. Note that this result differs
from
rt
the standard formula that uses the money market account Bt = e 0 rs ds for
hedging instead of Nt , cf. e.g. [39] pages 453-454. The notion of self-financing
portfolio is similar to that of Definition 5.1.
t t , 0 t T , the portfolio
Proposition 10.9. Letting t = Vt X
(t , t )t[0,T ] with value
Vt = t Xt + t Nt ,
0 t T,
is self-financing in the sense that

dVt = t dXt + t dNt ,
and it hedges the claim , i.e.
i
h rT

Vt = t Xt + t Nt = IE e t rs ds Ft ,
0 t T.
(10.29)
Proof. In order to check that the portfolio (t , t )t[0,T ] hedges the claim
it suffices to check that (10.29) holds since by (10.4) the price Vt at time
t [0, T ] of the hedging portfolio satisfies
i
h rT

Vt = Nt Vt = IE e t rs ds Ft ,
0 t T.
Next, we show that the portfolio (t , t )t[0,T ] is self-financing. By numeraire
invariance, cf. e.g. page 184 of [95], we have
"
331
N. Privault
dVt = d(Nt Vt )
= Vt dNt + Nt dVt + dNt dVt
t + t dNt dX
t
= Vt dNt + Nt t dX
t dNt + Nt t dX
t + t dNt dX
t + (Vt t X
t )dNt
= t X
t ) + (Vt t X
t )dNt
= t d(Nt X
= t dXt + t dNt .

We now consider an application to the forward Delta hedging of European
T ) where g : R R and (X
t )tR has the
type options with payoff = g(X
+
Markov property as in (10.24), where
: R+ R. Assuming that the function
x) defined by
C(t,
i
h
g(X
T ) Ft = C(t,
X
t)
Vt := IE
is C 2 on R+ , we have the following corollary of Proposition 10.9.
t ), 0 t T , the
X
t) X
t C (t, X
Corollary 10.1. Letting t = C(t,
x
C

t ), t
portfolio
(t, X
with value
x
t[0,T ]
Vt = t Nt + Xt
C
t ),
(t, X
x
t R+ ,
T ).
is self-financing and hedges the claim = NT g(X
Proof. This result follows directly from Proposition 10.9 by noting that
the stochastic
by Itos formula, and the martingale property of Vt under P
integral representation (10.28) is given by
C
t ),
(t, X
t =
x
0 t T.

+
In the case of a call option with payoff function = (XT NT ) =

with
T )+ on the geometric Brownian motion (X
t )t[0,T ] under P
NT (X
t) =
t,
t (X
(t)X
where (
(t))t[0,T ] is a deterministic function, we have the following corollary
on the hedging of exchange options based on the Margrabe formula (10.26).
Corollary 10.2. The decomposition
332
"

i
h rT
+
t ) Nt 0 (t, X
t)
IE e t rs ds (XT NT ) Ft = Xt 0+ (t, X
t ), 0 (t, X
t ))t[0,T ] in (Xt , Nt )
yields a self-financing portfolio (0+ (t, X
+
that hedges the claim = (XT NT ) .
Proof. We apply Corollary 10.1 and the classical relation
C
(t, x) = 0+ (t, x),
x
x R,
x) = x0 (t, x) 0 (t, x).

for the function C(t,
+
Note that the Delta hedging method requires the computation of the func x) and that of the associated finite differences, and may not apply
tion C(t,
to path-dependent claims.
Examples:
a) When the short rate process (r(t))t[0,T ] is a deterministic function and
rT
Nt = e t r(s)ds , Corollary 10.2 yields the usual Black-Scholes hedging
strategy

r
t ), e 0T r(s)ds (t, Xt )
+ (t, X
t[0,T ]

rT
r
r
t ), e 0T r(s)ds 0 (t, e tT r(s)ds Xt )
= 0+ (t, e t r(s)ds X
t[0,T ]
rt
in (Xt , e
r(s)ds
), that hedges the claim = (XT ) .
b) In case Nt = P (t, T ) and Xt = P (t, S), 0 t T < S, Corollary 10.2

shows that a bond call option with payoff (P (T, S) )+ can be hedged
as
i
h rT
+
t )P (t, T ) (t, X
t)
IE e t rs ds (P (T, S) ) Ft = P (t, S)+ (t, X
by the self-financing portfolio
t ), (t, X
t ))t[0,T ]
(+ (t, X
t ) of the
in (P (t, S), P (t, T )), i.e. one needs to hold the quantity + (t, X
t ) of the bond
bond maturing at time S, and to short a quantity (t, X
maturing at time T .
"
333
N. Privault
Exercises
Exercise 10.1 Let (Bt )tR+ be a standard Brownian motion started at 0

under the risk-neutral measure P . Consider a numeraire (Nt )tR+ given by
Nt := N0 eBt
t/2
t R+ ,
and a risky asset (Xt )tR+ given by

Xt := X0 eBt
t/2
t R+ .
denote the forward measure relative to the numeraire (Nt )tR , under
Let P
+
t := Xt /Nt of forward prices is known to be a martingale.
which the process X
1. Using the Ito formula, compute

t = d(Xt /Nt ) = (X0 /N0 )d e()Bt (2 2 )t/2 .
dX
2. Explain why the exchange option price IE[(XT NT )+ ] at time 0 has
the Black-Scholes form
erT IE[(XT NT )+ ]
!
/)
T
log(X
0
+
N0
= X0
2
(10.30)
!
T
log(X0 /)
.
2
Hints:
(i) Use the change of numeraire identity
X
T )+ ].
erT IE[(XT NT )+ ] = N0 IE[(
t is a martingale under the forward measure P

(ii) The forward price X
relative to the numeraire (Nt )tR+ .
3. Give the value of
in terms of and .
Exercise 10.2 Bond options. Consider two bonds with maturities T and S,
with prices P (t, T ) and P (t, S) given by
dP (t, T )
= rt dt + tT dWt ,
P (t, T )
and
dP (t, S)
= rt dt + tS dWt ,
P (t, S)
334
"

where ( T (s))s[0,T ] and ( S (s))s[0,S] are deterministic functions.
1. Show, using Itos formula, that

P (t, S)
P (t, S) S
t,
=
( (t) T (t))dW
d
P (t, T )
P (t, T )
t )tR is a standard Brownian motion under P.

where (W
+
2. Show that
w

wT
T
P (t, S)
s 1
exp
( S (s) T (s))dW
| S (s) T (s)|2 ds .
P (T, S) =
t
P (t, T )
2 t
denote the forward measure associated to the numeraire
Let P
Nt := P (t, T ),
0 t T.
3. Show that for all S, T > 0 the price at time t

i
h rT

IE e t rs ds (P (T, S) )+ Ft
of a bond call option on P (T, S) with payoff (P (T, S) )+ is equal to
i
h rT

IE e t rs ds (P (T, S) )+ Ft
(10.31)

1
P (t, S)
1
P (t, S)
v
v
+ log
P (t, T ) + log
,
= P (t, S)
2 v
P (t, T )
2 v
P (t, T )
where
v2 =
wT
t
| S (s) T (s)|2 ds.
4. Compute the self-financing hedging strategy that hedges the bond option
using a portfolio based on the assets P (t, T ) and P (t, S).
Exercise 10.3 Consider two risky assets S1 and S2 modeled by the geometric
Brownian motions
S1 (t) = e1 Wt +t
S2 (t) = e2 Wt +t ,
and
t R+ ,
(10.32)
where (Wt )tR+ is a standard Brownian motion under P.

1. Find a condition on r, and 2 so that the discounted price process
ert S2 (t) is a martingale under P.
2. Assume that r = 22 /2, and let
2
Xt = e(2 1 )t/2 S1 (t),
t R+ .
Show that the discounted process ert Xt is a martingale under P.

"
335
N. Privault
3. Taking Nt = S2 (t) as numeraire, show that the forward process X(t)

=
defined by
Xt /Nt is a martingale under the forward measure P
dP
NT
= erT
.
dP
N0
Recall that
t := Wt 2 t
W

4. Using the relation
1 (T ) S2 (T ))+ /NT ],
erT IE[(S1 (T ) S2 (T ))+ ] = N0 IE[(S
compute the price
erT IE[(S1 (T ) S2 (T ))+ ].
of the exchange option on the assets S1 and S2 .
Exercise 10.4 Extension of Proposition 10.3 to correlated Brownian motions.
Assume that (St )tR+ and (Nt )tR+ satisfy the stochastic differential equations
dSt = rt St dt + tS St dWtS ,
and dNt = t Nt dt + tN Nt dWtN ,
where (WtS )tR+ and (WtN )tR+ have the correlation

dWtS dWtN = dt,
where [1, 1].
1. Show that (WtN )tR+ can be written as
WtN = WtS +
1 2 Wt ,
t R+ ,
where (Wt )tR+ is a standard Brownian motion under P , independent

of (WtS )tR+ .
2. Letting Xt = St /Nt , show that dXt can be written as
dXt = (rt t + (tN )2 tN tS )Xt dt +
t Xt dWtX ,
where WtX is a standard Brownian motion under P and
t is to be
computed.
Exercise 10.5 Quanto options (Exercise 9.5 in [109]). Consider an asset priced
St at time t, with
dSt = rSt dt + S St dWtS ,
336
"

and an exchange rate (Rt )tR+ given by
dRt = (r rf )Rt dt + R Rt dWtR ,
from (10.16) in Proposition 10.5, where (WtR )tR+ is written as
WtR = WtS +
p
1 2 W t ,
t R+ ,
where (Wt )tR+ is a standard Brownian motion under P , independent of

(WtS )tR+ , i.e. we have
dWtR dWtS = dt,
where is a correlation coefficient.
1. Let
a = r rf + R S ( R )2
at
and Xt = e St /Rt , t R+ , and show by Exercise 10.4 that dXt can be

written as
dXt = rXt dt +
Xt dWtX ,
where (WtX )tR+ is a standard Brownian motion under P and
is to be
computed.
2. Compute the price
"
+ #
ST

er(T t) IE
Ft
RT
at time t of a quanto option.
"
337
Chapter 11
Forward Rate Modeling
This chapter is concerned with interest rate modeling, in which the mean
reversion property plays an important role. We consider the main short rate
models (Vasicek, CIR, CEV, affine models) and the computation of bond
prices in such models. Next we consider the modeling of forward rates in the
HJM and BGM models, as well as in two-factor models.
11.1 Short Term Models

Mean Reversion
The first model to capture the mean reversion property of interest rates, a
property not possessed by geometric Brownian motion, is the Vasicek [113]
model, which is based on the Ornstein-Uhlenbeck process. Here, the short
term interest rate process (rt )tR+ solves the equation
drt = (a brt )dt + dBt ,
(11.1)
where (Bt )tR+ is a standard Brownian motion, with solution

wt
a
rt = r0 ebt + (1 ebt ) + eb(ts) dBs ,
0
b
t R+ .
(11.2)
The law of rt is Gaussian at all times t, with mean r0 ebt + a(1 ebt )/b
and variance
2
wt
0
(eb(ts) )2 ds = 2
wt
0
e2bs ds =
2
(1 e2bt ),
2b
t R+ .
This model has the interesting properties of being statistically stationary in

time in the long run, and to admit a Gaussian N (a/b, 2 /(2b)) invariant dis"
N. Privault
tribution when b > 0, however its drawback is to allow for negative values of
rt .
Figure 11.1 presents a random simulation of t 7 rt in the Vasicek model
with r0 = a/b = 5%, i.e. the reverting property of the process is with respect
to its initial value r0 = 5%. Note that the interest rate in Figure 11.1 may
become negative, which is unusual for interest rates but may nevertheless
happen.
0.07
0.065
0.06
0.055
0.05
0.045
0.04
0.035
0.03
0.025
10
15
20
Fig. 11.1: Graph of t 7 rt in the Vasicek model.

The Cox-Ingersoll-Ross (CIR) [16] model brings a solution to the positivity problem encountered with the Vasicek model, by the use the nonlinear
equation
1/2
drt = ( rt )dt + rt dBt ,
, 0.
Note that rt remains strictly positive under the Feller condition 2 > 2 ,
cf. the study of the associated probability density in Lemma 4 of [32].
Other classical mean reverting models include the Courtadon (1982) model
drt = ( rt )dt + rt dBt ,
where , , are nonnegative, and the exponential Vasicek model
drt = rt ( a log rt )dt + rt dBt ,
where a, , are nonnegative.
340
"

Constant Elasticity of Variance (CEV)
Constant Elasticity of Variance models are designed to take into account nonconstant volatilities that can vary as a power of the underlying asset. The
Marsh-Rosenfeld (1983) model
(1)
drt = (rt
/2
+ rt )dt + rt
dBt
where , , , are nonnegative constants and is the variance (or diffusion)

elasticity coefficient, covers most of the CEV models. For = 1 this is the
CIR model, and for = 0 we get the standard CEV model
/2
drt = rt dt + rt
dBt .
If = 2 this yields the Dothan [24] model

drt = rt dt + rt dBt .
Affine Models
The class of short rate interest rate models admits a number of generalizations
that can be found in the references quoted in the introduction of this chapter,
among which is the class of affine models of the form
p
drt = ((t) + (t)rt )dt + (t) + (t)rt dBt .
(11.3)
Such models are called affine because the associated zero-coupon bonds can
be priced using an affine PDE of the type (11.11) below, as will be seen after
Proposition 11.2.
They also include the Ho-Lee model
drt = (t)dt + dBt ,
where (t) is a deterministic function of time, as an extension of the Merton
model drt = dt + dBt , and the Hull-White model
drt = ((t) (t)rt )dt + (t)dBt
which is a time-dependent extension of the Vasicek model.
"
341
N. Privault
11.2 Zero-Coupon Bonds

A zero-coupon bond is a contract priced P0 (t, T ) at time t < T to deliver
P0 (T, T ) = 1$ at time T . The computation of the arbitrage price P0 (t, T ) of
a zero-coupon bond based on an underlying short term interest rate process
(rt )tR+ is a basic and important issue in interest rate modeling.
In case the short term interest rate process (rt )tR+ is a deterministic
function of time, a standard arbitrage argument shows that the price P (t, T )
of the bond is given by
P (t, T ) = e
rT
t
rs ds
0 t T.
(11.4)
In case (rt )tR+ is an Ft -adapted random process the formula (11.4) is no

longer valid as it relies on future information, and we replace it with
i
h rT

0 t T,
(11.5)
under a risk-neutral measure P . It is natural to write P (t, T ) as a conditional
expectation under a martingale measure, as the use of conditional expectation
wT
helps to filter out the future information past time t contained in
rs ds.
t
Expression (11.5)
makes sense as the best possible estimate of the future
rT
quantity e t rs ds in mean square sense, given information known up to time
t.
Pricing bonds with non-zero coupon is not difficult in the case of a deterministic continuous-time coupon yield at rate c > 0. In this case the price
Pc (t, T ) of the coupon bound is given by
Pc (t, T ) = ec(T t) P0 (t, T ),
0 t T,
In the sequel we will only consider zero-coupon bonds, and let P (t, T ) =
P0 (t, T ), 0 t T .
The following proposition shows that Assumption (A) of Chapter 10 is
satisfied, i.e. the bond price process t 7 P (t, T ) can be taken as a numeraire.
Proposition 11.1. The discounted bond price process
t 7 e
rt
0
rs ds
P (t, T )
is a martingale under P .
Proof. We have
342
"

e
rt
0
rs ds
i
h rT

IE e t rs ds Ft
i
h rt
rT

= IE e 0 rs ds e t rs ds Ft
i
h rT

= IE e 0 rs ds Ft
P (t, T ) = e
rt
0
rs ds
and this suffices to conclude since by the tower property (16.25) of conditional expectations, any process of the form t 7 IE [F | Ft ], F L1 (), is
a martingale, cf. Relation (6.1).
Bond pricing PDE

We assume from now on that the underlying short rate process is solution
to the stochastic differential equation
drt = (t, rt )dt + (t, rt )dBt
(11.6)
where (Bt )tR+ is a standard Brownian motion under P .

Since all solutions of stochastic differential equations such as (11.6) have the
Markov property, cf e.g. Theorem V-32 of [96], the arbitrage price P (t, T )
can be rewritten as a function F (t, rt ) of rt , i.e.
i
i
h rT
h rT

P (t, T ) = IE e t rs ds Ft = IE e t rs ds rt = F (t, rt ),
and depends on rt only instead of depending on all information available in
Ft up to time t, meaning that the pricing problem can now be formulated as
a search for the function F (t, x).
Proposition 11.2. (Bond pricing PDE) The bond pricing PDE for P (t, T ) =
F (t, rt ) is written as
xF (t, x) =
F
F
1
2F
(t, x) + (t, x)
(t, x) + 2 (t, x) 2 (t, x), (11.7)
t
x
2
x
t R+ , x R, subject to the terminal condition

F (T, x) = 1,
x R.
(11.8)
Proof. From Itos formula we have

rt

rt
rt
d e 0 rs ds P (t, T ) = rt e 0 rs ds P (t, T )dt + e 0 rs ds dP (t, T )
= rt e
"
rt
0
rs ds
F (t, rt )dt + e
rt
0
rs ds
dF (t, rt )
343

N. Privault
= rt e
rt
rs ds
F (t, rt )dt + e
rt
rs ds F
(t, rt )((t, rt )dt + (t, rt )dBt )

x

1 2
F
F
+ e 0 rs ds
(t, rt ) 2 (t, rt )dt +
(t, rt )dt
2
x
t
r
F
0t rs ds
=e
(t, rt )
(t, rt )dBt
x

rt
1
F
2F
F
+ e 0 rs ds rt F (t, rt ) + (t, rt )
(t, rt ) + 2 (t, rt ) 2 (t, rt ) +
(t, rt ) dt.
x
2
x
t
(11.9)
rt
rt
Given that t 7 e 0 rs ds P (t, T ) is a martingale, the above expression (11.9)

should only contain terms in dBt (cf. Corollary II-1 of [96]), and all terms in
dt should vanish inside (11.9). This leads to the identities
r F (t, r ) + (t, r ) F (t, r ) + 1 2 (t, r ) F (t, r ) + F (t, r ) = 0
t
t
t
t
t
t
t
x
2
x2
t
rt

rt
d e 0 rs ds P (t, T ) = e 0 rs ds (t, rt )
(t, rt )dBt .
x
(11.10a)
Condition (11.8) is due to the fact that P (T, T ) = $1.
In the case of an interest rate process modeled by (11.3) we have

p
(t, x) = (t) + (t)x
and
(t, x) = (t) + (t)x,
hence (11.7) yields the affine PDE
xF (t, x) =
F
1
2F
F
(t, x) + ((t) + (t)x)
(t, x) + ((t) + (t)x) 2 (t, x),
t
x
2
x
(11.11)
t R+ , x R. By (11.10a), the above proposition also shows that

rt

rt
dP (t, T )
1
=
d e 0 rs ds e 0 rs ds P (t, T )
P (t, T )
P (t, T )

rt

rt
1
=
rt P (t, T )dt + e 0 rs ds d e 0 rs ds P (t, T )
P (t, T )
rt

rt
1
= rt dt +
e 0 rs ds d e 0 rs ds P (t, T )
P (t, T )
1
F
= rt dt +
(t, rt )(t, rt )dBt
F (t, rt ) x
344
"
= rt dt + (t, rt )
log F
(t, rt )dBt .
x
(11.12)
In the Vasicek case

drt = (a brt )dt + dWt ,
the bond price takes the form P (t, T ) = eA(T t)+C(T t)rt , cf. (11.15) below,
and by (11.12) we find
dP (t, T )
= rt dt (1 eb(T t) )dWt .
P (t, T )
b
(11.13)
Note that more generally, all affine short rate models as defined in Relation (11.3), including the Vasicek model, will yield a bond pricing formula of
the form
P (t, T ) = eA(T t)+C(T t)rt ,
cf. e.g. 3.2.4. of [9].
Probabilistic PDE Solution

Next we solve the PDE (11.7) by a direct computation of the conditional
expectation
i
h rT

(11.14)
in the Vasicek [113] model
drt = (a brt )dt + dBt ,
i.e. when the short rate (rt )tR+ has the expression
rt = g(t) +
where
wt
0
wt
a
h(t, s)dBs = r0 ebt + (1 ebt ) + eb(ts) dBs ,
0
b
a
g(t) := r0 ebt + (1 ebt ),
b
t R+ ,
and
h(t, s) := eb(ts) ,
0 s t,
are deterministic functions.

Letting ut = max(u, t), using the fact that Wiener integrals are Gaussian
random variables and the Gaussian moment generating function, we have
i
h rT

P (t, T ) = IE e t rs ds Ft
i
h rT
rs

= IE e t (g(s)+ 0 h(s,u)dBu )ds Ft
"
345
N. Privault
w
T
= exp
t
w
T
= exp
t
w
T
= exp
t
w
T
= exp
t
w
T
= exp

i
h rT rs

g(s)ds IE e t 0 h(s,u)dBu ds Ft

i
h rT rT

g(s)ds IE e 0 ut h(s,u)dsdBu Ft

i
h rT rT
wtwT

g(s)ds
h(s, u)dsdBu IE e t ut h(s,u)dsdBu Ft
0 ut

i
h rT rT
wtwT

g(s)ds
h(s, u)dsdBu IE e t u h(s,u)dsdBu Ft
0 t

h rT rT
i
wtwT
g(s)ds
h(s, u)dsdBu IE e t u h(s,u)dsdBu
t
0 t

2 !
wT
wtwT
1wT wT
= exp
g(s)ds
h(s, u)dsdBu +
h(s, u)ds du
t
0 t
u
2 t
w

w
w
t T
T
a
= exp
(r0 ebs + (1 ebs ))ds
eb(su) dsdBu
0 t
t
b

2 !
2 w T w T b(su)
e
ds du
exp
u
2 t
w

wt
T
a
= exp
(r0 ebs + (1 ebs ))ds (1 eb(T t) ) eb(tu) dBu
0
t
b
b

2 !
2 w T 2bu ebu ebT
exp
e
du
2 t
b

rt
1
a
= exp (1 eb(T t) ) + (1 eb(T t) )(r0 ebt + (1 ebt ))
b
b
b
bu
2 !
wT
2 wT
e
ebT
a
e2bu
du
exp
(r0 ebs + (1 ebs ))ds +
t
b
2 t
b
= eC(T t)rt +A(T t) ,
where
(11.15)
1
C(T t) := (1 eb(T t) ),
b
and
A(T t) :=
4ab 3 2 2 2ab
2 ab b(T t) 2 2b(T t)
+
(T t) +
e
3e
.
4b3
2b2
b3
4b
Analytical PDE Solution

In order to solve the PDE (11.7) analytically we may look for a solution of
the form
F (t, x) = eA(T t)+xC(T t) ,
(11.16)
346
"

where A and C are functions to be determined under the conditions A(0) = 0
and C(0) = 0. Plugging (11.16) into the PDE (11.7) yields the system of
Riccati and linear differential equations
2 2
A0 (s) = aC(s)
C (s)
2
C 0 (s) = bC(s) + 1,
which can be solved to recover the above value of P (t, T ).
Some Bond Price Simulations

In this section we consider again the Vasicek model, in which the short rate
(rt )tR+ is solution to (11.1). Figure 11.2 presents a random simulation of
t 7 P (t, T ) in the same Vasicek model. The graph of the corresponding
deterministic bond price obtained for a = b = = 0 is also shown on the
Figure 11.2.
1.1
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
10
15
20
Fig. 11.2: Graphs of t 7 P (t, T ) and t 7 er0 (T t) .

Figure 11.3 presents a random simulation of t 7 P (t, T ) for a non-zero
coupon bond with price Pc (t, T ) = ec(T t) P (t, T ), and coupon rate c > 0,
0 t T.
"
347
N. Privault
108.00
106.00
104.00
102.00
100.00
10
15
20
Fig. 11.3: Graph of t 7 P (t, T ) for a bond with a 2.3% coupon.
The above simulation can be compared to the actual market data of a coupon
bond in Figure 11.4 below.
Fig. 11.4: Bond price graph with maturity 01/18/08 and coupon rate 6.25%.
See Exercise 11.3 for bond a bond pricing formula in the CIR model.
Bond pricing in the Dothan model

In the Dothan [24] model, the short term interest rate process (rt )tR+ is
modeled according to a geometric Brownian motion
drt = rt dt + rt dBt ,
(11.17)
where the volatility > 0 and the drift R are constant parameters and
(Bt )tR+ is a standard Brownian motion. In this model the short term inter348
"

est rate rt remains always positive, while the proportional volatility term rt
accounts for the sensitivity of the volatility of interest rate changes to the
level of the rate rt .
On the other hand, the Dothan model is the only lognormal short rate
model that allows for an analytical formula for the zero coupon bond price
i
h rT

0 t T.
For convenience of notation we let p = 1 2/ 2 and rewrite (11.17) as
1
drt = (1 p) 2 rt dt + rt dBt ,
2
with solution

rt = r0 exp Bt p 2 t/2 ,
t R+ ,
(11.18)
where p/2 identifies to the market price of risk. By the Markov property of
(rt )tR+ , the bond price P (t, T ) is a function F (, rt ) of rt and of the time
to maturity = T t:
i
h rT

P (t, T ) = F (, rt ) = IE e t rs ds rt ,
0 t T.
(11.19)
By computation of the conditional expectation (11.19) using (8.48) we easily
obtain the following result, cf. [86], where the function (v, t) is defined in
(8.47).
Proposition 11.3. The zero-coupon bond price P (t, T ) = F (T t, rt ) is
given for all p R by
!

ww
1 + z2
4z 2 du dz
2 p2 /8
ux
,
,
F (, x) = e
e
exp 2
2
2
0
0
u
u 4
u z p+1
(11.20)
x > 0.
Proof. By Proposition 8.10 the probability distribution of the time integral
w T t
2
eBs p s/2 ds is given by
0
w
T t

2
eBs p s/2 ds dy
w

t
2
P
eBs p s/2 ds dy, Bt pt/2 dz dz
0

z/2 2
w pz/2p2 2 t/8
1 + ez
4e
t dy
=
e
exp 2
,
dz
2
2 y
2 y
4
y
"
349
N. Privault
= ep
2 (T t)/8
w
0

1 + z2
4z 2 (T t)
dz dy
exp 2 2
,
,
2
y
y
4
z p+1 y
y > 0,
where the exchange of integrals is justified by the Fubini theorem and the
non-negativity of integrands. Hence by (8.48) and (11.18) we find
F (T t, rt ) = P (t, T )

w

T

= IE exp
rs ds Ft
t

wT
2

= IE exp rt
e(Bs Bt ) p(st)/2 ds Ft
t

wT
2
= IE exp x
e(Bs Bt ) p(st)/2 ds
t

w T t
2
= IE exp x
eBs ps/2 ds
0
w
x=rt
x=rt

ds dy
0
0

w
w
2 2
4z 2 (T t)
dz dy
1 + z2
,
.
= ep (T t)/8
ert y
exp 2 2
2
0
0
y
y
4
z p+1 y
=
rt y
T t
Bs p 2 s/2

See [86] and [85] for more results on bond pricing in the Dothan model, and
[93] for numerical computations.
11.3 Forward Rates

A forward interest rate contract gives its holder a loan decided at present
time t and to be delivered over a future period of time [T, S] at a rate denoted by f (t, T, S), t T S, and called a forward rate.
Let us determine the arbitrage or fair value of this rate using the instruments available in a bond market, which are bonds priced at P (t, T ) for
various maturity dates T > t.
The loan can be realized using the bonds available on the market by proceeding in two steps:
1) at time t, borrow the amount P (t, S) by shortselling one unit of bond
with maturity S, which will mean refunding $1 at time S.
350
"

2) since one only needs the money at time T , it makes sense to invest
the amount P (t, S) over the period [t, T ] by buying a (possibly fractional)
quantity P (t, S)/P (t, T ) of a bond with maturity T priced P (t, T ) at time
t. This will yield the amount
$1
P (t, S)
P (t, T )
at time T .
As a consequence the investor will receive P (t, S)/P (t, T ) at time T , and will
refund $1 at time S.
The corresponding forward rate f (t, T, S) is then given by the relation
P (t, S)
exp ((S T )f (t, T, S)) = $1,
P (t, T )
0 t T S,
(11.21)
where we used exponential compounding, which leads to the following definition (11.22).
Definition 11.1. The forward rate f (t, T, S) at time t for a loan on [T, S]
is given by
f (t, T, S) =
log P (t, T ) log P (t, S)

.
ST
(11.22)
The spot forward rate f (t, t, T ) is given by

f (t, t, T ) =
log P (t, T )
,
T t
or P (t, T ) = e(T t)f (t,t,T ) ,
0 t T,
(11.23)
and is also called the yield.
"
351
N. Privault
Figure 11.5 presents a typical forward rate curve on the LIBOR (London
Interbank Offered Rate) market with t =07 may 2003, = six months.
5
Forward interest rate
4.5
3.5
2.5
2
0
10
15
years
20
25
30
TimeSerieNb
AsOfDate
2D
1W
1M
2M
3M
1Y
2Y
3Y
4Y
5Y
6Y
7Y
8Y
9Y
10Y
11Y
12Y
13Y
14Y
15Y
20Y
25Y
30Y
505
7mai03
2,55
2,53
2,56
2,52
2,48
2,34
2,49
2,79
3,07
3,31
3,52
3,71
3,88
4,02
4,14
4,23
4,33
4,4
4,47
4,54
4,74
4,83
4,86
Fig. 11.5: Graph of T 7 f (t, T, T + ).
The instantaneous forward rate f (t, T ) is defined by taking the limit of

f (t, T, S) as S & T , i.e.
f (t, T ) : = lim f (t, T, S)
S&T
log P (t, S) log P (t, T )

ST
log P (t, T + ) log P (t, T )
= lim
&0
log P (t, T )
=
T
1
P (t, T )
=
.
P (t, T ) T
= lim
S&T
(11.24)
The above equation can be viewed as a differential equation to be solved for

log P (t, T ) under the initial condition P (T, T ) = 1, which yields the following
proposition.
w

T
P (t, T ) = exp
f (t, s)ds ,
t
0 t T.
(11.25)
Proof. We check that

log P (t, T ) = log P (t, T ) log P (t, t) =
wT
w T log P (t, s)
ds =
f (t, s)ds.
t
t
s
352
"

As a consequence of (11.21) and (11.25) the forward rate f (t, T, S), 0 t
T S, can be recovered from (11.22) and the instantaneous forward rate
f (t, s), as:
log P (t, T ) log P (t, S)
ST
w

wS
T
1
=
f (t, s)ds
f (t, s)ds
t
t
ST
1 wS
=
f (t, s)ds,
0 t T < S.
ST T
f (t, T, S) =
(11.26)
Forward Swap Rates

The first interest rate swap occured in 1981 between IBM and the World
Bank. In particular, an interest rate swap makes it possible to exchange a
sequence of variable forward rates f (t, Tk , Tk+1 ), k = 1, . . . , n 1, against a
fixed rate over a time period [T1 , Tn ]. Over the succession of time intervals
[T1 , T2 ], . . . , [Tn1 , Tn ] defining a tenor structure, see Section 12.1 for details,
the combination of such exchanges will generate a cumulative discounted cash
flow
!
!
n1
n1
r Tk+1
r Tk+1
X
X
rs ds
rs ds
(Tk+1 Tk )e t
f (t, Tk , Tk+1 )
(Tk+1 Tk )e t
k=1
k=1
n1
X
(Tk+1 Tk )e
r Tk+1
t
rs ds
(f (t, Tk , Tk+1 ) ),
k=1
at time t, in which we used linear interest rate compounding. This cash flow
is used to make the contract fair, and it can be priced at time t as
"n1
#

rT
X

t k+1 rs ds
IE
(Tk+1 Tk )e
(f (t, Tk , Tk+1 ) ) Ft
k=1
n1
X
rT

k+1
rs ds
(Tk+1 Tk )(f (t, Tk , Tk+1 ) ) IE e t
Ft
k=1
n1
X
(Tk+1 Tk )P (t, Tk+1 )(f (t, Tk , Tk+1 ) ).
k=1
The swap rate S(t, T1 , Tn ) is by definition the value of that cancels this
cash flow and makes the contract fair. In other words, S(t, T1 , Tn ) is the rate
over [T1 , Tn ] that will be agreed in exchange for the family of forward rates
"
353
N. Privault
f (t, Tk , Tk+1 ), k = 1, . . . , n 1, and it satisfies
n1
X
(Tk+1 Tk )P (t, Tk+1 )(f (t, Tk , Tk+1 ) S(t, T1 , Tn )) = 0,
(11.27)
k=1
and is given by
S(t, T1 , Tn ) =
n1
X
1
(Tk+1 Tk )P (t, Tk+1 )f (t, Tk , Tk+1 ), (11.28)
P (t, T1 , Tn )
k=1
where
P (t, T1 , Tn ) =
n1
X
(Tk+1 Tk )P (t, Tk+1 ),
0 t T1 ,
k=1
is the annuity numeraire.
LIBOR Rates
Recall that the forward rate f (t, T, S), 0 t T S, is defined using
exponential compounding, from the relation
f (t, T, S) =

.
ST
(11.29)
In order to compute swaption prices one prefers to use forward rates as defined on the London InterBank Offered Rates (LIBOR) market instead of the
standard forward rates given by (11.29).
The forward LIBOR L(t, T, S) for a loan on [T, S] is defined using linear
compounding, i.e. by replacing (11.29) with the relation
1 + (S T )L(t, T, S) =
P (t, T )
,
P (t, S)
which yields the following definition.

Definition 11.2. The forward LIBOR rate L(t, T, S) at time t for a loan on
[T, S] is given by

1
P (t, T )
L(t, T, S) =
1 ,
0 t T < S.
(11.30)
S T P (t, S)
Note that (11.30) above yields the same formula for the instantaneous forward
rate
354
"

f (t, T ) : = lim L(t, T, S)
S&T
P (t, S) P (t, T )
(S T )P (t, S)
P (t, T + ) P (t, T )
lim
&0
P (t, T + )
1
P (t, T + ) P (t, T )
lim
P (t, T ) &0
1
P (t, T )
P (t, T ) T
log P (t, T )
,
T
= lim
S&T
=
=
=
=
as (11.24).
In addition, Relation (11.30) shows that the LIBOR rate can be viewed
t = Xt /Nt with numeraire Nt = (S T )P (t, S) and
as a forward price X
Xt = P (t, T ) P (t, S), according to Relation (10.7) of Chapter 10. As a
consequence, from Proposition 10.2, the LIBOR rate (L(t, T, S))t[T,S] is a
defined by
martingale under the forward measure P
rS
dP
1
=
e 0 rt dt .
dP
P (0, S)
LIBOR Swap Rates

The LIBOR swap rate S(t, T1 , Tn ) satisfies the same relation as (11.27) with
the forward rate f (t, Tk , Tk+1 ) replaced with the LIBOR rate L(t, Tk , Tk+1 ),
i.e.
n1
X
(Tk+1 Tk )P (t, Tk+1 )(L(t, Tk , Tk+1 ) S(t, T1 , Tn )) = 0.
k=1

S(t, T1 , Tn ) =
P (t, T1 ) P (t, Tn )
,
P (t, T1 , Tn )
0 t T1 .
(11.31)
Proof. By (11.28) and (11.30) we have

n1
X
1
(Tk+1 Tk )P (t, Tk+1 )L(t, Tk , Tk+1 )
P (t, T1 , Tn )
k=1

n1
X
1
P (t, Tk )
=
P (t, Tk+1 )
1
P (t, T1 , Tn )
P (t, Tk+1 )
S(t, T1 , Tn ) =
k=1
"
355
N. Privault
n1
X
1
(P (t, Tk ) P (t, Tk+1 ))
P (t, T1 , Tn )
P (t, T1 ) P (t, Tn )
P (t, T1 , Tn )
k=1
by a telescoping sum.
(11.32)
Clearly, a simple expression for the swap rate such as that of Proposition 11.5
cannot be obtained using the standard (i.e. non-LIBOR) rates defined in
(11.29).
When n = 2, the swap rate S(t, T1 , T2 ) coincides with the forward rate
L(t, T1 , T2 ):
S(t, T1 , T2 ) = L(t, T1 , T2 ),
(11.33)
and the bond prices P (t, T1 ) can be recovered from the forward swap rates
S(t, T1 , Tn ).
Similarly to the case of LIBOR rates, Relation (11.31) shows that the
LIBOR swap rate can be viewed as a forward price with (annuity) numeraire
Nt = P (t, T1 , Tn ) and Xt = P (t, T1 ) P (t, Tn ). Consequently the LIBOR
swap rate (S(t, T1 , Tn )t[T,S] is a martingale under the forward measure P

defined from (10.1) by
dP
P (T1 , T1 , Tn ) r0T1 rt dt
=
e
.
dP
P (0, T1 , Tn )
11.4 The HJM Model

In the previous chapter we have focused on the modeling of the short rate
(rt )tR+ and on its consequences on the pricing of bonds P (t, T ), from which
the forward rates f (t, T, S) and L(t, T, S) have been defined.
In this section we choose a different starting point and consider the problem of directly modeling the instantaneous forward rate f (t, T ). The graph
given in Figure 11.4 presents a possible random evolution of a forward interest
rate curve using the Musiela convention, i.e. we will write
g(x) = f (t, t + x) = f (t, T ),
under the substitution x = T t, x 0, and represent a sample of the
instantaneous forward curve x 7 f (t, t + x) for each t R+ .
356
"

Forward rate
5
4.5
4
3.5
3
2.5
2
1.5
1
0.5
20
15
10
0
10
x
5
15
20
Fig. 11.6: Stochastic process of forward curves.

In the HJM model, the instantaneous forward rate f (t, T ) is modeled under
P by a stochastic differential equation of the form
dt f (t, T ) = (t, T )dt + (t, T )dBt ,
(11.34)
where t 7 (t, T ) and t 7 (t, T ), 0 t T , are allowed to be random (adapted) processes. In the above equation, the date T is fixed and the
differential dt is with respect to t.
Under basic Markovianity assumptions, a HJM model with deterministic
coefficients (t, T ) and (t, T ) will yield a short rate process (rt )tR+ of the
form
drt = (a(t) b(t)rt )dt + (t)dBt ,
cf. 6.6 of [90], which is the [52] Hull-White model, cf. Section 11.1, with
explicit solution
wt
w t rt
rt
rt
(u)e u b( )d dBu ,
e u b( )d a(u)du +
rt = rs e s b( )d +
s
0 s t.
The HJM Condition

How to encode absence of arbitrage in the defining equation (11.34) is an
important question. Recall that under absence of arbitrage, the bond price
P (t, T ) has been defined as

w

T

P (t, T ) = IE exp
rs ds Ft ,
(11.35)
t
and the discounted bond price process

"
357
N. Privault
w

w

wT
t
t
t 7 exp rs ds P (t, T ) = exp rs ds
f (t, s)ds
0
0
t
w

t
= exp rs ds Xt
(11.36)
0
is a martingale by Proposition 11.1 and Relation (11.25). This latter property

will be used to characterize absence of arbitrage in the HJM model.
Proposition 11.6. (HJM Condition [49]). Under the condition
(t, T ) = (t, T )
wT
t
t [0, T ],
(t, s)ds,
(11.37)
which is known as the HJM absence of arbitrage condition, the process (11.36)
becomes a martingale.
Proof. Consider the spot forward rate
f (t, t, T ) =
and let
Xt =
wT
t
1 wT
f (t, s)ds,
T t t
f (t, s)ds = log P (t, T ),
0 t T,
with the relation

f (t, t, T ) =
1 wT
Xt
f (t, s)ds =
,
T t t
T t
0 t T,
(11.38)
where the dynamics of t 7 f (t, s) is given by (11.34). We note that when

f (t, s) = g(t)h(s) is a smooth function which satisfies the separation of variables property we have the relation
dt
wT
t
g(t)h(s)ds = g(t)h(t)dt + g 0 (t)
wT
t
g(t)h(s)dsdt,
which extends to f (t, s) as

dt
wT
t
f (t, s)ds = f (t, t)dt +
wT
t
dt f (t, s)ds,
which can be seen as a form of the Leibniz integral rule. Therefore we have
dt Xt = f (t, t)dt +
= f (t, t)dt +
wT
t
wT
t
dt f (t, s)ds
(t, s)dsdt +
wT
t
358
(t, s)dsdBt
"
= rt dt +
w
T

w

T
(t, s)ds dt +
(t, s)ds dBt ,
hence
|dt Xt |2 =
w
T
t
2
(t, s)ds dt.
Hence by Itos calculus we have

dt P (t, T ) = dt eXt
1
= eXt dt Xt + eXt (dt Xt )2
2
w
2
T
1
(t, s)ds dt
= eXt dt Xt + eXt
t
2

wT
wT
= eXt rt dt +
(t, s)dsdt +
(t, s)dsdBt
t
1
+ eXt
2
w
2
(t, s)ds dt,
and the discounted bond price satisfies

w

t
dt exp rs ds P (t, T )
0
w

w

t
t
= rt exp rs ds Xt dt + exp rs ds dt P (t, T )
0
0
w

w

t
t
= rt exp rs ds Xt dt exp rs ds Xt dt Xt
0
w
w
2
t
T
1
+ exp rs ds Xt
(t, s)ds dt
0
t
2
w

t
= rt exp rs ds Xt dt
0
w

wT
wT
t
exp rs ds Xt
rt dt +
(t, s)dsdt +
(t, s)dsdBt
0
w
2
w
t
T
1
(t, s)ds dt
+ exp rs ds Xt
0
t
2
w
w
T
t
(t, s)dsdBt
= exp rs ds Xt
0
w
w

2 !
T
t
1 wT
(t, s)dsdt
exp rs ds Xt
(t, s)ds
dt.
0
t
t
2
Thus the process P (t, T ) will be a martingale provided that
"
359
N. Privault
wT
t
(t, s)ds
1
2
w
T
t
2
(t, s)ds = 0,
0 t T.
(11.39)
Differentiating the above relation with respect to T , we get

(t, T ) = (t, T )
wT
t
(t, s)ds,
which is in fact equivalent to (11.39).
11.5 Forward Vasicek Rates

In this section we consider the Vasicek model, in which the short rate process
is the solution (11.2) of (11.1) as illustrated in Figure 11.1.
In this model the forward rate is given by
ST
rt (C(S t) C(T t)) + A(S t) A(T t))
=
ST
2 2ab
=
2
2b

1
rt
2 ab
2 2b(St)
2b(T t)
b(St)
b(T t)
+
(e
e
)
(e
e
)
.
ST
b
b3
4b3
f (t, T, S) =
In this model the forward rate t 7 f (t, T, S) can be represented as in

Figure 11.7, with here b/a > r0 .
0.01
f(t,T,S)
0.0095
0.009
0.0085
0.008
0.0075
0.007
0.0065
0.006
0.0055
0.005
10
Fig. 11.7: Forward rate process t 7 f (t, T, S).

360
"

Note that the forward rate cure t 7 f (t, T, S) is flat for small values of t.
The instantaneous short rate is given by
log P (t, T )
(11.40)
T
2
a
= rt eb(T t) + (1 eb(T t) ) 2 (1 eb(T t) )2 ,

b
2b
f (t, T ) : =
and the relation limT &t f (t, T ) = rt is easily recovered from this formula.
The instantaneous forward rate t 7 f (t, T ) can be represented as in
Figure 11.8, with here t = 0 and b/a > r0 :
0.14
f(t,T)
0.12
0.1
0.08
0.06
0.04
0.02
0
10
12
14
16
18
20
Fig. 11.8: Instantaneous forward rate process t 7 f (t, T ).
The HJM coefficients in the Vasicek model are in fact deterministic and
taking a = 0 we have
dt f (t, T ) = 2 eb(T t)
wT
t
eb(ts) dsdt + eb(T t) dBt ,
i.e.
(t, T ) = 2 eb(T t)
wT
t
eb(ts) ds,
and
(t, T ) = eb(T t) ,
and the HJM condition reads

wT
wT
(t, T ) = 2 eb(T t)
eb(ts) ds = (t, T )
(t, s)ds.
t
(11.41)
Random simulations of the Vasicek instantaneous forward rates are provided

in Figures 11.9 and 11.10.
"
361
N. Privault
rate %
7
6
5
4
3
2
1
0
40
30
0
20
10
15
x
20
10
25
30 0
Fig. 11.9: Forward instantaneous curve (t, x) 7 f (t, t + x) in the Vasicek model.
8
7
6
rate %
5
4
3
2
1
0
0
10
15
20
25
30
Fig. 11.10: Forward instantaneous curve x 7 f (0, x) in the Vasicek model.
For x = 0 the first slice of this surface is actually the short rate Vasicek
process rt = f (t, t) = f (t, t + 0) which is represented in Figure 11.11 using
another discretization.

362
"
0.07
0.065
0.06
0.055
0.05
0.045
0.04
0.035
0.03
0
10
15
20
Fig. 11.11: Short term interest rate curve t 7 rt in the Vasicek model.
Another example of market data is given in the next Figure 11.12, in which
the red and blue curves refer respectively to July 21 and 22 of year 2011.
Fig. 11.12: Market example of yield curves (11.23).
"
363
N. Privault
11.6 Modeling Issues

Parametrization of Forward Rates
In the Nelson-Siegel parametrization the forward interest rate curves are
parametrized by 4 coefficients z1 , z2 , z3 , z4 , as
g(x) = z1 + (z2 + z3 x)exz4 ,
x 0.
An example of a graph obtained by the Nelson-Siegel parametrization is given

in Figure 11.13, for z1 = 1, z2 = 10, z3 = 100, z4 = 10.
4
z1+(z2+xz3)exp(-xz4)
-2
-4
-6
-8
-10
0
0.2
0.4
0.6
0.8
Fig. 11.13: Graph of x 7 g(x) in the Nelson-Siegel model.
The Svensson parametrization has the advantage to reproduce two humps instead of one, the location and height of which can be chosen via 6 parameters
z1 , z2 , z3 , z4 , z5 , z6 as
g(x) = z1 + (z2 + z3 x)exz4 + z5 xexz6 ,
x 0.
A typical graph of a Svensson parametrization is given in Figure 11.14, for

z1 = 7, z2 = 5, z3 = 100, z4 = 10, z5 = 1/2, z6 = 1.
364
"

5
x->z1+(z2+z3*x)*exp(-x*z4)+z5*x*exp(-z6*x)
4.5
3.5
2.5
2
0
10
15
20
25
30
lambda
Fig. 11.14: Graph of x 7 g(x) in the Svensson model.

Figure 11.15 presents a fit of the market data of Figure 11.5 using a Svensson
curve.
5
4.5
3.5
2.5
Market data
Svensson curve
2
0
10
15
years
20
25
30
Fig. 11.15: Comparison of market data vs a Svensson curve.

It can be easily shown that the forward curves of the Vasicek model are
included neither in the Nelson-Siegel space, nor in the Svensson space, cf.
[90] and 3.5 of [5]. In addition, such curves do not appear to correctly model
the market forward curves considered above, cf. e.g. Figure 11.5.
In the Vasicek model we have

f
2
2 b(T t) b(T t)
(t, T ) = brt + a
+
e
e
,
T
b
b
and one can check that the sign of the derivatives of f can only change once
at most. As a consequence, the possible forward curves in the Vasicek model
are limited to one change of regime per curve, as illustrated in Figure 11.16
for various values of rt , and in Figure 11.17.
"
365
N. Privault
0.09
0.08
0.07
0.06
0.05
0.04
0.03
0.02
0.01
0
0
10
15
20
Fig. 11.16: Graphs of forward rates.
0.09
0.08
0.07
0.06
0.05
0.04
0.03
0.02
0
2
20
4
x
15
10
5
10
Fig. 11.17: Forward instantaneous curve (t, x) 7 f (t, t + x) in the Vasicek model.
One may think of constructing an instantaneous rate process taking values in

the Svensson space, however this type of modelization is not consistent with
absence of arbitrage, and it can be proved that the HJM curves cannot live
in the Nelson-Siegel or Svensson spaces, cf. 3.5 of [5].
Another way to deal with the curve fitting problem is to use deterministic
shifts for the fitting of one forward curve, such as the initial curve at t = 0,
cf. e.g. 8.2 of [90].
The Correlation Problem and a Two-Factor Model

The correlation problem is another issue of concern when using the affine
models considered so far. Let us compare three bond price simulations with
366
"

maturity T1 = 10, T2 = 20, and T3 = 30 based on the same Brownian path,
as given in Figure 11.18. Clearly, the bond prices P (t, T1 ) and P (t, T2 ) with
maturities T1 and T2 are linked by the relation
P (t, T2 ) = P (t, T1 ) exp(A(t, T2 ) A(t, T1 ) + rt (C(t, T2 ) C(t, T1 ))), (11.42)
meaning that bond prices with different maturities could be deduced from
each other, which is unrealistic.
1
0.9
0.8
0.7
0.6
0.5
0.4
P(t,T1)
P(t,T2)
P(t,T3)
0.3
0
10
15
t
20
25
30
Fig. 11.18: Graph of t 7 P (t, T1 ).

In affine short rates models, by (11.42), log P (t, T1 ) and log P (t, T2 ) are linked
by the linear relationship
log P (t, T2 ) = log P (t, T1 ) + A(t, T2 ) A(t, T1 ) + rt (C(t, T2 ) C(t, T1 ))
log P (t, T1 ) C(t, T1 )
= log P (t, T1 ) + A(t, T2 ) A(t, T1 ) + (C(t, T2 ) C(t, T1 ))
A(t, T1 )

C(t, T2 ) C(t, T1 )
= 1+
log P (t, T1 )
A(t, T1 )
C(t, T1 )
+A(t, T2 ) A(t, T1 ) (C(t, T2 ) C(t, T1 ))
A(t, T1 )
with constant coefficients, which yields the perfect correlation
Cor(log P (t, T1 ), log P (t, T2 )) = 1,
cf. 8.3 of [90], A solution to the correlation problem is to consider two
control processes (Xt )tR+ , (Yt )tR+ which are solution of
(1)
dXt = 1 (t, Xt )dt + 1 (t, Xt )dBt ,

(11.43)
dY = (t, Y )dt + (t, Y )dB (2) ,

t
2
t
2
t
t
"
367
N. Privault
(1)
(2)
where (Bt )tR+ , (Bt )tR+ have correlated Brownian motion with
(2)
Cov(Bs(1) , Bt ) = min(s, t),
s, t R+ ,
(11.44)
and
(1)
(2)
dBt dBt
= dt,
(11.45)
(1)
(2)
for some [1, 1]. In practice, (B )tR+ and (B )tR+ can be constructed from two independent Brownian motions (W (1) )tR+ and (W (2) )tR+ ,
by letting
(1)
(1)
Bt = Wt ,
B (2) = W (1) + p1 2 W (2) ,

t
t
t
t R+ ,
and Relations (11.44) and (11.45) are easily satisfied from this construction.
In two-factor models one chooses to build the short term interest rate rt via
t R+ .
rt = Xt + Yt ,
By the previous standard arbitrage arguments we define the price of a bond

with maturity T as

w

T

P (t, T ) : = IE exp
rs ds Ft
t

w

T

= IE exp
rs ds Xt , Yt
t
= F (t, Xt , Yt ),
(11.46)
since the couple (Xt , Yt )tR+ is Markovian. Using the Ito formula with two
variables and the fact that
t 7 e
rt
0
rs ds
P (t, T ) = e
rt
0
rs ds

w
T

IE exp
rs ds Ft
t
is an Ft -martingale under P we can derive a PDE

F
F
(t, x, y) + 2 (t, y)
(t, x, y)
x
y
2
2
1
F
1
F
+ 12 (t, x) 2 (t, x, y) + 22 (t, y) 2 (t, x, y)
2
x
2
y
2F
F
+1 (t, x)2 (t, y)
(t, x, y) +
(t, Xt , Yt ) = 0,
(11.47)
xy
t
(x + y)F (t, x, y) + 1 (t, x)
368
"

on R2 for the bond price P (t, T ). In the Vasicek model
(1)
dXt = aXt dt + dBt ,
dY = bY dt + dB (2) ,
t
t
t
this yields the solution F (t, x, y) of (11.47) as
P (t, T ) = F (t, Xt , Yt ) = F1 (t, Xt )F2 (t, Yt ) exp (U (t, T )) ,
(11.48)
where F1 (t, Xt ) and F2 (t, Yt ) are the bond prices associated to Xt and Yt in
the Vasicek model, and

ea(T t) 1 eb(T t) 1 e(a+b)(T t) 1
U (t, T ) =
T t+
ab
a
b
a+b
(1)
(2)
is a correlation term which vanishes when (Bt )tR+ and (Bt )tR+ are independent, i.e. when = 0, cf [9], Chapter 4, Appendix A, and [90]. [9].
Partial differentiation of log P (t, T ) with respect to T leads to the instantaneous forward rate
f (t, T ) = f1 (t, T ) + f2 (t, T )
(1 ea(T t) )(1 eb(T t) ),

ab
(11.49)
where f1 (t, T ), f2 (t, T ) are the instantaneous forward rates corresponding to

Xt and Yt respectively, cf. 8.4 of [90].
An example of a forward rate curve obtained in this way is given in Figure 11.19.
0.24
0.23
0.22
0.21
0.2
0.19
0.18
0
10
15
20
T
25
30
35
40
Fig. 11.19: Graph of forward rates in a two-factor model.
"
369
N. Privault
Next in Figure 11.20 we present a graph of the evolution of forward curve in
a two factor model.
0.24
0.235
0.23
0.225
0.22
0.215
1.4
1.2
1
0.8
t
0.6
0.4
0.2
0 0
Fig. 11.20: Random evolution of forward rates in a two-factor model.
11.7 The BGM Model

The models (HJM, affine, etc.) considered in the previous chapter suffer from
various drawbacks such as non-positivity of interest rates in Vasicek model,
and lack of closed form solutions in more complex models. The BGM [7]
model has the advantage of yielding positive interest rates, and to permit to
derive explicit formulas for the computation of prices for interest rate derivatives such as caps and swaptions on the LIBOR market.
In the BGM model we consider two bond prices P (t, T1 ), P (t, T2 ) with maturities T1 , T2 and the forward measure
r T2
dP2
e 0 rs ds
=
,
dP2
P (0, T2 )
with numeraire P (t, T2 ), cf. (10.6). The forward LIBOR rate L(t, T1 , T2 ) is
modeled as a geometric Brownian motion under P2 , i.e.
dL(t, T1 , T2 )
(2)
= 1 (t)dBt ,
L(t, T1 , T2 )
(11.50)
0 t T1 , i = 1, . . . , n 1, for some deterministic function 1 (t), with

solution
w

u
1wu
L(u, T1 , T2 ) = L(t, T1 , T2 ) exp
1 (s)dBs(2)
|1 |2 (s)ds ,
t
t
2
370
"

i.e. for u = T1 ,
L(T1 , T1 , T2 ) = L(t, T1 , T2 ) exp
w
T1
1 (s)dBs(2)

1 w T1
|1 |2 (s)ds .
2 t
Since L(t, T1 , T2 ) is a geometric Brownian motion under P2 , standard caplets

can be priced at time t [0, T1 ] from the Black-Scholes formula.
"
371
N. Privault
The following graph 11.21 summarizes the notions introduced in this chapter.
Bond price
P (t, T ) = e(T t)f(t,t,T )
2
Bondh price
i
rT
P (t, T ) = IE e t rs ds | Ft
Short rate1 rt
LIBOR rate3
(t,T )P (t,S)
L(t, T, S) = P(ST
)P (t,S)
Forward rate3
)log P (t,S)
f(t, T, S) = log P (t,TST
Bond price
rT
P (t, T ) = e t f(t,s)ds
Instantaneous forward rate4

P (t,T )
f(t, T ) = L(t, T ) = logT
Short rate
rt = f(t, t) = f(t, t, t)
Spot forward rate (yield)

rT
f(t, t, T ) = t f(t, s)ds/(T t)
Instantaneous forward rate4

f(t, T ) = L(t, T ) = limS&T f(t, T, S)
= limS&T L(t, T, S)
Can be modeled by Vasicek and other short rate models

Can be modeled from dP (t, T )/P (t, T ).
Can be modeled in the BGM model
4
Can be modeled in the HJM model
2
3
372
Fig. 11.21: Graph of stochastic interest rate modeling.

"
Exercises
Exercise 11.1 Consider a tenor structure {T1 , T2 } and a bond with maturity
T2 and price given at time t [0, T2 ] by
w

T2
P (t, T2 ) = exp
f (t, s)ds ,
t [0, T2 ],
t
where the instantaneous yield curve f (t, s) is parametrized as

s [t, T2 ].
f (t, s) = r1 1[0,T1 ] (s) + r2 1[T1 ,T2 ] (s),
Find a formula to estimate the values of r1 and r2 from the data of P (0, T2 )
and P (T1 , T2 ).
Exercise 11.2 Let (Bt )tR+ denote a standard Brownian motion started at
0 under the risk-neutral measure P . We consider a short term interest rate
process (rt )tR+ in a Ho-Lee model with constant deterministic volatility,
defined by
drt = adt + dBt ,
where a > 0 and > 0. Let P (t, T ) will denote the arbitrage price of a
zero-coupon bond in this model:

w

T

P (t, T ) = IE exp
rs ds Ft ,
0 t T.
(11.51)
t
a) State the bond pricing PDE satisfied by the function F (t, x) defined via

w

T

F (t, x) := IE exp
rs ds rt = x ,
0 t T.
t
b) Compute the arbitrage price F (t, rt ) = P (t, T ) from its expression (11.51)
as a conditional expectation.
Hint. One may use the integration by parts relation
wT
t
Bs ds = T BT tBt
wT
t
sdBs
wT
= (T t)Bt + T (BT Bt )
sdBs
t
wT
= (T t)Bt +
(T s)dBs ,
t
2 2
and the Laplace transform identity IE[eX ] = e

"
/2
for X ' N (0, 2 ).

373

N. Privault
c) Check that the function F (t, x) computed in question (b) does satisfy the
PDE derived in question (a).
d) Compute the forward rate f (t, T, S) in this model.
From now on we let a = 0.
e) Compute the instantaneous forward rate f (t, T ) in this model.
f) Derive the stochastic equation satisfied by the instantaneous forward rate
f (t, T ).
g) Check that the HJM absence of arbitrage condition is satisfied in this
equation.
Exercise 11.3 Consider the CIR process (rt )tR+ solution of
drt = art dt + rt dBt ,

where a, > 0 are constants (Bt )tR+ is a standard Brownian motion started
at 0.
a) Write down(1) the bond pricing PDE for the function F (t, x) given by

w

T

0 t T.
F (t, x) := IE exp
rs ds rt = x ,
t
Hint: Use Ito calculus and the fact that the discounted bond price is a
martingale.
b) Show that the PDE of Question (a) admits a solution of the form
F (t, x) = eA(T t)+xC(T t) where the functions A(s) and C(s) satisfy ordinary differential equations to be also written down(1) together with the
values of A(0) and C(0).
Exercise 11.4 Given (Bt )tR+ a standard Brownian motion, consider a HJM
model given by
dt f (t, T ) =
2
T (T 2 t2 )dt + T dBt .
2
(11.52)
a) Show that the HJM condition is satisfied by (11.52).

b) Compute f (t, T ) by solving (11.52).
rt
Hint: We have f (t, T ) = f (0, T ) + 0 ds f (s, T ) =
c) Compute the short rate rt = f (t, t) from the result of Question (b).
d) Show that the short rate rt satisfies a stochastic differential equation of
the form
drt = (t)dt + (rt f (0, t))(t)dt + (t)dBt ,
where (t), (t), (t) are deterministic functions to be determined.
374
"

Exercise 11.5 Let (rt )tR+ denote a short term interest rate process. For any
T > 0, let P (t, T ) denote the price at time t [0, T ] of a zero coupon bond
defined by the stochastic differential equation
dP (t, T )
= rt dt + tT dBt ,
P (t, T )
0 t T,
(11.53)
under the terminal condition P (T, T ) = 1, where (tT )t[0,T ] is an adapted

process. Let the forward measure PT be defined by

dPT
P (t, T ) r t rs ds
e 0
,
0 t T.
IE
Ft =
dP
P (0, T )
Recall that
BtT := Bt
wt
0
sT ds,
0 t T,
is a standard Brownian motion under PT .

b) Derive the stochastic differential equation satisfied by the discounted bond
price process
rt
t 7 e 0 rs ds P (t, T ),
0 t T,
and show that it is a martingale.
c) Show that
i
h rT
rt

IE e 0 rs ds Ft = e 0 rs ds P (t, T ),
0 t T.
d) Show that
i
h rT

0 t T.
e) Compute P (t, S)/P (t, T ), 0 t T , show that it is a martingale under

PT and that
w

T
P (t, S)
1wT S
exp
(sS sT )dBsT
(s sT )2 ds .
P (T, S) =
t
t
P (t, T )
2
f) Assuming that (tT )t[0,T ] and (tS )t[0,S] are deterministic functions,
compute the price
i
i
h rT
h
+
+
IE e t rs ds (P (T, S) ) Ft = P (t, T ) IET (P (T, S) ) Ft
of a bond option with strike .
Recall that if X is a centered Gaussian random variable with mean mt
and variance vt2 given Ft , we have
"
375
N. Privault

2
vt
1
IE[(eX K)+ | Ft ] = emt +vt /2
+ (mt + vt2 /2 log K)
2
vt

vt
1
K + (mt + vt2 /2 log K)
2
vt
where (x), x R, denotes the cumulative Gaussian distribution function.
Exercise 11.6 (Exercise 4.7 continued). Bridge model. Assume that the price
P (t, T ) of a zero coupon bond is modeled as
T
P (t, T ) = e(T t)+Xt ,
t [0, T ],
where > 0.
a) Show that the terminal condition P (T, T ) = 1 is satisfied.
b) Compute the forward rate
f (t, T, S) =
1
(log P (t, S) log P (t, T )).
ST
c) Compute the instantaneous forward rate

f (t, T ) = lim
S&T
1
ST
d) Show that the limit lim f (t, T ) does not exist in L2 ().
T &t
e) Show that P (t, T ) satisfies the stochastic differential equation

dP (t, T )
1
log P (t, T )
= dBt + 2 dt
dt,
P (t, T )
2
T t
t [0, T ].
f) Show, using the results of Exercise 11.5-(d), that

h rT T i

where (rtT )t[0,T ] is a process to be determined.
g) Compute the conditional density
#
"
dPT
P (t, T ) r t rsT ds
e 0
IE
Ft =
dP
P (0, T )
of the forward measure PT with respect to P.
h) Show that the process
t := Bt t,
B
0 t T,
376
"

i) Compute the dynamics of XtS and P (t, S) under PT .
Hint: Show that
wt 1
ST
(S T ) + (S T )
dBs =
log P (t, S).
0 Ss
St
j) Compute the bond option price
i
i
h rT T
h

IE e t rs ds (P (T, S) K)+ Ft = P (t, T ) IET (P (T, S) K)+ Ft ,
0 t < T < S.
Exercise 11.7 (Exercise 4.9 continued). Write down the bond pricing PDE
for the function

h rT
i

F (t, x) = IE e t rs ds rt = x
and show that in case = 0 the corresponding bond price P (t, T ) equals
P (t, T ) = eB(T t)rt ,
where
B(x) =
with =
0 t T,
2(ex 1)
,
2 + ( + )(ex 1)
p
2 + 2 2 .
Exercise 11.8 Stochastic string model [104]. Consider an instantaneous forward rate f (t, x) solution of
dt f (t, x) = x2 dt + dt B(t, x),
(11.54)
with a flat initial curve f (0, x) = r, where x represents the time to maturity,
and (B(t, x))(t,x)R2+ is a standard Brownian sheet with covariance
IE[B(s, x)B(t, y)] = (min(s, t))(min(x, y)),
s, t, x, y R+ ,
and initial conditions B(t, 0) = B(0, x) = 0 for all t, x R+ .

a) Solve the equation (11.54) for f (t, x).
b) Compute the short term interest rate rt = f (t, 0).
c) Compute the value at time t [0, T ] of the bond price
w

T t
P (t, T ) = exp
f (t, x)dx
0
with maturity T .
"
377
N. Privault
d) Compute the variance IE
"
w T t
0
2 #
B(t, x)dx
of the centered Gaussian
r T t
random variable 0
B(t, x)dx.
e) Compute the expected value IE[P (t, T )].
f) Find the value of such that the discounted bond price

w T t
B(t, x)dx ,
ert P (t, T ) = exp rT t(T t)3
0
3
t [0, T ].
satisfies ert IE[P (t, T )] = erT .

w

T
g) Compute the bond option price IE exp
rs ds (P (T, S) K)+ by
0
the Black-Scholes formula, knowing that

IE[(xem+X K)+ ] = xem+
v2
2
(v+(m+log(x/K))/v)K((m+log(x/K))/v),
when X is a centered Gaussian random variable with mean m = r v 2 /2

and variance v 2 .
Exercise 11.9 Let (rt )tR+ denote a short term interest rate process. For any
T > 0, let P (t, T ) denote the price at time t [0, T ] of a zero coupon bond
defined by the stochastic differential equation
dP (t, T )
= rt dt + tT dBt ,
P (t, T )
0 t T,
under the terminal condition P (T, T ) = 1, where (tT )t[0,T ] is an adapted

process. Let the forward measure PT be defined by

dPT
P (t, T ) r t rs ds
IE
e 0
,
0 t T.
Ft =
dP
P (0, T )
Recall that
BtT := Bt
wt
0
sT ds,
0 t T,

b) Derive the stochastic differential equation satisfied by the discounted bond
price process
rt
t 7 e 0 rs ds P (t, T ),
0 t T,
and show that it is a martingale.
c) Show that
i
h rT
rt

IE e 0 rs ds Ft = e 0 rs ds P (t, T ),
378
0 t T.
"

d) Show that
i
h rT

0 t T.
e) Compute P (t, S)/P (t, T ), 0 t T , show that it is a martingale under

PT and that
w

T
P (t, S)
1wT S
P (T, S) =
exp
(sS sT )dBsT
(s sT )2 ds .
t
P (t, T )
2 t
f) Assuming that (tT )t[0,T ] and (tS )t[0,S] are deterministic functions,
compute the price
i
i
h rT
h
+
+
IE e t rs ds (P (T, S) ) Ft = P (t, T ) IET (P (T, S) ) Ft
of a bond option with strike .
Recall that if X is a centered Gaussian random variable with mean mt
and variance vt2 given Ft , we have

2
1
vt
+ (mt + vt2 /2 log K)
IE[(eX K)+ | Ft ] = emt +vt /2
2
vt

vt
1
K + (mt + vt2 /2 log K)
2
vt
where (x), x R, denotes the Gaussian distribution function.
Exercise 11.10 (Exercise 11.6 continued).
a) Compute the forward rate
f (t, T, S) =
1
ST
b) Compute the instantaneous forward rate

f (t, T ) = lim
S&T
1
ST
c) Show that the limit lim f (t, T ) does not exist in L2 ().
T &t
d) Show that P (t, T ) satisfies the stochastic differential equation

dP (t, T )
1
log P (t, T )
= dBt + 2 dt
dt,
P (t, T )
2
T t
t [0, T ].
e) Show, using the results of Exercise 11.9-(d), that
"
379
N. Privault
h rT T i

where (rtT )t[0,T ] is a process to be determined.
f) Compute the conditional density

dPT
P (t, T ) r t rsT ds
IE
e 0
Ft =
dP
P (0, T )
of the forward measure PT with respect to P.
g) Show that the process
t := Bt t,
B
0 t T,

h) Compute the dynamics of XtS and P (t, S) under PT .
Hint: Show that
(S T ) + (S T )
wt
0
1
ST
dBs =
log P (t, S).
Ss
St
i) Compute the bond option price

i
i
h rT T
h

IE e t rs ds (P (T, S) K)+ Ft = P (t, T ) IET (P (T, S) K)+ Ft ,
0 t < T < S.
380
"
Chapter 12
Pricing of Interest Rate Derivatives
In this chapter we consider the pricing of caplets, caps, and swaptions, using
change of numeraire and forward swap measures.
12.1 Forward Measures and Tenor Structure

The maturity dates are arranged according to a discrete tenor structure
{0 = T0 < T1 < T2 < < Tn }.
An example of forward interest rate curve data is given in the table of Figure 12.1, which contains the values of (T1 , T2 , . . . , T23 ) and of {f (t, t + Ti , t +
Ti + )}i=1,...,23 , with t = 07/05/2003 and = six months.
2D
2.55
8Y
3.88
1W
2.53
9Y
4.02
1M
2.56
10Y
4.14
2M
2.52
11Y
4.23
3M
2.48
12Y
4.33
1Y
2.34
13Y
4.40
2Y
2.49
14Y
4.47
3Y
2.79
15Y
4.54
4Y
3.07
20Y
4.74
5Y
3.31
25Y
4.83
6Y 7Y
3.52 3.71
30Y
4.86
Fig. 12.1: Forward rates arranged according to a tenor structure.
Recall that by definition of P (t, Ti ) and absence of arbitrage the process

t 7 e
rt
0
rs ds
P (t, Ti ),
0 t Ti ,
i = 1, 2, . . . , n,
is an Ft -martingale under P, and as a consequence (P (t, Ti ))t[0,Ti ] can be

taken as numeraire in the definition
"
N. Privault
r Ti
i
dP
1
=
e 0 rs ds
dP
P (0, Ti )
(12.1)
i . The following proposition will allow us to price

of the forward measure P
i , it is a direct consequence of
contingent claims using the forward measure P
Proposition 10.1, noting that here we have P (Tt , Ti ) = 1.
Proposition 12.1. For all sufficiently integrable random variables F we have
i
h
r Ti

i [F | Ft ], 0 t T, i = 1, . . . , n.
IE F e t rs ds Ft = P (t, Ti )IE
(12.2)
Recall that for all Ti , Tj 0, the process
t 7
P (t, Tj )
,
P (t, Ti )
0 t min(Ti , Tj ),
i , cf. Proposition 10.2.

is an Ft -martingale under P
Dynamics under the forward measure

In order to apply Proposition 12.1 and to compute the price
i
h rT

i [F | Ft ],
IE e t rs ds F Ft = P (t, Ti )IE
it can be useful to determine the dynamics of the underlying processes rt ,
i.
f (t, T, S), and P (t, T ) under the forward measure P
Let us assume that the dynamics of the bond price P (t, Ti ) is given by
dP (t, Ti )
= rt dt + i (t)dWt ,
P (t, Ti )
(12.3)
for i = 1, 2, . . . , n, where (Wt )tR+ is a standard Brownian motion under

P and (rt )tR+ and (i (t))tR+ are adapted processes with respect to the
filtration (Ft )tR+ generated by (Wt )tR+ , i.e.

w
wt
t
1wt
|i (s)|2 ds ,
P (t, Ti ) = P (0, Ti ) exp
rs ds +
i (s)dWs
0
0
2 0
t [0, Ti ], i = 1, 2, . . . , n.
By the Girsanov theorem,
ti := Wt
W
wt
0
i (s)ds,
0 t Ti ,
382
(12.4)
"

i for all i = 1, 2, . . . , n, cf. e.g. (10.10),
is a standard Brownian motion under P
hence we have
ti = dWt i (t)dt,
dW
i = 1, 2, . . . , n,
and
tj = dWt j (t)dt = dW
ti + (i (t) j (t))dt,
dW
i . Hence the
tj )tR has drift (i (t) j (t))tR under P
which shows that (W
+
+
j is given by
dynamics of t 7 P (t, Tj ) under P
dP (t, Tj )
tj ,
= rt dt + |j (t)|2 dt + j (t)dW
P (t, Tj )
(12.5)
j , and we have
tj )tR is a standard Brownian motion under P
where (W
+
w

wt
wt
t
sj + 1
P (t, Tj ) = P (0, Tj ) exp
rs ds +
j (s)dW
|j (s)|2 ds
(12.6)
0
0
2 0
w

wt
wt
t
1wt
si +
= P (0, Tj ) exp
rs ds +
|j (s)|2 ds
j (s)dW
j (s)i (s)ds
0
0
0
2 0
w

wt
wt
t
1wt
si 1
= P (0, Tj ) exp
rs ds +
j (s)dW
|j (s) i (s)|2 ds
|i (s)|2 ds ,
0
0
2 0
2 0
t [0, Tj ], i, j = 1, 2, . . . , n.
In case the short rate process (rt )tR+ is given as the (Markovian) solution
to the stochastic differential equation
drt = (t, rt )dt + (t, rt )dWt ,
i by
its dynamics will be given under P
i.
drt = (t, rt )dt + (t, rt )i (t)dt + (t, rt )dW
t
(12.7)
In the Vasicek case we have

drt = (a brt )dt + dWt ,
and
i (t) = (1 eb(Ti t) ),
b
by (11.13), hence from (12.7) we have
drt = (a brt )dt
0 t Ti ,
2
i
(1 eb(Ti t) )dt + dW
t
b
(12.8)
and we obtain
"
383
N. Privault
dP (t, Ti )
2
i,
= rt dt + 2 (1 eb(Ti t) )2 dt (1 eb(Ti t) )dW
t
P (t, Ti )
b
b
from (11.13).
12.2 Bond Options

The next proposition can be obtained as an application of the Margrabe formula (10.26) of Proposition 10.8 by taking Xt = P (t, Tj ), Nt = P (t, Ti ), and
t = Xt /Nt = P (t, Tj )/P (t, Ti ). In the Vasicek model, this formula has been
X
first obtained in [57].
We work with a standard Brownian motion (Wt )tR+ under P, generating
the filtration (Ft )tR+ , and an Ft -adapted short rate process (rt )tR+ .
Proposition 12.2. Assume that the dynamics of the bond prices P (t, Ti ),
P (t, Tj ) are given by
dP (t, Ti )
= rt dt + i (t)dWt ,
P (t, Ti )
dP (t, Tj )
= rt dt + j (t)dWt ,
P (t, Tj )
where (i (t))tR+ and (j (t))tR+ are deterministic functions. Then the price
of a bond call option on P (Ti , Tj ) with payoff F = (P (Ti , Tj ) )+ can be
written as
i
h r Ti

IE e t rs ds (P (Ti , Tj ) )+ Ft
(12.9)

v
1
P (t, Tj )
v
1
P (t, Tj )
= P (t, Tj )
+ log
P (t, Ti ) + log
,
2 v
P (t, Ti )
2 v
P (t, Ti )
with v 2 =
w Ti
t
|j (s) i (s)|2 ds.
Proof. First we note that using Nt = P (t, Ti ) as a numeraire the price of a

bond call option on P (Ti , Tj ) with payoff F = (P (Ti , Tj ))+ can be written
or directly by (10.5), as
from Proposition 10.4 using the forward measure P,
i
i
h r Ti
h

i (P (Ti , Tj ) )+ Ft .
IE e t rs ds (P (Ti , Tj ) )+ Ft = P (t, Ti )IE
Next we use the Black-Scholes formula and the martingale property of the
forward price P (t, Tj )/P (t, Ti ), which can be written from (12.6) as the geometric Brownian motion
384
"
P (Ti , Tj ) =
w

w Ti
Ti
P (t, Tj )
si 1
exp
(j (s) i (s))dW
|i (s) j (s)|2 ds ,
t
P (t, Ti )
2 t
when (i (s))s[0,T ] and (j (s))s[0,T ] in (12.3)

under the forward measure P
i
j
are deterministic functions. The above relation can be obtained by solving
(10.12) in Proposition 10.3.

Note that from Corollary 10.1 the decomposition (12.9) gives the selffinancing portfolio in the assets P (t, Ti ) and P (t, Tj ) for the claim with payoff
(P (Ti , Tj ) )+ .
In the Vasicek case
the above bond
option price could also be computed

rT
from the joint law of rT , t rs ds , which is Gaussian, or from the dynamics
i , cf. 7.3 of [90], and [68] for the CIR
(12.5)-(12.8) of P (t, T ) and rt under P
and other short rate models with correlated Brownian motions.
12.3 Caplet Pricing

The caplet on the spot forward rate f (Ti , Ti , Ti+1 ) with strike is a contract
with payoff
(f (Ti , Ti , Ti+1 ) )+ ,
i
priced at time t [0, Ti ] from Proposition 10.4 using the forward measure P
as
i
h r Ti+1

rs ds
IE e t
(f (Ti , Ti , Ti+1 ) )+ Ft
(12.10)

i+1 (f (Ti , Ti , Ti+1 ) )+ | Ft ,
= P (t, Ti+1 )IE
by taking Nt = P (t, Ti+1 ) as a numeraire. Next, we consider the caplet with
payoff
(L(Ti , Ti , Ti+1 ) )+
on the LIBOR rate
L(t, Ti , Ti+1 ) =
1
Ti+1 Ti

P (t, Ti )
1 ,
P (t, Ti+1 )
0 t Ti < Ti+1 ,
i+1 defined in (12.1), from Proposition 10.2.

which is a martingale under P
The next formula (12.12) is known as the Black caplet formula. We assume
that L(t, Ti , Ti+1 ) is modeled in the BGM model of Section 11.7.
Proposition 12.3. Assume that L(t, Ti , Ti+1 ) is modeled as
"
385
N. Privault
dL(t, Ti , Ti+1 )
ti+1 ,
= i (t)dB
L(t, Ti , Ti+1 )
(12.11)
0 t Ti , i = 1, 2, . . . , n 1, where t 7 i (t) is a deterministic function,

i = 1, 2, . . . , n 1. The caplet on L(Ti , Ti , Ti+1 ) is priced as time t [0, Ti ]
as
i
h r Ti+1

rs ds
IE e t
(L(Ti , Ti , Ti+1 ) )+ Ft
(12.12)
= P (t, Ti+1 )L(t, Ti , Ti+1 )(d+ ) P (t, Ti+1 )(d ),
where
d+ =
log(L(t, Ti , Ti+1 )/) + i2 (t)(Ti t)/2
,
i (t) Ti t
d =
log(L(t, Ti , Ti+1 )/) i2 (t)(Ti t)/2
,
i (t) Ti t
and
and
|i (t)|2 =
1 w Ti
|i |2 (s)ds.
Ti t t

L(Ti , Ti , Ti+1 ) = L(t, Ti , Ti+1 ) exp
w
Ti

w Ti
si+1 1
i (s)dB
|i (s)|2 ds ,
2 t
0 t Ti , i.e. t 7 L(t, Ti , Ti+1 ) is a geometric Brownian motion with

i+1 . Hence by (12.10) we have
volatility i (t) under P
i
h r Ti+1

rs ds
IE e t
(L(Ti , Ti , Ti+1 ) )+ Ft

= P (t, Ti+1 ) IEi+1 (L(Ti , Ti , Ti+1 ) )+ | Ft
= P (t, Ti+1 )BS(, L(t, Ti , Ti+1 ), i (t), 0, Ti t),
t [0, Ti ], where
BS(, x, , r, ) = x(d+ ) er (d )
is the Black-Scholes function with
|i (t)|2 =
1 w Ti
|i |2 (s)ds.
Ti t t
386
"

The formula (12.12) is also known as the Black (1976) formula when applied
to options on underlying futures or forward contracts on commodities, which
are modeled according to (12.11). In this case, the bond price P (t, Ti+1 ) can
be simply modeled as P (t, Ti+1 ) = er(Ti+1 t) and (12.12) becomes
er(Ti+1 t) L(t, Ti , Ti+1 )(d+ ) er(Ti+1 t) (d ),
where L(t, Ti , Ti+1 ) is the underlying future price.
In general we may also write (12.12) as
i
h r Ti+1

rs ds
(Ti+1 Ti ) IE e t
(L(Ti , Ti , Ti+1 ) )+ Ft

P (t, Ti )
1 (d+ ) (Ti+1 Ti )P (t, Ti+1 )(d )
= P (t, Ti+1 )
P (t, Ti+1 )
= (P (t, Ti ) P (t, Ti+1 ))(d+ ) (Ti+1 Ti )P (t, Ti+1 )(d ),
and by Corollary 10.1 this gives the self-financing portfolio strategy
((d+ ), (d+ ) (Ti+1 Ti )(d ))
in the bonds (P (t, Ti ), P (t, Ti+1 )) with maturities Ti and Ti+1 , cf. Corollary 10.2 and [92]. Proposition 12.3 can also be proved by taking P (t, Ti+1 )
as numeraire and letting
t = P (t, Ti )/P (t, Ti+1 ) = 1 + (Ti+1 Ti )L(Ti , Ti , Ti+1 ).
X
Floorlets
Similarly, a floorlet on f (T, T, Ti ) with strike is a contract with payoff
( f (T, T, Ti ))+ , priced at time t [0, T ] as
i
h r Ti

i ( f (T, T, Ti ))+ | Ft .
IE e t rs ds ( f (T, T, Ti ))+ Ft = P (t, Ti )IE
Floorlets are analog to put options and can be similarly priced by the call/put
parity in the Black-Scholes formula.
Cap Pricing
More generally one can consider caps that are relative to a given tenor structure {T1 , . . . , Tn }, with discounted payoff
"
387
N. Privault
n1
X
(Tk+1 Tk )e
r Tk+1
t
rs ds
(f (Tk , Tk , Tk+1 ) )+ .
k=1
Pricing formulas for caps are easily deduced from analog formulas for caplets,
since the payoff of a cap can be decomposed into a sum of caplet payoffs. Thus
the price of a cap at time t [0, T1 ] is given by
"n1
#

r Tk+1
X

rs ds
IE
(Tk+1 Tk )e t
(f (Tk , Tk , Tk+1 ) )+ Ft
k=1
n1
X
rT

k+1

rs ds
(Tk+1 Tk ) IE e t
(f (Tk , Tk , Tk+1 ) )+ Ft
k=1
n1
X
i
h
k+1 (f (Tk , Tk , Tk+1 ) )+ Ft .
(Tk+1 Tk )P (t, Tk+1 )IE
k=1
(12.13)
In the above BGM model, the cap with payoff
n1
X
(Tk+1 Tk )(L(Tk , Tk , Tk+1 ) )+
k=1
can be priced at time t [0, T1 ] as

n1
X
(Tk+1 Tk )P (t, Tk+1 )BS(, L(t, Tk , Tk+1 ), k (t), 0, Tk t).
k=1
12.4 Forward Swap Measures

In this section we introduce the forward measures to be used for the pricing
of swaptions, and we study their properties. We start with the definition of
the annuity numeraire
P (t, Ti , Tj ) =
j1
X
(Tk+1 Tk )P (t, Tk+1 ),
0 t Ti ,
(12.14)
k=i
with in particular
P (t, Ti , Ti+1 ) = (Ti+1 Ti )P (t, Ti+1 ),
388
0 t Ti .
"

1 i < n. The annuity numeraire satisfies the following martingale
property,
rt
which can be proved by linearity and the fact that t 7 e 0 rs ds P (t, Tk ) is
a martingale for all k = 1, 2, . . . , n.
Remark 12.1. The discounted annuity numeraire
t 7 e
rt
0
rs ds
P (t, Ti , Tj ) = e
rt
0
rs ds
j1
X
(Tk+1 Tk )P (t, Tk+1 ),
0 t Ti ,
k=i
is a martingale under P.
i,j is defined by
The forward swap measure P
r Ti
i,j
dP
P (Ti , Ti , Tj )
= e 0 rs ds
,
dP
P (0, Ti , Tj )
(12.15)
1 i < j n. We have
"
#
i
h r Ti

1

dPi,j
IE
IE e 0 rs ds P (Ti , Ti , Tj ) Ft
Ft =
dP
P (0, Ti , Tj )
=
P (t, Ti , Tj ) r t rs ds
e 0
,
P (0, Ti , Tj )
0 t Ti , by Remark 12.1, and

i,j|F
r Ti
dP
P (Ti , Ti , Tj )
t
= e t rs ds
,
dP|Ft
P (t, Ti , Tj )
0 t Ti+1 ,
(12.16)
by Relation (10.3) in Lemma 10.1. We also know that the process

t 7 vki,j (t) :=
P (t, Tk )
P (t, Ti , Tj )
i,j by Proposition 10.2. It follows that the swap

is an Ft -martingale under P
rate
S(t, Ti , Tj ) :=
P (t, Ti ) P (t, Tj )
= vii,j (t) vji,j (t),
P (t, Ti , Tj )
0 t Ti ,
i,j .
defined in Proposition 11.5 is also a martingale under P
Using the forward swap measure we obtain the following pricing formula
for a given integrable claim with payoff of the form P (Ti , Ti , Tj )F :
"
#
i
h r Ti
i,j|F
dP

t
t rs ds
IE e
P (Ti , Ti , Tj )F Ft = P (t, Ti , Tj ) IE F
Ft
dP|Ft
"
389
N. Privault
h i
i,j F Ft ,
= P (t, Ti , Tj )IE
(12.17)
after applying (12.15) and (12.16) on the last line, or Proposition 10.1.
12.5 Swaption Pricing on the LIBOR

A swaption on the forward rate f (T1 , Tk , Tk+1 ) is a contract meant to protect
oneself against a risk based on an interest rate swap, and has payoff
!+
j1
rT
X
T k+1 rs ds
i
(Tk+1 Tk )e
(f (Ti , Tk , Tk+1 ) )
,
k=i
at time Ti .
This swaption can be priced at time t [0, Ti ] under the risk-neutral
measure P as
!+
j1

rT
rT
X

T k+1 rs ds
t i rs ds
i
(Tk+1 Tk )e
(f (Ti , Tk , Tk+1 ) )
IE e
Ft .
k=i
(12.18)
In the sequel and in practice the price (12.18) of the swaption will be evaluated as
!+
j1

rT
X

t i rs ds
IE e
(Tk+1 Tk )P (Ti , Tk+1 )(f (Ti , Tk , Tk+1 ) )
Ft ,
k=i
where we approximate the discount factor e

pectation P (Ti , Tk+1 ) given FTi .
r Tk+1
Ti
(12.19)
rs ds
by its conditional ex-
The above term (12.19) can be upper bounded by the cap price (12.13) written as
"
#
j1

r Ti
X
+
IE e t rs ds
(Tk+1 Tk )P (Ti , Tk+1 ) (f (Ti , Tk , Tk+1 ) ) Ft
k=i
= IE
"j1
X

#

r Tk+1

+
rs ds
(Tk+1 Tk ) IE (f (Ti , Tk , Tk+1 ) ) e t
FTi Ft
k=i
= IE
"j1
X
(Tk+1 Tk )e
r Tk+1
t
rs ds
(f (Ti , Tk , Tk+1 ) )

Ft
k=i
390
"
= IE
"j1
X
(Tk+1 Tk )P (t, Tk+1 ) (f (Ti , Tk , Tk+1 ) )
#

Ft .
k=i
In addition, when j = i + 1, the swaption price (12.19) coincides with the

price at time t of a caplet on [Ti , Ti+1 ] since
i
h r Ti
+
IE e t rs ds ((Ti+1 Ti )P (Ti , Ti+1 )(f (Ti , Ti , Ti+1 ) )) Ft
i
h r Ti
+
= (Ti+1 Ti ) IE e t rs ds P (Ti , Ti+1 ) ((f (Ti , Ti , Ti+1 ) )) Ft
r
rT

Ti
i+1 rs ds
+
= (Ti+1 Ti ) IE e t rs ds IE e Ti
FTi ((f (Ti , Ti , Ti+1 ) ))
r

rT
Ti
i+1 rs ds
+
= (Ti+1 Ti ) IE IE e t rs ds e Ti
((f (Ti , Ti , Ti+1 ) )) FTi
i
h r Ti+1
+
rs ds
= (Ti+1 Ti ) IE e t
(f (Ti , Ti , Ti+1 ) ) Ft ,
0 t Ti , which coincides with the caplet price (12.10) up to the factor
Ti+1 Ti .
In case we replace the forward rate f (t, T, S) with the LIBOR rate
L(t, T, S) defined in Proposition 11.5, the payoff of the swaption can be
rewritten as in the following lemma which is a direct consequence of the
definition of the swap rate S(Ti , Ti , Tj ).
Lemma 12.1. The payoff of the swaption in (12.19) can be rewritten as
j1
X
!+
(Tk+1 Tk )P (Ti , Tk+1 )(L(Ti , Tk , Tk+1 ) )
k=i
+
= (P (Ti , Ti ) P (Ti , Tj ) P (Ti , Ti , Tj ))

+
= P (Ti , Ti , Tj ) (S(Ti , Ti , Tj ) ) .
(12.20)
(12.21)
Proof. The relation

j1
X
(Tk+1 Tk )P (t, Tk+1 )(L(t, Tk , Tk+1 ) S(t, Ti , Tj )) = 0
k=i
that defines the forward swap rate S(t, Ti , Tj ) shows that

j1
X
(Tk+1 Tk )P (t, Tk+1 )L(t, Tk , Tk+1 )
k=i
= S(t, Ti , Tj )
j1
X
(Tk+1 Tk )P (t, Tk+1 )
k=i
"
391

Ft

Ft
N. Privault
= P (t, Ti , Tj )S(t, Ti , Tj )
= P (t, Ti ) P (t, Tj ),
by the definition (12.14) of P (t, Ti , Tj ), hence
j1
X
(Tk+1 Tk )P (t, Tk+1 )(L(t, Tk , Tk+1 ) )
k=i
= P (t, Ti ) P (t, Tj ) P (t, Ti , Tj )

= P (t, Ti , Tj ) (S(t, Ti , Tj ) ) ,
and for t = Ti we get
!+
j1
X
k=i
+
= P (Ti , Ti , Tj ) (S(Ti , Ti , Tj ) ) .

The next proposition simply states that a swaption on the LIBOR rate can
be priced as a European call option on the swap rate S(Ti , Ti , Tj ) under the
i,j .
forward swap measure P
Proposition 12.4. The price (12.19) of the swaption with payoff
j1
X
!+
(12.22)
k=i
i,j as
on the LIBOR market can be written under the forward swap measure P
i
h
i,j (S(Ti , Ti , Tj ) )+ Ft ,
P (t, Ti , Tj )IE
0 t Ti .
Proof. As a consequence of (12.17) and Lemma 12.1 we find
!+
j1

rT
X

t i rs ds
IE e
Ft
k=i
i
h r Ti
+
= IE e t rs ds (P (Ti , Ti ) P (Ti , Tj ) P (Ti , Ti , Tj )) Ft (12.23)
i
h r Ti
+
= IE e t rs ds P (Ti , Ti , Tj ) (S(Ti , Ti , Tj ) ) Ft
"
#

i,j|F
dP
1
+
t
IE
(S(Ti , Ti , Tj ) ) Ft
=
P (t, Ti , Tj )
dP|Ft
i
h
i,j (S(Ti , Ti , Tj ) )+ Ft .
= P (t, Ti , Tj )IE
(12.24)
392
"

In the next proposition we price a swaption with payoff (12.22) or equivalently
(12.21).
Proposition 12.5. Assume that the LIBOR swap rate (11.31) is modeled as
a geometric Brownian motion under Pi,j , i.e.
ti,j ,
dS(t, Ti , Tj ) = S(t, Ti , Tj )
(t)dW
where (
(t))tR+ is a deterministic function. Then the swaption with payoff
(P (T, Ti ) P (T, Tj ) P (Ti , Ti , Tj ))+ = P (Ti , Ti , Tj ) (S(Ti , Ti , Tj ) )
can be priced using the Black-Scholes formula as

i
h r Ti
+
IE e t rs ds P (Ti , Ti , Tj ) (S(Ti , Ti , Tj ) ) Ft
= (P (t, Ti ) P (t, Tj ))+ (t, S(t, Ti , Tj ))
(t, S(t, Ti , Tj ))
j1
X
(Tk+1 Tk )P (t, Tk+1 ),
k=i
where
d+ =
2
log(S(t, Ti , Tj )/) + i,j
(t)(Ti t)/2
,
i,j (t) Ti t
d =
2
log(S(t, Ti , Tj )/) i,j
(t)(Ti t)/2
,
i,j (t) Ti t
and
and
|i,j (t)|2 =
1 w Ti
(
(s))2 ds.
Ti t t
Proof. Since S(t, Ti , Tj ) is a geometric Brownian motion with variance

i,j , by (12.20)-(12.21) and (12.23)-(12.24) we have
(
(t))tR+ under P
i
h r Ti
+
IE e t rs ds P (Ti , Ti , Tj ) (S(Ti , Ti , Tj ) ) Ft
i
h rT

= IE e t rs ds (P (T, Ti ) P (T, Tj ) P (Ti , Ti , Tj ))+ Ft
i
h
i,j (S(Ti , Ti , Tj ) )+ Ft
= P (t, Ti , Tj )IE
= P (t, Ti , Tj )BS(, S(Ti , Ti , Tj ), i,j (t), 0, Ti t)
= P (t, Ti , Tj ) (S(t, Ti , Tj )+ (t, S(t, Ti , Tj )) (t, S(t, Ti , Tj )))
= (P (t, Ti ) P (t, Tj ))+ (t, S(t, Ti , Tj )) P (t, Ti , Tj ) (t, S(t, Ti , Tj ))
= (P (t, Ti ) P (t, Tj ))+ (t, S(t, Ti , Tj ))
"
393
N. Privault
(t, S(t, Ti , Tj ))
j1
X
(Tk+1 Tk )P (t, Tk+1 ).
k=i

In addition the hedging strategy
(+ (t, S(t, Ti , Tj )), (t, S(t, Ti , Tj ))(Ti+1 Ti ), . . .
. . . , (t, S(t, Ti , Tj ))(Tj1 Tj2 ), + (t, S(t, Ti , Tj )))
based on the assets (P (t, Ti ), . . . , P (t, Tj )) is self-financing by Corollary 10.2,
cf. also [92].
Swaption prices can also be computed by an approximation formula, from the
i,j , based on the bond
exact dynamics of the swap rate S(t, Ti , Tj ) under P
price dynamics of the form (12.3), cf. [106], page 17.
Exercises
Exercise 12.1 Consider two bonds with maturities T1 and T2 , T1 < T2 , which
follow the stochastic differential equations
dP (t, T1 ) = rt P (t, T1 )dt + 1 (t)P (t, T1 )dWt
and
dP (t, T2 ) = rt P (t, T2 )dt + 2 (t)P (t, T2 )dWt .
a) Using Ito calculus, show that the forward process P (t, T2 )/P (t, T1 ) is a
t := dWt 1 (t)dt under
driftless geometric Brownian motion driven by dW
the T1 -forward measure

P.
h r T1
i
b) Compute the price IE e 0 rs ds (K P (T1 , T2 ))+ of a bond put option
using change of numeraire and the Black-Scholes formula.
Hint: Given X a centered Gaussian random variable with variance v 2 , we
have:
IE [(xeXv
/2 +
) ] = (v/2+(log(/x))/v)x(v/2+(log(/x))/v).
Exercise 12.2 Given two bonds with maturities T , S and prices P (t, T ),
P (t, S), consider the LIBOR rate
L(t, T, S) =
P (t, T ) P (t, S)
(S T )P (t, S)
394
"

at time t, modeled as
0 t T,
dL(t, T, S) = t L(t, T, S)dt + L(t, T, S)dWt ,
(12.25)
where (Wt )t[0,T ] is a standard Brownian motion under the risk-neutral measure P , > 0 is a constant, and (t )t[0,T ] is an adapted process. Let
i
h rS

Ft = IE e t rs ds ( L(T, T, S))+ Ft
denote the price at time t of a floorlet option with strike , maturity T , and
payment date S.
S with maturity S.
a) Rewrite the value of Ft using the forward measure P
S ?
b) What is the dynamics of L(t, T, S) under the forward measure P
c) Write down the value of Ft using the Black-Scholes formula.
Hint. Given X a centered Gaussian random variable with variance v 2 we
have
IE [( em+X )+ ] = ((m log )/v) em+
v2
2
(v (m log )/v),
where denotes the Gaussian cumulative distribution function.

Exercise 12.3 Consider a European swap option giving its holder the right
to enter into a 3-year annual pay swap in four years, where a fixed rate of 5%
is paid and the LIBOR rate is received. Assume that the yield curve is flat
at 5% per annum with annual compounding and the volatility of the swap
rate is 20%.
a) What are the key assumptions in order to apply Blacks formula to value
this swaption ?
b) Compute the price of this swaption using Blacks formula as an application of Proposition 12.5, using the functions (d+ ) and (di ).
Exercise 12.4 We work in the short rate model

drt = dBt ,
where (Bt )tR+ is a standard Brownian motion under P, and P2 is the forward
measure defined by
r T2
1
dP2
=
e 0 rs ds .
dP
P (0, T2 )
a) State the expressions of 1 (t) and 2 (t) in
"
395
N. Privault
dP (t, Ti )
= rt dt + i (t)dBt ,
P (t, Ti )
i = 1, 2,
and the dynamics of the P (t, T1 )/P (t, T2 ) under P2 , where P (t, T1 ) and
P (t, T2 ) are bond prices with maturities T1 and T2 .
b) State the expression of the forward rate f (t, T1 , T2 ).
c) Compute the dynamics of f (t, T1 , T2 ) under the forward measure P2 with
r T2
1
dP2
=
e 0 rs ds .
dP
P (0, T2 )
d) Compute the price

i
h r T2

(T2 T1 ) IE e t rs ds (f (T1 , T1 , T2 ) )+ Ft
of a cap at time t [0, T1 ], using the expectation under the forward
measure P2 .
e) Compute the dynamics of the swap rate process
S(t, T1 , T2 ) =
P (t, T1 ) P (t, T2 )
,
(T2 T1 )P (t, T2 )
t [0, T1 ],
under P2 .
f) Compute the swaption price
i
h r T1

(T2 T1 ) IE e t rs ds P (T1 , T2 )(S(T1 , T1 , T2 ) )+ Ft
on the swap rate S(T1 , T1 , T2 ) using the expectation under the forward
swap measure P1,2 .
Exercise 12.5 Consider three zero-coupon bonds P (t, T1 ), P (t, T2 ) and
P (t, T3 ) with maturities T1 = , T2 = 2 and T3 = 3 respectively, and
the forward LIBOR L(t, T1 , T2 ) and L(t, T2 , T3 ) defined by

1
P (t, Ti )
L(t, Ti , Ti+1 ) =
1 ,
i = 1, 2.
P (t, Ti+1 )
Assume that L(t, T1 , T2 ) and L(t, T2 , T3 ) are modeled in the BGM model by
dL(t, T1 , T2 )
2,
= eat dW
t
L(t, T1 , T2 )
0 t T1 ,
(12.26)
t2 is a
and L(t, T2 , T3 ) = b, 0 t T2 , for some constants a, b > 0, where W
standard Brownian motion under the forward rate measure P2 defined by
396
"

r T2
dP2
e 0 rs ds
=
.
dP
P (0, T2 )
a) Compute L(t, T1 , T2 ), 0 t T2 by solving Equation (12.26).
b) Show that the price at time t of the caplet with strike can be written as
i
h r T2

2 (L(T1 , T1 , T2 ) )+ | Ft ,
IE e t rs ds (L(T1 , T1 , T2 ) )+ Ft = P (t, T2 )IE
2 denotes the expectation under the forward measure P2 .
where IE
c) Using the hint below, compute the price at time t of the caplet with strike
on L(T1 , T1 , T2 ).
d) Compute
P (t, T1 )
,
P (t, T1 , T3 )
0 t T1 ,
and
P (t, T3 )
,
P (t, T1 , T3 )
0 t T2 ,
in terms of b and L(t, T1 , T2 ), where P (t, T1 , T3 ) is the annuity numeraire

P (t, T1 , T3 ) = P (t, T2 ) + P (t, T3 ),
0 t T2 .
e) Compute the dynamics of the swap rate

t 7 S(t, T1 , T3 ) =
P (t, T1 ) P (t, T3 )
,
P (t, T1 , T3 )
0 t T1 ,
i.e. show that we have

2,
dS(t, T1 , T3 ) = 1.3 (t)S(t, T1 , T3 )dW
t
where 1,3 (t) is a process to be determined.
f) Using the Black-Scholes formula, compute an approximation of the swaption price
i
h r T1

IE e t rs ds P (T1 , T1 , T3 )(S(T1 , T1 , T3 ) )+ Ft

2 (S(T1 , T1 , T3 ) )+ | Ft ,
= P (t, T1 , T3 )IE
at time t [0, T1 ]. You will need to approximate 1,3 (s), s t, by freezing all random terms at time t.
have

v2
IE (em+X )+ = em+ 2 (v + (m log )/v) ((m log )/v),
"
397
N. Privault
Exercise 12.6 Consider a portfolio (tT , tS )t[0,T ] made of two bonds with
maturities T , S, and value
Vt = tT P (t, T ) + tS P (t, S),
0 t T,
at time t. We assume that the portfolio is self-financing, i.e.

dVt = tT dP (t, T ) + tS dP (t, S),
0 t T,
(12.27)
and that it hedges the claim (P (T, S) ) , so that

i
h rT
+
Vt = IE e t rs ds (P (T, S) ) Ft
i
h
+
= P (t, T ) IET (P (T, S) ) Ft ,
0 t T.
a) Show that
i
h rT
+
IE e t rs ds (P (T, S) K) Ft
h
i wt
wt
+
= P (0, T ) IET (P (T, S) K) +
sT dP (s, T ) +
sS dP (s, S).
0
b) Show that under the

self-financing condition (12.27), the discounted portrt
folio price Vt = e 0 rs ds Vt satisfies
dVt = tT dP (t, T ) + tS dP (t, S),
r
t
t
where P (t, T ) = e 0 rs ds P (t, T ) and P (t, S) = e 0 rs ds P (t, S) denote
the discounted bond prices.
c) Show that
h
i
h
i
+
+
IET (P (T, S) K) |Ft = IET (P (T, S) K)
w t C
(Xu , T u, v(u, T ))dXu .
+
0 x
Hint: use the martingale property and the Ito formula.

d) Show that the discounted portfolio price Vt = Vt /P (t, T ) satisfies
C
(Xt , T t, v(t, T ))dXt
x
P (t, S) C
tT .
=
(Xt , T t, v(t, T ))(tS tT )dB
P (t, T ) x
dVt =
e) Show that
dVt = P (t, S)
C
(Xt , T t, v(t, T ))(tS tT )dBt + Vt dP (t, T ).
x
398
"

f) Show that
C
dVt = P (t, S)
x
g) Compute the hedging strategy (tT , tS )t[0,T ] of the bond option.
h) Show that

C
log(x/K) + v 2 /2
(x, , v) =
.
x
v
Exercise 12.7 Given n bonds
with maturities T1 , . . . , Tn , consider the annuity
Pj1
numeraire P (t, Ti , Tj ) = k=i (Tk+1 Tk )P (t, Tk+1 ) and the swap rate
S(t, Ti , Tj ) =
P (t, Ti ) P (t, Tj )
P (t, Ti , Tj )
at time t [0, Ti ], modeled as

0 t Ti , (12.28)
dS(t, Ti , Tj ) = t S(t, Ti , Tj )dt + S(t, Ti , Tj )dWt ,
where (Wt )t[0,Ti ] is a standard Brownian motion under the risk-neutral measure P , (t )t[0,T ] is an adapted process and > 0 is a constant. Let
i
h r Ti

IE e t rs ds P (Ti , Ti , Tj )(S(Ti , Ti , Tj ))Ft
(12.29)
at time t [0, Ti ] of an option with payoff function .
a) Rewrite the option price (12.29) at time t [0, Ti ] using the forward swap
i,j defined from the annuity numeraire P (t, Ti , Tj ).
measure P
i,j ?
b) What is the dynamics of S(t, Ti , Tj ) under the forward swap measure P
c) Write down the value of the above option price (12.29) using a Gaussian
integral.
d) Apply the above to the computation at time t [0, Ti ] of the put swaption
price
i
h r Ti

IE e t rs ds P (Ti , Ti , Tj )( S(Ti , Ti , Tj ))+ Ft
with strike , using the Black-Scholes formula.
have
IE[( em+X )+ ] = ((m log )/v) em+
v2
2
(v (m log )/v),
"
399
N. Privault
Exercise 12.8 Consider a bond market with two bonds with maturities T1 ,
T2 , whose prices P (t, T1 ), P (t, T2 ) at time t are given by
dP (t, T1 )
= rt dt + 1 (t)dBt ,
P (t, T1 )
dP (t, T2 )
= rt dt + 2 (t)dBt ,
P (t, T2 )
where (rt )tR+ is a short term interest rate process, (Bt )tR+ is a standard
Brownian motion generating a filtration (Ft )tR+ , and 1 (t), 2 (t) are volatility processes. The LIBOR rate L(t, T1 , T2 ) is defined by
L(t, T1 , T2 ) =
P (t, T1 ) P (t, T2 )
.
P (t, T2 )
Recall that a caplet on the LIBOR market can be priced at time t [0, T1 ]
as
i
i
h r T2
h
+
(L(T1 , T1 , T2 ) )+ Ft ,
IE e t rs ds (L(T1 , T1 , T2 ) ) Ft = P (t, T2 )IE
(12.30)
defined by
r T1
P (T1 , T2 )
dP
= e 0 rs ds
,
dP
P (0, T2 )
under which
t := Bt
B
wt
0
t R+ ,
2 (s)ds,
(12.31)
is a standard Brownian motion.

In the sequel we let Lt = L(t, T1 , T2 ) for simplicity of notation.
a) Using Ito calculus, show that the LIBOR rate satisfies
t ,
dLt = Lt (t)dB
0 t T1 ,
(12.32)
where the LIBOR rate volatility is given by

(t) =
P (t, T1 )(1 (t) 2 (t))

.
P (t, T1 ) P (t, T2 )
b) Solve the equation (12.32) on the interval [t, T1 ], and compute LT1 from
the initial condition Lt .
c) Assuming that (t) in (12.32) is a deterministic function, show that the
price
i
h
(LT )+ Ft
P (t, T2 )IE
1
of the caplet can be written as P (t, T2 )C(Lt , v(t, T1 )), where v 2 (t, T1 ) =
w T1
|(s)|2 ds, and C(t, v(t, T1 )) is a function of Lt and v(t, T1 ).
t
400
"

d) Consider a portfolio (t1 , t2 )t[0,T1 ] made of bonds with maturities T1 , T2
and value
Vt = t1 P (t, T1 ) + t2 P (t, T2 ),
at time t [0, T1 ]. We assume that the portfolio is self-financing, i.e.
dVt = t1 dP (t, T1 ) + t2 dP (t, T2 ),
0 t T1 ,
(12.33)
and that it hedges the claim (LT1 )+ , so that

i
i
h r T1
h
+
(LT )+ Ft ,
Vt = IE e t rs ds (P (T1 , T2 )(LT1 )) Ft = P (t, T2 )IE
1
0 t T1 . Show that we have
i
h r T1
+
IE e t rs ds (P (T1 , T2 )(LT1 )) Ft
h
i wt
wt
(LT )+ +
= P (0, T2 )IE
s1 dP (s, T1 ) +
s2 dP (s, T1 ),
1
0
0 t T1 .
e) Show that under the
dVt = t1 dP (t, T1 ) + t2 dP (t, T2 ),
where P (t, Tk ) = e
prices.
f) Show that
rt
0
rs ds
P (t, Tk ), k = 1, 2, denote the discounted bond
i
h
h
i w t C
(LT )+ Ft = IE
(LT )+ +
(Lu , v(u, T1 ))dLu ,
IE
1
1
0 x
and that the discounted portfolio price Vt = Vt /P (t, T2 ) satisfies
dVt =
C
C
t .
(Lt , v(t, T1 ))dLt = Lt
(Lt , v(t, T1 ))(t)dB
x
x

g) Show that
dVt = (P (t, T1 ) P (t, T2 ))
C
(Lt , v(t, T1 ))(t)dBt + Vt dP (t, T2 ).
x
h) Show that
dVt =
"

C
C
(Lt , v(t, T1 ))d(P (t, T1 )P (t, T2 ))+ Vt Lt
(Lt , v(t, T1 )) dP (t, T2 ),
x
x
401
N. Privault
and deduce the values of the hedging portfolio (t1 , t2 )tR+ .
Exercise 12.9 Consider a bond market with tenor structure {Ti , . . . , Tj } and
bonds with maturities Ti , . . . , Tj , whose prices P (t, Ti ), . . . P (t, Tj ) at time t
are given by
dP (t, Tk )
= rt dt + k (t)dBt ,
P (t, Tk )
k = i, . . . , j,
where (rt )tR+ is a short term interest rate process and (Bt )tR+ denotes a standard Brownian motion generating a filtration (Ft )tR+ , and
i (t), . . . , j (t) are volatility processes.
The swap rate S(t, Ti , Tj ) is defined by
S(t, Ti , Tj ) =
where
P (t, Ti , Tj ) =
P (t, Ti ) P (t, Tj )
,
P (t, Ti , Tj )
j1
X
(Tk+1 Tk )P (t, Tk+1 )
k=i
is the annuity numeraire. Recall that a swaption on the LIBOR market can
be priced at time t [0, Ti ] as
!+
j1

rT
X

t i rs ds
(Tk+1 Tk )P (Ti , Tk+1 )(S(Ti , Tk , Tk+1 ) )
IE e
Ft
k=i
i
h
+
= P (t, Ti , Tj ) IEi,j (S(Ti , Ti , Tj ) ) Ft ,
(12.34)
under the forward swap measure Pi,j defined by

r Ti
dPi,j
P (Ti , Ti , Tj )
= e 0 rs ds
,
dP
P (0, Ti , Tj )
1 i < j n,
under which
ti,j := Bt
B
j1
X
k=i
(Tk+1 Tk )
P (t, Tk+1 )
k+1 (t)dt
P (t, Ti , Tj )
(12.35)
is a standard Brownian motion. Recall that the swap rate can be modeled as
ti,j ,
dS(t, Ti , Tj ) = S(t, Ti , Tj )i,j (t)dB
0 t Ti ,
(12.36)
where the swap rate volatilities are given by

402
"

j1
X
P (t, Tl+1 )
P (t, Tj )
(i (t)l+1 (t))+
(i (t)j (t))
P (t, Ti , Tj )
P (t, Ti ) P (t, Tj )
l=i
(12.37)
1 i, j n, cf. e.g. Proposition 10.8 of [89]. In the sequel we denote St =
S(t, Ti , Tj ) for simplicity of notation.
i,j (t) =
(Tl+1 Tl )
a) Solve the equation (12.36) on the interval [t, Ti ], and compute S(Ti , Ti , Tj )
from the initial condition S(t, Ti , Tj ).
b) Assuming that i,j (t) is a deterministic function of t for 1 i, j n, show
that the price (12.24) of the swaption can be written as
P (t, Ti , Tj )C(St , v(t, Ti )),
where
v 2 (t, Ti ) =
w Ti
t
|i,j (s)|2 ds,
and C(x, v) is a function to be specified using the Black-Scholes formula

BS(K, x, , r, ), with
IE[(xem+X K)+ ] = (v + (m + log(x/K))/v) K((m + log(x/K))/v),
where X is a centered Gaussian random variable with mean m = r v 2 /2
and variance v 2 .
c) Consider a portfolio (ti , . . . , tj )t[0,Ti ] made of bonds with maturities
Ti , . . . , Tj and value
j
X
Vt =
tk P (t, Tk ),
k=i
at time t [0, Ti ]. We assume that the portfolio is self-financing, i.e.

dVt =
j
X
tk dP (t, Tk ),
0 t Ti ,
(12.38)
k=i
and that it hedges the claim (S(Ti , Ti , Tj ) )+ , so that
!+
j1

rT
X

t i rs ds
Vt = IE e
Ft
k=i
i
h
+
= P (t, Ti , Tj ) IEi,j (S(Ti , Ti , Tj ) ) Ft ,
0 t Ti . Show that
!+
j1

rT
X

t i rs ds
IE e
Ft
k=i
"
403
N. Privault
j w
h
i X
t
+
= P (0, Ti , Tj ) IEi,j (S(Ti , Ti , Tj ) ) +
sk dP (s, Ti ),
k=i
0 t Ti .
d) Show that under the
dVt =
j
X
tk dP (t, Tk ),
k=i
rt
where P (t, Tk ) = e 0 rs ds P (t, Tk ), k = i, . . . , j, denote the discounted

bond prices.
e) Show that
i
h
+
IEi,j (S(Ti , Ti , Tj ) ) Ft
h
i w t C
+
= IEi,j (S(Ti , Ti , Tj ) ) +
(Su , v(u, Ti ))dSu .
0 x
f) Show that the discounted portfolio price Vt = Vt /P (t, Ti , Tj ) satisfies
dVt =
C
C
ti,j .
(St , v(t, Ti ))dSt = St
(St , v(t, Ti ))ti,j dB
x
x
g) Show that
dVt = (P (t, Ti ) P (t, Tj ))
C
(St , v(t, Ti ))ti,j dBt + Vt dP (t, Ti , Tj ).
x
h) Show that
j1
dVt = St i (t)
X
C
(St , v(t, Ti ))
(Tk+1 Tk )P (t, Tk+1 )dBt
x
k=i
j1
+(Vt St
X
C
(St , v(t, Ti )))
(Tk+1 Tk )P (t, Tk+1 )k+1 (t)dBt
x
k=i
C
+
(St , v(t, Ti ))P (t, Tj )(i (t) j (t))dBt .
x
i) Show that
dVt =
C
(St , v(t, Ti ))d(P (t, Ti ) P (t, Tj ))
x
C
+(Vt St
(St , v(t, Ti )))dP (t, Ti , Tj ).
x
404
"

j) Show that
C
(x, v(t, Ti )) =
x
log(x/K) v(t, Ti )
+
v(t, Ti )
2

.
k) Show that we have

log(St /K) v(t, Ti )
dVt =
+
d(P (t, Ti ) P (t, Tj ))
v(t, Ti )
2

dP (t, Ti , Tj ).
v(t, Ti )
2
l) Show that the hedging strategy is given by

+
,
ti =
v(t, Ti )
2

(Tj Tj1 )
,
tj =
+
v(t, Ti )
2
v(t, Ti )
2
and
tk = (Tk+1 Tk )
"
v(t, Ti )
2

,
i k j 2.
405
Chapter 13
Credit Risk, CDSs and CDOs
The bond pricing model of Chapter 11 is based on the terminal condition

P (T, T ) = $1, i.e. the bond payoff at maturity is always equal to $1, and
default never occurs. In this chapter we allow for the possibility of default
at a random time , in which case the terminal payoff of a bond vanishes at
maturity. We also consider the credit default options (swaps) that can act as
a protection against default.
13.1 Stochastic Default

Survival Probabilities and failure rate
Given t > 0, let P( t) denote the probability that a random system
with lifetime survives at least t years. Assuming that survival probabilities
P( t) are strictly positive for all t > 0, we can compute the conditional
probability for that system to survive up to time T , given that it was still
functioning at time t [0, T ], as
P( > T | > t) =
P( > T and > t)

P( > T )
=
,
P( > t)
P( > t)
0 t T,
with
P( < T | > t) = 1 P( > T | > t)
P( > t) P( > T )
=
P( > t)
P( < T ) P( < t)
=
P( > t)
P(t < < T )
=
,
0 t T,
P( > t)
"
N. Privault
and the conditional survival probability law
P( dx | > t) = P(x < < x + dx | > t)
= P( < x + dx | > t) P( < x | > t)
P( < x + dx) P( < x)
=
P( > t)
1
=
dP( < x)
P( > t)
1
=
dP( > x),
x > t.
P( > t)
From this we can deduce the failure rate function
(t) :=
=
=
=
=
P( < t + dt | > t)
dt
P(t < < t + dt)
1
P( > t)
dt
1
P( > t) P( > t + dt)
P( > t)
dt
d
log P( > t)
dt
1
d
P( > t),
t > 0,
P( > t) dt
which satisfies the differential equation

d
P( > t) = (t)P( > t),
dt
which can be solved as

w
t
P( > t) = exp (u)du ,
0
t R+ ,
(13.1)
under the initial condition P( > 0) = 1. This allows us to rewrite the survival
probability as
w

T
P( > T )
(u)du ,
0 t T,
P( > T | > t) =
= exp
t
P( > t)
with
P( > t + h | > t) = e(t)h ' 1 (t)h,
[h & 0],
(13.2)
and
408
"

P( < t + h | > t) = 1 e(t)h ' (t)h,
[h & 0],
(13.3)
as h tends to 0. When the failure rate (t) = > 0 is a constant function of

time, Relation (13.1) shows that
P( > T ) = eT ,
T 0,
i.e. has the exponential distribution with parameter . Note that given
(n )n1 a sequence of i.i.d. exponentially distributed random variables, letting
Tn = 1 + + n ,
n 1,
defines the sequence of jump times of a standard Poisson process with intensity > 0, cf. Section 14.1 below for details.
Stochastic Default
We now model the failure rate function (t )tR+ as a random process adapted
to a filtration (Ft )tR+ .
In case the random time is a stopping time with respect to (Ft )tR+ , i.e.
the knowledge of whether default already occurred at time t is contained in
Ft , t R+ , and we have
{ > t} Ft ,
t R+ ,
cf. Section 9.3, we have

P( > t | Ft ) = IE 1{ >t} | Ft = 1{ >t} ,
t R+ .
In the sequel we will not assume that is an Ft -stopping time, and by analogy
with (13.1) we will write P( > t | Ft ) as
w

t
P( > t | Ft ) = exp u du ,
t > 0.
(13.4)
0
This is the case in particular in [71] when u has the form u = h(Xu ), and
is given by

wt
= inf t R+ :
h(Xu )du L ,
0
where h is a non-negative function, (Xt )tR+ is a process generating a filtration (Ft )tR+ , and L is an independent exponentially distributed random
variable.
We let (Gt )tR+ be the filtration defined by G := F ( ), and
"
409
N. Privault

Gt := B G : A Ft such that A { > t} = B { > t} , (13.5)
t R+ , i.e. Gt contains the additional information on whether default at time
has occurred or not before time t. The process t can also be chosen among
the classical mean-reverting diffusion processes, including jump-diffusion processes.
Taking F = 1 in the next Lemma 13.1 shows that the survival probability
up to time T , given information known up to t, is given by

P( > T | Gt ) = IE 1{ >T } | Gt
(13.6)

w

T

= 1{ >t} IE exp
u du Ft , 0 t T.
t
Lemma 13.1. ([45]) For any FT -measurable integrable random variable F

we have

w

T

IE F 1{ >T } | Gt = 1{ >t} IE F exp
u du Ft .
t

w

T
w

exp
u du
T
0
P( > T | FT )
w
= exp
=
u du ,
t
t
P( > t | Ft )
exp u du
0
hence, since F is FT -measurable,

w

T
P( > T | FT )

1{ >t} IE F exp
u du Ft = 1{ >t} IE F
Ft
t
P( > t | Ft )
i
h
1{ >t}

=
IE F IE[1{ >T } | FT ] Ft
P( > t | Ft )
i
h
1{ >t}

=
IE IE[F 1{ >T } | FT ] Ft
P( > t | Ft )

IE F 1{ >T } Ft
= 1{ >t}
P( > t | Ft )

= 1{ >t} IE F 1{ >T } Gt

= IE F 1{ >T } Gt ,
0 t T.
In the last step of the above argument we used the key lemma
i
h

1{ >t} IE F 1{ >T } Gt =
i
h
1{ >t}

IE F 1{ >T } Ft ,
P( > t | Ft )
410
"

cf. Relation (75.2) in XX-75 of [20], Theorem VI-3-14 of [96], and Lemma 3.1
of [30], which parallels e.g. (16.23) in the appendix, under the probability
measure P|Ft , 0 t T . Indeed, according to (13.5), for any B Gt we
have, for some A Ft ,

IE 1B 1{ >t} F 1{ >T } = IE 1B{ >t} F 1{ >T }

= IE 1A{ >t} F 1{ >T }

= IE 1A 1{ >t} F 1{ >T }

IE[1{ >t} | Ft ]
= IE 1A 1{ >t}
F 1{ >T }
P({ > t} | Ft )

IE[1A 1{ >t} | Ft ]
= IE
F 1{ >T }
P({ > t} | Ft )

IE[1A 1{ >t} | Ft ]
= IE
IE F 1{ >T } | Ft
P({ > t} | Ft )

1A 1{ >t}
= IE
IE F 1{ >T } | Ft
P({ > t} | Ft )

IE[1A 1{ >t} | Ft ]
= IE
IE F 1{ >T } | Ft
P({ > t} | Ft )

1A 1{ >t}
IE F 1{ >T } | Ft
= IE
P({ > t} | Ft )

1B 1{ >t}
= IE
IE F 1{ >T } | Ft ,
P({ > t} | Ft )
hence by the characterization (16.32) we have

IE 1{ >t} F 1{ >T } | Gt =

1{ >t}
IE F 1{ >T } | Ft
P({ > t} | Ft )
The computation of P( > T | Gt ) according to (13.6) is then similar to that

of a bond price, by considering the failure rate (t) as a virtual short term
interest rate. In particular the failure rate (t, T ) can be modeled in the HJM
framework of Chapter 11.4, and

w

T

P( > T | Gt ) = IE exp
(t, u)du Ft
t
can then be computed by applying HJM bond pricing techniques.

The computation of expectations given Gt as in Lemma 13.1 can be useful
for pricing under insider trading, in which the insider has access to the augmented filtration Gt while the ordinary trader has only access to Ft , therefore
"
411
N. Privault
generating two different prices IE [F | Ft ] and IE [F | Gt ] for the same claim
F under the same risk-neutral measure P . This leads to the issue of computing the dynamics of the underlying asset price by decomposing it using a
Ft -martingale vs a Gt -martingale instead of using different forward measures
as in in 12.1. This can be obtained by the technique of enlargement of
filtration, cf. [61], [29], [55], [118].
13.2 Defaultable Bonds

The price of a default bond with maturity T , (random) default time and
(possibly random) recovery rate [0, 1] is given by

w

T

P (t, T ) = IE 1{ >T } exp
ru du Gt
t

w

T

+ IE 1{ T } exp
ru du Gt ,
0 t T,
t
r

T
Taking F = exp t ru du in Lemma 13.1, we get

w

w

T
T

IE 1{ >T } exp
ru du Gt = 1{ >t} IE exp
(ru + u )du Ft ,
t
cf. e.g. [71], [45], [26],
hence

w

T

P (t, T ) = 1{ >t} IE exp
(ru + u )du Ft
t

w

T

+ IE 1{ T } exp
ru du Gt , 0 t T.
t
In the case of complete default (zero-recovery) we have = 0 and

w

T

P (t, T ) = 1{ >t} IE exp
(rs + s )ds Ft , 0 t T.
t
(13.7)
From the above expression (13.7) we note that the effect of the presence of
a default time is to decrease the bond price, which can be viewed as an
increase of the short rate by the amount u .
This treatment of default risk parallels the pricing formula

w

T

P (t, T ) = ec(T t) IE exp
rs ds Gt ,
t
412
"

of coupon bonds where c > 0 is a continuous-time deterministic coupon rate,
under the identification c = .
Finally, from Proposition 12.1 the bond price (13.7) can also be expressed
with maturity T , as

w

T

P (t, T ) = 1{ >t} IE exp
(rs + s )ds Ft
t

w

w

T
T

= 1{ >t} IE exp
rs ds Ft IEP exp
s ds Ft
t
t

w

T

= 1{ >t} IE exp
rs ds Ft Q(t, T ),
t
where

w

T

Q(t, T ) = IEP exp
s ds Ft
t
cf. [13], [12].

denotes the survival probability under the forward measure P,
13.3 Credit Default Swaps

We work with a tenor structure {t = Ti < < Tj = T }.
A Credit Default Swap (CDS) is a contract consisting in
- A premium leg: the buyer is purchasing protection at time t against default at time Tk , k = i + 1, . . . , j, and has to make a fixed payment St at
times Ti+1 , . . . , Tj between t and T in compensation.
The discounted value at time t of the premium leg is
"j1
w
#
X
Tk+1

V (t, T ) = IE
St k 1{ >Tk+1 } exp
rs ds Gt
k=i
j1
X
k=i
= St

w

Tk+1

St k IE 1{ >Tk+1 } exp
rs ds Gt
j1
X
k P (t, Tk+1 )
k=i
= St P (t, Ti , Tj ),
where k = Tk+1 Tk , P (t, Ti , Tj ) is the annuity numeraire (12.14), and
"
413
N. Privault

w

Tk

P (t, Tk ) = 1{ >t} IE exp
(rs + s )ds Ft ,
t
0 t Tk ,
is the defaultable bond price with maturity Tk , k = i, . . . , j1. For simplicity we have ignored a possible accrual interest term over the time period
[Tk , ] when [Tk , Tk+1 ] in the above value of the premium leg.
- A protection leg: the seller or issuer of the contract makes a payment
1 k+1 to the buyer in case default occurs at time Tk+1 , k = i, . . . , j 1.
The value at time t of the protection leg is
"j1
w
#
X
Tk+1

IE
1(Tk ,Tk+1 ] ( )(1 k+1 ) exp
rs ds Gt ,
t
k=i
where k+1 is the recovery rate associated with the maturity Tk+1 , k =
i, . . . , j 1.
In the case of a non-random recovery rate k the value of the protection
leg becomes

w

j1
X
Tk+1

(1 k+1 ) IE 1(Tk ,Tk+1 ] ( ) exp
rs ds Gt .
t
k=i
The spread St is computed by equating the values of the protection and

premium legs, i.e. from the relation
V (t, T ) = St P (t, Ti , Tj )
"j1
w
#
X
Tk+1

= IE
1(Tk ,Tk+1 ] ( )(1 k+1 ) exp
rs ds Gt ,
t
k=i
which yields
St =

w

j1
X
Tk+1
1

rs ds Gt .
IE 1(Tk ,Tk+1 ] ( )(1 k+1 ) exp
t
P (t, Ti , Tj )
k=i
In the case of a constant recovery rate we find

St =

w

j1
X
Tk+1
1

rs ds Gt ,
IE 1(Tk ,Tk+1 ] ( ) exp
t
P (t, Ti , Tj )
k=i
and if is constrained to take values in the tenor structure {t = Ti , . . . , Tj },

we get
414
"
St =
h
w
i
1

IE 1(t,T ] ( ) exp
rs ds Gt .
t
P (t, Ti , Tj )
13.4 Correlated default times

Bernoulli random variables
Given a choice of modeling based on the distributions of two random variables X and Y , it is natural to consider a dependence structure between X
and Y .
Consider two Bernoulli random variables X and Y , with
pX = P(X = 1) = IE[1{X=1} ]
and pY = P(Y = 1) = IE[1{Y =1} ]
and correlation
=
IE[XY ] IE[X] IE[Y ]

IE[XY ] pX pY
p
=p
.
Var [X]Var [Y ]
pX (1 pX )pY (1 pY )
We note that in that case the joint distribution P(X = i, Y = j), i, j {0, 1}
is fully determined by P(X = 1), P(Y = 1) and the correlation , as
P(X = 1, Y = 1) = IE[XY ]
= pX pY + pX (1 pX )pY (1 pY ),
P(X = 0, Y = 1) = IE[(1 X)Y ]
= (1 pX )pY pX (1 pX )pY (1 pY ),
P(X = 1, Y = 0) = IE[X(1 Y )]
= pX (1 pY ) pX (1 pX )pY (1 pY ),
P(X = 0, Y = 0) = IE[(1 X)(1 Y )]
= (1 pX )(1 pY ) + pX (1 pX )pY (1 pY ).
Gaussian random variables

Consider now two centered Gaussian random variables X ' N (0, 2 ) and
Y ' N (0, 2 ) with probability density functions
fX (x) =
"
1
2 2
ex
/(2 2 )
and fY (x) = p
1
2 2
ex
/(2 2 )
x R.
415
N. Privault
When the covariance matrix
IE[X 2 ] IE[XY ]
2 IE[XY ]
=
=
IE[XY ] IE[Y 2 ]
IE[XY ] 2
with determinant
det = IE[X 2 ] IE[Y 2 ] (IE[XY ])2
= IE[X 2 ] IE[Y 2 ](1 ( corr (X, Y ))2 )
0,
is invertible, there exists a probability density function
T IE[X 2 ] IE[XY ] 1
1
1
x
x
f (x, y) =
exp
(13.8)
y
2 y
2 det
IE[XY ] IE[Y 2 ]
with respective marginals N (0, 2 ) and N (0, 2 ). The probability density
function (13.8) is called the centered joint Gaussian density with covariance
matrix .
When corr (X, Y ) = 1 the joint density function f (x, y) is not defined.
Note that, although f (x, y) always defines a couple of Gaussian random
variables with marginals N (0, 2 ) and N (0, 2 ), it may not coincide with the
actual joint density of the couple (X, Y ), when it admits such a density.
For example, (X, Y ) may have the joint density
f(x, y) :=
2
2
2
2
2
2
2
2
1
1
1 2 (x, y)ex /(2 )y /(2 ) +
1 2 (x, y)ex /(2 )y /(2 )
R
R+
which is not a joint Gaussian density and does not coincide with f (x, y),
although (X, Y ) has does have the Gaussian marginals N (0, 2 ) and N (0, 2 ),
according to the following computation:
2
2
2
2
1 w
1 2 2 (x, y)ex /(2 )y /(2 ) dy
R R+
w
w0
2
2
2
2
1 x2 /(22 )
1 x2 /(22 )
=
e
1R+ (x)
e
1R (x)
ey /(2 ) dy +
ey /(2 ) dy
0
2
2
2
2
1
1
= ex /(2 ) 1R+ (x) + ex /(2 ) 1R (x)
2
2
2
2
1
= ex /(2 ) ,
x R,
2
f (x, y)dy =
416
"

see here for a detailed discussion. Note that X + Y is not even Gaussian.
When = = 1 it has a density
1 w a w ax x2 /2y2 /2
1 w a x2 /2(ax)2 /2
e
e
dydx =
dx
a 0 0
0
a2 /2 w a
2
2
e
=
e(( 2xa/ 2) a /2)/2 dx
0
2
ea /4 w a 2 ((xa/2)2 )/2
=
dx
e
2 0
2
ea /4 w a( 21/ 2) x2 /2
=
e
dx
2 a 2
a2 /4 w a/ 2
e
x2 /2
=
dx
e
2 a 2
2
2
= ea /4 (a 2),
a 0,
which vanishes at a = 0. In general, the sum of two Gaussian variables is not

Gaussian unless (X, Y ) has a joint Gaussian distribution.
In this example the random variables X and Y are positively correlated,
as
w
2
2
2
2
1 w
yf (x, y)dy =
1 2 2 (x, y)yex /(2 )y /(2 ) dy
R R+
w
w0
2
2
2
2
1 x2 /(22 )
1 x2 /(22 )
e
1R+ (x)
yey /(2 ) dy +
e
1R (x)
yey /(2 ) dy
=
0
x2 /(22 )
x2 /(22 )
=
e
1R+ (x)
e
1R (x),
hence
w w
xyf (x, y)dydx

2
2
w0
w x2 /(22 )
xe
dx
xex /(2 ) dx
=
0

2
=
,
IE[XY ] =
and
=
Under a rotation

R=
IE[XY ]
2
= .

cos sin
,
sin cos
of angle [0, 2] we would find

"
417
N. Privault
IE[(X cos Y sin )(X sin + Y cos )]
= sin cos IE[X 2 ] + (cos2 sin2 ) IE[XY ] sin cos IE[Y 2 ]
2
= 2 sin cos + (cos2 sin2 )
2 sin cos
2
2
2
=
sin(2) + cos(2)
sin(2),
2
2
and
=
sin(2) + cos(2)
sin(2), ,
2
2
i.e. = /4 and = would lead to uncorrelated random variables.
Copulas and dependence structure

The word copula derives from the Latin noun for a link or tie that
connects two different things (Wikipedia). A two-dimensional copula is any
cumulative distribution function
C : [0, 1] [0, 1] [0, 1]
(u, v) 7 C(u, v)
of two uniform random variables on [0, 1]. In other words, any copula function
C(u, v) can be written as
C(u, v) = P(U u, V v),
u, v [0, 1],
where U and V are uniform random variables on [0, 1].
1.0
1.0
0.8
1.0
0.8
0.8
0.4
0.2
0.6
C(x,y)
0.6
C(x,y)
C(x,y)
0.6
0.4
0.2
0.0
1.0
0.8
0.0
1.0
0.8
0.6
1.0
1.0
0.8
0.4
0.6
0.4
0.2
0.0
1.0
0.8
0.4
0.6
0.4
0.2
0.2
0.0
0.6
y
0.4
0.8
0.6
y
0.8
0.4
0.2
0.0
1.0
0.6
0.4
0.2
0.2
0.0
0.0
0.2
0.0
0.0
Fig. 13.1: Different Gaussian copula graphs for = 0, = 0.85 and = 1.
In the above leftmost figure we have C(u, v) = uv, which is the copula
418
"

C(u, v) = P(U u, V v) = P(U u)P(U u) = uv,
u, v [0, 1],
of two independent uniformly distributed random variables U , V on [0, 1]. On

the other hand in the rightmost figure we have C(u, v) = min(u, v), which is
the copula
C(u, v) = P(U u, V v)
= P(U u, U v)
= P(U min(u, v))
u, v [0, 1],
= min(u, v)),
of two equal uniformly distributed random variables U = V on [0, 1].
1.0
3
15000
0.8
0.4
2
10000
c(x,y)
c(x,y)
c(x,y)
0.6
5000
0.2
0.0
1.0
0
1.0
0.8
0
1.0
0.8
0.6
1.0
0.8
0.4
0.6
0.4
0.2
0.0
1.0
0.8
0.4
0.6
0.4
0.2
0.2
0.0
0.6
1.0
0.8
0.4
0.8
0.6
0.6
0.4
0.2
0.2
0.0
0.2
0.0
0.0
0.0
Fig. 13.2: Different Gaussian copula density graphs for = 0, = 0.35 and = 0.999.
The leftmost figure above represents a uniform (product) probability density
on the square [0, 1] [0, 1], which corresponds to two independent uniformly
distributed random variables U , V on [0, 1]. The rightmost figure shows the
probability distribution of the fully correlated couple (U, U ), which does not
admit a probability density on the square [0, 1] [0, 1].
Any joint cumulative distribution function FX1 ,...,Xn (x1 , . . . , xn ) with
marginals
FXi (x) = FX1 ,...,Xn (, . . . , +, x, , . . . , +),
x R,
i = 1, . . . , n,
defines a dependence structure that can be encoded in an n-dimensional

copula C(u1 , . . . , un ) constructed as

1
1
(u1 ), . . . FX
(un ) ,
u1 , . . . , un R,
C(u1 , . . . , un ) := FX1 ,...,Xn FX
1
n
where C(u1 , . . . , un ) defines the joint cumulative distribution function of a
family of uniformly distributed random variables on [0, 1]n . Indeed it can be
"
419
N. Privault
checked that C(u1 , . . . , un ) has uniform marginal distributions on [0, 1], as
C(1, . . . , 1, x, 1, . . . , 1)

1
1
1
1
1
= FX1 ,...,Xn FX
(1), . . . FX
(1), FX
(x), FX
(1), . . . FX
(1)
1
i1
i
i+1
n

1
= FX1 ,...,Xn , . . . , FX
(x), , . . .
i
1
= FX i (FX
(x))
i
= x,
x R.
Relation (13.10) can be similarly extended to generate new dependence struc1, . . . , X

n ) of random variables with cumutures based on another family (X
lative distribution functions FX 1 , . . . , FX n , as

FC (x1 , . . . , xn ) := C FX 1 (x1 ), . . . FX n (xn ) ,
x1 , . . . , xn R.
In that case we note that the marginal distributions generated by FC (x1 , . . . , xn )

1 , . . . , X),
as we have
still coincide with the respective distributions of (X
FC (, . . . , +, x, , . . . , +)

= C FX 1 (), . . . FX i1 (), FX i (x), FX i+1 (), . . . FX n ()

= C 1, . . . 1, FX i (x), 1, . . . 1,
= FX i (x),
x R.
Gaussian copulas
The choice of (13.8) above as joint density function actually induces a particular dependence structure between the Gaussian random variables X and
Y , and corresponding to the joint cumulative distribution function
(x, y) := P(X x, Y y)

wx wy
1
1
u
u
exp
=
, 1
dudv,
v
v
2
2 det
x, y R. Letting
FX (x) := P(X x)
and FY (y) := P(Y y),
denote the cumulative distribution functions of X and Y , the random variables FX (X) and FY (Y ) are known to be uniformly distributed on [0, 1], and
(FX (X), FY (Y )) is a [0, 1] [0, 1]-valued random variable with joint cumulative distribution function
C (x, y) := P (FX (X) x, FY (Y ) y)
420
"

1
= P X FX
(x), Y FY1 (y)

1
= FX
(x), FY1 (y) ,
x, y R.
The function C (x, y), which is the joint cumulative distribution function
of a couple of uniformly distributed random variables on [0, 1], is called the
Gaussian copula generated by the jointly Gaussian distribution of (X, Y )
with covariance matrix .
The Gaussian copula C (x, y) admits a probability density function on
[0, 1] [0, 1] given by
2 C
(x, y)
xy
2
1
=
F FX
(x), FY1 (y)
xy

1
F
1
=
FX
(x), FY1 (y)
1
0
x FY (FY (y)) v

F
1
1
1
=
F
(x),
F
(y)
X
Y
x fY (FY1 (y)) v
c (x, y) =

1
2 F
F 1 (x), FY1 (y)
1
fX (FX
(x))fY (FY1 (y)) uv X

1
f FX
(x), FY1 (y)
,
=
1
1
fX (FX (x))fY (FY (y))
=
hence the Gaussian copula C (x, y) can be computed as

wxwy
C (x, y) =
c (u, v)dudv
0 0
w x w y f F 1 (u), F 1 (v)
X
Y
=
dudv,
x, y [0, 1].
0 0 fX (F 1 (u))fY (F 1 (v))
X
Y
The joint cumulative distribution function F (x, y) of (X, Y ) can be recovered as
F (x, y) = C (FX (x), FY (y)),
x, y R.
(13.9)
from the Gaussian copula C (x, y) and the respective cumulative distribution functions FX (x), FY (y) of X and Y .
In that sense, the Gaussian copula C (x, y) encodes the Gaussian dependence structure of the covariance matrix . Moreover, the Gaussian copula
C (x, y) can be used to generate a joint distribution function FC (x, y) by
letting
FC (x, y) := C (FX (x), FY (y)),
x, y R,
(13.10)
"
421
N. Privault
based on other, possibly non-gaussian cumulative distribution functions FX (x),
FY (y) of two random variables X and Y . In this case we note that the
marginals of the cumulative distribution function FC (x, y) are FX (x) and
FY (y) because C (x, y) has uniform marginals on [0, 1].
Correlated default times

Given 1 and 2 two default times we can also define the correlation
Cov(1 , 2 )
[1, 1].
= p
Var[1 ] Var[2 ]
The two default events A = {1 T } and B = {2 T } with probabilities
w

w

T
T
P({1 T }) = 1exp
1 (s)ds and P({2 T }) = 1exp
2 (s)ds
0
and the correlation

= p
P(A B) P(A)P(B)
p
[1, 1].
P(A)(1 P(A)) P(B)(1 P(B))
(13.11)
When trying to build a dependence structure for the default times 1 and 2 ,
the idea of [74] is to use the normalized Gaussian copula C (x, y), with

1
=
,
1
with correlation parameter [1, 1], and to model the joint default probability P(1 T, 2 T ) as
P(1 T, 2 T ) := C (P(1 T ), P(2 T )) .
From this expression, the default correlation (13.11) can be expressed as
C(P(1 T ), P(2 T )) P(1 T )P(2 T )
p
.
= p
P(1 T )(1 P(1 T )) P(2 T )(1 P(2 T ))
In [74] it is suggested to use a single average correlation estimate (8.1) page 82
of the Credit MetricsTM Technical Document, see also Appendix F therein.
The outcomes of this methodology have been discussed in a number of
magazine articles in recent years, to name a few:
Recipe for disaster: the formula that killed Wall Street, Wired Magazine,
23 February 2009, by Felix Salmon [103];
422
"

The formula that felled Wall Street, Financial Times Magazine, April
24 2009, by Sam Jones [62];
Formula from hell, Forbes.com, August 8 2009, by Susan Lee [72].
Estimating default rates

Recall the expression

P( > T | Gt ) = IE 1{ >T } | Gt

w

T

= 1{ >t} IE exp
u du Ft ,
t
(13.12)
0 t T,
of the survival probability up to time T , given information known up to t, in

terms of the hazard rate process (u )uR+ adapted to a filtration (Ft )tR+ ,
where
Gt = Ft ({ u} : 0 u t),
t R+ ,
i.e. Gt contains the additional information on whether default at time has
occurred or not before time t.
Defaultable bonds
The price of a default bond with maturity T , (random) default time and
(possibly random) recovery rate [0, 1] is given by

w

T

P (t, T ) = 1{ >t} IE exp
(ru + u )du Ft
t

w

T

+ IE 1{ T } exp
ru du Gt ,
0 t T,
t
where denotes the recovery rate.

In a simplified deterministic, step function model with zero deterministic
recovery rate and tenor structure {t = Ti < < Tj = T } we have
w

T
P (t, T ) = 1{ >t} exp
(ru + u )du
t
!
n
X
= 1{ >t} exp
(ri + i )(Ti Ti1 ) ,
i=1
0 t = T0 T1 Tn = T , with
"
423
N. Privault
r(t) =
n1
X
rl 1(Tl ,Tl+1 ] (t)
and (t) =
l=0
n1
X
l 1(Tl ,Tl+1 ] (t),
t R+ ,
l=0
(13.13)
from which we can infer
k = rk +
P (Tk+1 , T )
1
log
,
Tk+1 Tk
P (Tk , T )
k = 1, . . . , n 1,
and (13.12) yields

w

T
P( > T | GTk ) = 1{ >Tk } exp
u du
Tk
= 1{ >t} exp
n
X
!
i (Ti Ti1 ) ,
k = 1, . . . , n.
i=k
Credit Default Swaps

The buyer of a Credit Default Swap (CDS) is purchasing protection at time
t against default at time Tk , k = i + 1, . . . , j, by making a fixed payment St
(the premium leg) at times Ti+1 , . . . , Tj ranging from of t to T . On the other
hand, the issuer of the contract makes a payment 1 k+1 to the buyer in
case default occurs at time Tk+1 , k = i, . . . , j 1.
The contract is priced in terms of the swap rate St computed by equating
the values of the protection and premium legs, and acts as a compensation
that makes the deal fair to both parties. Recall that we have

w

j1
X
Tk+1
1

rs ds Gt
IE 1(Tk ,Tk+1 ] ( )(1 k+1 ) exp
t
P (t, Ti , Tj )
k=i

w

j1
X
Tk+1
1

=
rs ds Gt
IE (1{Tk < } 1{Tk+1 < } )(1 k+1 ) exp
t
P (t, Ti , Tj )
k=i

w

w
j1
Tk+1
Tk
1{ >t} X
s ds
=
IE (1 k+1 ) exp
s ds exp
t
t
P (t, Ti , Tj )
k=i
w

Tk+1

exp
rs ds Ft ,
St =
which will be written as

St P (t, Ti , Tj )
= 1{ >t}
j1
X
k=i
wT

wT

wT

k+1
k
k+1
(1 k ) exp
r(s)ds exp
s ds exp
s ds
t
424
"

when 1 k+1 , r(s) and (s) are deterministic. Given that
P (t, Ti , Tj ) =
j1
X
(Tk+1 Tk )P (t, Tk+1 ),
t [Ti , Tj ],
k=i
we can write
St
j1
X
wT

k+1
(r(s) + (s))ds
(Tk+1 Tk ) exp
t
k=i
= 1{ >t}
j1
X
(1 k ) exp
k=i
wT
k+1
r(s)ds

wT
wT
k+1
k
(s)ds .
(s)ds exp
exp
t
In particular, when r(t) and (t) are written as in (13.13) and assuming
that k = , k = i, . . . , j, we get, with t = Ti and writing k = Tk+1 Tk ,
k = i, . . . , j 1,
j1
k
X
X
p (rp + p )
STi
k+1 exp
p=i
k=i
= 1{ >t} (1 )
j1
X
exp
k
X
p (rp + p ) (1 exp(k k ))) .
p=i
k=i
Assuming further that k = , k = i, . . . , j, we have
j1
k
k
X
X
X
STi
k exp
p rp
p
p=i
k=i
= (1 )
j1
X
(1 exp(k ))) exp
k
X
p=i
k=i
(13.14)
p=i
p rp
k
X
p ,
p=i
which can be solved numerically for , cf. Sections 4 and 5 of [11] on the JP
Morgan model.
Note that the ratio
(1 )
P

Pk
k
k exp p=i p rp p=i p
P

Pk
k
p=i p
k=i (1 exp(k ))) exp
p=i p rp
STi
Pj1
Pj1
k=i
tends to STi k /(1 ) 1 as tends to , and tends to + as tends to

0. Therefore the equation (13.14) admits a numerical solution as we normally
have STi k /(1 ) 1 under standard market conditions.
"
425
N. Privault
13.5 Merton model of credit risk

The Merton [79] credit risk model reframes corporate debt as an option on
a firms underlying value. Precisely the value At of a firms asset is modeled
by a geometric Brownian motion
dAt = At dt + At dBt
under the historical measure P. Recall that At is modeled as
t
dAt = rAt dt + At dB
under the risk-neutral measure P . The company debt is represented by an
amount K in bonds to be paid at maturity T , cf. 4.1 of [44].
Default occurs if AT K with probability P(AT K), and otherwise the
company shareholder are entitled to receive AT K, which can be represented
as (AT K)+ in general. The default probability P(AT < K | Ft ) can be
computed from the lognormal distribution of AT as
P(AT < K | Ft ) = P(A0 eBT +(
/2)T
< K | Ft )
= P(BT < (( 2 /2)T + log(K/A0 ))/ | Ft )

= P(BT Bt + y < (( 2 /2)T + log(K/A0 ))/)y=Bt
w ((2 /2)(T t))+log(K/At ))/
2
1
ex /(2(T t)) dx
= p
2(T t)
1 w ((2 /2)(T t))+log(K/At ))/( T t) x2 /2

=
e
dx
2

( 2 /2))(T t) + log(At /K))
= 1
T t
= 1 (d )
= (d ),
where is the cumulative distribution function of the standard normal distribution, and
( 2 /2)(T t) + log(At /K)
.
d :=
T t
Note that under the risk-neutral measure P we have, replacing with r,
P (AT < K | Ft ) = (dr ),
with
dr =
(r 2 /2)(T t) + log(At /K)
,
T t
426
"

which implies the relation
1 (P(AT < K | Ft )) =
r
T t + 1 (P (AT < K | Ft )).
Denoting by a random default time with distribution given by

P( < T | Ft ) := P(AT < K | Ft ),
we find
P( < T | Ft ) = P(AT < K | Ft )

r
= 1 (P (AT < K | Ft ))
T t

r
1
= (P ( < T | Ft ))
T t
and
P ( < T | Ft ) = P (AT < K | Ft )

r
= 1 (P(AT < K | Ft )) +
T t

r
1
= (P( < T | Ft )) +
T t .
Note that we have

P( < T | Ft ) r, as in the next figure.
1
0.8
0.6
< r
0.4
> r
0.2
0.2
0.4
0.6
x
0.8
Fig. 13.3: Function x 7 1 (x) ( r) T t/ for > r, = r, and

< r.
"
427
N. Privault
The discounted cash flow er(T t) IE [(AT K)+ | Ft ] estimated at time
t [0, T ] can be computed from the Black-Scholes formula.
Similarly, the discounted amount received by the bond holders can be
estimated at time t [0, T ] from the value of a put option with strike K on
AT , as
er(T t) IE [min(AT , K) | Ft ] = er(T t) K er(T t) IE [(K AT )+ | Ft ]
= er(T t) K At (dr+ ) + Ker(T t) (dr )
= Ker(T t) (dr ) At (dr+ ),
and it can be interpreted at the value P (t, T ) at time t [0, T ] of a default
bond with maturity T . Writing
P (t, T ) = eyt,T (T t)
1
= er(T t) IE [min(AT , K) | Ft ]
K
At
(dr+ ),
= er(T t) (dr )
K
and interpreting min(AT /K, 1) as a recovery rate gives the yield
1
log(er(T t) IE [min(1, AT /K) | Ft ])
T t
1
=r
log(IE [min(1, AT /K) | Ft ])
T t
> r.
yt,T =
13.6 Modeling the default times

Consider now a sequence (k )k=i+1,...,j of random default times. A common
practice [114], [42], [53] inspired by the Merton [79] model is to parametrize
the default probabilities associated to each k , k = i + 1, . . . , j, by the conditioning
!
1 (P(k T )) ak m
p
P(k T | M = m) =
,
k = i + 1, . . . , j,
1 a2k
(13.15)
where ak (1, 1), k = i + 1, . . . , j, and M is a random variable with
probability density function (m). Note that we have
w
P(k T ) =
P(k T | M = m)(m)dm
428
"
1 (P(k T )) ak m
p
1 a2k
!
(m)dm,
and (m) can be typically chosen as the standard normal Gaussian density
function.
Default times k satisfying the conditional distribution (13.15) can be generated as follows:
i) Define the random samples Xi+1 , . . . , Xj by
q
Xk := ak M + 1 a2k Zk ,
where M and Zi+1 , . . . , Zj are independent centered random variables
with unit variance, and Zi+1 , . . . , Zj are normal random variables with
same cumulative distribution function .
ii) Let
k := F1
((Xk )),
k = i + 1, . . . , j,
(13.16)
k
Based on the above construction we can recover (13.15) as follows:
P(k T | M = m) = P(F1
((Xk )) T | M = m)
k
= P((Xk ) Fk (T ) | M = m)
= P(Xk 1 (Fk (T )) | M = m)

q
= P ak m + 1 a2k Zk 1 (Fk (T ))
q

=P
1 a2k Zk 1 (Fk (T )) ak m
!

1
1
(F
(T
))
a
m
k
k
1 a2k
!
1 (P(k T )) ak m
p
,
k = i + 1, . . . , j.
1 a2k
= P Zk p
=
Knowing that, given the sample M , the default times k , k = i + 1, . . . , j, are

independent random variables, we can compute the joint distribution
P(i+1 Ti+1 , . . . , j Tj | M = m)
= P(i+1 Ti+1 | M = m) P(j Tj | M = m),
conditionally to M = m. This yields
w
P(i+1 Ti+1 , . . . , j Tj ) =
P(i+1 Ti+1 , . . . , j Tj | M = m)(m)dm
"
429
N. Privault
w
P(i+1 Ti+1 | M = m) P(j Tj | M = m)(m)dm
w
1
1
(P(
T
))
a
m
(P(
T
))
a
m
i+1
i+1
i+1
j
j
j
(m)dm.
q
q
=
1 a2j
1 a2i+1
=
Note that the above recovers the correct marginal distributions

P(k Tk ) = P(i+1 , . . . , k1 , k Tk , k+1 , . . . , j )
!
w
1 (P(k Tk )) ak m
p
=
(m)dm
1 a2k
w
=
P(k T | M = m)(m)dm, k = i + 1, . . . , j.
In addition we have

P(i+1 Ti+1 , . . . , j Tj ) = C P(i+1 Ti+1 ), . . . , P(j Tj ) ,
where the function
C(xi+1 , . . . , xj ) :=
(xi+1 ) ai+1 m
1 (xj ) aj m
q
q
(m)dm
1 a2j
1 a2i+1
xi+1 , . . . , xj [0, 1], is a Gaussian copula on [0, 1]N with N = j i + 1, built

as
C(xi+1 , . . . , xj ) = F (1 (xi+1 ), . . . , 1 (xj )),
from the Gaussian cumulative distribution function
w
x
a
m
x
a
m
i+1
i+1
j
j
q
(m)dm
F (xi+1 , . . . , xj ) :=
q
1 a2j
1 a2i+1
w
x
a
m
x
a
m
i+1
i+1
j
j
P Zj q
(m)dm
=
P Zi+1 q
1 a2j
1 a2i+1
w
=
P(Xi+1 xi+1 , . . . , Xj xj | M = m)(m)dm
= P(Xi+1 xi+1 , . . . , Xj xj ),
xi+1 , . . . , xj [0, 1], of the vector (Xi+1 , . . . , Xj ), with covariance matrix
430
"
a2 a1
= ...
.
..
aN a1
..
.
..
.
a1 a2
1
..
.
a1 aN
..
.
..
.
..
.
1
aN aN 1
aN 1 aN
1
with N = j i + 1, and probability density function

j
Y
(1 a2k )1/2
k=i+1
(xi+1 , . . . , xj ) =
(2)N/2
(xi+1 ai+1 m)2

2(1a2
)
i+1
(xj aj m)2
2(1a2 )
j
em /2
dm
2
which is jointly Gaussian, with marginals given by

w
w
(xi+1 , . . . , xj )dxi+1 dxk1 dxk+1 dxj
w (xk ak m) em /2
1
2
p
e 2(1ak )
dm
2
2(1 a2k )
2
w (xk ak m) m2 /2
1
2
p
e 2(1ak )
dm
2
2 1 ak
w x2k 2ak xk m+m2
1
2(1a2 )
k
p
e
dm
2
2 1 ak
2
2
k xk )
exk /2 w (ma
2
p
e 2(1ak ) dm
2
2 1 ak
2
1
exk /2 ,
xk R.
2
2
=
=
=
=
=
More generally we have

j
Y
(xi+1 , . . . , xj ) =
j
Y
(1 a2k )1/2
k=i+1
(2)N/2
(1 a2k )1/2
k=i+1
(2)N/2
j
Y
1
=e
"
x2
i+1
1a2
i+1
++
x2
j
1a2
j
(xi+1 ai+1 m)2

2(1a2
)
i+1
(xj aj m)2
2(1a2 )
j
em /2
dm
2

2
2
x2 +a2 m2 2xj aj m
x2
i+1 +ai+1 m 2xi+1 ai+1 m
++ j j
+m2
1a2
1a2
j
i+1
1
dm
2
(1 a2k )1/2
k=i+1
(2)N/2 2
431
N. Privault
w
= q
1
2
x2
i+1
1a2
i+1
++
x2
++ j 2
1a
j
a2
j
1a2
j
a2i+1
a2i+1
2
xi+1
1a2
i+1

+2m
+ +
x a
xi+1 ai+1
++ j j2
)
2(1a2
2(1a )
i+1
j
2
exp
x2
i+1
1a2
i+1
x2
i+1
1a2
i+1

1
= p
1a2i+1
x2
j
1a2
j
+ +
2
a2j
1a2j
1
22
xi+1 ai+1
xj aj
+ +
1 a2i+1
1 a2j
!2 !
exp
1
22
xi+1 ai+1
xj aj
+ +
1 a2i+1
1 a2j
!2 !
exp

x2
a2
i+1
++ j 2
)
2 (1a2
1a
j
i+1
xj aj
1a2j
1 a2j
1 hx, 1 xi
(2)N det
dm
+ +
a2i+1

1

a2
j
2 (1a2 )
j
exp
q
(2)N 2 (1 a2i+1 ) (1 a2j )
1
!1/2
a2j
x2
++ j 2
1a
j
++
xi+1 ai+1
1a2i+1
1+
(2)N 2 (1 a2i+1 ) (1 a2j )
a2
i+1
1a2
i+1
(2)N 2 (1 a2i+1 ) (1 a2j )

1
= q
1+
(2)N (1 a2i+1 ) (1 a2j )
1+
= q
1
22
X
1p6=lN
xp xl ap al
(1 a2p )(1 a2l )
where
2 = 1 +
a2j
a2i+1
+ +
,
2
1 ai+1
1 a2j
and
= 2

2 (1a21 )a21
(1a21 )2
a1 a2
(1a21 )(1a22 )
a2 a1
(1a22 )(1a21 )
2 (1a22 )a22
(1a22 )2
..
.
..
.
aN a1
(1a2N )(1a21 )
..
.
..
.
..
..
2 (1a2N 1 )a2N 1
(1a4N 1 )
..
aN aN 1
(1a2N )(1a2N 1 )
a1 aN
(1a21 )(1a2N )
aN 1 aN
(1a2N 1 )(1a2N )
2
2
2
..
.
..
.
(1aN )aN
(1a2N )2
In that sense, we are in presence of a particular case of the Gaussian copula

correlation method [74].
When N = 2 we find

=

1 a1 a2
,
a2 a1 1
with
432
"

a21
a22
+
1 a21
1 a22
(1 a21 )(1 a22 ) + a21 (1 a22 ) + a22 (1 a21 )
=
(1 a21 )(1 a22 )
1 a22 a21
=
,
(1 a21 )(1 a22 )
2 = 1 +
and
2
(1a21 )a21
a2
(1aa21)(1a
2)
1
(1a21 )2
1
2
= 2
2 (1a22 )a22
a1
(1aa22)(1a
2)
(1a22 )2
2
1
2

(1a22 )a21
a2
(1aa21)(1a
2)
1 1a21 1 1a22 a21
1
2

= 2
(1a2 )a2
a1
2
(1aa22)(1a
1 1a21 a22
2)
2
1a
2
1
2
2 1
"
#
a2
2
(1aa21)(1a
2)
1
1a22 a21
1
2
= 2
a1
2
(1aa22)(1a
2)
1a22 a21
2
1
"
#
a2
2
(1aa21)(1a
2)
(1 a21 )(1 a22 )
1a22 a21
1
2
=
2
a1
1 a22 a21
(1aa22)(1a
2)
1a22 a21
2
1
"
#
2
a2
(1aa21)(1a
2)
1
1a22 a21
1
2
= 2
2
a
a
2
1
(1a2 )(1a2 )
2
2
1a2 a1
2
1

1
1 a1 a2
=
.
2
2
1 a2 a1 a1 a2 1
In particular, the case N = 2 is able to recover all two-dimensional copulas

by setting the correlation coefficient = a1 a2 . In the general case, is
parametrized by N numbers, which offers less degrees of freedom compared
with the joint Gaussian copula correlation method which relies on N (N 1)/2
coefficients.
13.7 Collateralized debt obligations (CDOs)

Consider a portfolio consisting of N = j i Credit Default Swaps with default times i and recovery rates i [0, 1], i = 1, . . . , N .
A synthetic CDO is a structured investment product constructed by dividing the above portfolio into n ordered tranches numbered i = 1, . . . , n, where
tranche n i represents a percentage pi % of the total portfolio value. We let
"
433
N. Privault
l = p1 + + pl ,
l = 1, . . . , n,
with n = p1 + + pn = 100% and 0 = 0.

The tranches are ordered according to increasing default risk, tranche n 1
being the riskiest one (equity tranche), and tranche n n being the safest
one (senior tranche). Intermediate tranches are referred to as mezzanine
tranches. In practice losses occur first to the equity tranches, next to the
mezzanine tranche holders, and finally to senior tranches.
AAA
Aaa
Aa
Baa
Equity
Fig. 13.4: A representation of CDO tranches.

CDOs can attract different types of investors. Unfunded investors (usually
for the higher tranches) are receiving premiums and make payments in case
of default. Funded investors (usually in the lower tranches) are investing in
risky bonds to receive principal payments at maturity, and they are the first
in line to incur losses. A CDO can also be used as a Credit Default Swap
(CDS) for the short investors who make premium payments in exchange
for credit protection in case of default.
The market for synthetic CDOs has been significantly reduced since the
2006-2008 subprime crisis.
Synthetic CDOs based on N Credit Default Swaps potentially generate a
cumulative loss
j1
X
Lt =
(1 l+1 )1{l+1 t} [0, N ],
l=i
at time t [Ti , Tj ], based on the default time l and recovery rate l of each
involved CDS, with N = j i.
When the first loss occurs, tranche n 1 is the first in line, and it loses the
amount
434
"

L1t = Lt 1{Lt /N p1 } + N p1 1{Lt /N >p1 } = N min(Lt /N, p1 ).
In case Lt /N > p1 , then tranche n 2 takes the remaining loss up to the
amount N p2 , that means the loss L2t of tranche n 2 is
L2t = (Lt N p1 )1{p1 <Lt /N 2 } + N p2 1{Lt /N >2 }
= (Lt N p1 )+ 1{Lt /N 2 } + N p2 1{Lt /N >2 }
= min((Lt N p1 )+ , N p2 )
= max(min(Lt , N p1 + N p2 ) N p1 , 0)
= max(min(Lt , N 2 ) N p1 , 0).
By induction, the potential loss taken by tranche n i is given by
Lit = (Lt N i1 )1{i1 <Lt /N i } + N pi 1{Lt /N >i }
= (Lt N i1 )+ 1{Lt /N i } + N pi 1{Lt /N >i }
= min((Lt N i1 )+ , N pi )
= max(min(Lt , N i ) N i1 , 0),
i = 1, . . . , n.
In the end, tranche n n will take the loss

Lnt = (Lt N n1 )1{n1 <Lt /N } = (Lt N n1 )+ .
Super senior AAA
Senior Aaa
Mezzanine Aa
Mezzanine Baa
Equity
Fig. 13.5: A Titanic-style representation of cumulative tranche losses.

The CDO tranche n l, l = 1, . . . , n, can be decomposed into:
- a premium leg: the short investor is purchasing protection at time t
against default at time Tk , k = i + 1, . . . , j, by making fixed payments
St at times Ti+1 , . . . , Tj between t and T in compensation. This premium
can also be received by the unfunded investor.
The discounted value at time t of the premium leg for the tranche n l is
"
435
N. Privault
Vlp (t, T ) = IE
"j1
X
w
#
Tk+1

Stl k (N pl LlTk+1 ) exp
rs ds Gt
t
k=i
= Stl
j1
X

w

Tk+1

k IE (N pl LlTk+1 ) exp
rs ds Gt ,
t
k=i
l = 1, . . . , N , where Stl is the premium spread paid at time Tk+1 until

k = j 1 or LTk+1 = 1, whichever comes first, and k = Tk+1 Tk ,
k = i, . . . , j 1.
- a protection leg: the short investor receives protection against default
from the premium leg, which can also be paid by funded investors. The
value at time t of the protection leg is
"j1
#
w k

X

Vld (t, T ) = IE
1[Ti ,Tj ] (k )(1 k+1 ) exp
ru du Gt
w
Tj
k=i

ws

= IE
exp
ru du dLls Gt
Ti
t

w

w

Ti
Tj

ru du LlTi Gt
ru du LlTj exp
= IE exp
t
t
w

ws

Tj

+ IE
rs exp
ru du Lls ds Gt
Ti
t

w

w

ws

Tj
Tj

= IE exp
ru du LlTj Gt + IE
rs exp
ru du Lls ds Gt ,
t
Ti
where we applied integration by parts on [Ti , Tj ] and used the fact that
LTi = 0.
The spread Stl paid by tranche n l is computed by equating the values of the
protection and premium legs, i.e. from the relation Vlp (t, T ) = Vld (t, T ), as
i
hr

rs
T

IE Tij exp t ru du dLls Gt
i
h
r
Stl = P
Tk+1

j1
l
rs ds Gt
k=i k IE (N pl LTk+1 ) exp t
i
i
h
r

hr

rs
T
T

IE exp t j ru du LlTj Gt + IE Tij rs exp t ru du Lls ds Gt
r
i
h
=
Pj1
Tk+1

l
rs ds Gt
k=i k IE (N pl LTk+1 ) exp t
0,
l = 1, . . . , N .
436
"

Expected tranche loss
The expected cumulative loss can be computed by linearity as
IE[Lt | M = m] =
j1
X

IE (1 l+1 )1{k+1 t} | M = m
l=i
j1
X
(1 l+1 )P (k+1 t | M = m)
=
l=i
j1
X
=
(1 l+1 )
1 (P(k+1 T )) + ak m
p
1 a2k
l=i
!
,
and the expected cumulative loss can be written as

IE[Lt ] =
2
1 w
IE [Lt | M = m] (m)dm =
IE [Lt | M = m] em /2 dm.
The situation is different for the expected loss of tranche n k is written as

the expected value

IE[Lkt ] = IE min((Lt N k1 )+ , N pi ) ,
k = 1, . . . , n.
of the nonlinear function fk (x) := min((x N k1 )+ , N pk ) of Lt .
3
N pi
2.5
f(x)
1.5
1
0.5
Ni-1
4 Ni-1+N pi 5
Fig. 13.6: Function fk (x) = min((x N k1 )+ , N pk ).

The expected tranche loss IE[Lkt ] n k can be estimated by the Monte Carlo
method when the default times are generated according to (13.16).
In order to expected tranche losses analytically we can use the fact that
the cumulative loss Lt is a discrete random variable, with for example
"
437
N. Privault
P Lt = N
j1
X
!
k+1
= P(i+1 t, . . . , j t),
k=i
and
P(Lt = 0) = P(i+1 > t, . . . , j > t),
which require the knowledge of the joint distribution of the default times
i+1 , . . . , j .
Bond rating
Moodys
S&P
categories
Municipal Corporate Municipal Corporate
Aaa/AAAs
0.00
0.52
0.00
0.60
Aa/AA
0.06
0.52
0.00
1.50
A/A
0.03
1.29
0.23
2.91
Baa/BBB
0.13
4.64
0.32
10.29
Ba/BB
2.65
19.12
1.74
29.93
B/B
11.86
43.34
8.48
53.72
Caa-C/CCC-C
16.58
69.18
44.81
69.19
Investment Grade
0.07
2.09
0.20
4.14
Non-Invest Grade
4.29
31.37
7.37
42.35
All
0.10
9.70
0.29
12.98
Fig. 13.7: Cumulative historic default rates (in percentage).

If the k0 s are identically distributed with common cumulative distribution
function F and ak = a, k = , k = i + 1, . . . , j, then the cumulative loss Lt
has a Bernoulli distribution given M , given by

N
(1 P( T | M ))N k (P( T | M ))k
P(Lt = (1 )k | M ) =
k

1
N k 1
k
N
(F (T )) aM
(F (T )) aM
=
1
,
k
1 a2
1 a2
k = 0, 1, . . . , N .
The expected loss of tranche n k can then be expressed as
IE[Lkt ] =
2
1 w
IE [fk (Lt ) | M = m] (m)dm =
IE [fk (Lt ) | M = m] em /2 dm,
2
k = 1, . . . , n, where IE [fk (Lt ) | M = m] is computed either by the Monte

Carlo method, from the distribution of Lt .
Source: Moodys, S&P.
438
"

In [112], the tranche loss has been approximated by a Gaussian random
variable for very large portfolios with N .
The -percentile loss of the portfolio can be estimated as
IE[Lt | M = m] =
N
1
X
1 (P(k T )) ak 1 ()
p
1 a2k
(1 k+1 )
k=0
!
,
where m = 1 ().
Such (Gaussian) Merton [79] and Vasicek [112] type models have been
implemented in Basel II [83]. Namely in Basel II, banks are expected to hold
capital in prevision of unexpected losses in a worst case scenario, according
to the Internal Ratings-Based (IRB) formula
!
!
N
1
X
1 (P(k T )) ak 1 ()
p
(1 k+1 )
P(k T ) ,
1 a2k
k=0
where the confidence level is set at = 0.999 i.e. m = 1 (0.999) =
3.09,
page 10 of [63]. Recall that the function x 7

cf. Relation (2.4)
1 (x)+ak 1 ()
2
1ak
always lies above the graph of x when ak < 0, as in
the next figure.

1
( ( -1(x) - a -1(0.999)))/ (1-a2)1/2)

x
0.8
0.6
0.4
0.2
0.2
0.4
0.6
0.8
Fig. 13.8: Internal Ratings-Based formula.
Exercises
Exercise 13.1 Defaultable bonds. Consider a (random) default time with
law
w

t
P( > t | Ft ) = exp u du ,
0
"
439
N. Privault
where t is a (random) default rate process which is adapted to the filtration (Ft )tR+ . Recall that the probability of survival up to time T , given
information known up to time t, is given by

w

T

P( > T | Gt ) = 1{ >t} IE exp
u du Ft ,
t
where Gt = Ft ({ < u} : 0 u t), t R+ , is the filtration defined by adding the default time information to the history (Ft )tR+ . In this
framework, the price P (t, T ) of defaultable bond with maturity T , short term
interest rate rt and (random) default time is given by

w

T

P (t, T ) = IE 1{ >T } exp
ru du Gt
(13.17)
t

w

T

= 1{ >t} IE exp
(ru + u )du Ft .
t
In the sequel we assume that the processes (rt )tR+ and (t )tR+ are modeled
according to the Vasicek processes
(1)
drt = art dt + dBt ,
d = b dt + dB (2) ,
t
t
t
(1)
(2)
where (Bt )tR+ and (Bt )tR+ are two standard Ft -Brownian motions with
(2)
(1)
correlation [1, 1], and dBt dBt = dt.
a) Give a justification for the fact that

w

T

IE exp
(ru + u )du Ft
t
can be written as a function F (t, rt , t ) of t, rt and t , t [0, T ].

b) Show that
w

w

T
t

(ru + u )du Ft
t 7 exp (rs + s )ds IE exp
0
is an Ft -martingale under P.
c) Use the Ito formula with two variables to derive a PDE on R2 for the
function F (t, x, y).
d) Show that we have
wT
t
rs ds = C(a, t, T )rt +
wT
t
C(a, s, T )dBs(1) ,
and
440
"

wT
t
s ds = C(b, t, T )t +
wT
t
C(b, s, T )dBs(2) ,
where
1
C(a, t, T ) = (ea(T t) 1).
a
e) Show that the random variable
wT
t
rs ds +
wT
t
s ds
is Gaussian and compute its conditional mean

w

wT
T

IE
rs ds +
s dsFt
t
and variance
Var
w
T
rs ds +
wT
t

s dsFt ,
conditionally to Ft .
f) Compute P (t, T ) from its expression (13.17) as a conditional expectation.
g) Show that the solution F (t, x, y) to the 2-dimensional PDE of Question (c)
is
F (t, x, y) = exp (C(a, t, T )x C(b, t, T )y)

2w
T
2 w T 2
C 2 (a, s, T )ds +
C (b, s, T )ds
exp
2 t
2 t

wT
exp
C(a, s, T )C(b, s, T )ds .
t
h) Show that the defaultable bond price P (t, T ) can also be written as

w
T

P (t, T ) = eU (t,T ) P( > T | Gt ) IE exp
rs ds Ft ,
t
where
U (t, T ) =
(T t C(a, t, T ) C(b, t, T ) + C(a + b, t, T )) .

ab
i) By partial differentiation of log P (t, T ) with respect to T , compute the

log P (t, T )
corresponding instantaneous short rate f (t, T ) =
.
T
j) Show that P( > T | Gt ) can be written using an HJM type default rate
as

w
T
P( > T | Gt ) = 1{ >t} exp
f2 (t, u)du ,
t
where
"
441
N. Privault
f2 (t, u) = t eb(ut)
2 2
C (b, t, u).
2
(1)
k) Show how the result of Question (h) can be simplified when (Bt )tR+
(2)
and (Bt )tR+ are independent.
Exercise 13.2 Copulas. In the sequel, U denotes a uniformly distributed
random variable on [0, 1].
a) To which couple (U, V ) of uniformly distributed random variables on [0, 1]
does the copula function
CM (u, v) = min(u, v),
u, v [0, 1],
correspond?
b) Show that the function
Cm (u, v) := (u + v 1)+ ,
u, v [0, 1],
is the copula on [0, 1] [0, 1] corresponding to (U, V ) = (U, 1 U ).

c) Show that for any copula function C(u, v) on [0, 1] [0, 1] we have
C(u, v) CM (u, v),
u, v [0, 1].
(13.18)
d) Show that for any copula function C(u, v) on [0, 1] [0, 1] we also have
Cm (u, v) C(u, v),
u, v [0, 1].
(13.19)
Hint: For fixed v [0, 1], let h(u) := C(u, v) (u + v 1) and show that
h(1) = 0 and h0 (u) 0.
Exercise 13.3 Credit default swaps. Estimate the first default rate 1 and
the associated default probability in the framework of (13.14), based on CDS
market data McDonalds Corp, cf. also [11].
442
"
Chapter 14
Stochastic Calculus for Jump Processes
The modelling of risky asset by stochastic processes with continuous paths,

based on Brownian motions, suffers from several defects. First, the path continuity assumption does not seem reasonable in view of the possibility of
sudden price variations (jumps) resulting of market crashes. Secondly, the
modeling of risky asset prices by Brownian motion relies on the use of the
Gaussian distribution which tends to underestimate the probabilities of extreme events.
A solution is to use stochastic processes with jumps, that will account for
sudden variations of the asset prices. On the other hand, such jump models
are generally based on the Poisson distribution which has a slower tail decay
than the Gaussian distribution. This allows one to assign higher probabilities
to extreme events, resulting in a more realistic modeling of asset prices.
14.1 The Poisson Process

The most elementary and useful jump process is the standard Poisson process
which is a stochastic process (Nt )tR+ with jumps of size +1 only, and whose
paths are constant in between two jumps, i.e. at time t, the value Nt of the
process is given by
Nt =
1[Tk ,) (t),
t R+ ,
(14.1)
k=1
where
The notation Nt is not to be confused with the same notation used for num
eraire
processes in Chapter 10.
"
N. Privault
1[Tk ,) (t) =
1 if t Tk ,
0 if 0 t < Tk ,
k 1, and (Tk )k1 is the increasing family of jump times of (Nt )tR+ such
that
lim Tk = +.
k
In addition, (Nt )tR+ satisfies the following conditions:

1. Independence of increments: for all 0 t0 < t1 < < tn and n 1 the
random variables
Nt1 Nt0 , . . . , Ntn Ntn1 ,
are independent.
2. Stationarity of increments: Nt+h Ns+h has the same distribution as
Nt Ns for all h > 0 and 0 s t.
The meaning of the above stationarity condition is that for all fixed k N
we have
P(Nt+h Ns+h = k) = P(Nt Ns = k),
for all h > 0, i.e. the value of the probability
P(Nt+h Ns+h = k)
does not depend on h > 0, for all fixed 0 s t and k N.
The next figure represents a sample path of a Poisson process.
7
Nt
0
0
10
Fig. 14.1: Sample path of a Poisson process (Nt )tR+ .
444
"

Based on the above assumption, given a time value T > 0 a natural question
arises:
what is the probability distribution of the random variable NT ?
We already know that Nt takes values in N and therefore it has a discrete
distribution for all t R+ .
It is a remarkable fact that the distribution of the increments of (Nt )tR+ ,
can be completely determined from the above conditions, as shown in the
following theorem.
As seen in the next result, cf. [6], Nt Ns has the Poisson distribution
with parameter (t s).
Theorem 14.1. Assume that the counting process (Nt )tR+ satisfies the
above Conditions 1 and 2. Then for all fixed 0 s t we have
P(Nt Ns = k) = e(ts)
((t s))k
,
k!
k N,
(14.2)
for some constant > 0.

The parameter > 0 is called the intensity of the Poisson process (Nt )tR+
and it is given by
1
:= lim P(Nh = 1).
(14.3)
h0 h
The proof of the above Theorem 14.1 is technical and not included here,
cf. e.g. [6] for details, and we could in fact take this distribution property
(14.2) as one of the hypotheses that define the Poisson process.
Precisely, we could restate the definition of the standard Poisson process
(Nt )tR+ with intensity > 0 as being a process defined by (14.1), which is
assumed to have independent increments distributed according to the Poisson
distribution, in the sense that for all 0 t0 t1 < < tn ,
(Nt1 Nt0 , . . . , Ntn Ntn1 )
is a vector of independent Poisson random variables with respective parameters
((t1 t0 ), . . . , (tn tn1 )).
In particular, Nt has the Poisson distribution with parameter t, i.e.
"
445
N. Privault
P(Nt = k) =
(t)k t
e ,
k!
t > 0.
The expected value IE[Nt ] of Nt can be computed as

IE[Nt ] = t,
(14.4)
cf. Exercise 16.1.
Short Time Behaviour
From (14.3) above we deduce the short time asymptotics

P(Nh = 1) = heh ' h,
h 0,
P(Nh = 0) = eh ' 1 h,
h 0.
and
By stationarity of the Poisson process we find more generally that
P(Nt+h Nt = 1) = heh ' h,
h 0,
P(Nt+h Nt = 0) = eh ' 1 h,
h 0,
and
for all t > 0.
This means that within a short interval [t, t + h] of length h, the increment Nt+h Nt behaves like a Bernoulli random variable with parameter
h. This fact can be used for the random simulation of Poisson process paths.
We also find that
P(Nt+h Nt = 2) ' h2
2
,
2
h 0,
t > 0,
k
,
k!
h 0,
t > 0.
and more generally

P(Nt+h Nt = k) ' hk
The intensity of the Poisson process can in fact be made time-dependent (e.g.
by a time change), in which case we have
We use the notation f (h) ' hk to mean that limh0 f (h)/hk = 1.
446
"

k
w
r t (u)du
t
s
P(Nt Ns = k) = exp (u)du
,
s
k!
k 0.
In particular,
(t)dt
' 1 (t)dt,
P(Nt+dt Nt = k) = (t)e(t)dt dt ' (t)dt,
o(dt),
k = 0,
k = 1,
k 2,
and P(Nt+dt Nt = 0), P(Nt+dt Nt = 1) coincide respectively with (13.2)

and (13.3) above. The intensity process ((t))tR+ can also be made random
in the case of Cox processes.
Poisson Process Jump Times

In order to prove the next proposition we note that we have the equivalence
{T1 > t} {Nt = 0},
and more generally
{Tn > t} {Nt n 1},
for all n 1.
In the next proposition we compute the distribution of Tn with its density.
It coincides with the gamma distribution with integer parameter n 1, also
known as the Erlang distribution in queueing theory.
Proposition 14.1. For all n 1 the probability distribution of Tn has the
density function
tn1
t 7 n et
(n 1)!
on R+ , i.e. for all t > 0 the probability P(Tn t) is given by
P(Tn t) = n
w
t
es
sn1
ds.
(n 1)!
Proof. We have
P(T1 > t) = P(Nt = 0) = et ,
t R+ ,
and by induction, assuming that
"
447
N. Privault
P(Tn1 > t) =
w
t
es
(s)n2
ds,
(n 2)!
n 2,
we obtain
P(Tn > t) = P(Tn > t Tn1 ) + P(Tn1 > t)
= P(Nt = n 1) + P(Tn1 > t)
w
(s)n2
(t)n1
es
+
ds
= et
t
(n 1)!
(n 2)!
w
(s)n1
ds,
t R+ ,
=
es
t
(n 1)!
where we applied an integration by parts to derive the last line.
In particular, for all n Z and t R+ , we have

P(Nt = n) = pn (t) = et
(t)n
,
n!
i.e. pn1 : R+ R+ , n 1, is the density function of Tn .

Similarly we could show, using the strong Markov property (see e.g. Theorem 6.5.4 of [81]) that the times
k := Tk+1 Tk
spent in state k N, with T0 = 0, form a sequence of independent identically distributed random variables having the exponential distribution with
parameter > 0, i.e.
P(0 > t0 , . . . , n > tn ) = e(t0 +t1 ++tn ) ,
t0 , . . . , tn R+ .
Since the expectation of the exponentially distributed random variable k

with parameter > 0 is given by
IE[k ] =
1
,
we can check that the higher the intensity (i.e. the higher the probability
of having a jump within a small interval), the smaller is the time spent in
each state k N on average.
In addition, conditionally to {NT = n}, the n jump times on [0, T ] of
the Poisson process (Nt )tR+ are independent uniformly distributed random
variables on [0, T ]n , cf. e.g. 12.1 of [91]. This fact can be useful for the
random simulation of the Poisson process.
448
"

Compensated Poisson Martingale
From (14.4) above we deduce that
IE[Nt t] = 0,
(14.5)
i.e. the compensated Poisson process (Nt t)tR+ has centered increments.
Since in addition (Nt t)tR+ also has independent increments we get
the following proposition.
Proposition 14.2. The compensated Poisson process
(Nt t)tR+
is a martingale with respect to its own filtration (Ft )tR+ .
Extensions of the Poisson process include Poisson processes with timedependent intensity, and with random time-dependent intensity (Cox processes). Renewal processes are counting processes
X
Nt =
1[Tn ,) (t),
t R+ ,
n1
in which k = Tk+1 Tk , k N, is a sequence of independent identically

distributed random variables. In particular, Poisson processes are renewal
processes.
14.2 Compound Poisson Processes

The Poisson process itself appears to be too limited to develop realistic price
models as its jumps are of constant size. Therefore there is some interest in
considering jump processes that can have random jump sizes.
Let (Zk )k1 denote an i.i.d. sequence of square-integrable random variables with probability distribution (dy) on R, independent of the Poisson
process (Nt )tR+ . We have
P(Zk [a, b]) = ([a, b]) =
wb
a
(dy),
< a b < .
Definition 14.1. The process

Yt =
Nt
X
Zk ,
t R+ ,
(14.6)
k=1
"
449
N. Privault
is called a compound Poisson process.
The next figure represents a sample path of a compound Poisson process,
with here Z1 = 0.9, Z2 = 0.7, Z3 = 1.4, Z4 = 0.6, Z5 = 2.5, Z6 = 1.5,
Z7 = 1.2.
2.5
Yt
1.5
0.5
-0.5
0
10
Fig. 14.2: Sample path of a compound Poisson process (Yt )tR+ .

Given that {NT = n}, the n jump sizes of (Yt )tR+ on [0, T ] are independent
random variables which are distributed on R according to (dx). Based on
this fact, the next proposition allows us to compute the moment generating
function of the increment YT Yt .
Proposition 14.3. For any t [0, T ] we have

w
IE [exp ((YT Yt ))] = exp (T t)
(ey 1)(dy) ,
R.
Proof. Since Nt has a Poisson distribution with parameter t > 0 and is
independent of (Zk )k1 , for all R we have by conditioning:
"
IE [exp ((YT Yt ))] = IE exp
!#
NT
X
Zk
k=Nt +1
"
= IE exp
NT
Nt
X
!#
Zk
k=1
"
IE exp
n=0
= e(T t)
n
X
!#
Zk
P(NT Nt = n)
k=1
"
!#
n
X
X
n
(T t)n IE exp
Zk
n!
n=0
k=1
Recall the convention
Pn
k=1
Zk = 0 if n = 0.
450
"
= e(T t)
X
n
n
(T t)n (IE [exp (Z1 )])
n!
n=0
= exp ((T t) IE [exp (Z1 )])

w
= exp (T t)
(ey 1)(dy) ,
since (dy) is the probability distribution of Z1 and
(dy) = 1.
From the characteristic function we can compute the expectation and variance of Yt for fixed t, as
IE[Yt ] = t IE[Z1 ]
and
Var [Yt ] = t IE[|Z1 |2 ].
For the expectation we have

IE[Yt ] = i
w
d
IE[eiYt ]|=0 = t
y(dy) = t IE[Z1 ].
This relation can also be directly recovered as

##
" "N
t

X

Zk Nt
IE[Yt ] = IE IE
k=1
" n
#
X
X
n tn
=e
IE
Zk Nt = n
n!
n=1
k=1
" n
#
X
X
n tn
t
=e
IE
Zk
n!
n=1
t
k=1
X
(t)n1
= tet IE[Z1 ]
(n 1)!
n=1
= t IE[Z1 ].
More generally one can show that for all 0 t0 t1 tn and 1 , . . . , n
R we have
" n
#
!
n
w
Y i (Y Y
X
IE
e k tk tk1 ) = exp
(tk tk1 )
(eik y 1)(dy)
k=1
k=1
=
=
n
Y
k=1
n
Y
exp (tk tk1 )
(eik y 1)(dy)
h
i
IE ei(Ytk Ytk1 ) .
k=1
"
451
N. Privault
This shows in particular that the compound Poisson process (Yt )tR+ has
independent increments, as the standard Poisson process (Nt )tR+ .
Since the compensated Poisson process also has centered increments by
(14.5), we have the following proposition.
Proposition 14.4. The compensated compound Poisson process
Mt := Yt t IE[Z1 ],
t R+ ,
is a martingale.
By construction, compound Poisson processes only have a finite number
of jumps on any interval. They belong to the family of Levy processes which
may have an infinite number of jumps on any finite time interval, cf. [15].
14.3 Stochastic Integrals with Jumps

Given (t )tR+ a stochastic process we let the stochastic integral of (t )tR+
with respect to (Yt )tR+ be defined by
wT
0
t dYt :=
NT
X
Tk Zk .
k=1
wT
Note that this expression
t dYt has a natural financial interpretation as
0
the value at time T of a portfolio containing a (possibly fractional) quantity
t of a risky asset at time t, whose price evolves according to random returns
Zk at random times Tk .
In particular the compound Poisson process (Yt )tR+ in (14.1) admits the
stochastic integral representation
wt
ZNs dNs .
Yt = Y0 +
0
Next, given (Wt )tR+ a standard Brownian motion independent of (Yt )tR+
and (Xt )tR+ a jump-diffusion process of the form
Xt =
wt
0
us dWs +
wt
0
vs ds + Yt ,
452
t R+ ,
"

where (t )tR+ is a process which is adapted to the filtration (Ft )tR+ generated by (Wt )tR+ and (Yt )tR+ , and such that
w

w

T
T
IE
2s |us |2 ds < and IE
|s vs |ds < ,
T > 0,
0
we let the stochastic integral of (s )sR+ with respect to (Xs )sR+ be defined
by
wT
0
s dXs :=
wT
0
s us dWs +
wT
0
s vs ds +
NT
X
Tk Z k ,
T > 0.
k=1
The coumpound Poisson compensated stochastic integral can be shown to

satisfy the Ito isometry
IE
"
w
2 #
t (dYt IE[Z1 ]dt)
= IE[|Z1 |2 ] IE
w
T
0

||2t dt ,
(14.7)
provided the process (t )tR+ is adapted to the filtration generated by

(Yt )tR+ , which makes the left limit process (s )sR+ predictable. The proof
of (14.7) can be written using simple predictable processes, similarly to the
proof of Proposition 4.3. It also follows by taking expectations on both sides
of the stochastic Fubini type theorem
w
=2
2
t (dYt IE[Z1 ]dt)
wT
0
w t
0
s (dYs IE[Z1 ]ds)(dYt IE[Z1 ]dt) + IE
w
T
0

2
||2t ZN
dN
,
t
t
in which the diagonal has been excluded in the double integral, and using the
fact that the expectation of the double stochastic integral vanishes.
For the mixed continuous-jump martingale
wt
Xt =
us dWs + Yt t IE[Z1 ],
0
t R+ ,
we have the isometry
"
453
N. Privault
IE
"
wT
0
2 #
s dXs
= IE
w
T
0

w

T
|s |2 |us |2 ds + IE[|Z1 |2 ] IE
|s |2 ds .
0
(14.8)
provided (s )sR+ is adapted to the filtration (Ft )tR+ generated by (Wt )tR+
and (Yt )tR+ .
This isometry formula will be used in Section 15.5 for the computation of
hedging strategies in jump models.
When (Xt )tR+ takes the form
Xt = X0 +
wt
0
us dWs +
wt
0
vs ds +
wt
0
s dYs ,
t R+ ,
the stochastic integral of (t )tR+ with respect to (Xt )tR+ satisfies

wT
0
s dXs :=
wT
wT
s us dWs +
wT
s vs ds +
s us dWs +
wT
s vs ds +
wT
0
NT
X
s s dYs
Tk Tk Zk ,
T > 0.
k=1
14.4 It
o Formula with Jumps
Let us first consider the case of a standard Poisson process (Nt )tR+ with
intensity . We have the telescoping sum
f (Nt ) = f (0) +
Nt
X
(f (k) f (k 1))
k=1
= f (0) +
= f (0) +
= f (0) +
wt
0
wt
0
wt
0
(f (1 + Ns ) f (Ns ))dNs
(f (Ns ) f (Ns 1))dNs
(f (Ns ) f (Ns ))dNs .
Here, Ns denotes the left limit of the Poisson process at time s, i.e.
Ns = lim Nsh .
h&0
454
"

k 1.
k = NTk = 1 + NT ,
k
By the same argument we find, in the case of the compound Poisson process
(Yt )tR+ ,
f (Yt ) = f (0) +
= f (0) +
= f (0) +
Nt
X
(f (YT + Zk ) f (YT ))
k
k=1
wt
0
wt
0
(f (ZNs + Ys ) f (Ys ))dNs
(f (Ys ) f (Ys ))dNs ,
which can be decomposed using a compensated Poisson stochastic integral

as
wt
wt
f (Yt ) = f (0) + (f (Ys ) f (Ys ))(dNs ds) + (f (Ys ) f (Ys ))ds.
0
More generally, for a process of the form

wt
wt
wt
Xt = X0 +
us dWs +
vs ds +
s dYs ,
0
t R+ ,
we find, by combining the Ito formula for Brownian motion with the above
argument we get
f (Xt ) = f (X0 ) +
+
wt
0
wt
0
us f 0 (Xs )dWs +
vs f 0 (Xs )ds +
wt
NT
X
1 w t 00
f (Xs )|us |2 ds
2 0
(f (XT + Tk Zk ) f (XT ))
k
wt
1 w t 00
vs f 0 (Xs )ds
f (Xs )|us |2 ds +
us f 0 (Xs )dWs +
0
2 0
= f (X0 ) +
0
wt
+ (f (Xs + s ZNs ) f (Xs ))dNs
0
k=1
t R+ .
i.e.
wt
wt
1 w t 00
f (Xs )|us |2 ds +
vs f 0 (Xs )ds
f (Xt ) = f (X0 ) +
us f 0 (Xs )dWs +
0
0
2 0
wt
+ (f (Xs ) f (Xs ))dNs ,
t R+ .
(14.9)
0
"
455
N. Privault
For example, in case

wt
wt
wt
Xt =
us dWs +
vs ds +
s dNs ,
0
t R+ ,
we get
wt
1wt
f (Xt ) = f (0) +
us f 0 (Xs )dWs +
|us |2 f 00 (Xs )dWs
0
2 0
wt
wt
+ vs f 0 (Xs )ds + (f (Xs + s ) f (Xs ))dNs
0
0
wt
1wt
0
|us |2 f 00 (Xs )dWs
(14.10)
= f (0) +
us f (Xs )dWs +
0
2 0
wt
wt
0
+ vs f (Xs )ds + (f (Xs ) f (Xs ))dNs .
0
Given two processes (Xt )tR+ and (Yt )tR+ written as

Xt =
wt
us dWs +
wt
vs ds +
wt
s dNs ,
t R+ ,
Yt =
wt
as dWs +
wt
bs ds +
wt
cs dNs ,
t R+ ,
and
the Ito formula for jump processes also shows that

d(Xt Yt ) = Xt dYt + Yt dXt + dXt dYt
where the product dXt dYt is computed according to the extension
dt
dBt
dNt
dt
0
0
0
dBt
0
dt
0
dNt
0
0
dNt
of the Ito multiplication table (4.21), i.e. we have

dXt dYt = (vt dt + ut dBt + t dNt )(bt dt + at dBt + ct dNt )
= bt vt (dt)2 + bt ut dt dBt + bt t dt dNt
+at vt dtdBt + at ut (dBt )2 + at t dBt dNt
+ct vt dNt dBt + ct ut (dBt )2 + ct t dNt dNt
= at ut dt + ct t dNt ,
and in particular
(dXt )2 = (vt dt + ut dBt + t dNt )2 = u2t dt + t2 dNt .
For a process of the form
456
"

Xt = X0 +
wt
0
us dWs +
wt
0
s dYt ,
t R+ ,
the Ito formula with jumps (14.10) can be rewritten as

wt
wt
f (Xt ) = f (X0 ) +
vs f 0 (Xs )ds +
us f 0 (Xs )dWs
0
0
wt
1 w t 00
+
f (Xs )|us |2 ds +
s f 0 (Xs )dYs
0
2 0
wt
+ (f (Xs ) f (Xs ) Xs f 0 (Xs )) d(Ns s)
0
wt
+ (f (Xs ) f (Xs ) Xs f 0 (Xs )) ds,
t R+ ,
0
where we used the relation dYs = Xs f 0 (Xs )dNs , which implies

wt
0
s f 0 (Xs )dYs =
wt
0
Xs f 0 (Xs )dNs ,
t 0.
This above formulation is at the basis of the extension of It

os formula to
Levy processes with an infinite number of jumps on any interval, using the
bound
|f (x + y) f (x) yf 0 (x)| Cy 2 ,
for f a Cb2 (R) function. Such processes, also called infinite activity Levy
processes [15] are also useful in financial modeling and include the gamma
process, stable processes, variance gamma processes, inverse Gaussian processes, etc, as in the following illustrations.
1. Gamma process.
0
t
Fig. 14.3: Sample trajectories of a gamma process.
"
457
N. Privault
2. Stable process.
Fig. 14.4: Sample trajectories of a stable process.
3. Variance Gamma process.
Fig. 14.5: Sample trajectories of a variance gamma process.
458
"

4. Inverse Gaussian process.
0
t
Fig. 14.6: Sample trajectories of an inverse Gaussian process.
5. Negative Inverse Gaussian process.
Fig. 14.7: Sample trajectories of a negative inverse Gaussian process.
14.5 Stochastic Differential Equations with Jumps

Let us start with the simplest example
dSt = St dNt ,
(14.11)
of a stochastic differential equation with respect to the standard Poisson process, with constant coefficient R.
When
Nt = Nt Nt = 1,
"
459
N. Privault
i.e. when the Poisson process has a jump at time t, the equation (14.11) reads
dSt = St St = St ,
t > 0.
which can be solved to yield

St = (1 + )St ,
t > 0.
By induction, applying this procedure for each jump time gives us the solution
St = S0 (1 + )Nt ,
t R+ .
Next, consider the case where is time-dependent, i.e.

dSt = t St dNt .
(14.12)
At each jump time Tk , Relation (14.12) reads

dSTk = STk ST = Tk ST ,
k
i.e.
STk = (1 + Tk )ST ,
k
and repeating this argument for all k = 1, . . . , Nt yields the product solution
St = S0
Nt
Y
(1 + Tk ) = S0
(1 + s ),
t R+ .
Ns =1
0st
k=1
The equation
dSt = t St dt + t St (dNt dt),
(14.13)
is then solved as
St = S0 exp
w
s ds
wt
0
Y
Nt
s ds
(1 + Tk ),
t R+ .
k=1
A random simulation of the numerical solution of the above equation (14.13)

is given in Figure 14.8 for constant = t , t R+ .
460
"

2
St
1.5
0.5
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
2.0
Fig. 14.8: Geometric Poisson process.
The above simulation can be compared to the real sales ranking data of
Figure 14.9.
Fig. 14.9: Ranking data.

A random simulation of the geometric compound Poisson process
St = S0 exp
w
s ds IE[Z1 ]
wt
0
Y
Nt
s ds
(1 + Tk Zk )
t R+ ,
k=1
solution of
dSt = t St dt + t St (dYt IE[Z1 ]dt),
is given in Figure 14.10.
"
461
N. Privault
2
St
1.5
0.5
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
2.0
Fig. 14.10: Geometric compound Poisson process.
In the case of a jump-diffusion stochastic differential equation of the form

dSt = t St dt + t St (dYt IE[Z1 ]dt) + t St dWt ,
we get
St = S0 exp
Nt
Y
w
s ds IE[Z1 ]
wt
0
s ds +
wt
0
s dWs

1wt
|s |2 ds
2 0
(1 + Tk Zk ),
k=1
t R+ . A random simulation of the geometric Brownian motion with compound Poisson jumps is given in Figure 14.11.
462
"

3
2.5
St
1.5
1
0.5
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
2.0
Fig. 14.11: Geometric Brownian motion with compound Poisson jumps.
By rewriting St as
w

wt
wt
t
1wt
St = S0 exp
s ds +
s (dYs IE[Z1 ]ds) +
s dWs
|s |2 ds
0
0
0
2 0
Nt
Y
(eTk (1 + Tk Zk )),
k=1
t R+ , one can extend this jump model to processes with an infinite number
of jumps on any finite time interval, cf. [15]. The next Figure 14.12 shows
a number of downward and upward jumps occuring in the historical price
of the SMRT stock, with a typical geometric Brownian behavior in between
jumps.
Fig. 14.12: SMRT Stock price.
"
463
N. Privault
14.6 Girsanov Theorem for Jump Processes

Recall that in its simplest form, the Girsanov theorem for Brownian motion
follows from the calculation
2
1 w
IE[f (WT T )] =
f (x T )ex /(2T ) dx
2T
2
1 w
f (x)e(x+T ) /(2T ) dx
=
2T
2
2
1 w
=
f (x)ex T /2 ex /(2T ) dx
2T
2
= IE[f (WT )eWT

(WT )],
= IE[f
T /2
]
(14.14)
for any bounded measurable function f on R, which shows that WT is a
Gaussian random variable with mean T under the probability measure P

defined by
= eWT 2 T /2 dP,
dP
cf. Section 6.2. Equivalently we have
(WT + T )],
IE[f (WT )] = IE[f
(14.15)
hence
under the probability measure
2
= eWT
dP
T /2
dP,
the random variable WT +T has the centered Gaussian distribution

N (0, T ).
More generally, the Girsanov theorem states that (Wt + t)t[0,T ] is a stan
dard Brownian motion under P.
When Brownian motion is replaced with a standard Poisson process
(Nt )tR+ , the above space shift
Wt 7 Wt + t
may not be used because Nt + t cannot be a Poisson process, whatever the
change of probability applied, since by construction, the paths of the standard Poisson process has jumps of unit size and remain constant between
464
"

jump times.
The correct way to proceed in order to extend (14.15) to the Poisson case
is to replace the space shift with a time contraction (or dilation) by a certain
factor 1 + c with c > 1, i.e.
Nt 7 Nt/(1+c)
or Nt 7 N(1+c)t .
Assume that (Nt )tR+ is a standard Poisson process with intensity under
P. By analogy with (14.14) we can write
((1 + c)T )k
= ecT (1 + c)k P(NT = k),
k!
P(N(1+c)T = k) = e(1+c)T
k N, and for f any bounded function on N we have

IE[f (N(1+c)T )] =
f (k)P(N(1+c)T = k)
(14.16)
k=0
= ecT
f (k)(1 + c)k P(NT = k)
k=0
= ecT IE[f (NT )(1 + c)NT ]

w
= ecT
(1 + c)NT f (NT )dP
=
f (NT )dP
[f (NT )],
= IE
is defined by
where the probability measure P
:= ecT (1 + c)NT dP.

dP
Consequently,
:= ecT (1 + c)NT dP,
dP
the law of the random variable NT is that of N(1+c)T under P, i.e. it is a

Poisson random variable with intensity (1 + c)T .
Equivalently to (14.16) we have

[f (NT /(1+c) )],
IE[f (NT )] = IE
"
465
N. Privault
the law of NT /(1+c) is that of a standard Poisson random varii.e. under P
able with parameter T . As a consequence, (Nt/(1+c) )tR+ is a standard

, and since (Nt/(1+c) t)tR has
Poisson process with intensity under P
+
independent increments, the compensated process

Nt/(1+c) t,
t R+ ,
by (6.2). In addition we have

is a martingale under P
Nt/(1+c) =
1[Tn ,) (t/(1 + c))
n1
t R+ ,
1[(1+c)Tn ,) (t),
n1
, the jump times of (Nt/(1+c) )t[0,T ] are given by

which shows that under P
((1 + c)Tn )n1 ,

and we know that they are distributed as the jump times of a Poisson process
.
with intensity under P
> 0 and letting

Next, taking
c := 1 +
= (1 + c) we can rewrite the above by saying that

i.e.
P(N(1+c)T = k) = ecT (1 + c)k P(NT = k) = eT
)k
(T
(NT = k),
=P
k!
k N, and
:= ecT (1 + c)NT dP = e()T

dP
!NT
dP,
the law of NT is that of a Poisson random variable with intensity

= (1 + c)T.
T
Consequently, since
466
"

tR
(Nt (1 + c)t)tR+ = (Nt t)
+
has independent increments, the compensated Poisson process
Nt (1 + c)t = Nt t
by (6.2), although when c 6= 0 it is not a martingale
is a martingale under P
under P.
In the case of compound Poisson processes the Girsanov theorem can be
extended to variations in jump sizes in addition to time variations, and we
have the following more general result.
Theorem 14.2. Let (Yt )t0 be a compound Poisson process with intensity > 0 and jump distribution (dx). Consider another jump distribution
(dx), and let
d
(x) :=
(x) 1,
x R.
d
Then,
NT
Y
:= e()T
dP
,
, ,
(1 + (Zk ))dP
k=1
the process
Yt =
Nt
X
Zk ,
t R+ ,
k=1
is a compound Poisson process with

> 0, and
- modified intensity
- modified jump distribution (dx).
Proof. For any bounded measurable function f on R, we extend (14.16) to
the following change of variable
"
#
NT
Y
()T
IE,
IE, f (YT ) (1 + (Zi ))
[f (YT )] = e
i=1
=e
()T
X
k=0
"
"
IE, f
k
X
i=1
!
Zi
k
Y
#

(1 + (Zi ))NT = k P(NT = k)
i=1
467
N. Privault
= eT
X
(T )k
k=0
=e
= eT
k!

X
(T )k w
k=0
X
k=0
k
X
!
Zi
i=1
X
(T )k w
k=0
= eT
k!
"
IE, f
k!
k
Y
#
(1 + (Zi ))
i=1
f (z1 + + zk )
k
Y
(1 + (zi ))(dz1 ) (dzk )
i=1
f (z1 + + zk )
k
Y
d
i=1
!
(zi ) (dz1 ) (dzk )
w
) w
(T
f (z1 + + zk )
(dz1 ) (dzk ).
k!
k
This shows that under P,

, YT has the distribution of a compound Poisson
and jump distribution . We refer to Proposition 9.6
process with intensity
.
of [15] for the independence of increments of (Yt )tR+ under P

,
> 0 and jump

Note that the compound Poisson process with intensity
distribution can be built as
Nt/
Xt :=
h(Zk ),
k=1
provided is the image measure of by the function h : R R, i.e.

P(h(Zk ) A) = P(Zk h1 (A)) = (h1 (A)) = (A),
for all measurable subset A of R.
Compensated Compound Poisson Martingale

As a consequence of Theorem 14.2, the compensated process
IE [Z1 ]
Yt t
defined by
becomes a martingale under the probability measure P
,
= e()T
dP
,
NT
Y
, .
(1 + (Zk ))dP
k=1
Finally, the Girsanov theorem can be extended to the linear combination

of a standard Brownian motion (Wt )tR+ and an independent compound
Poisson process (Yt )tR+ , as in the following result which is a particular case
of Theorem 33.2 of [105].
468
"

Theorem 14.3. Let (Yt )t0 be a compound Poisson process with intensity
> 0 and jump distribution (dx). Consider another jump distribution (dx)
> 0, and let
and intensity parameter
(x) :=
(x) 1,
d
x R,
and let (ut )tR+ be a bounded adapted process. Then the process

wt
IE [Z1 ]t
Wt +
us ds + Yt
0
tR+
is a martingale under the probability measure

Y
NT
wT
1wT
)T
= exp (
, .
dP
us dWs
|us |2 ds
(1+(Zk ))dP
u,,
0
2 0
k=1
(14.17)
As a consequence of Theorem 14.3, if
wt
Wt +
vs ds + Yt
0
, , it will become a martingale under P

is not a martingale under P
u,,
and are chosen in such a way that

provided u,
IE [Z1 ],
vs = us
s R,
(14.18)
in which case we will have the martingale decomposition

IE [Z1 ]dt,
dWt + ut dt + dYt

wt
IE [Z1 ]
in which both Wt +
us ds
and Yt t
0
tR+
are both martR+

tingales under P
u,,
= = 0, Theorem 14.3 coincides with the usual Girsanov theorem

When
for Brownian motion, in which case (14.18) admits only one solution given
u,0,0 . Note that uniqueness occurs also
by u = v and there is uniqueness of P
when u = 0 in the absence of Brownian motion with Poisson jumps of fixed
size a (i.e. (dx) = (dx) = a (dx)) since in this case (14.18) also admits
= v and there is uniqueness of P
. These remarks will
only one solution
0,,a
be of importance for arbitrage pricing in jump models in Chapter 15.
"
469
N. Privault
Exercises
Exercise 14.1 Let (Nt )tR+ be a standard Poisson process with intensity
> 0, started at N0 = 0.
dSt = St dNt St dt = St (dNt dt).
b) Using the first Poisson jump time T1 , solve the stochastic differential equation
dSt = St dt + dNt ,
t (0, T2 ).
Exercise 14.2 Consider a standard Poisson process (Nt )tR+ with intensity
> 0.
a) Solve the stochastic differential equation dXt = Xt dNt for (Xt )tR+ ,
where > 0 and X0 = 1.
b) Show that the solution (St )tR+ of the stochastic differential equation
dSt = rdt + St dNt ,
wt
is given by St = S0 Xt + rXt
Xs1 ds.
0
c) Compute IE[Xt ] and IE[Xt /Xs ], 0 s t.
d) Compute IE[St ], t R+ .
Exercise 14.3 Consider the compound Poisson process Yt :=
Nt
X
Zk , where
k=1
(Nt )tR+ is a standard Poisson process with intensity > 0, (Zk )k1 is an
i.i.d. sequence of N (0, 1) Gaussian random variables. Solve the stochastic
dSt = rSt dt + St dYt ,
where , r R.
Exercise 14.4 Show, by direct computation or using the characteristic function, that the variance of the compound Poisson process Yt with intensity
> 0 satisfies
w
Var [Yt ] = t IE[|Z1 |2 ] = t
x2 (dx).
Exercise 14.5 Consider an exponential compound Poisson process of the form

470
"

St = S0 et+Wt +Yt ,
t R+ ,
where (Yt )tR+ is a compound Poisson process of the form (14.6).

a) Derive the stochastic differential equation with jumps satisfied by (St )tR+ .
) of probability measures under which the
b) Let r > 0. Find a family (P
u,,
discounted asset price ert St is a martingale.

Exercise 14.6 Consider (Nt )tR+ a standard Poisson process with intensity
> 0, independent of (Wt )tR+ , under a probability measure P. Let (St )tR+
be defined by the stochastic differential equation
dSt = St dt + YNt St dNt ,
(14.19)
where (Yk )k1 is an i.i.d. sequence of random variables of the form

Yk = eXk 1,
where
Xk ' N (0, 2 ),
k 1.
a) Solve the equation (14.19).

b) We assume that and the risk-free rate r > 0 are chosen such that the
discounted process (ert St )tR+ is a martingale under P. What relation
does this impose on and r?
c) Under the relation of Question (b), compute the price at time t of a European call option on ST with strike and maturity T , using a series
expansion of Black-Scholes functions.
> 0 under a probability measure P. Let (St )tR+ be the mean reverting
process defined by the stochastic differential equation
dSt = St dt + (dNt dt),
(14.20)
where S0 > 0 and , > 0.

a) Solve the equation (14.20) for St .
b) Compute f (t) := IE[St ] for all t R+ .
c) Under which condition on , , and does the process St become a
submartingale ?
d) Propose a method for the calculation of expectations of the form IE[(ST )]
where is a payoff function.
"
471
Chapter 15
Pricing and Hedging in Jump Models
In this chapter we consider option pricing and hedging in jump-diffusion

models. In comparison with the continuous case the situation is further complicated by the existence of multiple risk-neutral measures. As a consequence,
perfect replicating hedging strategies cannot be computed in general.
15.1 Risk-Neutral Measures

Consider an asset price modeled by the equation,
dSt = St dt + St dWt + St dYt ,
(15.1)
where (Yt )tR+ is the compound Poisson process defined in Section 14.2, with
jump size distribution (dx) under P . The equation (15.1) has for solution

Y
Nt
1
St = S0 exp t + Wt 2 t
(1 + Zk ),
2
(15.2)
k=1
t R+ . An important issue for non-abitrage pricing is to determine a

risk-neutral probability measure P under which the discounted process
(ert St )tR+ is a martingale, and this goal can be achieved using the Girsanov theorem for jump processes, cf. Section 14.6.
We have
d(ert St ) = rert St dt + ert dSt
= ( r)ert St dt + ert St dWt + ert St dYt
= ( r + IE [Z1 ])ert St dt + ert St dWt + ert St (dYt IE [Z1 ]dt),
which yields a martingale under P provided
"
N. Privault
r + IE [Z1 ] = 0,
however that condition may not be satisfied under P by the market parameters.
In that case a change of measure might be needed. In order for the discounted process (ert St )tR+ to be a martingale, we may choose a drift pa > 0, and a jump distribution satisfying
rameter u R, and intensity
IE [Z1 ].
r = u
(15.3)
The Girsanov theorem for jump processes then shows that

IE [Z1 ]dt
dWt + udt + dYt
is a martingale under the probability measure Pu,,
defined in (14.17). Consequently the discounted asset price
d(ert St ) = ( r)ert St dt + ert St dWt + ert St dYt
IE [Z1 ]dt),
= ert St (dWt + udt) + ert St (dYt
is a martingale under Pu,,
.
In this setting the non-uniqueness of the risk neutral measure is apparent
since additional degrees of freedom are involved in the choices of u, and
the measure , whereas in the continuous case the choice of u = ( r)/ in
(6.4) was unique.
15.2 Pricing in Jump Models

Recall that a market is without arbitrage if and only it admits at least one
risk-neutral measure.
Consider the probability measure Pu,,
built in the previous section, under
which the discounted asset price
IE [Z1 ]dt) + ert St dW
t,
d(ert St ) = ert St (dYt
t = Wt + udt is a standard Brownian motion under
is a martingale, and W
Pu,,
.
Then the arbitrage price of a claim with payoff C is given by
er(T t) IEu,,
[C | Ft ]
474
(15.4)
"

under Pu,,
.
Clearly the price (15.4) of C is no longer unique in the presence of jumps
due to the infinity of choices satisfying the martingale condition (15.3), and
= = 0, or ( = 0 and
such a market is not complete, except if either
= = 1 ).
Pricing of Vanilla Options

The price of a vanilla option with payoff of the form (ST ) on the underlying
asset ST can be written from (15.4) as
er(T t) IEu,,
[(ST ) | Ft ],
(15.5)
where the expectation can be computed as

IEu,,
[(ST ) | Ft ]
"
!
#
N

T

Y
1 2

= IEu,,
(1 + Zk ) Ft
S0 exp T + WT T
2
k=1

N

T

Y
1 2

(T
t)
(1
+
Z
)
S
exp
(T
t)
+
(W
W
)
= IEu,,
F
t
t
t

T
k
2
k=Nt

N
T
Y
1 2
= IEu,,
(1 + Zk )
x exp (T t) + (WT Wt ) (T t)
2
k=N
t
x=St
Pu,,
(NT Nt = n)
n=0
NT
(T t)+(WT Wt ) 2 (T t)/2
IEu,,
xe
(1 + Zk ) NT Nt = n
k=Nt
=e
(T
t)
t))n
X
((T
n!
n=0
"
(T t)+(WT Wt )
IEu,,
xe
(T t)/2
n
Y
x=St
!#
(1 + Zk )
k=1
=e
(T
t)
x=St
w
t))n w
X
((T
n!
n=0
"
(T t)+(WT Wt )
IEu,,
xe
(T t)/2
n
Y
k=1
!#
(dz1 ) (dzn ),
(1 + zk )
x=St
hence
"
475
N. Privault
er(T t) IE0,,
[(ST ) | Ft ]
w
X
t))n w
1
((T
= p
e(r+)(T t)
n!
2(T t)
n=0
!
n
Y
2
2
St e(T t)+x (T t)/2
(1 + zk ) ex /(2(T t)) (dz1 ) (dzn )dx.
k=1
15.3 Black-Scholes PDE with Jumps

Recall that by the Markov property of (St )tR+ the price (15.5) at time t of
the option with payoff (ST ) can be written as a function f (t, St ) of t and
St , i.e.
f (t, St ) = er(T t) IEu,,
(15.6)
[(ST ) | Ft ],
with the terminal condition f (T, x) = (x). In addition,
t 7 er(T t) f (t, St )
is a martingale under Pu,,
by the same argument as in (6.1).
In this section we derive a partial integro-differential equation (PIDE) for
the function (t, x) 7 f (t, x). We have
IE [Z1 ]dt),
t + St (dYt
dSt = rSt dt + St dW
(15.7)
t = Wt + ut is a standard Brownian motion under P .

where W
u,,
Hence the Ito formula with jumps (14.9) shows that

2
f
f
f
t + 1 2 St2 f (t, St )dt
(t, St )dt + rSt (t, St )dt + St (t, St )dW
t
x
x
2
x2
f
IE [Z1 ]St (t, St )dt + (f (t, St (1 + ZN )) f (t, St ))dNt
t
x
f
t
= St (t, St )dW
x
IE [(f (t, x(1 + Z1 )) f (t, x))]x=S dt
+(f (t, St (1 + ZNt )) f (t, St ))dNt
t
f
1 2 2 2f
f
+ (t, St )dt + rSt (t, St )dt + St 2 (t, St )dt
t
x
2
x
IE [Z1 ]St f (t, St )dt.

+ IE [(f (t, x(1 + Z1 )) f (t, x))]x=St dt
x
df (t, St ) =
Based on the relation
476
"

d(er(T t) f (t, St )) = rer(T t) f (t, St )dt + er(T t) df (t, St )
and the facts that
t )tR ,
Brownian motion (W
+
the process given by the differential
IE [(f (t, x(1+Z1 ))f (t, x))]x=S dt,
(f (t, St (1+ZNt ))f (t, St ))dNt
t
and
the process t 7 er(T t) f (t, St ),
are all martingales under Pu,,
, we conclude to the vanishing
f
f
1
2f
(t, St ) + rSt (t, St ) + 2 St2 2 (t, St )
t
x
2
x
IE [(f (t, x(1 + Z1 )) f (t, x))]x=S
IE [Z1 ]St f (t, St ) = 0,
+
t
x
rf (t, St ) +
or
f
f
1
2f
(t, x) + rx (t, x) + 2 x2 2 (t, x)
t
x
2
x
IE [(f (t, x(1 + Z1 )) f (t, x))]
IE [Z1 ]x f (t, x) = 0,
+
x
rf (t, x) +
which leads to the Partial Integro-Differential Equation (PIDE)
rf (t, x) =
f
f
1
2f
(t, x) + rx (t, x) + 2 x2 2 (t, x)
t
x
2
x

w
f
+
f (t, x(1 + y)) f (t, x) yx (t, x) (dy),
x
(15.8)
under the terminal condition f (T, x) = (x).

A major technical difficulty when solving the PIDE (15.8) numerically is
that the operator

w
f
f 7
f (t, x(1 + y)) f (t, x) yx (t, x) (dy)
x
is nonlocal, therefore adding significant difficulties to the application of standard discretization schemes.
"
477
N. Privault
In addition we have found that the change df (t, St ) in the portfolio price
(15.6) is given by
f
t + rf (t, St )dt
(t, St )dW
(15.9)
x
IE [(f (t, x(1 + Z1 )) f (t, x))]x=S dt.
+(f (t, St (1 + ZNt )) f (t, St ))dNt
t
df (t, St ) = St
In the case of Poisson jumps with fixed size a, i.e. Yt = aNt , (dx) = a (dx),
the PIDE (15.8) reads
rf (t, x) =
f
1
2f
f
(t, x) + rx (t, x) + 2 x2 2 (t, x)
t
x
2
x

f (t, x(1 + a)) f (t, x) ax f (t, x) ,
+
x
and we have
f
t + rf (t, St )dt
(t, St )dW
x
(t, St (1 + a)) f (t, St ))dt.
+(f (t, St (1 + a)) f (t, St ))dNt (f
df (t, St ) = St
15.4 Exponential Models

Instead of modeling the asset price (St )tR+ through a stochastic exponential
(15.2) solution of the stochastic differential equation with jumps of the form
(15.1), we may consider an exponential price process of the form
!
Nt
X
St = S0 et+Wt +Yt = S0 exp t + Wt +
Zk ,
t R+ ,
k=1
rt
and choose a risk-neutral measure Pu,,
St )tR+ is a mar under which (e
tingale. Then the expectation
er(T t) IEu,,
[(ST ) | Ft ]
also becomes a (non-unique) arbitrage price at time t [0, T ] for the contingent claim with payoff (ST ).
Such an arbitrage price can be expressed as

r(T t)
T +WT +YT

er(T t) IEu,,
IEu,,
) Ft
(ST ) Ft = e
(S0 e

(T t)+(WT Wt )+YT Yt
= er(T t) IEu,,
) Ft
(St e

(T t)+(WT Wt )+YT Yt
)
= er(T t) IEu,,
(xe
x=St
478
"

"
NT
X
= er(T t) IEu,,
x exp (T t) + (WT Wt ) +
!!#
Zk
k=Nt +1
x=St
= er(T t)(T t)
"
!!#
n
X
X
t))n
((T
(T t)+(WT Wt )
IEu,,
xe
exp
Z
k
n!
n=0
k=1
.
x=St
In the exponential model

St = S0 et+Wt +Yt = S0 e(+
/2)t+Wt 2 t/2+Yt
the process St satisfies

1
dSt =
+ 2 St dt + St dWt + St (eYt 1)dNt ,
2
hence St has jumps of size ST (eZk 1), k 1, and (15.3) reads
k
1
IE [eZ1 1].
+ 2 r = u
2
The Merton Model
We assume that (Zk )k1 is a family of independent identically distributed
Gaussian N (, 2 ) random variables under Pu,,
with
1
IE [eZ1 1] = u (e
+2 /2 1),
+ 2 r = u
2
from (15.3), hence is a standard Brownian motion under Pu,,
. For simplicity
we choose u = 0, i.e.
1
+2 /2 1),
= r 2 (e
2
Hence we have
er(T t) IE,
[(ST ) | Ft ]
= er(T t)(T t)
X
t))n
((T
n!
n=0
"
(T t)+(WT Wt )
IE,
exp
xe
n
X
k=1
"
!!#
Zk
x=St
479
N. Privault
= er(T t)(T t)
i
X
t))n h
((T
IE (xe(T t)+n+X )
,
n!
x=St
n=0
where
X = (WT Wt ) +
n
X
(Zk ) ' N (0, 2 (T t) + n 2 )
k=1
is a centered Gaussian random variable with variance

v 2 = 2 (T t) +
n
X
Var Zk = 2 (T t) + n 2 .
k=1
Hence when (x) = (x )+ , using the relation

BS(x, , v 2 /, r, ) = er IE[(xeXv
/2+r
K)+ ]
we get
+
er(T t)(T t) IE,
[(ST ) | Ft ]
= er(T t)(T t)
= er(T t)(T t)
i
X
t))n h
((T
IE (xe(T t)+n+X )+
n!
x=St
n=0
X
t))n
((T
n!
n=0
h
i
2
+2 /2 1))(T t)+n+X
IE (xe(r /2(e
)+
x=St
t)
r(T t)(T
=e
X
t))n
((T
n!
n=0
h
i
2
+2 /2 1)(T t)+Xv 2 /2+r(T t)
IE (xen+n /2(e
)+
x=St
t)
(T
=e
X
t))n
((T
n!
n=0
BS(St en+n
+
/2(e
2 /2
1)(T t)
, , 2 + n 2 /(T t), r, T t).
We may also write
+
er(T t)(T t) IE,
[(ST ) | Ft ]
= e(T t)
X
t))n
2
((T
+2 /2 1)(T t)
en+n /2(e
n!
n=0

2
+2 /2 1)(T t)
, 2 + n 2 /(T t), r, T t
BS St , enn /2+(e
480
"
+ 2 /2
= ee
(T t)
X
+n2 /2 (T t))n
(e
n!
n=0

+ 2 /2 +2 /2
BS St , , 2 + n 2 /(T t), r + n
(e
1), T t .
T t
15.5 Self-Financing Hedging with Jumps

Consider a portfolio with value
Vt = t ert + t St
at time t, and satisfying the self-financing condition
dVt = rt ert dt + t dSt ,
cf. Relation (5.1). Assuming that the portfolio price takes the form Vt =
f (t, St ) for all times t [0, T ], by (15.7) we have
dVt = df (t, St )
= rt ert dt + t dSt
IE [Z1 ]dt))
t + St (dYt
= rt ert dt + t (rSt dt + St dW
IE [Z1 ]dt)
t + t St (dYt
= rVt dt + t St dW
IE [Z1 ]dt),
t + t St (dYt
(15.10)
= rf (t, St )dt + t St dW
has to match
f
t
(t, St )dW
(15.11)
x
IE [(f (t, x(1 + Z1 )) f (t, x))]x=S dt,
+ (f (t, St (1 + ZNt )) f (t, St ))dNt
t
df (t, St ) = rf (t, St )dt + St
which is obtained from (15.9).

In such a situation we say that the claim C can be exactly replicated.
Exact replication is possible in essentially only two situations:
= 0. In this case we find the usual Black(i) Continuous market, =
Scholes Delta:
f
t =
(t, St ).
(15.12)
x
(ii) Poisson jump market, = 0 and Yt = aNt , (dx) = a (dx). In this case
we find
"
481
N. Privault
t =
1
(f (t, St (1 + a)) f (t, St )).
aSt
(15.13)
Note that in the limit a 0 this expression recovers the Black-Scholes

Delta formula (15.12).
When Conditions (i) or (ii) above are not satisfied, exact replication is not
possible and this results into an hedging error given from (15.10) and (15.11)
by
VT (ST ) = VT f (T, ST )
wT
wT
= V0 f (0, S0 ) +
dVt
df (t, St )
0
0

wT
wT
f
IE [Z1 ]dt)
t +
(t, St ) dW
t St (ZNt dNt
= V0 f (0, S0 ) +
St t
0
0
x
wT
wT
(f (t, St (1 + ZN )) f (t, St ))dNt +

IE [(f (t, x(1 + Z1 )) f (t, x))]x=S dt.
t
Assuming for simplicity that Yt = aNt , i.e. (dx) = a (dx), we get

wT
f
t
VT f (T, ST ) = V0 f (0, S0 ) +
St t
(t, St ) dW
0
x
wT
(f (t, St (1 + a)) f (t, St ) at St )(dNt dt),

0
hence the mean square hedging error is given from the Ito isometry (14.8) by
IEu, [(VT f (T, ST ))2 ]
= (V0 f (0, S0 ))2 + 2 IEu,
+ IEu,
"
wT
0
"
"
w
T
0

2 #
f
t
(t, St ) dW
St t
x
2 #
(f (t, St (1 + a)) f (t, St ) at St )(dNt dt)

wT

2 #
f
St2 t
(t, St ) dt
0
x

2
((f (t, St (1 + a)) f (t, St ) at St )) dt .
= (V0 f (0, S0 ))2 + 2 IEu,

IE
+
u,
wT
Clearly, the initial portfolio value V0 that minimizes the above quantity is
V0 = f (0, S0 ) = erT IEu,,
[(ST )].
When hedging only the risk generated by the Brownian part we let
t =
f
(t, St )
x
482
"

as in the Black-Scholes model, and in this case the hedging error due to the
presence of jumps becomes
w

T
2
IE
IEu, [(VT f (T, ST ))2 ] =
((f (t, St (1 + a)) f (t, St ) at St )) dt .
u,
0
Next, let us find the optimal strategy (t )tR+ that minimizes the remaining
hedging error
! #
"
2

wT
f
((f (t, St (1 + a)) f (t, St ) at St ))2 dt .
(t, St ) +
IEu,
2 St2 t
0
x
For all t [0, T ], the almost-sure minimum of

2
f
((f (t, St (1 + a)) f (t, St ) at St ))2
t 7 2 St2 t
(t, St ) +
x
is given by differentiation with respect to t , as the solution of

f
t ((f (t, St (1 + a)) f (t, St ) at St )) = 0,
2 St2 t
(t, St ) aS
x
i.e.
2
t =
f
a
(t, St ) +
(f (t, St (1 + a)) f (t, St ))
x
St
,
2 + a2
t [0, T ].
(15.14)
We note that the optimal strategy (15.14) is a weighted average of the

Brownian and jump hedging strategies (15.12) and (15.13) according to the
of the continuous and jump comrespective variance parameters 2 and a2
ponents.
= 0 we get
Clearly, if a
t =
f
(t, St ),
x
t [0, T ],
which is the Black-Scholes perfect replication strategy, and when = 0 we

recover
f (t, St (1 + a)) f (t, St )
t =
,
t [0, T ].
aSt
"
483
N. Privault
which is (15.13).
Note that the fact that perfect replication is not possible in a jumpdiffusion model can be interpreted as a more realistic feature of the model,
as perfect replication is not possible in the real world.
See [59] for an example of a complete market model with jumps, in which
continuous and jump noise are mutually excluding each other over time.
Exercises
> 0 under a probability measure P. Let (St )tR+ be defined by the stochastic differential equation
dSt = rSt dt + St (dNt dt),
where > 0.
1. Find the value of R such that the discounted process (ert St )tR+ is
a martingale under P.
2. Compute the price at time t of a power option with payoff |ST |2 at maturity T .
Exercise 15.2 Consider a long forward contract with payoff ST K on a
jump diffusion risky asset (St )tR+ given by
dSt = St dt + St dWt + St dYt .
1. Show that the forward claim admits a unique arbitrage price to be computed in a market with risk-free rate r > 0.
2. Show that the forward claim admits an exact replicating portfolio strategy based on the two assets St and ert .
3. Show that the portfolio strategy of Question 2 coincides with the optimal
portfolio strategy (15.14).
Exercise 15.3 Consider (Wt )tR+ a standard Brownian motion and (Nt )tR+
a standard Poisson process with intensity > 0, independent of (Wt )tR+ ,
under a probability measure P . Let (St )tR+ be defined by the stochastic
dSt = St dt + St dNt + St dWt .
484
(15.15)
"

1. Solve the equation (15.15).
2. We assume that , and the risk-free rate r > 0 are chosen such that the
discounted process (ert St )tR+ is a martingale under P . What relation
does this impose on , , and r?
3. Under the relation of Question (2), compute the price at time t of a
European call option on ST with strike and maturity T , using a series
expansion of Black-Scholes functions.
dSt = rSt dt + YNt St dNt ,
where (Yk )k1 is an i.i.d. sequence of uniformly distributed random variables
on [1, 1].
1. Show that the discounted process (ert St )tR+ is a martingale under P.
2. Compute the price at time 0 of a European call option on ST with strike
and maturity T , using a series of multiple integrals.
dSt = rSt dt + YNt St (dNt dt),
where (Yk )k1 is an i.i.d. sequence of uniformly distributed random variables
on [0, 1].
(a) Find the value of R such that the discounted process (ert St )tR+
(b) Compute the price at time t of a long forward contract with maturity T
and payoff ST .
> 0 under a risk-neutral probability measure P. Let (St )tR+ be defined by
the stochastic differential equation
dSt = rSt dt + St (dNt dt),
where > 0. Consider a portfolio with value
Vt = t ert + t St
at time t, and satisfying the self-financing condition
"
485
N. Privault
dVt = rt ert dt + t dSt .
We assume that the portfolio hedges the claim (ST ), i.e. we have Vt =
f (t, St ) for all times t [0, T ].
1. Show that under self-financing the portfolio value Vt satisfies
dVt = rf (t, St )dt + t St (dNt dt).
(15.16)
2. Show that the claim C can be exactly replicated by the hedging strategy
t =
1
(f (t, St (1 + )) f (t, St )).
St
486
"
Chapter 16
Basic Numerical Methods
This chapter is an elementary introduction to finite difference methods for

the resolution of PDEs and stochastic differential equations. We cover the
explicit and implicit finite difference schemes for the heat equations and the
Black-Scholes PDE, as well as the Euler and Milshtein schemes for stochastic
differential equations.
16.1 Discretized Heat Equation

Consider the heat equation
2
(t, x) =
(t, x)
t
x2
(16.1)
with initial condition

(0, x) = f (x)
on a compact interval [0, T ] [0, X] divided into the grid points
(ti , xj ) = (it, jx),
i = 0, . . . , N,
j = 0, . . . , M,
with t = T /N and x = X/M . Our goal is to solve the heat equation

(16.1) with initial condition (0, x) and lateral boundary conditions (t, 0),
(t, X), via a discrete approximation
((ti , xj ))0iN,
0jM
of the solution to (16.1), by evaluating derivatives using finite differences.
"
N. Privault
Explicit method
Using the forward time difference approximation
(ti+1 , xj ) (ti , xj )
(ti , x) '
t
t
of the time derivative, we discretize (16.1) as
(ti+1 , xj ) (ti , xj )
(ti , xj+1 ) + (ti , xj1 ) 2(ti , xj )
=
t
(x)2
and letting = (t)/(x)2 this yields
(ti+1 , xj ) = (ti , xj+1 ) + (1 2)(ti , xj ) + (ti , xj1 ),
1 j M 1, 1 i N , i.e.
(ti , x0 )
..
i+1 = Ai +
,
.
0
(ti , xM )
with
and
The vector
i = 0, 1, . . . , N 1,
(ti , x1 )
..
i =
,
.
(ti , xM 1 )
1 2
0
1 2
0
1 2
..
..
A = ...
.
.
0
0
0
0
0
0
0
0
0
(ti , x0 )
0
..
.
0
(ti , xM )
i = 0, 1, . . . , N,
..
.
0
0
0
..
.
0
0
0
..
.
0
0
0
..
.
1 2
0
1 2
0
1 2
i = 0, . . . , N,
can be given by the lateral boundary conditions (t, 0) and (t, X).
488
"

Implicit method
Using the backward time difference approximation
(ti , xj ) (ti1 , xj )
(ti , x) '
t
t
(ti , xj ) (ti1 , xj )
(ti , xj+1 ) + (ti , xj1 ) 2(ti , xj )
=
t
(x)2
and letting = (t)/(x)2 we get
(ti1 , xj ) = (ti , xj+1 ) + (1 + 2)(ti , xj ) (ti , xj1 ),
1 j M 1, 1 i N , i.e.
i1
with
= Bi +
(ti , x0 )
0
..
.
0
(ti , xM )
i = 1, 2, . . . , N,
1 + 2
0
0
0
0
1 + 2
0
0
0
0
1 + 2
0
0
0
..
.
.
.
.
.
..
..
..
..
..
..
B= .
.
.
0
0
1
+
2
0
0
0 1 + 2
0
0
0
0
1 + 2
By inversion of the matrix B, i is given in terms of i1 as
(ti , x0 )
..
i = B 1 i1 B 1
i = 1, . . . , N.
,
.
0
(ti , xM )
16.2 Discretized Black-Scholes PDE

Consider the Black-Scholes PDE
"
489
N. Privault
r(t, x) =
1
2
(t, x) + rx (t, x) + x2 2 2 (t, x),
t
x
2
x
(16.2)
under the terminal condition (T, x) = (x K)+ , resp. (T, x) = (K x)+ ,

for a European call, resp. put, option. The constant volatility coefficient
may also be replaced with a function (x) of the underlying, in the case local
volatility models.
Note that in the solution of the Black-Scholes PDE, time is run backwards
as we start from a terminal condition (T, x) at time T . Thus here the explicit
method uses backward differences while the implicit method uses forward
differences.
Explicit method
Using here the backward time difference approximation
(ti , xj ) (ti1 , xj )
(ti , x) '
t
t
r(ti , xj ) =
(ti , xj+1 ) (ti , xj1 )

(ti , xj ) (ti1 , xj )
+ rxj
t
2x
1
(ti , xj+1 ) + (ti , xj1 ) 2(ti , xj )
+ x2j 2
,
2
(x)2
1 j M 1, 0 i N 1, i.e.
(ti1 , xj ) =
1
t( 2 j 2 rj)(ti , xj1 ) + (1 t( 2 j 2 + r))(ti , xj )
2
1
+ t( 2 j 2 + rj)(ti , xj+1 ),
2
1 j M 1, where the lateral boundary conditions (ti , 0) and (ti , X)

are given by
(ti , x0 ) = 0,
(ti , xM ) = xM Ker(T ti ) ,
0 i N,
for a European call option, and

(ti , x0 ) = Ker(T ti ) ,
(ti , xM ) = 0
0 i N,
for a European put option.
490
"

The explicit finite difference method is known to have a divergent behaviour when time runs backwards, as illustrated in Figure 16.1.
Explicit method
100
50
-50
-100
0
0.1
0.2
0.3
0.4
0.5
time to maturity
0.6
0.7
0.8
0.9
20
40
60
80
100
120
140
160
180
200
strike
Fig. 16.1: Divergence of the explicit finite difference method.
Implicit method
Using the forward time difference approximation
(ti+1 , xj ) (ti , xj )
(ti , x) '
t
t
r(ti , xj ) =
(ti+1 , xj ) (ti , xj )
(ti , xj+1 ) (ti , xj1 )
+ rxj
t
2x
1
(ti , xj+1 ) + (ti , xj1 ) 2(ti , xj )
+ x2j 2
,
2
(x)2
1 j M 1, 0 i N 1, i.e.
1
(ti+1 , xj ) = t( 2 j 2 rj)(ti , xj1 ) + (1 + t( 2 j 2 + r))(ti , xj )
2
1
t( 2 j 2 + rj)(ti , xj+1 ),
2
1 j M 1, i.e.
"
491
N. Privault
1
2 t
i+1
= Bi +

r 2 (ti , x0 )
0
..
.
0

21 t r(M 1) + 2 (M 1)2 (ti , xM )
i = 0, 1, . . . , N 1, with
Bj,j1 =

1
t rj 2 j 2 ,
2
Bj,j = 1 + 2 j 2 t + rt,
and

1
Bj,j+1 = t rj + 2 j 2 ,
2
for j = 1, . . . , M 1, and B(i, j) = 0 otherwise.
By inversion of the matrix B, i is given in terms of i+1 as
1
2
(ti , x0 )
2 t r
..
i = B 1 i+1 B 1
,
.
0

21 t r(M 1) + 2 (M 1)2 (ti , xM )
i = 0, 1, . . . , N 1, where the boundary conditions (ti , x0 ) and (ti , xM )

can be provided as in the case of the explicit method. Note that for all
j = 1, . . . , M 1 we have
Bj,j1 + Bj,j + Bj,j+1 = 1 + rt,
hence when the terminal condition is a constant (T, x) = c > 0 we get
(ti , x) = c(1 + rt)(N i) = c(1 + rT /N )(N i) ,
i = 0, . . . , N,
hence for all s [0, T ],

(s, x) = lim (t[N s/T ] , x)
N
= c lim (1 + rT /N )(N [N s/T ])

N
= c lim (1 + rT /N )[N (T s)/T ]

N
= c lim (1 + rT /N )(T s)/T

N
= cer(T s) ,
492
"

as expected, where [x] denotes the integer part of x R. The implicit finite
difference method is known to be more stable than the explicit method, as
illustrated in Figure 16.2, in which the discretization parameters have been
taken to be the same as in Figure 16.1.
Implicit method
140
120
100
80
60
40
20
0
time to maturity
10
20
40
60
80
100
120
140
160
180
200
strike
Fig. 16.2: Stability of the implicit finite difference method.
16.3 Euler Discretization

In order to apply the Monte Carlo method in option pricing, we need to
1, . . . , X
N } of sample values of a random variable X,
generate a sequence {X
such that the empirical mean
IE[(X)] '
1 ) + + (X
N )
(X
N
can be used according to the strong law of large number for the evaluation of
the expected value IE[(X)]. Despite its apparent simplicity, the Monte Carlo
method can be delicate to implement and the optimization of Monte Carlo
algorithms and random number generation have been the object of numerous
works which are outside the scope of this text, cf. e.g. [43], [69].
Random samples for the solution of a stochastic differential equation of
the form
dXt = b(Xt )dt + a(Xt )dWt
(16.3)
can be generated by discretization. More precisely, the Euler discretization
scheme for the stochastic differential equation (16.3) is given by
tN = X
tN +
X
k+1
k
'
"
N
X
tk
w tk+1
tk
b(Xs )ds +
tN )(tk+1
b(X
k
w tk+1
tk
a(Xs )dWs
tN )(Wt
tk ) + a(X
Wtk ).
k+1
k
493

N. Privault
In particular, when Xt is the geometric Brownian motion given by
dXt = rXt dt + Xt dWt
we get
N = X
N + rX
N (tk+1 tk ) + X
N (Wt
X
Wtk ),
tk+1
tk
tk
tk
k+1
which can be computed as
tN = X
tN
X
0
k
k
Y

1 + r(ti ti1 ) + (Wti Wti1 ) .
i=1
16.4 Milshtein Discretization

In the Milshtein scheme we expand a(Xs ) as
a(Xs ) ' a(Xtk ) + a0 (Xtk )b(Xtk )(s tk ) + a0 (Xtk )a(Xtk )(Ws Wtk ).
As a consequence we get
tN = X
tN +
X
k+1
k
tN +
'X
k
w tk+1
tk
w tk+1
b(Xs )ds +
w tk+1
tk
a(Xs )dWs
b(Xs )ds + a(Xtk )(Wtk+1 Wtk )

w tk+1
+a0 (Xtk )b(Xtk )
(s tk )dWs
tk
w tk+1
0
+a (Xtk )a(Xtk )
(Ws Wtk )dWs
tk
w tk+1
tN +
'X
k
tk
w tk+1
+a0 (Xtk )a(Xtk )
(Ws Wtk )dWs .
tk
tk
Next, using Itos formula we note that

w tk+1
w tk+1
(Wtk+1 Wtk )2 = 2
(Ws Wtk )dWs +
ds,
tk
tk
hence
w tk+1
tk
(Ws Wtk )dWs =
1
((Wtk+1 Wtk )2 (tk+1 tk )),
2
and
tN ' X
tN +
X
k+1
k
w tk+1
tk
494
"

1
+ a0 (Xtk )a(Xtk )((Wtk+1 Wtk )2 (tk+1 tk ))
2
tN + b(Xt )(tk+1 tk ) + a(Xt )(Wt
'X
W tk )
k
k
k+1
k
1 0
2
+ a (Xtk )a(Xtk )((Wtk+1 Wtk ) (tk+1 tk )).
2
As a consequence the Milshtein scheme is written as
N ' X
N + b(X
N )(tk+1 tk ) + a(X
N )(Wt
X
W tk )
tk+1
tk
tk
tk
k+1
1 0 N
tN )((Wt
Wtk )2 (tk+1 tk )),
+ a (Xtk )a(X
k+1
k
2
i.e. in the Milshtein scheme we take into account the small difference
(Wtk+1 Wtk )2 (tk+1 tk )
existing between (Wt )2 and t. Taking (Wt )2 equal to t brings us back
to the Euler scheme.
When Xt is the geometric Brownian motion given by
dXt = rXt dt + Xt dWt
we get
tN (Wt Wt )2 ,
tN = X
tN +(r 2 /2)X
tN (tk+1 tk )+ X
tN (Wt Wt )+ 1 2 X
X
k+1
k
k+1
k
k
k+1
k
k
k
2
which can be computed as
tN
tN = X
X
0
k
k
Y
i=1
"

1
1 + (r 2 /2)(ti ti1 ) + (Wti Wti1 ) + 2 (Wti Wti1 )2 .
2
495
Appendix: Background on Probability

Theory
In this appendix we review a number of basic probabilistic tools that are

needed in option pricing and hedging. We refer to [56], [21], [87] for more on
the needed probability background.
Probability Spaces and Events

We will need the following notation coming from set theory. Given A and B
to abstract sets, A B means that A is contained in B, and the property
that belongs to the set A is denoted by A. The finite set made of n
elements 1 , . . . , n is denoted by {1 , . . . , n }, and we will usually make a
distinction between the element and its associated singleton set {}.
A probability space is an abstract set that contains the possible outcomes of a random experiment.
Examples:
i) Coin tossing: = {H, T }.
ii) Rolling one die: = {1, 2, 3, 4, 5, 6}.
iii) Picking on card at random in a pack of 52: = {1, 2, 3, . . . , 52}.
iv) An integer-valued random outcome: = N.
In this case the outcome N can be the random number of trials
needed until some event occurs.
"
N. Privault
v) A non-negative, real-valued outcome: = R+ .
In this case the outcome R+ may represent the (non-negative) value
of a continuous random time.
vi) A random continuous parameter (such as time, weather, price or wealth,
temperature, ...): = R.
vii) Random choice of a continuous path in the space = C(R+ ) of all continuous functions on R+ .
In this case, is a function : R+ R and a typical example is
the graph t 7 (t) of a stock price over time.
Product spaces:
Probability spaces can be built as product spaces and used for the modeling
of repeated random experiments.
i) Rolling two dice: = {1, 2, 3, 4, 5, 6} {1, 2, 3, 4, 5, 6}.
In this case a typical element of is written as = (k, l) with
k, l {1, 2, 3, 4, 5, 6}.
ii) A finite number n of real-valued samples: = Rn .
In this case the outcome is a vector = (x1 , . . . , xn ) Rn with n
components.
Note that to some extent, the more complex is, the better it fits a practical
and useful situation, e.g. = {H, T } corresponds to a simple coin tossing
experiment while = C(R+ ) the space of continuous functions on R+ can be
applied to the modeling of stock markets. On the other hand, in many cases
and especially in the most complex situations, we will not attempt to specify
explicitly.
Events
An event is a collection of outcomes, which is represented by a subset of .
The collections G of events that we will consider are called -algebras, and
assumed to satisfy the following conditions.
(i) G,
498
"
Background on Probability Theory

(ii) For all countable sequences An G, n 1, we have
An G,
n1
(iii) A G = ( \ A) G,
where \ A := { :
/ A}.
The collection of all events in will often be denoted by F. The empty set
and the full space are considered as events but they are of less importance
because corresponds to any outcome may occur while corresponds to
an absence of outcome, or no experiment.
In the context of stochastic processes, two -algebras G and F such that
G F will refer to two different amounts of information, the amount of information associated to G being here lower than the one associated to F.
The formalism of -algebras helps in describing events in a short and precise
way.
Examples:
i) = {1, 2, 3, 4, 5, 6}.
The event A = {2, 4, 6} corresponds to
the result of the experiment is an even number.
ii) Taking again = {1, 2, 3, 4, 5, 6},
F := {, , {2, 4, 6}, {1, 3, 5}}
defines a -algebra on which corresponds to the knowledge of parity
of an integer picked at random from 1 to 6.
Note that in the set-theoretic notation, an event A is a subset of , i.e.
A , while it is an element of F, i.e. A F. For example, we have
{2, 4, 6} F, while {{2, 4, 6}, {1, 3, 5}} F.
Taking
G := {, , {2, 4, 6}, {2, 4}, {6}, {1, 2, 3, 4, 5}, {1, 3, 5, 6}, {1, 3, 5}} F,
defines a -algebra on which is bigger than F and corresponds to the
knowledge whether the outcome is equal to 6 or not, in addition to the
parity information contained in F.
iii) Take
= {H, T } {H, T } = {(H, H), (H.T ), (T, H), (T, T )}.
"
499
N. Privault
In this case, the collection F of all possible events is given by
F = {, {(H, H)}, {(T, T )}, {(H, T )}, {(T, H)},
(16.4)
{(T, T ), (H, H)}, {(H, T ), (T, H)}, {(H, T ), (T, T )},

{(T, H), (T, T )}, {(H, T ), (H, H)}, {(T, H), (H, H)},
{(H, H), (T, T ), (T, H)}, {(H, H), (T, T ), (H, T )},
{(H, T ), (T, H), (H, H)}, {(H, T ), (T, H), (T, T )}, } .
Note that the set F of all events considered in (16.4) above has altogether

n
1=
event of cardinal 0,
0

n
4=
events of cardinal 1,
1

n
6=
2

n
4=
3

n
1=
event of cardinal 4,
4
with n = 4, for a total of
16 = 2n =
4
X
4
k=0
=1+4+6+4+1
events.
The collection of events
G := {, {(T, T ), (H, H)}, {(H, T ), (T, H)}, }
defines a sub -algebra of F, associated to the information the results of
two coin tossings are different.
Exercise: Write down the set of all events on = {H, T }.
Note also that (H, T ) is different from (T, H), whereas {(H, T ), (T, H)} is
equal to {(T, H), (H, T )}.
In addition we will usually make a distinction between the outcome
and its associated event {} F, which satisfies {} .
500
"
Probability Measures
A probability measure is a mapping P : F [0, 1] that assigns a probability
P(A) [0, 1] to any event A, with the properties
a) P() = 1, and
!
[
X
b) P
An =
P(An ), whenever Ak Al = , k 6= l.
n=1
n=1
A property or event is said to hold P-almost surely (also written P-a.s.) if it

holds with probability equal to one.
P(A1 An ) = P(A1 ) + + P(An )
when the subsets A1 , . . . , An of are disjoints, and
P(A B) = P(A) + P(B)
if A B = . In the general case we can write
P(A B) = P(A) + P(B) P(A B).
The triple
(, F, P)
(16.5)
was introduced by A.N. Kolmogorov (1903-1987), and is generally referred

to as the Kolmogorov framework.
In addition we have the following convergence properties.
1. Let (An )nN be a nondecreasing sequence of events, i.e. An An+1 ,
n N. Then we have
!
[
P
An = lim P(An ).
(16.6)
nN
2. Let (An )nN be a nonincreasing sequence of events, i.e. An+1 An ,

n N. Then we have
!
\
P
An = lim P(An ).
(16.7)
nN
"
501
N. Privault
Conditional Probabilities and Independence

Given any two events A, B with P(B) 6= 0, we call
P(A | B) :=
P(A B)
P(B)
the probability of A given B, or conditionally to B. Note that if P(B) = 1

we have P(AB c ) P(B c ) = 0 hence P(AB) = P(A) and P(A | B) = P(A).
We also recall the following property:
!
[
X
X
X
P B
An =
P(BAn ) =
P(B | An )P(An ) =
P(An | B)P(B),
n=1
n=1
n=1
n=1
for any family of events (An )n1 , B, provided Ai Aj = , i 6= j, and

P(An ) > 0, n 1. This also shows that conditional probability measures are
probability measures, in the sense that whenever P(B) > 0 we have
a) P( | B) = 1, and
!

X
[

b) P
An B =
P(An | B), whenever Ak Al = , k 6= l.
n=1
n=1
In particular if
An = , (An )n1 becomes a partition of and we get
n=1
the law of total probability

P(B) =
X
n=1
P(B An ) =
P(An | B)P(B) =
n=1
P(B | An )P(An ), (16.8)
n=1
provided Ai Aj = , i 6= j, and P(An ) > 0, n 1. However we have in

general
!
[
X

P A
Bn 6=
P(A | Bn ),
n=1
n=1
even when Bk Bl = , k 6= l. Indeed, taking for example A = = B1 B2

with B1 B2 = and P(B1 ) = P(B2 ) = 1/2, we have
1 = P( | B1 B2 ) 6= P( | B1 ) + P( | B2 ) = 2.
Finally, two events A and B are said to be independent if
P(A | B) = P(A),
502
"

i.e. if
P(A B) = P(A)P(B).
In this case we find
P(A | B) = P(A).
Random Variables
A real-valued random variable is a mapping
X : R
7 X()
from a probability space into the state space R.
Given X : R a random variable and A a (measurable) subset of R,
we denote by {X A} the event
{X A} = { : X() A}.
Given G a -algebra on G, the mapping X : R is said to be Gmeasurable if
{X x} = { : X() x} G,
for all x R. In this case we will also say that the knowledge of X depends
only on the information contained in G.
Examples:
i) Let = {1, 2, 3, 4, 5, 6} {1, 2, 3, 4, 5, 6}, and consider the mapping
X : R
(k, l) 7 k + l.
Then X is a random variable giving the sum of the two numbers appearing on each die.
ii) the time needed everyday to travel from home to work or school is a
random variable, as the precise value of this time may change from day
to day under unexpected circumstances.
iii) the price of a risky asset is a random variable.
Measurability of subsets of R refers to Borel measurability, a concept which will

not be defined in this text.
"
503
N. Privault
In the sequel we will often use the notion of indicator function 1A of an event
A. The indicator function 1A is the random variable
1A : {0, 1}
7 1A ()
defined by

1A () =
1 if A,
0 if
/ A,
with the property

1AB () = 1A ()1B (),
(16.9)
since
A B { A and B}
{1A () = 1 and 1B () = 1}
1A ()1B () = 1.
We also have
1AB = 1A + 1B 1AB = 1A + 1B 1A 1B ,
and
1AB = 1A + 1B ,
(16.10)
if A B = . In addition, any Bernoulli random variable X : {0, 1}

can be written as an indicator function
X = 1A
on with A = {X = 1} = { : X() = 1}. For example if = N and
A = {k}, for all l N we have

1 if k = l,
1{k} (l) =
0 if k 6= l.
If X is a random variable we also let

1 if X = n,
1{X=n} =
0 if X 6= n,
and

1{X<n} =
1 if X < n,
0 if X n.
504
"
Probability Distributions
The probability distribution of a random variable X : R is the collection
{P(X A) : A measurable subset of R}.
In fact the distributions of X can be reduced to the knowledge of either
{P(a < X b) = P(X b) P(X a) : a < b R},
or
{P(X a) : a R},
or
{P(X a) : a R}.
Two random variables X and Y are said to be independent under the
probability P if their probability distributions satisfy
P(X A , Y B) = P(X A)P(Y B)
for all (measurable) subsets A and B of R.
Distributions Admitting a Density

In this case the distribution of X is given by
P(a X b) =
wb
a
f (x)dx
where the function f : R R+ is called the density of the distribution of

X. We also say that the distribution of X is absolutely continuous, or that X
is an absolutely continuous random variable. This, however, does not imply
that the density function f : R R+ is continuous.
In particular we always have
w
f (x)dx = P( X ) = 1
for all probability density functions f : R R+ .

The density fX can be recovered from the distribution functions
wx
x 7 P(X x) =
fX (s)ds,
x R,
and
"
505
N. Privault
x 7 P(X x) =
as
fX (x) =
w
x
x R,
fX (s)ds,
wx
w
fX (s)ds =
fX (s)ds,
x
x x
x R.
Examples:
i) The uniform distribution on an interval.
The density of the uniform distribution on the interval [a, b], a < b, is
given by
1
f (x) =
1[a,b] (x),
x R.
ba
ii) The Gaussian distribution.
The density of the standard normal distribution is given by
2
1
f (x) = ex /2 ,
2
x R.
More generally, X has a Gaussian distribution with mean R and

variance 2 > 0 (in this case we write X ' N (, 2 )) if
f (x) =
2 2
e(x)
/(2 2 )
x R.
iii) The exponential distribution with parameter > 0.

In this case we have
f (x) = 1[0,) (x)e
x
,
e
0,
x0
(16.11)
x < 0.
We also have
P(X > t) = et ,
t R+ .
(16.12)
In addition, if X1 , . . . , Xn are independent exponentially distributed random variables with parameters 1 , . . . , n we have
P(min(X1 , . . . , Xn ) > t) = P(X1 > t, . . . , Xn > t)
= P(X1 > t) P(Xn > t)
= et(1 ++n ) ,
t R+ , (16.13)
hence min(X1 , . . . , Xn ) is an exponentially distributed random variable

with parameter 1 + + n .
We also have
506
"

P(X1 < X2 ) = P(X1 X2 ) = 1 2
wwy
0
e1 x2 y dxdy =
1
,
1 + 2
(16.14)
and we note that

P(X1 = X2 ) = 1 2
{(x,y)R2+ : x=y}
e1 x2 y dxdy = 0.
iv) The gamma distribution.


a
x1 eax ,
()
1 ax
1[0,) (x)x
e
=
f (x) =
()
0,
where a > 0 and > 0 are parameters and
w
() =
x1 ex dx,
x0
x < 0,
> 0,
is the Gamma function.

v) The Cauchy distribution.
f (x) =
1
,
(1 + x2 )
x R.
vi) The lognormal distribution.

In this case,
f (x) = 1[0,) (x)
x 2
e(log x)
/(2 )
2
2
1
e(log x) /(2 ) , x 0
x 2
0,
x < 0.
Exercise: For each of the above probability density functions, check that the
condition
w
f (x)dx = 1
is satisfied.
Remark 16.1. Note that if the distribution of X admits a density then for
all a R, we have
"
507
N. Privault
P(X = a) =
wa
a
f (x)dx = 0,
(16.15)
and this is not a contradiction.

In particular, Remark 16.1 shows that
P(a X b) = P(X = a) + P(a < X b) = P(a < X b) = P(a < X < b),
for a b.
In practice, Property (16.15) appears for example in the framework of
lottery games with a large number of participants, in which a given number
a selected in advance has a very low (almost zero) probability to be chosen.
Given two absolutely continuous random variables X : R and Y :
R we can form the R2 -valued random variable (X, Y ) defined by
(X, Y ) : R2
7 (X(), Y ()).
We say that (X, Y ) admits a joint probability density
f(X,Y ) : R2 R+
when
P((X, Y ) A B) =
w w
A
f(X,Y ) (x, y)dxdy
for all measurable subsets A, B of R. The density f(X,Y ) can be recovered

from the distribution functions
wx wy
(x, y) 7 P(X x, Y y) =
f(X,Y ) (s, t)dsdt,
and
(x, y) 7 P(X x, Y y) =
ww
x
f(X,Y ) (s, t)dsdt,
as
2 w x w y
f(X,Y ) (s, t)dsdt
xy
2 ww
=
f(X,Y ) (s, t)dsdt,
xy x y
f(X,Y ) (x, y) =
(16.16)
x, y R.
The probability densities fX : R R+ and fY : R R+ of X :
R and Y : R are called the marginal densities of (X, Y ) and are given
by
508
"

fX (x) =
f(X,Y ) (x, y)dy,
x R,
fY (y) =
f(X,Y ) (x, y)dx,
y R.
and
(16.17)
The conditional density fX|Y =y : R R+ of X given Y = y is defined by

fX|Y =y (x) :=
f(X,Y ) (x, y)
,
fY (y)
x, y R,
(16.18)
provided fY (y) > 0.
Discrete Distributions
We only consider integer-valued random variables, i.e. the distribution of X
is given by the values of P(X = k), k N.
Examples:
i) The Bernoulli distribution.
We have
P(X = 1) = p
and
P(X = 0) = 1 p,
(16.19)
where p [0, 1] is a parameter.

ii) The binomial distribution.
We have
P(X = k) =

n k
p (1 p)nk ,
k
k = 0, 1, . . . , n,
where n 1 and p [0, 1] are parameters.

iii) The geometric distribution.
We have
P(X = k) = (1 p)pk ,
k N,
(16.20)
where p (0, 1) is a parameter.

Note that if (Xk )kN is a sequence of independent Bernoulli random
variables with distribution (16.19), then the random variable
"
509
N. Privault
X := inf{k N : Xk = 0}
has the geometric distribution (16.20).
iv) The negative binomial distribution (or Pascal distribution).
We have
P(X = k) =

k+r1
(1 p)r pk ,
r1
k N,
where p (0, 1) and r 1 are parameters. Note that the negative binomial distribution recovers the geometric distribution when r = 1.
v) The Poisson distribution.
We have
P(X = k) =
k
e ,
k!
k N,
where > 0 is a parameter.

Remark 16.2. The distribution of a discrete random variable cannot admit
a density. If this were the case, by Remark 16.1 we would have P(X = k) = 0
for all k N and
1 = P(X R) = P(X N) =
P(X = k) = 0,
k=0
which is a contradiction.
Given two discrete random variables X and Y , the conditional distribution
of X given Y = k is given by
P(X = n | Y = k) =
P(X = n and Y = k)
,
P(Y = k)
n N,
provided P(Y = k) > 0, k N.
Expectation of a Random Variable

The expectation of a random variable X is the mean, or average value, of
X. In practice, expectations can be even more useful than probabilities. For
example, knowing that a given equipment (such as a bridge) has a failure
probability of 1.78493 out of a billion can be of less practical use than know510
"

ing the expected lifetime (e.g. 200000 years) of that equipment.
For example, the time T () to travel from home to work/school can be a
random variable with a new outcome and value every day, however we usually refer to its expectation IE[T ] rather than to its sample values that may
change from day to day.
In general, the expectation of the indicator function 1A is defined as
IE[1A ] = P(A),
for any event A. For a Bernoulli random variable X : {0, 1} with
parameter p [0, 1], written as X = 1A with A = {X = 1}, we have
p = P(X = 1) = P(A) = IE[1A ] = IE[X].
Discrete Distributions
Next, let X : N be a discrete random variable. The expectation IE[X]
of X is defined as the sum
IE[X] =
kP(X = k),
k=0
in which the possible values k N of X are weighted by their probabilities.

More generally we have
IE[(X)] =
(k)P(X = k),
k=0
for all sufficiently summable functions : N R.

Given a non-negative random variable X, the finiteness of IE[X] <
implies P(X < ) < 1, however the converse is not true. For example the
expectation IE[(X)] may be infinite even when (X) is always finite, take
for example
(X) = 2X
and
P(X = k) = 1/2k ,
k 1.
(16.21)
The expectation of the indicator function X = 1A can be recovered as

IE[1A ] = 0 P( \ A) + 1 P(A) = P(A).
Note that the expectation is a linear operation, i.e. we have
"
511
N. Privault
IE[aX + bY ] = a IE[X] + b IE[Y ],
a, b R,
(16.22)
provided
IE[|X|] + IE[|Y |] < .
The notion of expectation takes its full meaning under conditioning. For example, the expected return of a random asset usually depends on information
such as economic data, location, etc. In this case, replacing the expectation by
a conditional expectation will provide a better estimate of the expected value.
For instance, life expectancy is a natural example of a conditional expectation since it typically depends on location, gender, and other parameters.
The conditional expectation of X : N given an event A is defined
by
IE[X | A] =
kP(X = k | A).
k=0
Lemma 16.1. Given an event A such that P(A) > 0, we have

IE[X | A] =
1
IE [X1A ] .
P(A)
(16.23)
Proof. By Relation (16.9) we have

1 X
1 X
kP({X = k} A) =
k IE 1{X=k}A
P(A)
P(A)
k=0
k=0
"
#
X

1 X
1
=
k IE 1{X=k} 1A =
IE 1A
k1{X=k}
P(A)
P(A)
IE[X | A] =
k=0
k=0
1
IE [1A X] ,
=
P(A)
(16.24)
where we used the relation

X=
k1{X=k}
k=0
which holds since X takes only integer values.
If X is independent of A (i.e. P({X = k} A) = P({X = k})P(A), k N)

we have IE[X1A ] = IE[X]P(A) and we naturally find
IE[X | A] = IE[X].
If X = 1B we also have in particular
IE[1B | A] = 0 P(X = 0 | A) + 1 P(X = 1 | A)
512
"

= P(X = 1 | A)
= P(B | A).
One can also define the conditional expectation of X given that {Y = k}, as
IE[X | Y = k] =
nP(X = n | Y = k),
n=0
where Y : N is a discrete random variable. In general we have

IE[IE[X | Y ]] =
IE[X | Y = k]P(Y = k)
k=0
=
=
=
nP(X = n | Y = k)P(Y = k)
k=0 n=0
X
X
n=0
P(X = n and Y = k)
k=0
nP(X = n) = IE[X],
n=0
where we used the marginal distribution

P(X = n) =
P(X = n and Y = k),
n N,
k=0
that follows from the law of total probability (16.8) by taking Ak = {Y = k}.
Hence we have the relation
IE[X] = IE[IE[X | Y ]],
(16.25)
which is sometimes referred to as the tower property. Taking

Y =
k1Ak ,
k=0
i.e. Ak = {Y = k}, k N, we also get the law of total expectation

IE[X] = IE[IE[X | Y ]] =
IE[X | Ak ]P(Ak ),
(16.26)
k=0
whenever (Ak )kN is a partition of .
"
513
N. Privault
Random sums
Based on the tower property or ordinary conditioning, the expectation of a
Y
X
random sum
Xk , where (Xk )kN is a sequence of random variables, can
k=1
be computed from the tower property (16.25) as

" Y
#
" " Y
##
X
X
Xk = IE IE
IE
Xk Y
k=1
k=1
X
n=0
"
IE
Y
X
k=1
n
X
"
IE
n=0
#

Xk Y = n P(Y = n)
#

Xk Y = n P(Y = n),
k=1
and if Y is independent of (Xk )kN this yields

#
" n
#
" Y
X
X
X
IE
Xk P(Y = n).
Xk =
IE
k=1
n=0
k=1
Similarly, for a random product we will have

" Y
#
" n
#
Y
X
Y
IE
Xk P(Y = n).
IE
Xk =
k=1
n=0
(16.27)
k=1
Example:
The life expectancy in Singapore is IE[T ] = 80 years overall, where T
denotes the lifetime of a given individual chosen at random. Let G
{m, w} denote the gender of that individual. The statistics show that
IE[T | G = w] = 78
and
IE[T | G = m] = 81.9,
and we have
80 = IE[T ]
= IE[IE[T |G]]
= P(G = w) IE[T | G = w] + P(G = m) IE[T | G = m]
= 81.9 P(G = w) + 78 P(G = m)
= 81.9 (1 P(G = m)) + 78 P(G = m),
showing that
514
"

80 = 81.9 (1 P(G = m)) + 78 P(G = m),
i.e.
P(G = m) =
81.9 80
1.9
=
= 0.487.
81.9 78
3.9
Distributions Admitting a Density

Given a random variable X whose distribution admits a density fX : R
R+ we have
w
IE[X] =
xfX (x)dx,
and more generally,

IE[(X)] =
(x)fX (x)dx,
for all sufficiently integrable function on R. For example, if X has a standard

normal distribution we have
w
2
dx
IE[(X)] =
(x)ex /2 .
2
In case X has a Gaussian distribution with mean R and variance 2 > 0
we get
w
2
2
1
IE[(X)] =
(x)e(x) /(2 ) dx.
2
2
In case (X, Y ) : R2 is a R2 -valued couple of random variables whose
distribution admits a density fX,Y : R R+ we have
w w
(x, y)fX,Y (x, y)dxdy,
IE[(X, Y )] =
for all sufficiently integrable function on R2 .

The expectation of an absolutely continuous random variable satisfies the
same linearity property (16.22) as in the discrete case.
The variance of the random variable X is defined by
Var[X] := IE[X 2 ] (IE[X])2 ,
provided IE[|X|2 ] < . If (Xk )kN is a sequence of independent random
variables we have
"
515
N. Privault
"
Var
n
X
#
Xk
n
X
= IE
k=1
"
= IE
"
n
X
Xk
n
X
=
=
n
X
Xk
IE
k=1
l=1
n X
n
X
n
X
Xk IE
" n
X
k=1
n
X
#!2
Xk
k=1
"
Xl IE
k=1 l=1
"
k=1
Xk Xl
= IE
!2
n X
n
X
#
Xl
l=1
IE[Xk ] IE[Xl ]
k=1 l=1
IE[Xk2 ] +
k=1
n
X
IE[Xk Xl ]
1k6=ln
n
X
(IE[Xk ])2
k=1
IE[Xk ] IE[Xl ]
1k6=ln
(IE[Xk2 ] (IE[Xk ])2 )
k=1
n
X
Var [Xk ].
(16.28)
k=1
Exercise: In case X has a Gaussian distribution with mean R and variance

2 > 0, check that
= IE[X]
2 = IE[X 2 ] (IE[X])2 .
and
The conditional expectation of an absolutely continuous random variable can

be defined as
w
IE[X | Y = y] =
xfX|Y =y (x)dx
where the conditional density fX|Y =y (x) is defined in (16.18), with the relation
IE[X] = IE[IE[X | Y ]]
(16.29)
as in the discrete case, since
w
w w
IE[IE[X | Y ]] =
IE[X | Y = y]fY (y)dy =
xfX|Y =y (x)fY (y)dxdy

w
w w
xfX (x)dx = IE[X],
x
f(X,Y ) (x, y)dydx =
=
where we used Relation (16.17) between the density of (X, Y ) and its marginal
X.
For example, an exponentially distributed random variable X with probability density function (16.11) has the expected value
IE[X] =
w
0
xex dx =
516
1
.
"

If X and Y are independent exponentially distributed random variables with
parameters and , using (16.23) and (16.14) we can also compute the
conditional expectation

1
IE X1{X<Y }
IE[min(X, Y ) | X < Y ] = IE[X | X < Y ] =
P(X < Y )
w
wy
= ( + )
ey
xex dxdy
0
0
w
y
e
(1 (1 + y)ey )dy
= ( + ) 2
0
w

w
w
= ( + ) 2
ey dy
ey ey dy
ey yey )dy
0
0
0

1
1
= ( + ) 2
+ ( + )

( + )
( + )2
= ( + ) 2
( + )2
( + )2
( + )2

( + )2 ( + )
= ( + ) 2
( + )2
1
=
= IE[min(X, Y )].
(16.30)
+
Conditional Expectation Revisited

The construction of conditional expectation given above for discrete and absolutely continuous random variables can be generalized to -algebras.
For any p 1 we let
Lp () = {F : R : IE[|F |p ] < }
(16.31)
denote the space of p-integrable random variables F : R. Given G F

a sub -algebra of F and F L2 (, F, P), the conditional expectation of F
given G, and denoted
IE[F | G],
can be defined to be the orthogonal projection of F onto L2 (, G, P).
For this we define a scalar product h, iL2 () between elements of L2 (, G, P),
as
hF, GiL2 () := IE[F G],
F, G L2 (, G, P).
This scalar product is associated to the norm k kL2 () by the relation

"
517
N. Privault
kF k = hF, F iL2 () ,
F L2 (, G, P),
and it induces a notion of orthogonality, namely F is orthogonal to G in

L2 (, G, P) if and only if hF, GiL2 () = 0.
Namely, IE[F | G] is characterized by the relation
hG, F IE[F | G]i = 0,
which rewrites as
IE[G(F IE[F | G])] = 0,
i.e.
IE[GF ] = IE[G IE[F | G]],
(16.32)
for all bounded and G-measurable random variables G, where h, i is the

scalar product in L2 (, F, P).
In other words, anytime the relation
IE[GF ] = IE[GX]
holds for all bounded and G-measurable random variables G, and a given
G-measurable random variable X, we can claim that
X = IE[F | G]
by uniqueness of the orthogonal projection onto the subspace L2 (, G, P) of
L2 (, F, P).
The conditional expectation operator has the following properties.
i) IE[F G | G] = G IE[F | G] if G depends only on the information contained
in G.
Proof: By the characterization (16.32) it suffices to show that
IE[HF G] = IE[HG IE[F |G]],
(16.33)
for all bounded and G-measurable random variables G, H, which implies

IE[F G | G] = G IE[F | G].
Relation (16.33) holds from (16.32) because the product HG is Gmeasurable hence G can be replaced with HG in (16.32).
ii) IE[G|G] = G when G depends only on the information contained in G.
518
"

Proof: This is a consequence of point (i) above by taking F = 1.
iii) IE[IE[F |G] | H] = IE[F |H] if H G, called the tower property.
Proof: First we note that (iii) holds when H = {, } because taking
G = 1 in (16.32) yields
IE[F ] = IE[IE[F | G]].
(16.34)
Next, by the characterization (16.32) it suffices to show that

IE[H IE[F |G]] = IE[H IE[F |H]],
(16.35)
for all bounded and G-measurable random variables H, which will imply
(iii) from (16.32).
In order to prove (16.35) we check that by (16.34) and point (i) above
we have
IE[H IE[F |G]] = IE[IE[HF |G]] = IE[HF ]
= IE[IE[HF |H]] = IE[H IE[F |H]],
and we conclude by the characterization (16.32).
iv) IE[F |G] = IE[F ] when F does not depend on the information contained
in G or, more precisely stated, when the random variable F is independent of the -algebra G.
Proof: It suffices to note that for all bounded G-measurable G we have
IE[F G] = IE[F ] IE[G] = IE[G IE[F ]],
and we conclude again by (16.32).
v) If G depends only on G and F is independent of G, then
IE[h(F, G)|G] = IE[h(x, F )]x=G .
This relation can be proved using the tower property and the characterization (16.32), by noting that for any K L2 (, G, P) we have
IE[K IE[h(x, F )]x=G ] = IE[K IE[h(x, F ) | G]x=G ]
= IE[K IE[h(G, F ) | G]]
= IE[IE[Kh(G, F ) | G]]
= IE[h(G, F )].
"
519
N. Privault
The notion of conditional expectation can be extended from square-integrable
random variables in L2 () to integrable random variables in L1 (), cf. e.g.
[64], Theorem 5.1.
Moment Generating Functions

The characteristic function of a random variable X is the function X :
R C defined by
X (t) = IE[eitX ],
t R.
The Laplace transform (or moment generating function) of a random variable X is the function X : R R defined by
X (t) = IE[etX ],
t R,
provided the expectation is finite.

IE[X n ] =
n
X (0),
t
n 1,
provided IE[|X|n ] < . The Laplace transform X of a random variable X

with density f : R R+ satisfies
w
X (t) =
etx f (x)dx,
t R.
Note that in probability we are using the bilateral Laplace transform for which
the integral is from to +.
The characteristic function X of a random variable X with density f :
R R+ satisfies
w
X (t) =
eitx f (x)dx,
t R.
On the other hand, if X : N is a discrete random variable we have

X (t) =
eitn P(X = n),
t R.
n=0
The main applications of characteristic functions lie in the following theorems:

Theorem 16.1. Two random variables X : R and Y : R have
same distribution if and only if
520
"

t R.
X (t) = Y (t),
Theorem 16.1 is used to identify or to determine the probability distribution

of a random variable X, by comparison with the characteristic function Y
of a random variable Y whose distribution is known.
The characteristic function of a random vector (X, Y ) is the function
X,Y : R2 C defined by
X,Y (s, t) = IE[eisX+itY ],
s, t R.
Theorem 16.2. Given two independent random variables X : R and

Y : R are independent if and only if
s, t R.
X,Y (s, t) = X (s)Y (t),
A random variable X is Gaussian with mean and variance 2 if and only

if its characteristic function satisfies
2
IE[eiX ] = ei
2 /2
R.
(16.36)
In terms of Laplace transforms we have, replacing i by ,

2
IE[eX ] = e+
2 /2
R.
(16.37)
From Theorems 16.1 and 16.2 we deduce the following proposition.

2
Proposition 16.1. Let X ' N (, X
) and Y ' N (, Y2 ) be independent
Gaussian random variables. Then X + Y also has a Gaussian distribution
2
X + Y ' N ( + , X
+ Y2 ).
Proof. Since X and Y are independent, by Theorem 16.2 the characteristic

function X+Y of X + Y is given by
X+Y (t) = X (t)Y (t)
= eitt
=e
2
2
X
/2 itt2 Y
/2
2
2
it(+)t2 (X
+Y
)/2
t R,
where we used (16.36). Consequently, the characteristic function of X + Y is

2
that of a Gaussian random variable with mean + and variance X
+ Y2
and we conclude by Theorem 16.1.
"
521
N. Privault
Exercises
Exercise 1 Compute the expected value IE[X] of a Poisson random variable
X with parameter > 0.
Exercise 2 Let X denote a centered Gaussian random variable with variance
2 , > 0. Show that the probability P (eX > c) is given by
P (eX > c) = ((log c)/),
where log = ln denotes the natural logarithm and
1 w x y2 /2
e
dy,
(x) =
2
x R,
denotes the Gaussian cumulative distribution function.

Exercise 3 Let X ' N (, 2 ) be a Gaussian random variable with parameters
> 0 and 2 > 0, and probability density function
f (x) =
1
2 2
e(x)
/(2 2 )
x R.
a) Write down IE[X] as an integral and show that

= IE[X].
b) Write down IE[X 2 ] as an integral and show that
2 = IE[(X IE[X])2 ].
c) Consider the function x 7 (x K)+ from R to R+ , defined as
x K if x K,
+
(x K) =
0
if x K,
where K R be a fixed real number. Write down IE[(X K)+ ] as an
integral and compute this integral.
Hints: (x K)+ is zero when x < K, and when = 0 and = 1 the
result is
2
1
IE[(X K)+ ] = eK /2 K(K),
2
where
(x) :=
wx
ey
/2
dy
,
2
522
x R.
"

d) Write down IE[eX ] as an integral, and compute IE[eX ].
Exercise 4 Let X be a centered Gaussian random variable with variance
2
2
1
ex /(2 ) and let R.
2 > 0 and probability density function x 7 2
2
a) Write down IE[( X)+ ] as an integral. Hint: ( x)+ is zero when x > .
b) Compute this integral to show that
2
2
IE[( X)+ ] = e /(2 ) + (/),

2
where
(x) =
wx
ey
/2
dy
,
2
x R.
Exercise 5 Let X be a centered Gaussian random variable with variance

2
2
1
ex /(2 ) and let R.
2 > 0 and probability density function x 7 2
2
a) Write down IE[( + X)+ ] as an integral. Hint: ( + x)+ is zero when
x < .
b) Compute this integral to show that

2
2
IE[( + X)+ ] = e /(2 ) +

,
2
where
(x) =
wx
ey
/2
dy
,
2
x R.
Exercise 6 Let X be a centered Gaussian random variable with variance v 2 .

a) Compute

2
2
1 w
IE eX 1[K,[ (xeX ) =
eyy /(2v ) dy.
2v 2 1 log(K/x)
Hint: use the decomposition
y
y2
v 2 2 y v 2
=
.
2
v
4
v
2
b) Compute
IE[(em+X K)+ ] =
1
2v 2
(em+x K)+ ex
/(2v 2 )
dx.
c) Compute the expectation (16.37) above.
"
523
Exercise Solutions
Chapter 1
Exercise 1.1
a) The possible values of R are a and b.
b) We have
IE [R] = aP (R = a) + bP (R = b)
ra
br
+b
=a
ba
ba
= r.
c) By Theorem 1.1, there do not exist arbitrage opportunities in this market
since there exists a risk-neutral measure P from Question b.
d) The risk-neutral measure is unique hence the market model is complete
by Theorem 1.3.
e) Taking
(1 + b) (1 + a)
and =
,
=
1 (b a)
S0 (b a)
we check that
1 + S0 (1 + a) =
1 + S0 (1 + b) = ,
which shows that

1 + S1 = C.
f) We have
(C) = 0 + S0
(1 + b) (1 + a)
=
+
(1 + r)(b a)
ab
525
N. Privault
(1 + b) (1 + a) (1 + r)( )
(1 + r)(b a)
b a r( )
=
.
(1 + r)(b a)
(16.38)
g) We have
IE [C] = P (R = a) + P (R = b)
ra
br
+
.
=
ba
ba
(16.39)
h) Comparing (16.38) and (16.39) above we do obtain

(C) =
1
IE [C]
1+r
i) The initial value (C) of the portfolio is interpreted as the arbitrage price
of the option contract and it equals the expected value of the discounted
payoff.
j) We have
11 S1 if K > S1 ,
C = (K S1 )+ = (11 S1 )+ =
0 if K S1 .
k) We have = 2, = 0 and
=
(11 (1 + a))
2
= ,
ba
3
(1 + b)(11 (1 + a))
8
=
.
(1 + r)(b a)
1.05
l) The arbitrage price (C) of the contingent claim C is

(C) = 0 + S0 = 6.952.
Chapter 2
Exercise 2.1
a) The possible values of Rt are a and b.
b) We have
IE [Rt+1 | Ft ] = aP (Rt+1 = a | Ft ) + bP (Rt+1 = b | Ft )
br
ra
=a
+b
= r.
ba
ba
526
"

c) We have
IE [St+k | Ft ] =
i
ki
k
X
ra
br
k
(1 + b)i (1 + a)ki St
ba
ba
i
i
ki
k
X
br
k
ra
(1 + b)
(1 + a)
= St
ba
ba
i
i=0
k

ra
br
= St
(1 + b) +
(1 + a)
ba
ba
i=0
= (1 + r)k St .
Assuming that the formula holds for k = 1, its extension to k 2 can also
be proved recursively from the tower property (16.25) of conditional
expectations, as follows:
IE [St+k | Ft ] = IE [IE [St+k | Ft+k1 ] | Ft ]
= (1 + r) IE [St+k1 | Ft ]
= (1 + r) IE [IE [St+k1 | Ft+k2 ] | Ft ]
= (1 + r)2 IE [St+k2 | Ft ]
= (1 + r)2 IE [IE [St+k2 | Ft+k3 ] | Ft ]
= (1 + r)3 IE [St+k3 | Ft ]
=
= (1 + r)k2 IE [St+2 | Ft ]
= (1 + r)k2 IE [IE [St+2 | Ft+1 ] | Ft ]
= (1 + r)k1 IE [St+1 | Ft ]
= (1 + r)k St .
Chapter 3
Exercise 3.1
a) The condition VN = C reads
N N + N (1 + a)SN 1 = (1 + a)SN 1 K
N N + N (1 + b)SN 1 = (1 + b)SN 1 K
from which we deduce N = 1 and N = K(1 + r)N /0 .

b) We have
"
527
N. Privault
N 1 N 1 + N 1 (1 + a)SN 1 = N N 1 + N (1 + a)SN 1
N 1 N 1 + N 1 (1 + b)SN 1 = N N 1 + N (1 + b)SN 1 ,
which yields N 1 = N = 1 and N 1 = N = K(1 + r)N /0 . Similarly, solving the self-financing condition
t t + t (1 + a)St = t+1 t + t+1 (1 + a)St
t t + t (1 + b)St = t+1 t + t+1 (1 + b)St ,
at time t yields t = 1 and t = K(1 + r)N /0 , t = 1, 2, . . . , N .

c) We have
t (C) = Vt = t t + t St = St K(1 + r)N t /0 = St K(1 + r)(N t) .
d) For all t = 0, 1, . . . , N we have
(1 + r)(N t) IE [C | Ft ] = (1 + r)(N t) IE [SN K | Ft ],
= (1 + r)(N t) IE [SN | Ft ] (1 + r)(N t) IE [K | Ft ]
= (1 + r)(N t) (1 + r)N t St K(1 + r)(N t)
= St K(1 + r)(N t)
= Vt = t (C).
Exercise 3.2
a) This model admits a unique risk-neutral measure P because we have
a < r < b. We have
P (Rt = a) =
0.07 0.05
br
=
,
ba
0.07 (0.02)
P(Rt = b) =
ra
0.05 (0.02)
=
,
ba
0.07 (0.02)
and
t = 1, 2, . . . , N .
b) There are no arbitrage opportunities in this model, due to the existence
of a risk-neutral measure.
c) This market model is complete because the risk-neutral measure is unique.
d) We have
C = (SN )2 ,
hence
H = (SN )2 /(1 + r)N = h(XN ),
with
h(x) = x2 (1 + r)N .
528
"

Now we have
Vt = vt (Xt ),
where the function vt (x) is given by
N
t
X
(N t)!
k!(N t k)!
k=0
k
N tk
k
N tk !

br
1+a
ra
1+b
h x
ba
ba
1+r
1+r
vt (x) =
N
t
X
(N t)!
k!(N t k)!

k
N tk
2k
2(N tk)
ra
br
1+b
1+a
ba
ba
1+r
1+r
= x2 (1 + r)N
k=0
N
t
X
(N t)!
k!(N t k)!
k
N tk

(b r)(1 + a)2
(r a)(1 + b)2
2
2
(b a)(1 + r)
(b a)(1 + r)

N t
(r
a)(1
+
b)2
(b r)(1 + a)2
= x2 (1 + r)N
+
2
2
(b a)(1 + r)
(b a)(1 + r)
N t
x2 (r a)(1 + b)2 + (b r)(1 + a)2
=
N
2t
N
t
(1 + r)
(b a)
N t
x2 (r a)(1 + 2b + b2 ) + (b r)(1 + 2a + a2 )
=
(1 + r)N 2t (b a)N t
= x2 (1 + r)N
k=0
N t
x2 r(1 + 2b + b2 ) a(1 + 2b + b2 ) + b(1 + 2a + a2 ) r(1 + 2a + a2 )
(1 + r)N 2t (b a)N t
N t
(1
+
r(2
+
a
+
b)
ab)
= x2
.
(1 + r)N 2t
e) We have
t1
"
vt
=
1+b
1+r Xt1
vt
1+a
1+r Xt1
Xt1 (b a)/(1 + r)

2
2
1+b
1+a
1+r
1+r
(1 + r(2 + a + b) ab)N t
= Xt1
(b a)/(1 + r)
(1 + r)N 2t
(1 + r(2 + a + b) ab)N t
,
t = 1, 2, . . . , N,
= St1 (2 + b + a)
(1 + r)N t
529
N. Privault
representing the quantity of the risky asset to be present in the portfolio
at time t. On the other hand we have
Vt t1 Xt
Xt0
Vt t1 Xt
=
0
t0 =
Xt Xt1 (2 + b + a)/(1 + r)
0 (1 + r)N 2t
St St1 (2 + b + a)
= St (1 + r(2 + a + b) ab)N t
0 (1 + r)N
(1 + a)(1 + b)
,
= (St1 )2 (1 + r(2 + a + b) ab)N t
0 (1 + r)N
= Xt (1 + r(2 + a + b) ab)N t
t = 1, 2, . . . , N .
f) Let us check that the portfolio is self-financing. We have
0
1
t+1 St = t+1
St0 + t+1
St1
(1 + a)(1 + b) 0
S
0 (1 + r)N t
(1 + r(2 + a + b) ab)N t1
+(St )2 (2 + b + a)
(1 + r)N t1
N t1
(1
+
r(2
+
a
+
b)
ab)
= (St )2
(1 + r)N t
((2 + b + a)(1 + r) (1 + a)(1 + b))
1
= (Xt )2 (1 + r(2 + a + b) ab)N t
(1 + r)N 3t
= (1 + r)t Vt
= t St ,
t = 1, 2, . . . , N.
= (St )2 (1 + r(2 + a + b) ab)N t1
Exercise 3.3
a) We have
Vt = t St + t t
= t (1 + Rt )St1 + t (1 + r)t1 .
b) We have
IE [Rt |Ft1 ] = aP (Rt = a | Ft1 ) + bP (Rt = b | Ft1 )
ra
br
+b
=a
ba
ba
r
r
=b
a
ba
ba
530
"

= r.
c) By the result of Question a we have
IE [Vt | Ft1 ] = IE [t (1 + Rt )St1 | Ft1 ] + IE [t (1 + r)t1 | Ft1 ]
= t St1 IE [1 + Rt | Ft1 ] + (1 + r) IE [t t1 | Ft1 ]
= (1 + r)t St1 + (1 + r)t t1
= (1 + r)t St + (1 + r)t t
= (1 + r)Vt1 ,
where we used the self-financing condition.
d) We have
1
IE [Vt | Ft1 ]
1+r
3
8
=
P (Rt = a | Ft1 ) +
P (Rt = b | Ft1 )
1+r
1+r

1
0.25 0.15
0.15 0.05
=
3
+8
1 + 0.15
0.25 0.05
0.25 0.05

1
3 8
=
+
1.15 2 2
= 4.78.
Vt1 =
Exercise 3.4
a) Using the formulas
1
1
IE [V2 | F1 ] =
IE [V2 | S1 ]
1+r
1+r
1
p
=
P (S2 = S0 (1 + b)2 | S1 ) =
1{S1 =S0 (1+b)} ,
1+r
1+r
V1 =
and
1
IE [V1 | F0 ]
1+r

1
p
=
P (S1 = S0 (1 + b)) + 0 P (S1 = S0 (1 + a))
1+r 1+r
p
=
,
(1 + r)2
V0 =
we find the table
"
531
N. Privault
S2
S1 = 3, V1 = 1/4 V2
S0 = 1
S2
V0 = 1/16
V2
S1 = 1, V1 = 0 S2
V2
=9
=1
=3
=0
=1
=0
b) The equation 2 S2 + 2 2 = V2 reads
2 S0 (1 + b)2 + 2 (1 + r)2 = 1
2 S0 (1 + b)(1 + a) + 2 (1 + r)2 = 0
when S1 = S0 (1 + b), which yields

2 =
1
S0 (b a)(1 + b)
and 2 =
1+a
.
(b a)(1 + r)2
On the other hand, the equation 1 S1 + 1 1 = V1 reads
,
1 S0 (1 + b) + 1 (1 + r) =
1+r
1 S0 (1 + a) + 1 (1 + r) = 0
which yields
1 =
p
S0 (b a)(1 + r)
and 1 =
p (1 + a)
.
(b a)(1 + r)2
This can be summarized in the following table:

S1 = 3, V1 = 1/4 S2
S0 = 1
V2
V0 = 1/16 2 = 1/6, 2 = 1/8 S2
1 = 1/8
S1 = 1, V1 = 0
V2
1 = 1/16
S2
2 = 0, 2 = 0
V2
=9
=1
=3
=0
=1
=0
When S1 = S0 (1 + a) at time t = 1 the option price is V1 = 0 and the

hedging strategy is to cut all positions: 2 = 2 = 0. On the other hand,
if S1 = S0 (1 + b) then there is a chance of being in the money at maturity
and we need to increase our position in the underlying from 1 = 1/8 to
2 = 1/6.
Note that the self-financing condition
1 S1 + 1 1 = 2 S1 + 2 1 ,
532
"

is always verified. For example when S1 = S0 (1 + a) we have
1
1
S1 1 = 0 S1 + 0 1 = 0,
8
16
while when S1 = S0 (1 + b) we find
1
1
1
1
1
S1 1 = S1 1 = .
8
16
6
8
4
Chapter 4
Exercise 4.1
a) We need to check whether the four properties of the definition of Brownian
motion are satisfied. Checking Conditions (i) to (iii) does not pose any
particular problem since the time changes t 7 c + t, t 7 t/c2 and t 7
ct2 are deterministic, continuous, and increasing. As for Condition (iv),
Bc+t Bc+s clearly has a centered Gaussian distribution with variance t,
and the same property holds for cBt/c2 since
Var (c(Bt/c2 Bs/c2 )) = c2 Var (Bt/c2 Bs/c2 ) = c2 (t s)/c2 = t s.
As a consequence, (a) and (b) are standard Brownian motions.
Concerning (c), we note that Bct2 is a centered Gaussian random variable
with variance ct2 - not t, hence (Bct2 )tR+ is not a standard Brownian
motion. w
T
b) We have
2dBt = 2(BT B0 ) = 2BT , which has a Gaussian law with
0
mean 0 and variance 4T . On the other hand,
wT
0
(21[0,T /2] (t)+1(T /2,T ] (t))dBt = 2(BT /2 B0 )+(BT BT /2 ) = BT +BT /2 ,
which has a Gaussian law with mean 0 and variance 4(T /2)+T /2 = 5T /2.
w 2
c) The stochastic integral
sin(t) dBt has a Gaussian distribution with
0
mean 0 and variance
w 2
w 2 1 cos(2t)
dt = .
sin2 (t)dt =
0
0
2
d) If 0 s t we have
IE[Bt Bs ] = IE[(Bt Bs )Bs ]+IE[Bs2 ] = IE[(Bt Bs )] IE[Bs ]+IE[Bs2 ] = 0+s = s,
"
533
N. Privault
and similarly we obtain IE[Bt Bs ] = t when 0 t s, hence in general we
have
IE[Bt Bs ] = min(s, t),
s, t R+ .
e) We have
d(f (t)Bt ) = f (t)dBt + Bt df (t) + df (t) dBt
= f (t)dBt + Bt f 0 (t)dt + f 0 (t)dt dBt
= f (t)dBt + Bt f 0 (t)dt,
and by integration on both sides we get
0 = f (T )BT f (0)B0
wT
=
d(f (t)Bt )
0
wT
wT
=
f (t)dBt +
Bt f 0 (t)dt,
0
hence the conclusion.

Exercise 4.2 Applying Itos formula to the function f (x) = x3 , we have
BT3 = f (BT )
wT
1 w T 00
f (Bt )dt
= f (B0 ) +
f 0 (Bt )dBt +
0
2 0
wT
wT
=3
Bt2 dBt + 3
Bt dt.
0
Next, by a restriction to [0, T ] of the integration by parts formula (4.9) we

find
wT
wT
wT
(T t)dBt , ,
Bt dt = T BT
tdBt =
0
hence
BT3 = 3
wT
=3
wT

wT
tdBt
Bt2 dBt + 3 T BT
0
(T t +
Bt2 )dBt ,
and we find t = 3(T t + Bt2 ), t [0, T ].

Exercise 4.3 Let f L2 ([0, T ]). We have
i
i
h rT
h rT
rt

E e 0 f (s)dBs Ft = e 0 f (s)dBs E e t f (s)dBs Ft
h rT
i
rt
= e 0 f (s)dBs E e t f (s)dBs
534
"
= exp
w
f (s)dBs +

1wT
|f (s)|2 ds ,
2 t
0 t T.
Exercise 4.4 We have

w

T

E exp
Bt dBt
= E exp (BT2 T )/2
0
h
i
2
= eT /2 E e(BT ) /2
eT /2 w x2 /2 x2 /(2T )
=
e
e
dx
2T
eT /2 w (1/T )x2 /2
=
e
dx
2T
eT /2
.
=
1 T
for all < 1/T .
Exercise 4.5
2
a) We have f (t) = f (0)ect (interest rate compounding) and St = S0 eBt t/2+rt ,

t R+ , (geometric Brownian motion).
b) Those quantities can be directly computed from the expression of St as a
function of the N (0, t) random variable Bt . Alternatively, taking expectations in the stochastic differential equations dSt = rSt dt + St dBt that
u(t) := IE[St ] satisfies the ordinary differential equation u0 (t) = ru(t) with
u(0) = S0 and solution u(t) = IE[St ] = S0 ert . On the other hand, taking
expectations on both sides of
dSt2 = 2St dSt + (dSt )2 = 2rSt2 dt + 2 St2 dt + 2St dBt
shows that v(t) := IE[St2 ] satisfies the ordinary differential equation v 0 (t) =
2
( 2 + 2r)v(t), with v(0) = S02 and solution v(t) = IE[St2 ] = S02 e( +2r)t ,
hence
Var[St ] = IE[St2 ] (IE[St ])2
= v(t) u2 (t)
= S02 e(
=
+2r)t
2
S02 e2rt (e t
S02 e2rt
1),
t R+ .
c) We have
"
535
N. Privault
d log St =
1
1
2
dSt
(dSt )2 = rdt + dBt
t,
St
2St2
2
t R+ .
d) We find
f
f
(St , Yt )dSt +
(St , Yt )dYt
x
y
1 2f
1 2f
2f
(St , Yt )(dSt )2 +
(St , Yt )(dYt )2 +
(St , Yt )dSt dYt
+
2 x2
2 y 2
xy
f
f
(St , Yt )(rSt dt + St dBt ) +
(St , Yt )(Yt dt + Yt dWt )
=
x
y
2 2 2
2 2 2
St f
Yt f
2f
+
(St , Yt )dt +
(St , Yt )dt + St Yt
(St , Yt )dt.
2
2
2 x
2 y
xy
df (St , Yt ) =
Exercise 4.6
a) Letting Yt = ebt Xt , we have
dYt = d(ebt Xt )
= bebt Xt dt + ebt dXt
= bebt Xt dt + ebt (bXt dt + ebt dBt )
= dBt ,
hence
Yt = Y0 +
wt
0
dYs = Y0 +
wt
0
dBs = Y0 + Bt ,
and
Xt = ebt Yt = ebt Y0 + ebt Bt = ebt X0 + ebt Bt .
b) Letting Ys = ebs Xs , we have
dYs = d(ebs Xs )
= bebs Xs ds + ebs dXs
= bebs Xs ds + ebs (bXs ds + eas dBs )
= e(ba)s dBs ,
hence we can solve for Yt by integrating on both sides as
wt
wt
dYs = Y0 + e(ba)s dBs ,
t R+ .
Yt = Y0 +
0
This yields the solution

Xt = ebt Yt = ebt Y0 + ebt
wt
0
e(ba)s dBs ,
t R+ .
Comments:
536
"

(i) This type of computation appears anywhere discounting by the factor
ebt is involved.
wt
(ii) The stochastic integral
e(ba)s dBs cannot be computed in closed
0
form. It is a centered Gaussian random variable with variance
wt
0
e2(ba)s ds =
e2(ba)t 1
2(b a)
if b 6= a, and variance t if a = b.
Exercise 4.7
a) Note that the stochastic integral
wT
0
1
dBs
T s
is not defined in L2 (), as by the Ito isometry we have

"

2 # w
wT 1
T
1
1
= +.
dBs
ds =
IE
=
2
0 T s
0 (T s)
T s 0
By (4.29) we have
T
Xt
XtT
dBt
dXtT
d
+
dt =
,
=
T t
T t (T t)2
T t
hence by integration using the initial condition X0 = 0 we have
wt 1
XtT
=
dBs ,
0 T s
T t
t [0, T ).
b) We have
IE[XtT ] = (T t) IE
w
t
0

1
dBs = 0,
T s
t [0, T ).
c) By the Ito isometry we have

w

t
1
dBs
Var[XtT ] = 2 (T t)2 Var
0 T s
"
2 #
wt 1
= 2 (T t)2 IE
dBs
0 T s
wt
1
ds
= 2 (T t)2
0 (T s)2
"
537
N. Privault

1
1
= 2 (T t)2
T t T

t
2
= 1
,
t [0, T ).
T
d) We have
lim kXtT kL2 () = lim Var[XtT ] = 0.
t0
t0
Exercise 4.8 Exponential Vasicek model.

wt
a) We have zt = eat z0 + ea(ts) dBs .
0
wt
b) We have yt = eat y0 + (1 eat ) + ea(ts) dBs .

a

0
2
c) We have dxt = xt +
a log xt dt + xt dBt .
2

wt
d) We have rt = exp eat log r0 + (1 eat ) + ea(ts) dBs , with

0
a
= + 2 /2.
e) We have

2
IE[rt ] = r0 exp eat log r0 + (1 eat ) +
(1 e2at ) .
a
4a

2
f) We have lim IE[rt ] = r0 exp
+
.
t
a 4a
Exercise 4.9 Cox-Ingersoll-Ross (CIR) model.
wt
wt
a) We have rt = r0 + ( rs )ds +
rs dBs .
0
0
b) Using the fact that the expectation of the stochastic integral with respect
to Brownian motion is zero, we get, taking expectations on both sides of
the above integral equation: u0 (t) = u(t).
c) Apply Itos formula to

wt
wt
rs dBs ,
rt2 = f r0 + ( rs )ds +
0
with f (x) = x2 , to obtain
d(rt )2 = rt ( 2 + 2 2rt )dt + 2rt rt dBt .
(16.40)
d) Taking again the expectation on both sides of (16.40) we get

IE[rt2 ] = IE[r02 ] +
wt
0
( 2 IE[rt ] + 2 IE[rt ] 2 IE[rt2 ])dt,
and after differentiation with respect to t this yields

538
"

vt0 = ( 2 + 2)u(t) 2v(t).
Exercise 4.10
a) We have
St = eXt
wt
1 w t 2 Xs
vs eXs ds +
us eXs dBs +
u e ds
0
0
2 0 s
wt
wt
2 wt
= eX0 + eXs dBs +

eXs ds +
eXs ds
0
0
2 0
wt
wt
2 w t
= S0 + Ss dBs +
Ss ds +
Ss ds.
0
0
2 0
= e X0 +
wt
b) Let r > 0. The process (St )tR+ satisfies the stochastic differential equation
dSt = rSt dt + St dBt
when r = + 2 /2.
c) Let the process (St )tR+ be defined by St = S0 eBt +t , t R+ . Using the
decomposition ST = St e(BT Bt )+ , we have
P(ST > K | St = x) = P(St e(BT Bt )+(T t) > K | St = x)
= P(xe(BT Bt )+(T t) > K)
= P(e(BT Bt ) > Ke(T t) /x)

log(Ke(T t) /x)
=

log(x/K) +
=
,

where = T t.
d) We have
2 = Var[X] = Var[(BT Bt )] = 2 Var[BT Bt ] = 2 (T t),
hence = T t.
Problem 4.11
a) We have
n
h
i X
(n)
IE QT =
IE[(BkT /n B(k1)T /n )2 ]
k=1
n
X
(kT /n (k 1)T /n)
k=1
"
539
N. Privault
n 1.
= T,
b) We have
!2
n
i
h
X
(n)
IE (QT )2 = IE
(BkT /n B(k1)T /n )2
k=1
= IE
n
X
(BkT /n B(k1)T /n ) (BlT /n B(l1)T /n )
k,l=1
n
X

IE (BkT /n B(k1)T /n )4
k=1
+2

IE (BkT /n B(k1)T /n )2 IE (BlT /n B(l1)T /n )2
1k<ln
=3
n
X
(kT /n (k 1)T /n)2
k=1
+2
(kT /n (k 1)T /n)(lT /n (l 1)T /n)
1k<ln
n(n 1)T 2
T2
+
n
n2
2T 2
2
=T +
,
n 1,
n
=3
hence
h
i h
i2
2T 2
(n)
(n)
(n)
Var[QT ] = IE (QT )2 IE QT
,
=
n
n 1.
c) We have
h
i
(n)
(n)
(n)
kQT T k2L2 () = IE (QT IE[QT ])2
h
i
(n)
= Var QT
n(n + 2)T 2
T2
n2
2T 2
=
,
n
hence
(n)
lim kQT T k2L2 () = lim
2T 2
= 0,
n
showing that
540
"

(n)
lim QT = T
in L2 ().
d) We have
n
X
(BkT /n B(k1)T /n )B(k1)T /n =
k=1
1X 2
2
BkT /n B(k1)T
/n
2
k=1
1X
(BkT /n B(k1)T /n )(BkT /n B(k1)T /n )
2
k=1
1
= ((BT )2 (B0 )2 )
2
n
1X
(BkT /n B(k1)T /n )(BkT /n B(k1)T /n )

2
k=1
1
(n)
= ((BT )2 QT ),
2
which converges to ((BT )2 T )/2 in L2 () as n tends to infinity, hence
wT
0
Bt dBt = lim
n
X
(BkT /n B(k1)T /n )B(k1)T /n =
k=1
(BT )2 T
.
2
e) We have
n
h
i X
(n) =
IE Q
IE[(B(k1/2)T /n B(k1)T /n )2 ]
T
k=1
n
X
((k 1/2)T /n (k 1)T /n)
k=1
T
,
2
n 1.
Next, we have
!2
n
h
i
X
(n)
)2 = IE
IE (Q
(B(k1/2)T /n B(k1)T /n )2
T
k=1
= IE
n
X
(B(k1/2)T /n B(k1)T /n ) (BlT /n B(l1)T /n )
k,l=1
n
X

IE (B(k1/2)T /n B(k1)T /n )4
k=1
"
541
N. Privault
X
+2

IE (B(k1/2)T /n B(k1)T /n )2 IE (B(l1/2)T /n B(l1)T /n )2
1k<ln
=3
n
X
((k 1/2)T /n (k 1)T /n)2
k=1
+2
((k 1/2)T /n (k 1)T /n)((l 1/2)T /n (l 1)T /n)
1k<ln
T2
n(n 1)T 2
+
4n
4n2
n(n + 2)T 2
,
n 1.
=
4n2
=3
Finally we find
h
i
(n) T /2k2 2
(n)
(n) 2
kQ
L () = IE (QT IE[QT ])
T
i
h
(n)
= Var Q
T
n(n + 2)T 2
T2
=
4n2
4
T2
=
,
2n
hence
(n)
T /2k2 2
lim kQ
L () = lim
T
T2
= 0,
2n
showing that
(n) = T
lim Q
T
2
in L2 ().
f) We have
n
X
(BkT /n B(k1)T /n )B(k1/2)T /n
k=1
n
X
(BkT /n B(k1/2)T /n )B(k1/2)T /n
k=1
n
X
+
=
1
2
(B(k1/2)T /n B(k1)T /n )B(k1/2)T /n
k=1
n
X
2
2
BkT
/n B(k1/2)T /n
k=1
n
1X
(BkT /n B(k1/2)T /n )(BkT /n B(k1/2)T /n )
2
k=1
542
"

n
1X 2
2
B(k1/2)T /n B(k1)T
/n
2
k=1
n
1X
+
(B(k1/2)T /n B(k1)T /n )(B(k1/2)T /n B(k1)T /n )
2
k=1
1
1X
(BkT /n B(k1/2)T /n )(BkT /n B(k1/2)T /n )
= (BT )2
2
2
k=1
1X
(B(k1/2)T /n B(k1)T /n )(B(k1/2)T /n B(k1)T /n ),
+
2
k=1
which converges to ((BT )2 T + T )/2 = (BT )2 /2 in L2 () as n tends to

infinity, hence
wT
Bt dBt = lim
n
X
(BkT /n B(k1)T /n )B(k1/2)T /n =
k=1
(BT )2
,
2
see Section 2.4 of [80] for further details on the Stratonovich integral.
g) We have
n
i X
h
(n) =
IE[(B(k)T /n B(k1)T /n )2 ]
IE Q
T
k=1
n
X
((k )T /n (k 1)T /n)
k=1
T
= (1 ) ,
2
n 1.
Next, we have
!2
n
h
i
X
(n) 2
2
IE (QT ) = IE
(B(k)T /n B(k1)T /n )
k=1
= IE
n
X
(B(k)T /n B(k1)T /n ) (BlT /n B(l1)T /n )
k,l=1
n
X

IE (B(k)T /n B(k1)T /n )4
k=1
+2

IE (B(k)T /n B(k1)T /n )2 IE (B(l)T /n B(l1)T /n )2
1k<ln
=3
n
X
((k )T /n (k 1)T /n)2
k=1
"
543
N. Privault
X
+2
((k )T /n (k 1)T /n)((l )T /n (l 1)T /n)
1k<ln
T2
n(n 1)T 2
+ (1 )2
n
n2
2
2 n(n + 2)T
,
n 1.
= (1 )
n2
= 3(1 )2
Finally we find
h
i
(n) (1 )T /2k2 2
(n)
(n) 2
kQ
L () = IE (QT IE[QT ])
T
i
h
(n)
= Var Q
T
n(n + 2)T 2
(1 )2 T 2
n2
T2
= 2(1 )2 ,
n
= (1 )2
hence
(n)
2
(1 )T k2 2
lim kQ
L () = (1 ) lim
T
T2
= 0.
n
Next we have
n
X
(BkT /n B(k1)T /n )B(k)T /n
k=1
=
=
n
X
(BkT /n B(k)T /n )B(k)T /n +
k=1
n
X
1
2
n
X
(B(k)T /n B(k1)T /n )B(k)T /n
k=1
n
2
2
BkT
/n B(k)T /n
k=1
n
1X
(BkT /n B(k)T /n )(BkT /n B(k)T /n )
2
k=1
1X 2
2
+
B(k)T /n B(k1)T
/n
2
k=1
n
1X
+
(B(k)T /n B(k1)T /n )(B(k)T /n B(k1)T /n )
2
k=1
1
1X
= (BT )2
(BkT /n B(k)T /n )(BkT /n B(k)T /n )
2
2
k=1
1X
+
(B(k)T /n B(k1)T /n )(B(k)T /n B(k1)T /n ),
2
k=1
which converges to ((BT )2 T + (1 )T )/2 = ((BT )2 + (1 2)T )/2

in L2 () as n tends to infinity, hence
544
"

wT
0
Bt d Bt = lim
n
X
(BkT /n B(k1)T /n )B(k)T /n =
k=1
(BT )2 + (1 2)T
.
2
In particular we find
wT
0
n
X
Bt d0 Bt = lim
(BkT /n B(k1)T /n )BkT /n =
k=1
(BT )2 + T
,
2
and we note that

wT
wT
1 wT
Bt dBt +
Bt d1 Bt .
Bt dBt =
0
0
0
2
h) We have
lim
n
X
(k )
k=1
T
n

k
T
T
(k 1)
n
n
n
T X
T
(k )
n n
n
= lim
k=1
n
n
T X T
T XT
= lim
k lim
n n
n n
n
n
k=1
k=1
n(n + 1)
T2
= T lim
lim
n
n n
2n2
T2
=
,
2
2
which does not depend on [0, 1] hence the stochastic phenomenon of

the previous questions does not occur when approximating the determinrT
istic integral 0 tdt = T 2 /2 by Riemann sums.
In mathematical finance we choose to use the Ito integral (which corresponds to the choice = 1) because it is suitable for the modeling of
market returns as
St+t St
dSt
'
= t + Bt = t + (Bt+t Bt )
St
St
or
dSt ' St+t St = St t + St Bt , = St t + St (Bt+t Bt ),
based on the value St at the the left endpoint of the discretized time
interval [t, t + t].
"
545
N. Privault
Chapter 5
Exercise 5.1
a) We have
St = S0 et +
wt
0
e(ts) dBs .
b) The function C(t, x) satisfies the Black-Scholes PDE

rC(t, x) =
C
C
1
2C
(t, x)+rx
(t, x)+ x2 2 2 (t, x),
t
x
2
x
x > 0,
t [0, T ],
with terminal condition C(T, x) = ex .

c) We find

2 2r(T t)
C(t, x) = exp r(T t) + xer(T t) +
(e
1) .
4r
d) We have
t =

C
2 2r(T t)
(t, St ) = exp St er(T t) +
(e
1) .
x
4r
Exercise 5.2
a) We have, counting approximately 46 days to maturity,
(r 12 2 )(T t) + log SKt
T t
(0.04377 12 (0.9)2 )(46/365) + log
p
=
0.9 46/365
= 2.46,
d =
17.2
36.08
and
p
d+ = d + 0.9 46/365 = 2.14.
From the attached table we get
(d+ ) = (2.14) = 0.0162
and
(d ) = (2.46) = 0.0069,
hence
f (t, St ) = St (d+ ) Ker(T t) (d )
= 17.2 0.0162 36.08 e0.0437746/365 0.0069
546
"

= HK$ 0.031.
b) We have
Print
t =
f
(t, St ) = (d+ ) = (2.14) = 0.0162,
x
(16.41)
hence one should only hold a fractional quantity 16.2 of the risky asset in
order to hedge 1000 such call options when = 0.90.
c) From the curve it turns out that when f (t, St ) = 10 0.023 = HK$ 0.23,
the volatility is approximately equal to = 122%.
This approximate value of implied volatility can be found under the col-
Derivative
Warrant
umn Implied Volatility
(IV.)Search
on this set of market data from the Hong
Kong Stock Exchange:
http://www.hkex.com.hk/dwrc/se
Updated: 6 November 2008
Basic Data
DW Issuer
Code
UL
Call
/Put
DW
Type
Listing
(D-M-Y)
Maturity Strike Entitle(D-M-Y)

ment
Ratio^
Total
Issue
Size
O/S
(%)
01897
FB
Link to Relevant
Exchange
Traded
Options 10 138,000,000 16.43
00066 Call Standard
18-12-2007
23-12-2008
36.08
04348
BP
00066 Call Standard 18-12-2007 23-02-2009
38.88
10 300,000,000
0.25
04984
AA
00066 Call Standard 02-06-2005Market

22-12-2008
Data 12.88
10 300,000,000
0.36
Strike Entitle- 05931

Total SB O/S
Delta
IV.
Day
Day
Closing 27.868
T/O
UL10 200,000,000
Base Listing 0.04
Su
00066
Call Standard
27-03-2008
29-12-2008
ment
Issue
(%) (%)
(%) High Low
Price #
('000) Price Document
Ratio^ 09133
Size CT
($)31-01-2008
($)
($)
($)
Ann
00066 Call Standard
08-12-2008
36.88
10 200,000,000 0.15
36.08
10 138,000,000
0.000
0.000
0.023
13436 SG16.43
000660.780
Call125.375
Standard
14-05-2008
30-04-2009
32 0 17.200
10 200,000,000
0.10
38.88
10 300,000,000
0.25
0.767
88.656 0.000
0.000
0.024
13562
BP Market
00066data
Call
26-05-2008
30 0 17.200
10 150,000,000
Fig. S.1:
for Standard
the warrant
#01897 on08-12-2008
the MTR Corporation.
0.00
12.88
10 300,000,000
0.000
0.000
0.540 26.6 0 17.200
13688 RB 0.36
000668.075
Call128.202
Standard
04-06-2008
20-02-2009
10 200,000,000
7.17
27.868
10 200,000,000
0.04
2.239
126.132
0.000
0.000
0.086
0 17.200
13764
SGaround
0006620%.
CallThe
Standard
13-06-2008
28per
10 is300,000,000
would be
observed
volatility26-02-2009
value = 1.22
year
0.31
36.88
actually quite high.

10 200,000,000
0.000
0.000
0.010 27.38 0 17.200
13785 ML 0.15
000660.416
Call133.443
Standard
17-06-2008
19-01-2009
10 100,000,000
0.50
Remark: a typical value for the volatility in standard market conditions
Exercise 5.3
32
10 200,000,000
0.10 1.059
0.000
0.000
0.031 28.8 0 17.200
Call 61.785
Standard
18-06-2008
18-12-2008
10 100,000,000
a) 13821
We find JP
h(x) 00066
= x K.
30
Letting g(t, x), the0.987

PDE rewrites
as 0.000
10b)150,000,000
0.000
0.026 23.88 0 17.200
14111 UB 0.00
00066 Call127.080
Standard
09-07-2008 16-02-2009
10 500,000,000 0.88
26.6
28
27.38
0.81
r(x (t))
=
(t) + rx, 0.013
10 200,000,000
0.000
0.000
14264
BI 7.17
000660.706
Call 49.625
Standard
16-07-2008
25-02-2009 26.38 0 17.200
10 200,000,000
"
10 300,000,000
0.010
0.010
0.010
14305 DB 0.31
000660.549
Call 49.880
Standard
22-07-2008
09-03-2009
547
0.03
27 6 17.200
10 300,000,000
0.00

10 100,000,000
63.598
0.000
0.000
0.014 28.08 0 17.200
14489
FB 0.50
000660.714
Call http://www.ntu.edu.sg/home/nprivault/indext.html
Standard
06-08-2008
29-06-2009
10 175,000,000
0.15
N. Privault
hence (t) = (0)ert and g(t, x) = x (0)ert . The final condition
g(T, x) = h(x) = x K
yields (0) = Ke
c) We have
rT
and g(t, x) = x Ker(T t) .

t =
g
(t, St ) = 1,
x
hence
t =
Vt t St
g(t, St ) St
St Ker(T t) St
=
=
= KerT .
At
At
At
Note that we could also have directly used the identification

Vt = g(St , t) = St Ker(T t) = St KerT At = t St + t At ,
which immediately yields t = 1 and t = KerT .
Exercise 5.4
a) We develop two approaches.
(i) By financial intuition. We need to replicate a fixed amount of $1 at
maturity T , without risk. For this there is no need to invest in the
stock. Simply invest g(t, St ) := er(T t) at time t [0, T ] and at
maturity T you will have g(T, ST ) = er(T t) g(t, St ) = $1.
(ii) By analysis and the Black-Scholes PDE. Given the hint, we try plugging a solution of the form g(t, x) = f (t), not depending on the variable x, into the Black-Scholes PDE (5.25). Given that here we have
g
(t, x) = 0,
x
2g
(t, x) = 0,
x2
and
g
(t, x) = f 0 (t),
t
we find that the Black-Scholes PDE reduces to rf (t) = f 0 (t) with

the terminal condition f (T ) = g(T, x) = 1. This equation has
for solution f (t) = er(T t) and this is also the unique solution
g(t, x) = f (t) = er(T t) of the Black-Scholes PDE (5.25) with terminal condition g(T, x) = 1.
b) We develop two approaches.
(i) By financial intuition. Since the terminal payoff $1 is riskless we do
not need to invest in the risky asset, hence we should keep t = 0.
Our portfolio value at time t becomes
Vt = g(t, St ) = er(T t) = t St + t At = t At
548
"

with At = ert , so that we find t = erT , t [0, T ]. This portfolio
strategy remains constant over time, hence it is clearly self-financing.
(ii) By analysis. The Black-Scholes theory tells us that
t =
and
t =
g
(t, x) = 0,
x
Vt
er(T t)
Vt t St
=
=
= erT .
At
At
ert
Exercise 5.5
a) We have
Ct = er(T t) IE [ST K | Ft ]
= er(T t) IE [ST | Ft ] Ker(T t)
= ert IE [erT ST | Ft ] Ker(T t)
= ert ert St Ker(T t)
= St Ker(T t) .
We can check that the function g(x, t) = xKer(T t) satisfies the BlackScholes PDE
rg(x, t) =
g
2 2 2 g
g
(x, t) + rx (x, t) +
x
(x, t)
t
x
2
x2
with terminal condition g(x, T ) = xK, since g(x, t)/t = rKer(T t)

and g(x, t)/x = 1.
b) We simply take t = 1 and t = KerT in order to have
Ct = t St + t ert = St Ker(T t) ,
t [0, T ].
Note again that this hedging strategy is constant over time, and the relation t = g(St , t)/x for the Delta, cf. (16.41), is satisfied.
Exercise 5.6 Using It
os formula and the fact that the expectation of the
stochastic integral with respect to (Wt )tR+ is zero, cf. Relation (4.13), we
have

h
i

C(x, T ) = erT IE (ST )S0 = x
w

w

T
T
1

ert 00 (St ) 2 (St )dtS0 = x
= (x) IE
rers (St )dtS0 = x + IE
0
0
2

h
i
h
i
wT
wT

rt
rt
e
IE St 0 (St )S0 = x dt
= (x)
re
IE (St )S0 = x dt + r
0
"
549
N. Privault

i
1 w T rt h 00

e
IE (St ) 2 (St )S0 = x dt,
0
2
hence by differentiation with respect to T we find

i
rT h

e
IE (ST )S0 = x
T

i
i
h
h

= rerT IE (ST )S0 = x + rerT IE ST 0 (ST )S0 = x

h
i
1

+ erT IE 00 (ST ) 2 (ST )S0 = x .
2
ThetaT =
Exercise 5.7
a) Letting Vt := t St + t At denote the value of the hedging portfolio, the
self-financing condition reads
dVt = t dAt + t dSt = rVt dt + ( r)t St dt + t St dBt ,
t R+ . and by the Itos formula we have
dg(t, St )
g
g
1
2g
g
=
(t, St )dt + ( D)St (t, St )dt + 2 St2 2 (t, St )dt + St (t, St )dBt .
t
x
2
x
x
g
(t, St )
By identification of the terms in dBt and dt above we get t =
x
and the Black-Scholes PDE with dividend
rg(t, x) =
g
g
1
2g
(t, x) + (r D)x (t, x) + 2 x2 2 (t, x).
t
x
2
x
(16.42)
b) In order to solve (16.42) we note that the function f (t, x) := e(T t)D g(t, x)
satisfies the standard Black-Scholes PDE, i.e. we have
rf (t, x) =
f
1
2f
f
(t, x) + rx (t, x) + 2 x2 2 (t, x),
t
x
2
x
with the same terminal condition f (T, x) = g(T, x) = (x K)+ , hence we

have
f (t, x) = BS(K, x, , r D, T t) = x(d+ ) Ker(T t) (d ),
where
d =
log(x/K) + (r D 2 /2)(T t)
.
T t
Consequently the pricing function of the European call option with dividend rate D is
550
"

g(t, x) = e(T t)D f (t, x) = e(T t)D BS(K, x, , rD, T t),
t [0, T ].
Chapter 6
IE [(pST1 + qST2 )] IE [p(ST1 ) + q(ST2 )]
since is convex,
= p IE [(ST1 )] + q IE [(ST2 )]
because (St )tR+ is a martingale,
by Jensens inequality,
= p IE [(IE [ST2 | FT1 ])] + q IE [(ST2 )]
p IE [IE [(ST2 ) | FT1 ]] + q IE [(ST2 )]
= p IE [(ST2 )] + q IE [(ST2 )]
by the tower property,
= IE [(ST2 )],
because p + q = 1,
see Exercise 8.9 for an extension to arbitrary summations.

Remark: This type of technique is useful to get an upper price estimate from
Black-Scholes when the actual option price is difficult to compute: here the
closed form computation would involve a double integration of the form
h
i
2
2
IE [(pST1 + qST2 )] = IE pS0 eBT1 T1 /2 + qS0 eBT2 T2 /2
h

i
2
2
= IE S0 eBT1 T1 /2 p + qe(BT2 BT1 ) (T2 T1 )/2

2
2
1 w w
S0 ex T1 /2 p + qey (T2 T1 )/2
=
2
2
2
dxdy
ex /(2T1 )y (2(T2 T1 )) p
T1 (T2 T1 )

+
2
2
1 w w
S0 ex T1 /2 p + qey (T2 T1 )/2 K
=
2
2
2
dxdy
ex /(2T1 )y (2(T2 T1 )) p
T1 (T2 T1 )
1 w
=
2 {(x,y)R2 : S0 ex (p+qey2 (T2 T1 )/2 )Ke2 T1 /2 }
(S0 ex
T1 /2
(p + qey
e
(T2 T1 )/2
) K)
dxdy
x2 /(2T1 )y 2 (2(T2 T1 ))
T1 (T2 T1 )
= ...
Exercise 6.2
"
551
N. Privault
a) By Jensens inequality we have
+
er(T t) IE [(ST K)+ | Ft ] er(T t) (IE [ST K | Ft ])

+
= er(T t) er(T t) St K
+

= St Ker(T t) ,
t [0, T ].
b) By Jensens inequality we have
+
er(T t) IE [(K ST )+ | Ft ] er(T t) (IE [K ST | Ft ])

+
= er(T t) K er(T t) St

+
= Ker(T t) St ,
t [0, T ].

er(T1 t) IE er(T2 T1 ) IE (ST2 ST1 )+ | FT1 | Ft
= er(T1t)
IE ST1
r + 2 /2 p
T2 T1
ST1 er(T2 T1 )
r 2 /2 p
T2 T1

| Ft
1 t)
= er(T

r + 2 /2 p
r 2 /2 p
T2 T1 ST1 er(T2 T1 )
T2 T1
IE ST1 | Ft

r + 2 /2 p
r 2 /2 p
= St
T2 T1 ST1 er(T2 T1 )
T2 T1
t [0, T1 ].
Exercise 6.4
a) For all t [0, T ] we have
C(t, St ) = er(T t) St2 IE
ST2
St2
h
i
2
= er(T t) St2 IE e2(BT Bt ) (T t)+2r(T t)
= St2 e(r+
)(T t)
b) For all t [0, T ] we have

t =
2
C
(t, x)|x=St = 2St e(r+ )(T t) ,
x
and
552
"
t =
2
C(t, St ) t St
ert 2 (r+2 )(T t)
=
(S e
2St2 e(r+ )(T t) )
At
A0 t
S2 2
= t e (T t)+r(T 2t) .
A0
As for the self-financing condition we have

dC(t, St ) = d(St2 e(r+
)(T t)
(r+ 2 )(T t)
St2 dt + e(r+
(r+ 2 )(T t)
St2 dt
= (r + )e
= (r + )e
(r+ 2 )(T t)
= re
St2 dt
+e
)(T t)
(r+ 2 )(T t)
(r+ 2 )(T t)
+ 2St e
d(St2 )
(2St dSt + 2 St2 dt)
dSt ,
and
St2 2 (T t)+r(T 2t)
e
At dt
A0
t dSt + t dAt = 2St e(r+
)(T t)
dSt r
= 2St e(r+
)(T t)
dSt rSt2 e
(T t)+r(T t)
dt,
which recovers dC(t, St ) = t dSt + t dAt , i.e. the portfolio strategy is

self-financing.
Exercise 6.5
a) By (4.28) we have
St = S0 et +
wt
e(ts) dBs .
St = S0 +
wt
ers dBs ,
b) For = r we have
which is a martingale, being a stochastic integral with respect to Brownian

motion. This fact can also be proved directly by computing the conditional
expectation IE[St | Fs ] and showing it is equal to Ss , i.e.:

wt
IE[St | Fs ] = IE S0 + eru dBu | Fs
0
w

t
= IE[S0 ] + IE
eru dBu | Fs
0
w

i
hw s
t
ru
eru dBu | Fs
= S0 + IE
e
dBu | Fs + IE
s
0
w

ws
t
= S0 +
eru dBu + IE
eru dBu
0
s
ws
= S0 +
eru dBu
0
"
553
N. Privault
= Ss ,
s [0, t].
c) We rewrite the stochastic differential equation satisfied by (St )tR+ as

t ,
dSt = St dt + dBt = rSt dt + dB
where
r
dt + dBt .
t )tR is a standard BrowBy the Girsanov Theorem 6.1, the process (B

+
nian motion under the probability measure P defined by

2 !
r
T r
dP
:= exp
BT
.
dP
t :=
dB
Consequently, (St )tR+ is a martingale under P .

d) We have
C(t, St ) = er(T t) IE [exp(ST )|Ft ]

wT
u Ft
= er(T t) IE exp erT S0 +
er(T u) dB
0

wt
wT
u +
u Ft
= er(T t) IE exp erT S0 + er(T u) dB
er(T u) dB
0
t

w

T

r(T t)
r(T u)
= exp r(T t) + e
St IE exp
e
dBu Ft
t

w

T
u
= exp r(T t) + er(T t) St IE exp
er(T u) dB
t

2 w T r(T u) 2
(e
) du
= exp r(T t) + e
St exp
2 t

2
= exp r(T t) + er(T t) St +

(e2r(T t) 1) .
4r

r(T t)
e) We have
t =

C
2 2r(T t)
(t, St ) = exp St er(T t) +
(e
1)
x
4r
and
C(t, St ) t St
At

er(T t)
2 2r(T t)
=
exp St er(T t) +
(e
1)
At
4r
t =
554
"

St
2 2r(T t)
exp St er(T t) +
(e
1) .
At
4r
f) We have

2 2r(T t)
dC(t, St ) = rer(T t) exp St er(T t) +
(e
1) dt
4r

2 2r(T t)
r(T t)
(e
1) dt
rSt exp St e
+
4r

2 2r(T t)
2 r(T t)
exp St er(T t) +
(e
1) dt
e
2
4r

2 2r(T t)
r(T t)
+ exp St e
+
(e
1) dSt
4r

1 r(T t)
2 2r(T t)
+ e
exp St er(T t) +
(e
1) 2 dt
2
4r

2 2r(T t)
r(T t)
r(T t)
= re
exp St e
+
(e
1) dt
4r

2
rSt exp St er(T t) +

(e2r(T t) 1) dt
4r
+t dSt .
On the other hand we have
t dSt + t dAt = t dSt

2 2r(T t)
(e
1) dt
+rer(T t) exp St er(T t) +
4r

2 2r(T t)
r(T t)
rSt exp St e
+
(e
1) dt,
4r
showing that
dC(t, St ) = t dSt + t dAt ,
and confirming that the strategy (t , t )tR+ is self-financing.
Exercise 6.6
a) We have
f
(t, x) = (r 2 /2)f (t, x),
t
and
"
f
(t, x) = f (t, x),
x
2f
(t, x) = 2 f (t, x),
x2
555
N. Privault
hence
dSt = df (t, Bt )
f
f
1 2f
=
(t, Bt )dt
(t, Bt )dt +
(t, Bt )dBt +
t
x
2
x2

1
1
= r 2 f (t, Bt )dt + f (t, Bt )dBt + 2 f (t, Bt )dt
2
2
= rf (t, Bt )dt + f (t, Bt )dBt
= rSt dt + St dBt .
b) We have
IE[eBT |Ft ] = IE[e(BT Bt +Bt ) |Ft ]
= eBt IE[e(BT Bt ) |Ft ]
= eBt IE[e(BT Bt ) ]
= eBt +
(T t)/2
c) We have
IE[ST |Ft ] = IE[eBT +rT
=e
rT T /2
= erT
IE[e
T /2
|Ft ]
BT
|Ft ]
T /2 Bt + 2 (T t)/2
= erT +Bt
=e
t/2
r(T t)+Bt +rt 2 t/2
= er(T t) St ,
t [0, T ].
d) We have
Vt = er(T t) IE[C|Ft ]
= er(T t) IE[ST K|Ft ]
= er(T t) IE[ST |Ft ] er(T t) IE[K|Ft ]
= St er(T t) K,
t [0, T ].
e) We take t = 1 and t = KerT /A0 , t [0, T ].

f) We find
VT = IE[C | FT ] = C.
Exercise 6.7 Digital options.

a) By definition of the indicator functions 1[K,) and 1[0,K] we have
556
"
1[K,) (x) =
1 if x K,
resp. 1[0,K] (x) =
0 if x < K,
1 if x K,
0 if x > K,
which shows the claimed result by the definition of Cd and Pd .

b) We have
t (Cd ) = er(T t) IE[Cd | Ft ]
= er(T t) IE[1[K,) (ST ) | St ]
= er(T t) P(ST K | St )
= Cd (t, St ).
c) We have t (Cd ) = Cd (t, St ), where
Cd (t, x) = er(T t) P(ST > K | St = x)

r(T t) 2 (T t)/2 + log(St /K)
= er(T t)
T t
= er(T t) (d ) ,
with
d =
(r 2 /2)(T t) + log(St /K)
.
T t
d) The price of this modified contract with payoff

C = 1[K,) (ST ) + 1[0,K) (ST )
is given by
t (C ) = er(T t) IE[1[K,) (ST ) + 1[0,K) (ST ) | St ]
= er(T t) P(ST K | St ) + er(T t) P(ST K | St )
= er(T t) P(ST K | St ) + er(T t) (1 P(ST K | St ))
= er(T t) er(T t) + (1 )P(ST K | St )

r(T t) 2 (T t)/2 + log(St /K)
.
= er(T t) + (1 )er(T t)
T t
e) We note that
1[K,) (ST ) + 1[0,K] (ST ) = 1[0,) (ST ),
almost surely since P(ST = K) = 0, hence
t (Cd ) + t (Pd ) = er(T t) IE[Cd | Ft ] + er(T t) IE[Pd | Ft ]
= er(T t) IE[Cd + Pd | Ft ]
"
557
N. Privault
2
1.5
1
0.5
020
15
200
10
150

100
5
50
0 0
underlying
Fig. S.2: Price of a digital call option.

= er(T t) IE[1[K,) (ST ) + 1[0,K] (ST ) | Ft ]
= er(T t) IE[1[0,) (ST ) | Ft ]
= er(T t) IE[1 | Ft ]
= er(T t) ,
0 t T.
f) We have
t (Pd ) = er(T t) t (Cd )
= er(T t) er(T t)
r(T t) 2 (T t)/2 + log(x/K)
T t
= er(T t) (1 (d ))
= er(T t) (d ).
g) We have
Cd
(t, St )
x

r(T t) 2 (T t)/2 + log(x/K)
= er(T t)
x
T t
x=St
1
r(T t)
(d )2 /2
p
=e
e
2(T t)St
> 0.
t =
The Black-Scholes hedging strategy of such a call option does not involve
short-selling because t > 0 for all t, cf. Figure S.3 which represents the
risky investment in the hedging portfolio of a digital call option.
558
"

5
4
3
2
1
05
4
3
2
1
0 20
40
60
80
100
120
140
160
180
underlying
Fig. S.3: Risky hedging portfolio value for a digital call option.
Figure S.4 presents the riskless hedging portfolio value for a digital call
option.
2
1
0
-1
-2
-3
-4 4
3.5
3
2.5
160
2
140
1.5
120
100
1
80
0.5
0 40
60
underlying
Fig. S.4: Riskless hedging portfolio value for a digital call option.
h) Here we have
Pd
(t, St )
x

r(T t) 2 (T t)/2 + log(x/K)
= er(T t)
x
T t
x=St
1
r(T t)
(d )2 /2
p
= e
e
2(T t)St
< 0.
t =
The Black-Scholes hedging strategy of such a put option does involve

short-selling because t < 0 for all t.
"
559
N. Privault
Exercise 6.8 The price process with reinvested dividends is given by
(eDt St )tR+ , and after discount (ert eDt St )tR+ = (e(rD)t St )tR+ is a
martingale, hence (St )tR+ has the dynamics
t ,
dSt = (r D)St dt + St dB
t )tR is a standard Brownian motion under P . Hence we have,
where (B
+
after discounting the payoff (ST K)+ at the rate r,
Vt = g(t, St )

= er(T t) IE (ST K)+ | Ft

+

2
= er(T t) IE S0 eBT +(rD /2)T K | Ft

= e(T t)D BS(K, x, , r D, T t).
Chapter 7
w

T 1
1
(dSt )2
VST = IE
2
0 St
T
w
2
p
T 1
1
(1)
= IE
(r vt )St dt + St + vt dBt
2
0 St
T
w

T
1
= IE
( + vt )dt
0
T
1 wT
=+
IE[vt ]dt,
T 0
with
IE[vt ] = IE[v0 ]et + m(1 et ),
t R+ ,
hence
1 wT
(IE[v0 ]et + m(1 et ))dt
T 0
w
1 T
=+
(IE[v0 ]et + m(1 et ))dt
T 0
wT
1
= + m + (IE[v0 ] m)
et dt
0
T
eT 1
= + m + (IE[v0 ] m)
.
T
VST = +
Note that if the process (vt )tR+ is started in the gamma stationary distribution then we have IE[v0 ] = IE[vt ] = m, t R+ , and the variance swap rate
560
"

VST = + m becomes independent of the time T .
Exercise 7.2
Chapter 8
Exercise 8.1
a) We have St = S0 eBt , t R+ .
b) We have
2
IE[ST ] = S0 IE[eBT ] = S0 e T /2 .
c) We have
!
P
sup Bt a
=2
t[0,T ]
w
a
ex
/(2T )
dx
,
2T
a > 0,
i.e. the probability density function of sup Bt is given by

t[0,T ]
r
(a) =
2 a2 /(2T )
e
1[0,) (a),
T
a R.
d) We have
"
!#
IE[ST ] = S0 IE exp sup Bt

t[0,T ]
= S0
w
0
ex (x)dx
2S0 w (xT )2 /(2T )+2 T /2

2S0 w xx2 /(2T )
e
dx
e
dx =
=
2T 0
2 2 T 0
w
w
2
2S0 2 T /2
2S0 2
x2 /2
=
e
ex /(2T ) dx = e T /2
dx
e
T
T
2
2T
T
2
2
2
= 2S0 e T /2
ex /2 dx = 2S0 e T /2 ( T ) = 2 IE[ST ]( T ).
Remark: We note that the ratio between the expectedgains by selling at

the maximum and selling at time T is given by 2( T ), which cannot
be greater than 2.
"
561
N. Privault
2
1/2
2 (T
ratio
1.5
0.5
0
0
0.5
1.5
2
time T
2.5
3.5
Fig. S.5: Average return by selling at the maximum vs selling at maturity T

as a function of T .
e) By a symmetry argument we have

P

inf Bt a = P
sup (Bt ) a
t[0,T ]
!
=P
sup Bt a
t[0,T ]
t[0,T ]
!
sup Bt a
=P
=2
t[0,T ]
ex
/(2T )
dx
,
2T
a < 0,

t[0,T ]
r
(a) =
2 a2 /(2T )
e
1(,0] (a),
T
a R.
f) We have

w0
E[ST ] = S0 E exp inf Bt
= S0
t[0,T ]
ex (x)dx
2S0 w 0 (xT )2 /(2T )+2 T /2

2S0 w 0 xx2 /(2T )
e
dx =
e
dx
=
2T
2T
w
w
T
2
2S0 2 T /2 T x2 /(2T )
2S0 2
e
dx = e T /2
=
e
ex /2 dx
2
2T
2
= 2S0 e T /2 ( T ) = 2E[ST ]( T ).
Remarks:
(i) From the inequality
0 IE[(T BT )+ ]
562
"

2
1 w
(T x)+ ex /(2T ) dx
2T
2
1 w T
(T + x)ex /(2T ) dx
=
2T
T w T x2 /(2T )
1 w T x2 /(2T )
=
e
dx
xe
dx
2T
2T
r
w
w
T
2
T
T T x2 /2
ex /2 dx
xe
dx
=
2
2
r
T h x2 /2 i T
= T ( T ) +
e
2
T 2 T /2
= T ( T ) +
e
,
2
we get
e
T /2
( T )
,
2T
hence
E[ST ]
2S
0 .
2T
(ii) The ratio between the expected

gains by maturity T vs selling at the
minimum is given by 2( T ), which is at most 1 and tends to 0
as and T tend to infinity.
1/2
2 (-T1/2
)
2 (T
)
ratio
1.5
0.5
0
0
0.5
1.5
2
time T
2.5
3.5
Fig. S.6: Average return by selling at the minimum vs selling at maturity T

as a function of T .
(iii) Given that IE[ST ] = 2E[ST ]( T ), we find the bound
2E[ST ]( T ) E[ST ] 2E[ST ]( T ),

with equality if = 0 or T = 0. We also have
"
563
N. Privault
2
2
2E[ST ]E[ST ] = 2e T /2 (1( T )) = 2e T /2 ( T ) = E[ST ]
hence
E[ST ] + E[ST ] = 2E[ST ],
and
or E[ST ] E[ST ] = E[ST ] E[ST ],
2S0
2E[ST ]
E[ST ] 2E[ST ].
2T
Exercise 8.2
a) We have
P (a t) = P (Xt > a) =
w
a
r
Xt (x)dx =
2 w x2 /(2t)
e
dx,
t y
y > 0.
b) We have
a (t) =
=
=
=
=
d
P (a t)
dt
d w
Xt (x)dx
a
dt r
r
1 2 3/2 w x2 /(2t)
1 2 3/2 w x2 x2 /(2t)
t
e
dx +
t
e
dx
a
a
2
2
t
r

w
w
2
2
2
1 2 3/2
t
ex /(2t) dx + aea /(2t) +

ex /(2t) dx
a
a
2
2
a
ea /(2t) ,
t > 0.
2t3
c) We have
a w 5/2 a2 /(2t)
IE[(a )2 ] =
t
e
dt
2 0
w
2 2
2a
=
x2 ea x /2 dx
2 0
1
= 2,
a
by the change of variable x = t1/2 , x2 = 1/t, t = x2 , dt = 2x3 dx.
Remark: We have
a w 1/2 a2 /(2t)
IE[a ] =
t
e
dt = +.
2 0
564
"

Exercise 8.3 Barrier options.
a) By (8.37) and (8.22) we find

g
St
St
T t
T t
t =
(t, St ) = +
+
y
K
B

2r/2

2
K r(T t)
2r
St
B
B
T t
T t

+ e
1 2

B
B
KSt
St
12r/2
2

2r St
B
B
T t
T t
+ 2
+
+
B
KSt
St

2 !
K
1
2
St
T t
1
exp
+
,
p
B
2
B
2(T t)
0 < St B, 0 t T , cf. also Exercise 7.1-(ix) of [109] and Figure 8.16
above.
b) We find
P(YT a & BT b) = P(BT 2a b),
a b 0.
c) We find
fYT ,BT (a, b) = 1(,b0] (a)
"
1
T
2
2
2
(b 2a)e T /2+b(2ab) /(2T )
T
565
N. Privault
r
2
2
2
1
(2a b)e T /2+b(2ab) /(2T ) ,

T
T
=
0,
a b 0.
d) The function g(t, x) is given in Relations (8.14) and (8.15) above.

Exercise 8.4 Barrier forward contracts.
a) Up-and-in barrier long forward contract. We have
er(T t) IE[C | Ft ] = er(T t) IE (ST K) 1(
max Su > B
0uT
= 1(
max Su > B
) (S
t
Ker(T t) ) + 1(
0ut
max Su B

Ft

) (t, S ),
t
0ut
(16.43)
where the function

T t
T t
(t, x) := x +
(x/B) Ker(T t)
(x/B)

2
T t
+B(B/x)2r/ +
(B/x)

T t
r(T t)
1+2r/ 2
Ke
(B/x)

(B/x)
solves the Black-Scholes PDE with the terminal condition
2
(T, x) = (x K + (B/x)2r/ (B Kx/B))1[B,) (x),

as in the proof of Proposition 8.2. Note that only the values of (t, x) with
x [0, B] are used for pricing.
566
"
25
20
15
10
5
0
80
75
70
65
underlying
60
55
220
200
180
120
140
160
Time in days
100
Fig. S.7: Graph of the up-and-in long forward contract price with K = 60 < B = 80.
As for the hedging strategy we find

T t
2
1
T t
(t, St ) = +
(x/B) + e(+ (x/B)) /2
x
2

T t
2
2
2r
1
T t
Ker(T t)( (x/B)) /2 2 (B/x)1+2r/ +
(B/x)
x 2
T t
2
2
1
+ (B/x)1+2r/ e(+ (B/x)) /2
2

2
K(1 2r/ 2 ) r(T t)
T t
e
(B/x)2r/
(B/x)
B
T t
2
2
K
(B/x)2r/ er(T t)( (B/x)) /2
B 2
2r

2
T t
T t
= +
(x/B) 2 (B/x)1+2r/ +
(B/x)

T t
T t
2
2
1
B
+ (1 K/B) e(+ (x/B)) /2 + er(T t)( (x/B)) /2
x
2

K
T t
2 r(T t)
2r/ 2
(B/x)

(B/x) ,
(1 2r/ )e
B
t =
since by (8.42) we have

T t
e(
and
T t
e(
"
T t
T t
(B/x))2 /2
= er(T t) (x/B)2r/ e(+
(x/B))2 /2
= er(T t) (B/x)2r/ e(+
(x/B))2 /2
(B/x))2 /2
567
N. Privault
0.3
0.25
0.2
0.15
0.1
0.05
0
80
75
70
65
underlying
60
55
220
120
140
160
Time in days
180
200
100
Fig. S.8: Delta of the down-and-in long forward contract with K = 60 < B = 80.
b) Up-and-out barrier long forward contract. We have
r(T t)
r(T t)
IE[C | Ft ] = e
IE (ST K) 1(
max Su < B
0uT
1(
max Su B

Ft

) (t, S ),
t
(16.44)
0ut
where the function

T t
T t
(x/B)
(t, x) := x +
(x/B) Ker(T t)

2
T t
B(B/x)2r/ +
(B/x)

2
T t
+Ker(T t) (B/x)1+2r/
(B/x)
2
(T, x) = (x K)1[0,B] (x) (B/x)2r/ (B Kx/B)1[B,) (x).

Note that only the values of (t, x) with x [B, ) are used for pricing.
568
"

Up-and-out barrier long forward contract price
20
15
10
5
0
-5
220
200
180
Time in days 160
140
120
100 60
65
75
70
underlying
80
Fig. S.9: Graph of the up-and-out long forward contract price with K = 60 < B = 80.
As for the hedging strategy, we find

T t
2
1
T t
(t, St ) = +
(x/B) e(+ (x/B)) /2
x
2

T t
2
2
2r
1
T t
+ Ker(T t)( (x/B)) /2 + 2 (B/x)1+2r/ +
(B/x)
x 2
T t
2
2
1
(B/x)1+2r/ e(+ (B/x)) /2
2

2
K(1 2r/ 2 ) r(T t)
T t
+
e
(B/x)2r/
(B/x)
B
T t
2
2
K
+ (B/x)2r/ er(T t)( (B/x)) /2
B 2
2r

2
T t
T t
= +
(x/B) + 2 (B/x)1+2r/ +
(B/x)
T t
T t
2
1
1 B r(T t)(
(x/B))2 /2
e(+ (x/B)) /2
e
2
2 x
T t
T t
2
K
1 K r(T t)(
(x/B))2 /2
+ e(+ (x/B)) /2 +
e
B 2
2 x

2
K
T t
(B/x)
+ (1 2r/ 2 )er(T t) (B/x)2r/
B
2r

2
T t
T t
= +
(x/B) + 2 (B/x)1+2r/ +
(B/x)

T t
T t
2
2
1
B
(1 K/B) e(+ (x/B)) /2 + er(T t)( (x/B)) /2
x
2

K
T t
2 r(T t)
2r/ 2
+ (1 2r/ )e
(B/x)

(B/x) ,
B
t =
by (8.42).
"
569
N. Privault
Delta of the up-and-out barrier long forward contract
1
0.95
0.9
0.85
0.8
0.75
0.7
0.65
0.6
0.55
0.5
60
65
underlying 70
75
80 100
120
140
200
180
160
Time in days
220
Fig. S.10: Delta of the up-and-out long forward contract price with K = 60 < B =
80.
c) Down-and-in barrier long forward contract. We have
er(T t) IE[C | Ft ] = er(T t) IE (ST K) 1(
min Su < B
0uT
= 1(
min Su < B
) (S
t
Ker(T t) ) + 1(
0ut

Ft

min Su B
) (t, S )
t
0ut
(16.45)
where the function

T t
T t
(t, x) := x +
(x/B) Ker(T t)
(x/B)

2
T t
+B(B/x)2r/ +
(B/x)

2
T t
Ker(T t) (B/x)1+2r/
(B/x)
2
(T, x) = (x K + (B/x)2r/ (B Kx/B))1[0,B] (x).
570
"

Down-and-in barrier long forward contract price
18
16
14
12
10
8
6
4
2
0100
120
140
160
Time in days
180
200
220
100
95
85
90
underlying
80
Fig. S.11: Graph of the down-and-in long forward contract price with K = 60 <
B = 80.

t =
(t, St )
x
2r
2
T t
(x/B) + 2 (B/x)1+2r/
= +

T t
2
1
(1 K/B) e(+ (x/B)) /2 +
2
2
K
+ (1 2r/ 2 )er(T t) (B/x)2r/
B

T t
(B/x)
+
T t
B r(T t)(
(x/B))2 /2
e
x

T t
(B/x) .
Delta of the own-and-in barrier long forward contract
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0100
120
140
160
Time in days
180
200
220
100
95
85
90
underlying
80
Fig. S.12: Delta of the down-and-in long forward contract with K = 60 B
0uT
= 1(
min Su B

Ft

) (t, S )
t
(16.46)
0ut
where the function

T t
T t
(x/B) Ker(T t)
(x/B)
(t, x) := x +

2
T t
B(B/x)2r/ +
(B/x)
!

T t
r(T t)
1+2r/ 2
+Ke
(B/x)
(B/x)
2
(T, x) = (x K)1[B,) (x) (B Kx/B)(B/x)2r/ 1[0,B] (x).

Note that (t, x) above coincides with the price of (8.15) of the standard
down-and-out barrier call option in the case K < B, cf. Exercise 8.3-(d).
Down-and-out barrier long forward contract price
40
35
30
25
20
15
10
5
0
220
200 180
160
Time in days
140
120
100 80
85
95
90
underlying
100
Fig. S.13: Graph of the down-and-out long forward contract price with K = 60 <
B = 80.

t =
(t, St )
x
2r

2
T t
T t
(x/B) 2 (B/x)1+2r/ +
(B/x)
= +

T t
T t
2
2
1
B
+ (1 K/B) e(+ (x/B)) /2 + er(T t)( (x/B)) /2
x
2
572
"

2
K
T t
(1 2r/ 2 )er(T t) (B/x)2r/
(B/x) .
B
Delta of the down-and-out barrier long forward contract
1.1
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
220
95
90
underlying
200
180 160
Time in days
140
120
100 80
85
100
Fig. S.14: Delta of the down-and-out long forward contract with K = 60 < B = 80.
e) Up-and-in barrier short forward contract. The price of the up-and-in barrier short forward contract is identical to (16.43) with a negative sign.
f) Up-and-out barrier short forward contract. The price of the up-and-out
barrier short forward contract is identical to (16.44) with a negative sign.
Note that (t, x) coincides with the price of (8.13) of the standard upand-out barrier put option in the case B < K.
g) Down-and-in barrier short forward contract. The price of the down-and-in
barrier short forward contract is identical to (16.45) with a negative sign.
h) Down-and-out barrier short forward contract. The price of the down-andout barrier short forward contract is identical to (16.46) with a negative
sign.
Exercise 8.5
a) We have

wa
2
dx
ex /(2T )
min Bt a = 2
,
t[0,T ]
2T

P
a < 0,

t[0,T ]
r
(a) =
"
2 a2 /(2T )
e
1(,0] (a),
T
a R.
573

N. Privault
b) We have

IE min St = S0 IE exp min Bt
t[0,T ]
t[0,T ]
2S0 w 0 xx2 /(2T )

2S0 w 0 (xT )2 /(2T )+2 T /2
=
e
dx
e
dx =
2T
2 2 T
w T
2
2S0 2 T /2 w T x2 /(2T )
2S0 2
e
e
dx = e T /2
ex /2 dx
=
2
2T

2
= 2S0 e T /2 T = 2 IE[ST ] 1 T ,
hence

IE ST min St = IE[ST ] IE min St = IE[ST ] 2 IE[ST ] 1 T
t[0,T ]
t[0,T ]

1
2
= IE[ST ] 2 T 1 = 2S0 e T /2 T
,
2

and
e
T /2

IE ST min St = S0 2 T 1 = S0 12 T .
t[0,T ]
Remark: We note that as T goes to infinity, the price of the lookback

option converges to S0 .
1/2
2 ((T
)-1)
0.8
price
0.6
0.4
0.2
0
0
time T
Fig. S.15: Price of the lookback call option as a function of T with S0 = 1.

Exercise 8.6 Lookback options. By (8.27) we find
t =
f
(t, St , M0t )
x
574
"

2r
St
T t
= 1 + 1 + 2 +
t
M0
t 2r/2

t
M
2
M0
0
T t
+er(T t)
1

,
St
2r
St
t [0, T ], and
t At = f (t, St , M0t ) t St

t 1+2r/2
t

M0
M0
St
T t
T t
r(T t)
e

.
= M0t er(T t)
t
M0
St
St
Exercise 8.7
a) The integral
rT
rs ds is centered Gaussian with variance

"
2 #
w w

wT
T
T
IE
rs ds
= 2 IE
Bs Bt dsdt
0
= 2
= 2
wT wT
0
IE[Bs Bt ]dsdt
wT wT
min(s, t)dsdt
0
0
wT wt
= 2 2
sdsdt
0
0
wT
= 2
t2 dt
0
= 2 T 3 /3.
b) Since the integral
rT
0
rs ds is a random variable with probability density

(x) = p
1
2T 3 /3
e3x
/(2T 3 )
we have
erT IE
"
wT
0
rT
+ #
w
ru du
= erT
(x )+ (x)dx
2
2 3
e
= p
(x )e3x /(2 T ) dx
2 2 T 3 /3
p
2
erT w
2 3 (x 2 T 3 /3 )ex /2 dx
=
/
T
/3
2
p
2
2
erT 2 T 3 /3 w
erT w
=
xex /2 dx
ex /2 dx
/ 2 T 3 /3
2
2 / 2 T 3 /3
"
575
N. Privault
p
p
2 T 3 /3 h x2 /2 i
erT
(1 (/ 2 T 3 /3))
e
/ 2 T 3 /3
2
2
p
p
erT 2 T 3 /3 32 /(22 T 3 )
erT
e
=
(1 (/ 2 T 3 /3))
2
2
!
r
r
2 T 3 32 /(22 T 3 )
3
erT
rT
=e
.
e

6
2 T 3
2
erT

"
+ #
w

1 wT
1 T

r(T t)
e
IE
Su du
Su du Ft
Ft = er(T t) IE
T 0
T 0
w

1 T

Su duFt er(T t)
= er(T t) IE
T 0
w
w

t
T
1
1

Su duFt + er(T t) IE
Su duFt er(T t)
= er(T t) IE
0
t
T
T
w

T
1 wt
1

= er(T t)
Su du + er(T t) IE
Su duFt er(T t)
t
T 0
T
1 wT
1 wt
Su du + er(T t)
IE[Su | Ft ]du er(T t)
= er(T t)
T 0
T t
w
w
1 T
1 t
Su du + er(T t)
St er(ut) du er(T t)
= er(T t)
T 0
T t
w
w
1 t
St T t ru
= er(T t)
Su du + er(T t)
e du er(T t)
T 0
T 0
1 wt
St
= er(T t)
Su du + er(T t) (er(T t) 1) er(T t)
T 0
rT
1 wt
1 er(T t)
= er(T t)
er(T t) ,
Su du + St
T 0
rT
t [0, T ], cf. [40] page 361. We check that the function f (t, x, y) =
er(T t) (y/T ) + x(1 er(T t) )/(rT ) satisfies the PDE
rf (t, x, y) =
f
f
1
2f
f
(t, x, y) + x (t, x, y) + rx (t, x, y) + x2 2 2 (t, x, y),
t
y
x
2
x
t, x > 0, and the boundary conditions f (t, 0, y) = er(T t) (y/T ),

0 t T , y R+ , and f (T, x, y) = y/T , x, y R+ . However, the
condition limy f (t, x, y) = 0 is not satisfied because we need to take
y > 0 in the above calculation.
Exercise 8.9 This question extends Exercise 6.1 to n 3. We have

ST1 + + STn
(ST1 ) + + (STn )
IE
IE
n
n
576
since is convex,
"

IE [(ST1 )] + + IE [(STn )]
n
IE [(IE [STn | FT1 ])] + + IE [(IE [STn | FT1 ])]
=
n
(IE [IE [STn | FT1 ]]) + + (IE [IE [STn | FT1 ]])
n
(IE [STn ]) + + (IE [STn ])
=
n
= IE [(STn )].
=
because (St )tR+ is a martingale,

by Jensens inequality,
by the tower property,
Exercise 8.10 Taking t [, T ], under the condition

At :=
1 wt
Ss ds K,
T
we have
+ #
1 wT

Ss ds K
Ft
T
"
+ #
1 wT

= er(T t) IE
At +
Ss ds K
Ft
T t

w
T
1

= er(T t) IE At +
Ss ds K Ft
T t

er(T t) w T

IE
Ss dsFt
= er(T t) (At K) +
t
T
er(T t) w T
IE [Ss | Ft ]ds
= er(T t) (At K) +
t
T
r(T t) w T
e
er(st) ds
= er(T t) (At K) + St
t
T
er(T t) w T t rs
e ds
0
T
er(T t) r(T t)
(e
1)
r(T )
er(T t) IE
"
1 er(T t)
,
r(T )
t [, T ].
Exercise 8.11 The Asian option price can be written as

"
+ #

1 wT

r(T t)
(UT )+ | Ut
Su du K
e
IE
Ft = St IE
T 0
= St h(t, Ut ) = St g(t, Zt ),
"
577
N. Privault
which shows that
g(t, Zt ) = h(t, Ut ),
and it remains to use the relation
Ut =
1 er(T t)
+ er(T t) Zt ,
rT
t [0, T ].
Chapter 9
Exercise 9.1 Stopping times.
a) When 0 t < 1 the question is > t ? cannot be answered at time t
without waiting to know the value of B1 at time 1. Therefore is not a
stopping time.
b) For any t R+ , the question is > t ? can be answered based on
the observation of the paths of (Bs )0st and of the (deterministic) curve
(es/2 )0st up to the time t. Therefore is a stopping time.
Since is a stopping time and (Bt )tR+ is a martingale, the stopping

time theorem shows that eBt (t )/2 tR+ is also a martingale and in
particular its expectation

IE eBt (t )/2 = IE eB0 (0 )/2 = IE eB0 0/2 = 1
is constantly equal to 1 for all t. This shows that
h
i

IE eB /2 = IE lim eBt (t )/2 = lim IE[eBt (t )/2 ] = 1.
t
Next, we note that we have eB = e /2 , hence

IE[e ] = IE[eB /2 ] = 1,
i .e.
IE[e ] = 1/ 1.
Remark: note that this argument fails when < 1 because in that case
is not a.s. finite.
c) For any t R+ , the question is > t ? can be answered based on
the observation of the paths of (Bs )0st and of the (deterministic) curve
(1 + s)0st up to the time t. Therefore is a stopping time.
Since is a stopping time and (Bt )tR+ is a martingale, the stopping
2
time theorem shows that (Bt
(t ))tR+ is also a martingale and
2
2
in particular its expectation IE[Bt
(t )] = IE[B0
(0 )] =
IE[B02 0] = 0 is constantly equal to 0 for all t. This shows that
h
i
2
2
IE[B2 ] = IE lim (Bt
(t )) = lim IE[(Bt
(t ))] = 0.
t
578
"

Next, we note that we have B2 = 1 + , hence
1 + IE[ ] = IE[1 + ] = IE[B2 ] = IE[ ] = 0,
i.e.
IE[ ] = 1/(1 ).
Remark: Note that this argument is valid whenever 1 and yields
IE[ ] = + when = 1, however it fails when > 1 because in that case
is not a.s. finite.
Exercise 9.2
a) By the stopping time theorem, for all n 0 we have
h
i
1 = IE e 2rBL n r(L n)
i
h
i
h
= IE e 2rBL n r(L n) 1{L <n} + IE e 2rBL n r(L n) 1{L n}
h
i
h
i
= IE e 2rBL rL 1{L <n} + IE e 2rBn rn 1{L n}
h
i

= eL 2r IE erL 1{L <n} + IE e 2rBn rn 1{L n} .
The first term above converges to
eL
2r

IE erL 1{L <} = eL 2r IE erL
as n tends to infinity, by dominated or monotone convergence and the fact

that r > 0. The second term can be bounded as
h
i
h
i
0 IE e 2rBn rn 1{L n} ern IE eL 2r 1{L n} ern eL 2r ,

which tends to 0 as n tends to infinity because r > 0. Therefore we have
h
i
1 = lim IE e 2rBL n r(L n) = eL 2r IE[erL ],

n
which yields IE[erL ] = eL 2r for any r 0. When r < 0 we could in

fact show that IE[erL ] = +.
b) In order to maximize

IE erL BL = L IE erL = LeL 2r
we differentiate
(LeL 2r ) = eL 2r L 2reL 2r = 0,
L
"
579
N. Privault
which yields the optimal level L = 1/ 2r.

This shows that when the value of r is large
the better strategy is to
opt for a small gain at the level L = 1/ 2r rather than to wait for a
longer time.
Exercise 9.3
a) Letting A0 = 0,
An+1 = An + IE[Mn+1 Mn | Fn ],
n 0,
and
Nn = Mn An ,
n N,
(16.47)
we have,
(i) for all n N,
IE[Nn+1 | Fn ] = IE[Mn+1 An+1 | Fn ]
= IE[Mn+1 An IE[Mn+1 Mn | Fn ] | Fn ]
= IE[Mn+1 An | Fn ] IE[IE[Mn+1 Mn | Fn ] | Fn ]
= IE[Mn+1 An | Fn ] IE[Mn+1 Mn | Fn ]
= IE[An | Fn ] + IE[Mn | Fn ]
= M n An
= Nn ,
hence (Nn )nN is a martingale with respect to (Fn )nN .
(ii) We have
An+1 An = IE[Mn+1 Mn | Fn ]
= IE[Mn+1 | Fn ] IE[Mn | Fn ]
= IE[Mn+1 | Fn ] Mn 0,
n N,
since (Mn )nN is a submartingale.

(iii) By induction we have
An = An1 + IE[Mn Mn1 | Fn1 ],
n 1,
which is Fn1 -measurable provided An is Fn1 -measurable, n 1.

(iv) This property is obtained by construction in (16.47).
b) For all bounded stopping times and such that a.s., we have
IE[M ] = IE[N ] + IE[A ]
IE[N ] + IE[A ]
580
"

= IE[N ] + IE[A ]
= IE[M ],
by (9.11), since (Mn )nN is a martingale and (An )nN is nondecreasing.
Exercise 9.4 American digital options.
a) The optimal strategy is as follows:
(i) if St K, then exercise immediately.
(ii) if St < K, then wait.
b) The optimal strategy is as follows:
(i) if St > K, then wait.
(ii) if St K, exercise immediately.
c) Based on the answers to Question a we set
CdAm (t, K) = 1,
0 t < T,
CdAm (T, x) = 0,
0 x < K.
and
d) Based on the answers to Question b, we set
PdAm (t, K) = 1,
0 t < T,
and
PdAm (T, x) = 0,
x > K.
e) Starting from St K, the maximum possible payoff is clearly reached

as soon as St hits the level K before the expiration date T , hence the
discounted optimal payoff of the option is er(K t) 1{K <T } .
f) From Relation (8.7) we find

a u
a u
e2a
,
P(a u) =
u
u
and by differentiation with respect to u this yields the probability density
function
fa (u) =
(au)2
a
P(a u) =
e 2u 1[0,) (u)
u
2u3
of the first hitting time of level a by Brownian motion with drift . Given
the relation
Su = St e(Bu Bt )
"
(ut)/2+(ut)
u t,
581

N. Privault
we find that the probability density function of the first hitting time of
level K after time t by (Su )u[t,) is given by
2
a
u 7 p
e(a(ut)) /(2(ut)) ,
2(u t)3
u t,
with = 1 (r 2 /2) and

a :=
K
1
log ,
given that St = x. Hence for x (0, K) we have, letting = T t,

CdAm (t, x) = IE[er(K t) 1{K <T } | St = x]
wT
2
a
=
er(st) p
e(a(st)) /(2(st)) ds
t
2(s t)3
w
2
a
=
ers
e(as) /(2s) ds
0
2s3

2 !
w log(K/x)
1
2
K
exp rs 2
=
r
s + log
ds
0 2s3
2 s
2
x
(r/2 1/2)(r/2 +1/2)
K
=
x

2 !
w log(K/x)
1
2
K
exp 2
r+
s + log
ds
0 2s3
2 s
2
x
2r/2 w
2
1 x w y2 /2
1
K
=
e
dy +
ey /2 dy
y
y
K
x
+
2
2

x
(r + 2 /2) + log(x/K)
=
K

x 2r/2 (r + 2 /2) + log(x/K)
+
,
0 < x < K,
K

where
y =

2
K
r+
+ log
,
2
x

1
and used the decomposition

K
1
2
K
1
2
K
=
log
r+
s + log
+
r+
s + log
.
x
2
2
x
2
2
x
We check that
CdAm (t, K) = () + () = 1,
582
"

and
2
CdAm (T, x) =
x 2r/
x
() +
() = 0,
K
K
x < K,
since = 0, which is consistent with the answers to Question c.

g) Starting from St K, the maximum possible payoff is clearly reached
as soon as St hits the level K before the expiration date T , hence the
discounted optimal payoff of the option is er(K t) 1{K <T } .
h) Using the notation and answer to Question f, for x > K we find, letting
= T t,
PdAm (t, x) = IE[er(K t) 1{K <T } | St = x]
w
(as)2
a
e 2s ds
=
ers
0
2s3

2 !
w log(x/K)
1
2
x
=
exp rs 2
r
s + log
ds
0 2s3
2 s
2
K
( r2 12 )( r2 + 12 )
K
=
x

2 !
w log(x/K)
1
2
x
r+
s + log
ds
exp 2
0 2s3
2 s
2
K
2
1 x w y2 /2
1 x 2r/ w y2 /2
=
e
dy +
e
dy
y+
2 K y
2 K

(r + 2 /2) log(x/K)
x
=
K

x 2r/2 (r + 2 /2) log(x/K)
+
,
x > K,
K

with
y =

2
x
r+
+ log
,
2
K

1
We check that
PdAm (t, K) = () + () = 1,
and
2
PdAm (T, x) =
x 2r/
x
() +
() = 0,
K
K
0 < x < K,
since = 0, which is consistent with the answers to Question c.

i) The call-put parity does not hold for American digital options since for
x (0, K) we have
"
583
N. Privault

x
(r + 2 /2) + log(x/K)
CdAm (t, x) + PdAm (t, x) = 1 +

K

x 2r/2 (r + 2 /2) + log(x/K)
,
K

while for x > K we find

x
(r + 2 /2) log(x/K)
CdAm (t, x) + PdAm (t, x) = 1 +

K

x 2r/2 (r + 2 /2) log(x/K)
.
K

Exercise 9.5 American forward Contracts.
a) For all stopping times such that t T we have
i
h

IE er( t) (K S )St = K IE er( t) St IE er( t) S St
= er( t) K St ,
since [t, T ] is bounded and (ert St )tR+ is a martingale, and the above
quantity is clearly maximized by taking = t. Hence we have

f (t, St ) =
sup
IE er( t) (K S )St = K St ,
t T
stopping time
and the optimal strategy is to exercise immediately (or avoiding to buy

the option) at time t.
b) Similarly we have

IE er( t) (S K)St = IE er( t) S St K IE er( t) St

= St K IE er( t) St ,
since [t, T ] is bounded and (ert St )tR+ is a martingale, and the above
quantity is clearly maximized by taking = T . Hence we have

f (t, St ) =
sup
IE er( t) (S K)St = St er(T t) K,
t T
stopping time
and the optimal strategy is to wait until the maturity time T in order to
exercise.
c) Concerning the perpetual American long forward contract, since u 7
er(ut) Su is a martingale, for all stopping times we have

IE er( t) (S K)St = IE er( t) S St K IE er( t) St
by Fatous Lemma.
584
"

= St K IE er( t) St
St ,
t 0.
On the other hand, for all fixed T > 0 we have

IE er(T t) (ST K)St = er(T t) IE ST St er(T t) IE K St
= St er(T t) K,
t [0, T ],
hence
sup
t T
stopping time

IE er( t) (S K)St (St er(T t) K),
T [t, ),
and letting T we get

sup
IE er( t) (S K)St lim (St er(T t) K)
T
t
stopping time
= St ,
hence we have
f (t, St ) =
sup
t T
stopping time

IE er( t) (S K)St = St ,
and the optimal strategy = + is to wait indefinitely.

Concerning the perpetual American short forward contract we have

f (t, St ) =
sup
IE er( t) (K S )St
t T
stopping time
sup
t T
stopping time

= fL (St ).
On the other hand, for = L we have
(K SL ) = (K L ) = (K L )
since 0 < L = 2Kr/(2r + 2 ) < K, hence

fL (St ) = IE er( t) (K SL )+ St

= IE er( t) (K SL )St

sup
IE er( t) (K S )St
t T
stopping time
= f (t, St ),
"
585
N. Privault
which shows that
f (t, St ) = fL (St ),
i.e. the perpetual American short forward contract has same price and
exercise strategy as the perpetual American put option.
Exercise 9.6
a) We have
Yt = ert (S0 ert+Bt
t/2 2r/ 2
t /+rt
2r/ 2 rt2r 2 t/ 2 +2r B
= S0
t /(2r/)2 t/2
2r/ 2 2r B
= S0
and
Zt = ert St = S0 eBt
t/2
which are both martingales under P because they are standard geometric
Brownian motions with respective volatilities and 2r/.
b) Since Yt and Zt are both martingales and L is a stopping time we have
2r/ 2
S0
= IE [Y0 ]
= IE [YL ]
2
= IE [erL S2r/
]
L
2
= IE [erL L2r/ ]
2
= L2r/ IE [erL ],
hence
IE [erL ] = (x/L)2r/
if S0 = x L (note that in this case YL t remains bounded by L2r/ ),

and
S0 = IE [Z0 ] = IE [ZL ] = IE [erL SL ] = IE [erL L] = L IE [erL ],
hence
IE [erL ] = x/L
if S0 = x L (note that in this case ZL t remains bounded by L).
c) We find

IE erL (K SL )S0 = x = (K L) IE erL S0 = x
586
"
K L
x L ,
0 < x L,
2r/2
(K L) x
,
L
x L.
(16.48)
d) By differentiating

IE erL (K SL )S0 = x
L

2
2r K
(x/L)2r/
1
,
2 L
=
Kx
2,
L
0 < L < x,
L > x,
and check that the minimum occurs for L = x.

e) The value L = x shows that the optimal strategy for the American finite
expiration short forward contract is to exercize immediately starting from
S0 = x, which is consistent with the result of Exercise 9.5-(a), since given
any stopping time upper bounded by T we have
IE[er (K S )] = K IE[er ] IE[er S ] = K IE[er ] S0 K S0 .
Exercise 9.7
a) The option payoff equals ( St )p if St L.
b) We have
i
h

fL (St ) = IE er(L t) (( SL )+ )p St
i
h

= IE er(L t) (( L)+ )p St
i
h

= ( L)p IE er(L t) St .
c) We have

h
i

fL (x) = IE er(L t) ( SL )+ St = x
( x)p ,
0 < x L,
=
2

( L)p x 2r/ , x L.
L
d) By differentiating
"
d
dx (
(16.49)
x)p = p( x)p1 we find
587
N. Privault
2
fL0 (L ) =
2r
(L )2r/ 1
( L )p
= p( L )p1 ,
2
(L )2r/2
i.e.
2r
( L ) = pL ,
2
or
L =
2r
< .
2r + p 2
e) By (16.49) the price can be computed as
( St )p ,
0 < St L ,
p
2r/2
f (t, St ) = fL (St ) =
p 2
2r + p 2 St
,
St L ,
2
2r + p
2r
using (9.12) as in the proof of Proposition 9.4, since

u 7 eru fL (Su ),
u t,
is a nonnegative supermartingale.
Exercise 9.8
a) The payoff will be (St )p .
b) We have
i
h

fL (St ) = E er(L t) ( (SL )p )St
i
h

= E er(L t) ( Lp )St

h
i

= ( Lp )E er(L t) St .
c) We have

h
i

fL (x) = E er(L t) ( (SL )p )St = x
xp ,
0 < x L,
2

( Lp ) x 2r/ , x L.
L
d) We have
2
fL0 (L ) =
(L )2r/ 1
2r
( (L )p )
= p(L )p1 ,
2
(L )2r/2
588
"

i.e.
2r
( (L )p ) = p(L )p ,
2
or
L =
2r
2r + p 2
1/p
< ()1/p .
(16.50)
Remark: We may also compute L by maximizing L 7 fL (x) for all

fixed x. The derivative fL (x)/L can be computed as
2r/2 !
L
fL (x)
p
=
( L )
L
L
x
2r/2
2r/2
L
2r
L
= pLp1
+ 2 L1 ( Lp )
,
x
x
and equating fL (x)/L to 0 at L = L yields
p(L )p1 +
2r 1
(L ) ( (L )p ) = 0,
2
which recovers (16.50).

e) We have
(St )p ,
0 < St L ,
2
fL (St ) =
(S )2r/
( (L )p ) t 2r/2 ,
St L
(L )
(St )p ,
0 < St L ,
=
2
p(S )2r/2 (L )p+2r/2 ,

St L ,
t
2r
(St )p ,
0 < St L ,

2r/(p2 )
=
p 2
2r + p 2 Stp
< ,
St L ,
2r + p 2
2r
however we cannot conclude as in Exercise 9.7-(e) since the process

u 7 eru fL (Su ),
u t,
does not remain nonnegative when p > 1, so that (9.12) cannot be applied
as in the proof of Proposition 9.4.
Exercise 9.9
"
589
N. Privault
a) We have that

Zt :=
St
S0
e(ra)t+
t/22 2 t/2
= eBt
2 t/2
t R+ ,
is a geometric Brownian motion without drift under the risk-neutral probability measure P , hence it is a martingale.
b) By the stopping time theorem we have
IE [ZL ] = IE [Z0 ] = 1,
which rewrites as
"
IE
SL
S0
#
((ra) 2 /2+2 2 /2)L
= 1,
or, given the relation SL = L,

L
S0
h
i
2
2 2
IE e((ra) /2+ /2)L = 1,
i.e.

IE erL =
S0
L

,
provided we choose such that

((r a) 2 /2 + 2 2 /2) = r,
(16.51)
i.e.
0 = 2 2 /2 + (r a 2 /2) r.
This equation admits two solutions
p
(r a 2 /2) (r a 2 /2)2 + 4r 2 /2
=
,
2
and we choose the negative solution
p
(r a 2 /2) (r a 2 /2)2 + 4r 2 /2
=
2
since S0 /L = x/L > 1 and the expectation IE [erL ] < 1 is lower than 1
as r 0.
c) Noting that L = 0 if S0 L, for all L (0, K) we have

h
i

IE erL (K SL )+ S0 = x
590
"
0 < x L,
K x,
=

i
h
E erL (K L)+ S0 = x , x L.
0 < x L,
K x,
=

i
h
(K L)E erL S0 = x , x L.
K x,
=
2 /2)2 +4r 2 /2
(ra2 /2) (ra
2
(K L) x
,
L
0 < x L,
x L.
d) In order to compute L we observe that, geometrically, the slope of fL (x)

at x = L is equal to 1, i.e.
fL0 (L ) = (K L )
(L )1
= 1,
(L )
or
(K L ) = L ,
or L =
K < K.
1
e) For x L we have
x
L
!

x
= K
K
1
1 K

K
x( 1)
=
1
K

K
x
1
=
1
K

1
x
1
=
K
x 1 K
=
.
K
fL (x) = (K L )
(16.52)
f) Let us check that the relation

fL (x) (K x)+
(16.53)
holds. For all x K we have
"
591
N. Privault
fL (x) (K x) =
x 1
K
=K
K
+xK
1
!

x 1 1
x
+
1 .
K
1 K
Hence it suffices to take K = 1 and to show that for all

L =
x1
1
we have
x
1
0.
fL (x) (1 x) =

+x1
Equality to 0 holds for x = /( 1). By differentiation of this relation

we get

1
1
+1
1
1

1
= x1
+1
0,
fL0 (x) (1 x)0 = x1
hence the function fL (x) (1 x) is nondecreasing and the inequality

holds throughout the interval [/( 1), K].
On the other hand, using (16.51) it can be checked by hand that fL given
by (16.52) satisfies the equality
1
(r a)xfL0 (x) + 2 x2 fL00 (x) = rfL (x)
2
for x L =
(16.54)
K. In case
1
0 x L =
K < K,
1
we have
fL (x) = K x = (K x)+ ,
hence the relation

1
rfL (x) (r a)xfL0 (x) 2 x2 fL00 (x) (fL (x) (K x)+ ) = 0
2
592
"

always holds. On the other hand, in that case we also have
1
(r a)xfL0 (x) + 2 x2 fL00 (x) = (r a)x,
2
and to conclude we need to show that
1
(r a)xfL0 (x) + 2 x2 fL00 (x) rfL (x) = r(K x),
2
(16.55)
which is true if
ax rK.
Indeed by (16.51) we have
(r a) = r + ( 1) 2 /2
r,
hence
a
r,
1
since < 0, which yields

ax aL a
K rK.
1
g) By Itos formula and the relation

t
dSt = (r a)St dt + St dB
we have
d(fL (St )) = rert fL (St )dt + ert dfL (St )
1
= rert fL (St )dt + ert fL0 (St )dSt + ert 2 St2 fL00 (St )
2

1
= ert rfL (St ) + (r a)St fL0 (St ) + 2 St2 fL00 (St ) dt
2
t ,
+ert St f 0 (St )dB
L
and from Equations (16.54) and (16.55) we have

1
(r a)xfL0 (x) + 2 x2 fL00 (x) rfL (x),
2
hence
t 7 ert fL (St )
is a supermartingale.
h) By the supermartingale property of
"
593
N. Privault
t 7 ert fL (St ),
for all stopping times we have
i
i
h
h

fL (S0 ) IE er fL (S )S0 IE er (K S )+ S0 ,
by (16.53), hence
fL (S0 )
sup
stopping time
i
h

IE er (K S )+ S0 .
(16.56)
i) The stopped process

t 7 ertL fL (StL )
is a martingale since it has vanishing drift up to time L by (16.54),
and it is constant after time L , hence by the martingale stopping time
Theorem (9.1) we find
i
h

fL (S0 ) = IE er fL (SL )S0
i
h

= IE er fL (L )S0
i
h

= IE er (K SL )+ S0
i
h

sup
IE er (K S )+ S0 .
stopping time
j) By combining the above results and conditioning at time t instead of time

0 we deduce that
i
h

K,
K St ,
0 < St
=
1

St
,
St
K,
1
for all t R+ , where
L = inf{u t : Su L}.
We note that the perpetual put option price does not depend on the value
of t 0.
Exercise 9.10
594
"

a) By the definition (9.34) of S1 (t) and S2 (t) we have

S1 (t)
Zt = ert S2 (t)
S2 (t)
= ert S1 (t) S2 (t)1
2
= S1 (0) S2 (0)1 e(1 +(1)2 )Wt 2 t/2 ,

which is a martingale when
22 = (1 + (1 )2 )2 ,
i.e.
1 + (1 )2 = 2 ,
which yields either = 0 or
=
22
> 1,
2 1
since 0 1 < 2 .
b) We have
IE[erL (S1 (L ) S2 (L ))+ ] = IE[erL (LS2 (L ) S2 (L ))+ ]
= (L 1)+ IE[erL S2 (L )].
(16.57)
c) Since L t is a bounded stopping time we can write

S1 (L t)
S1 (0)
= IE er(L t) S2 (L t)
(16.58)
S2 (0)
S2 (0)
S2 (L t)

S1 (L )
S1 (t)
= IE erL S2 (L )
1{L <t} + IE ert S2 (t)
1{L >t}
S2 (L )
S2 (t)
We have
ert S2 (t)
S1 (t)
S2 (t)
1{L >t} ert S2 (t)L 1{L >t} ert S2 (t)L ,
hence by a uniform integrability argument,

S1 (t)
lim IE ert S2 (t)
1{L >t} = 0,
t
S2 (t)
and letting t go to infinity in (16.58) shows that

S1 (0)
S1 (L )
S2 (0)
= IE erL S2 (L )
= L IE erL S2 (L ) ,
S2 (0)
S2 (L )
"
595
N. Privault
since S1 (L )/S2 (L ) = L/L = 1. The conclusion
IE[erL (S1 (L ) S2 (L ))+ ] = (L 1)+ L S2 (0)
S1 (0)
S2 (0)

(16.59)
then follows by an application of (16.57).

d) In order to maximize (16.59) as a function of L we consider the derivative
L1
1
= (L 1)L1 = 0,
L L
L
which vanishes for
L =
,
1
and we substitute L in (16.59) with the value of L .

e) In addition to r = 22 /2 it is sufficient to let S1 (0) = and 1 = 0 which
yields = 2, L = 2, and we find
sup
IE[er ( S2 ( ))+ ] =
stopping time
1 2
,
S2 (0) 2
which coincides with the result of Proposition 9.4.

Exercise 9.11
a) It suffices to check the sign of the quantity
( 1)( + 2r/ 2 ),
(16.60)
2
in (9.36), which is positive when (, 2r/ ] [1, ), and negative

when 2r/ 2 1.
b) The sign of (16.60) is positive when (, 1] [2r/ 2 , ), and
negative when 1 2r/ 2 .
c) By the stopping time theorem, for any n 0 we have
h
i
()
x = IE er(L n) ZL n | S0 = x
h
i
h
i
= IE Z()
1{L <n} | S0 = x + ern IE Zn() 1{L >n} | S0 = x
L

IE erL (SL ) 1{L <n} | S0 = x

= L IE erL 1{L <n} | S0 = x ,

and by letting n to infinity and applying monotone convergence this yields
2
x (x/L)min(1,2r/ ) , x L,

IE erL 1{L <} | S0 = x
2
L
(x/L)max(1,2r/ ) , 0 < x L.
596
"

d) We note that P (L < ) = 1 by (9.14), hence if 2 /2 r < 0 we have

IE erL (K SL )+ 1{L <} | S0 = x
2r/ 2
, x L,
(K L)(x/L)

= (K L) IE erL | S0 = x
(K L)x/L, 0 < x L.
Similarly, if r 2 /2 we have

IE erL (K SL )+ 1{L <} | S0 = x

(K L)(x/L), x L,
= (K L) IE erL | S0 = x
2
(K L)(x/L)2r/ , 0 < x L.
2
e) This follows by noting that (K L)(L/x)2r/ and (K L)(x/L) increase

to when L tends to zero.
f) If 2 /2 r < 0 we have

IE erL (SL K)+ 1{L <} | S0 = x
2r/ 2
, x L,
rL
(L K)(x/L)
= (L K) IE e
1{L <} | S0 = x
(L K)x/L, 0 < x L.
If r 2 /2 we have

IE erL (SL K)+ 1{L <} | S0 = x
rL
(L K)(x/L), x L,
= (L K) IE e
1{L <} | S0 = x
2
(L K)(x/L)2r/ , 0 < x L.
g) This follows by noting that for fixed x > 0, (L K)x/L and (L

2
K)(x/L)2r/ both increase to x when L tends to infinity.
Chapter 10
Exercise 10.1
a) We have

Xt
Nt
X0 ()Bt (2 2 )t/2
d e
N0
2
2
2
2
X0
X0
=
( )e()Bt ( )t/2 dBt +
( )2 e()Bt ( )t/2 dt
N0
2N0
t = d
dX
"
597
N. Privault
2
2
X0 2
( 2 )e()Bt ( )t/2 dt
2N0
Xt 2
Xt
Xt
=
( 2 )dt +
( )dBt +
( )2 dt
2Nt
Nt
2Nt
Xt
Xt
= ( )dt +
( )dBt
Nt
Nt
Xt
( )(dBt dt)
=
Nt
Xt

= ( ) dB
t = ( )Xt dBt ,
Nt
t = dBt dt is a standard Brownian motion under P.

where dB
b) By change of numeraire we have

N0 (XT NT )+ = N0 IE
(X
T )+ .
erT IE[(XT NT )+ ] = IE
NT
t is a driftless geometric Brownian
Next, by the result of Question (a), X
hence we have
motion with volatility under P,
!
!
/)
/)
T
log(X
T
X
T )+ ] = X
0 log(X
0
0
+

,
IE[(
2
2
T
by the Black-Scholes formula with zero interest rate and volatility param 0 we
eter
= . By multiplication by N0 and the relation X0 = N0 X
conclude to (10.30), i.e.

(X
T )+
erT IE (XT NT )+ = N0 IE
0 (d+ ) N0 (d )
= N0 X
= X0 (d+ ) N0 (d ).
c) We have
= .
Exercise 10.2 Bond options.
a) Itos formula yields

P (t, S)
P (t, S) S
d
=
( (t) T (t))(dWt T (t)dt)
P (t, T )
P (t, T )
P (t, S) S
t,
=
( (t) T (t))dW
P (t, T )
(16.61)
by the Girsanov
t )tR is a standard Brownian motion under P
where (W
+
theorem.
b) From (16.61) we have
598
"

w

wt
t
P (t, S)
P (0, S)
s 1
=
exp
( S (s) T (s))dW
| S (s) T (s)|2 ds ,
0
P (t, T )
P (0, T )
2 0
hence
w

wu
u
P (u, S)
P (t, S)
s 1
=
exp
( S (s) T (s))dW
| S (s) T (s)|2 ds ,
t
P (u, T )
P (t, T )
2 t
t [0, u], and for u = T this yields
w

wT
T
P (t, S)
s 1
exp
( S (s) T (s))dW
| S (s) T (s)|2 ds ,
P (T, S) =
t
P (t, T )
2 t
denote the forward measure associated to the
since P (T, T ) = 1. Let P
numeraire
Nt := P (t, T ),
0 t T.
c) For all S T > 0 we have
i
h rT

IE e t rs ds (P (T, S) K)+ Ft
"

+ #
P (t, S)
1wT S

= P (t, T )IE
exp X
| (s) T (s)|2 ds K
Ft
P (t, T )
2 t

+
eX+m(t,T,S) K Ft ,
= P (t, T )IE
where X is a centered Gaussian random variable with variance
wT
v 2 (t, T, S) =
| S (s) T (s)|2 ds
t
given Ft , and
1
P (t, S)
.
m(t, T, S) = v 2 (t, T, S) + log
2
P (t, T )
Recall that when X is a centered Gaussian random variable with variance
v 2 , the expectation of (em+X K)+ is given, as in the standard BlackScholes formula, by
IE[(em+X K)+ ] = em+
v2
2
where
(z) =
"
(v + (m log K)/v) K((m log K)/v),
wz
ey
/2
dy
,
2
z R,
599
N. Privault
denotes the Gaussian cumulative distribution function and for simplicity
of notation we dropped the indices t, T, S in m(t, T, S) and v 2 (t, T, S).
Consequently we have
i
h rT

IE e t rs ds (P (T, S) K)+ Ft

v
1
P (t, S)
v
1
P (t, S)
= P (t, S)
+ log
KP (t, T ) + log
.
2 v
KP (t, T )
2 v
KP (t, T )
d) The self-financing hedging strategy that hedges the bond option is obtained by holding a (possibly fractional) quantity

1
P (t, S)
v
+ log
2 v
KP (t, T )
of the bond with maturity S, and by shorting a quantity

v
1
P (t, S)
K + log
2 v
KP (t, T )
of the bond with maturity T .
Exercise 10.3
a) The process
ert S2 (t) = S2 (0)e2 Wt +(r)t
is a martingale if
r=
1 2
.
2 2
b) We note that
2
ert Xt = ert e(r)t1 t/2 S1 (t)

= ert e
(22 12 )t/2
t12 t/2
=e
S1 (t)
S1 (t)
2
= S1 (0)et1 t/2 e1 Wt +t
= S1 (0)e
1 Wt 12 t/2
is a martingale, where
2
Xt = e(r)t1 t/2 S1 (t) = e(2 1 )t/2 S1 (t).

c) By (10.32) we have
Xt
X(t)
=
Nt
600
"

2
= e(2 1 )t/2
S1 (t)
S2 (t)
S1 (0) (22 12 )t/2+(1 2 )Wt

e
S2 (0)
S1 (0) (22 12 )t/2+(1 2 )W
t +2 (1 2 )t
e
=
S2 (0)
S1 (0) (1 2 )W
t +2 1 t( 2 + 2 )t/2
2
1
e
=
S2 (0)
S1 (0) (1 2 )W
t (1 2 )2 t/2
,
=
e
S2 (0)
=
where
t := Wt 2 t
W
defined by
is a standard Brownian motion under the forward measure P
rT
dP
NT
= e 0 rs ds
dP
N0
rT S2 (T )
=e
S2 (0)
= erT e2 WT +T
= e2 WT +(r)T
2
= e2 WT 2 t/2 .
2
d) Given that Xt = e(2 1 )t/2 S1 (t) and X(t)

= Xt /Nt = Xt /S2 (t), we have
2
erT IE[(S1 (T ) S2 (T ))+ ] = erT IE[(e(2 1 )T /2 XT S2 (T ))+ ]

rT (22 12 )T /2
(22 12 )T /2
IE[(XT e
S2 (T ))+ ]
2
2
X
T e(2 1 )T /2 )+ ]
= S2 (0)e
IE[(
2
2
T (1 2 )2 T /2
(22 12 )T /2
0 e(1 2 )W
= S2 (0)e
IE[(X
e(2 1 )T /2 )+ ]

2
2
0 0+ (T, X
0 ) e(22 12 )T /2 0 (T, X
0)
= S2 (0)e(2 1 )T /2 X
=e
(22 12 )T /2
0 0+ (T, X
0)
= S2 (0)e(2 1 )T /2 X
S2 (0)e
(22 12 )T /2
2
2
0)
e(2 1 )T /2 0 (T, X
2
2
0 ) S2 (0)0 (T, X
0)
= e(2 1 )T /2 X0 0+ (T, X
0
0
= S1 (0)+ (T, X0 ) S2 (0) (T, X0 ),
where
0+ (T, x) =
"
log(x/)
( 2 )2 (22 12 )
+ 1
T
2|1 2 |
|1 2 | T
601
N. Privault

log(x/)
+ 1 T ,
|1 2 | T
1 > 2 ,

log(x/)
1 T ,
|1 2 | T
1 < 2 ,
and
log(x/)
( 2 )2 + (22 12 )
1
T
2|1 2 |
|1 2 | T

log(x/)
+ 2 T , 1 > 2 ,
|1 2 | T
=

log(x/)
2 T , 1 < 2 ,
|1 2 | T
0 (T, x) =
if 1 6= 2 . In case 1 = 2 we find
erT IE[(S1 (T ) S2 (T ))+ ] = erT IE[S1 (T )(1 S2 (0)/S1 (0))+ ]
= (1 S2 (0)/S1 (0))+ erT IE[S1 (T )]
= (S1 (0) S2 (0))1{S1 (0)>S2 (0)} .
Exercise 10.4
a) It suffices to check that the definition of (WtN )tR+ implies the correlation
identity dWtS dWtN = dt by Itos calculus.
b) We let
q
t =
and
dWtX =
(tS )2 2tR tS + (tR )2
p
tS tN
N
dWtS 1 2 t dWt ,
t R+ ,
which defines a standard Brownian motion under P due to the definition

of
t .
Exercise 10.5
p
a) We have
= ( S )2 2 R S + ( R )2 .
t = ert Xt = e(ar)t St /Rt , t R+ , we have
b) Letting X
"
+ #
h
+ i
ST

IE
Ft = eaT IE XT eaT Ft
RT

+
T e(ar)T Ft
= e(ar)T IE X
602
"

2 /2)(T t)
1
St
t (r a +
X
+
log
Rt
T t
T t

1
St
(r a
2 /2)(T t)
(ar)T
+
log
e
Rt
T t
T t

(r a +
2 /2)(T t)
St (ra)(T t)
1
St
=
+
log
Rt
Rt
T t
T t

(r a
2 /2)(T t)
1
St
+
log
,
Rt
T t
T t
= e(ar)T
hence the price of the quanto option is

"
+ #
ST

er(T t) IE
Ft
RT

1
St
(r a +
2 /2)(T t)
St a(T t)
+
log
e
=
Rt
Rt
T t
T t

1
St
(r a
2 /2)(T t)
r(T t)
+
log
.
e
Rt
T t
T t
Chapter 11
w

T2
P (0, T2 ) = exp
f (t, s)ds = er1 T1 r2 (T2 T1 ) ,
0
t [0, T2 ],
and
w

T2
P (T1 , T2 ) = exp
f (t, s)ds = er2 (T2 T1 ) ,
T1
t [0, T2 ],
from which we deduce

r2 =
1
log P (T1 , T2 ),
T2 T1
and
T2 T1
1
log P (0, T2 )
T1
T1
log P (T1 , T2 )
1
=
log P (0, T2 )
T1
T1
1
P (0, T2 )
= log
.
T1
P (T1 , T2 )
r1 = r2
"
603
N. Privault
Exercise 11.2
a) We have rt = r0 + at + Bt , and
F (t, rt ) = F (t, r0 + at + Bt ),
hence by Proposition 11.2 the PDE satisfied by F (t, x) is
xF (t, x) +
F
F
1 2F
(t, x) + a
(t, x) + 2 2 (t, x) = 0,
t
x
2 x
(16.62)
with terminal condition F (T, x) = 1.

b) We have rt = r0 + at + Bt and

w

T

F (t, rt ) = IE exp
rs ds Ft
t

wT
wT

= IE exp r0 (T t) a
sds
Bs ds Ft
t
t

wT
2
2

= IE er0 (T t)a(T t )/2 exp (T t)Bt
(T s)dBs Ft
t

wT
2
2

= er0 (T t)a(T t )/2(T t)Bt IE exp
(T s)dBs Ft
t

wT
= er0 (T t)a(T t)(T +t)/2(T t)Bt IE exp
(T s)dBs
t

2 wT
(T s)2 ds
= exp (T t)rt a(T t)2 /2 +
2 t

= exp (T t)rt a(T t)2 /2 + 2 (T t)3 /6 ,

hence F (t, x) = exp (T t)x a(T t)2 /2 + 2 (T t)3 /6 .
Note that the PDE (16.62) can also be solved by looking for a solution of
the form F (t, x) = eA(T t)+xC(T t) , in which case one would find A(s) =
as2 /2 + 2 s3 /6 and C(s) = s.
c) We check that the function F (t, x) of Question b satisfies the PDE (16.62)
of Question a, since F (T, x) = 1 and

2
(T t)2 F (t, x) a(T t)F (t, x)
xF (t, x) + x + a(T t)
2
1 2
+ (T t)2 F (t, x) = 0.
2
d) We have
604
"

1
f (t, T, S) =
(log P (t, T ) log P (t, S))
ST

1
2
2
=
(T t)rt +
(T t)3 (S t)rt +
(S t)3
ST
6
6
1 2
3
3
((T t) (S t) ).
= rt +
ST 6
e) We have
f (t, T ) =
2
log P (t, T ) = rt
(T t)2 .
T
2
f) We have
dt f (t, T ) = 2 (T t)dt + adt + dBt .
g) The HJM condition (11.37) is satisfied since the drift of dt f (t, T ) equals
rT
t ds.
Exercise 11.3
rt
a) The process e 0 rs ds F (t, rt ) is a martingale and we have

rt

d e 0 rs ds F (t, rt )

rt
F
F
2 2 F
2
= e 0 rs ds rt F (t, rt )dt +
(t, rt )dt +
(t, rt )drt +
(t,
r
)(dr
)
t
t
t
x
2 x2

rt
F
F
= e 0 rs ds rt F (t, rt )dt +
(t, rt )dt +
(t, rt )(art dt + rt dBt )
t
x
2 2
r
0t rs ds F
+rt e
(t, rt )dt,
2 x2
hence
xF (t, x) +
F
2 2 F
F
(t, x) ax
(t, x) + x
(t, x) = 0.
t
x
2 x2
(16.63)
b) Plugging F (t, x) = eA(T t)+xC(T t) into the PDE (16.63) shows that

2 x 2
eA(T t)+xC(T t) x A0 (T t) xC 0 (T t) axC(T t) +
C (T t)
2
= 0,
hence
"
2 2
1 C 0 (T t) aC(T t) +
C (T t) = 0,
2
0
A (T t) = 0.
605
N. Privault
Remark: The initial condition A(0) = 0 shows that A(s) = 1, and it can
be shown from the condition C(0) = 0 that
C(T t) =
with =
2(1 e(T t) )
,
2 + (a + )(e(T t) 1)
t [0, T ],
a2 + 2 2 , see e.g. Eq. (3.25) page 66 of [9].
Exercise 11.4
a) We have (t, s) = s and we check that
(t, T ) = 2 T (T 2 t2 )/2 = T
wT
t
sds = (t, T )
wT
t
(t, s)ds.
b) We have
f (t, T ) = f (0, T ) +
wt
ds f (s, T )
wt
wt 2
= f (0, T ) +
T (T s2 )ds + T
dBs
0
0
2
w
w
wt
2
2
t
3 t
T
ds
T
s2 ds + T
dBs
= f (0, T ) +
0
0
0
2
2
2 3
2
3
= f (0, T ) + T t/2 T t /6 + T Bt
0
2
= f (0, T ) + 2 T t(T 2 /2 t2 /6) + T Bt .

c) We have
rt = f (t, t) = f (0, t) + 2 t2 (t2 /2 t2 /6) + tBt = f (0, t) + 2 t4 /3 + tBt .
d) We have
4 2 3
t dt + Bt dt + tdBt
3
1
= 4 2 t3 /3dt + (rt f (0, t) 2 t4 /3)dt + tdBt
t
1
= (rt f (0, t) + 2 t4 )dt + tdBt
t
1
= 2 t3 dt + (rt f (0, t))dt + tdBt ,
t
drt =
which is a Hull-White type short term interest rate model with the timedependent deterministic coefficients (t) = 2 t3 , (t) = 1/t and (t) = t.
Note that t 7 f (0, t) is the initial rate curve data.
Exercise 11.5
a) We have
606
"
P (t, T ) = P (s, T ) exp

0 s t T.
b) We have

d e
rt
0
rs ds
w
ru du +
wt
s
uT dBu

1wt T 2
|u | du ,
2 s

rt
P (t, T ) = e 0 rs ds tT P (t, T )dBt ,
which gives a martingale after integration, from the properties of the It

o
integral.
c) By the martingale property of the previous question we have
i
i
h rT
h
rT

IE e 0 rs ds Ft = IE P (T, T )e 0 rs ds Ft
= P (t, T )e
rt
0
rs ds
0 t T.
d) By the previous question we have

i
h rT

IE e 0 rs ds Ft
i
h rt
rT

= IE e 0 rs ds e 0 rs ds Ft
i
h rT

= IE e t rs ds Ft ,
0 t T,
P (t, T ) = e
rt
since e
e) We have
rs ds
rt
0
rs ds
is an Ft -measurable random variable.
w

t
P (t, S)
P (s, S)
1wt S 2
=
exp
(uS uT )dBu
(|u | |uT |2 )du
s
P (t, T )
P (s, T )
2 s
w

t
P (s, S)
1wt S
S
T
T
=
exp
(u u )dBu
(u uT )2 du ,
s
P (s, T )
2 s
0 t T , hence letting s = t and t = T in the above expression we have

w
T
P (t, S)
1wT S
P (T, S) =
(sS sT )dBsT
(s sT )2 ds .
exp
t
P (t, T )
2 t
f) We have
h
i
+
P (t, T ) IET (P (T, S) )
"
+ #
P (t, S) r T (sS sT )dBsT 1 r T (sS sT )2 ds
2 t
= P (t, T ) IET
et
P (t, T )
= P (t, T ) IE[(eX )+ | Ft ]

2
vt
1
= P (t, T )emt +vt /2
+ (mt + vt2 /2 log )
2
vt
"
607
N. Privault

vt
1
P (t, T ) + (mt + vt2 /2 log ) ,
2
vt
with
mt = log(P (t, S)/P (t, T ))
and
vt2 =
wT
t
1wT S
(s sT )2 ds
2 t
(sS sT )2 ds,
i.e.
h
i
+
P (t, T ) IET (P (T, S) )

1
vt
1
vt
P (t, S)
P (t, S)
+ log
P (t, T ) + log
.
= P (t, S)
2
vt
P (t, T )
2
vt
P (t, T )
Exercise 11.6
T
a) We check that P (T, T ) = eXT = 1.

b) We have

1
XtS XtT (S T )
ST

wt 1
wt 1
1
(S t)
dBs (T t)
dBs
0 Ss
0 T s
ST

T t
1 wt St
dBs
ST 0 Ss T s
1 w t (T s)(S t) (T t)(S s)
dBs
ST 0
(S s)(T s)
w
t
(s t)(S T )
dBs .
+
S T 0 (S s)(T s)
f (t, T, S) =
=
=
=
=
c) We have
f (t, T ) =
wt
0
ts
dBs .
(T s)2
d) We note that
lim f (t, T ) =
T &t
wt
0
1
dBs
ts
does not exist in L2 ().

e) By Itos calculus we have
dP (t, T )
1
XtT
= dBt + 2 dt + dt
dt
P (t, T )
2
T t
608
"

1
log P (t, T )
= dBt + 2 dt
dt,
2
T t
t [0, T ].
f) Let
XtS
1
rtS = + 2
2
St
wt 1
1
= + 2
dBs ,
0 Ss
2
and apply the result of Exercise 11.11.9-(d).
g) We have

2
dPT
IE
Ft = eBt t/2 .
dP
t := Bt t is a standard Browh) By the Girsanov theorem, the process B
nian motion under PT .
i) We have
wT 1
dBs
log P (T, S) = (S T ) + (S T )
0 Ss
wT 1
wt 1
dBs + (S T )
dBs
= (S T ) + (S T )
t Ss
0 Ss
wT 1
ST
=
log P (t, S) + (S T )
dBs
t Ss
St
w
wT 1
T
ST
1
s + 2 (S T )
=
log P (t, S) + (S T )
dB
ds
t Ss
t Ss
St
w
T
ST
1
s + 2 (S T ) log S t ,
=
log P (t, S) + (S T )
dB
t Ss
St
ST
0 < T < S.
j) We have
i
h

P (t, T ) IET (P (T, S) K)+ Ft
= P (t, T ) IE[(eX )+ | Ft ]

2
1
vt
+ (mt + vt2 /2 log )
= P (t, T )emt +vt /2
2
vt

vt
1
P (t, T ) + (mt + vt2 /2 log )
2
vt

1
1
mt +vt2 /2
= P (t, T )e
vt + (mt log ) P (t, T )
(mt log ) ,
vt
vt
with
mt =
"
ST
St
log P (t, S) + 2 (S T ) log
St
ST
609
N. Privault
and
wT
1
ds
(S s)2

1
1
= 2 (S T )2
ST
St
(T t)
2
,
= (S T )
(S t)
vt2 = 2 (S T )2
hence
i
h

P (t, T ) IET (P (T, S) K)+ Ft
2 (ST )

2
St
(ST )(St)
evt /2
= P (t, T ) (P (t, S))
ST
2 (ST ) !!
(ST )(St)
1
St
(P (t, S))
vt + log
vt
ST
2 (ST ) !!
(ST )(St)
St
1
(P (t, S))
.
P (t, T )
log
vt
ST
Exercise 11.7 From Proposition 11.2 the bond pricing PDE is
F
F
1
2F
(t, x) = xF (t, x) ( x)
(t, x) 2 x2 2 (t, x)
t
x
2
x
F (T, x) = 1.
Let us search for a solution of the form
F (t, x) = eA(T t)xB(T t) ,
with A(0) = B(0) = 0, which implies
0
A (s) = 0
B 0 (s) + B(s) + 12 2 B 2 (s) = 1.
hence in particular A(s) = 0, s R, and B(s) solves a Riccatti equation,

whose solution is easily checked to be
B(s) =
with =
2(es 1)
,
2 + ( + )(es 1)
p
2 + 2 2 .
610
"

Exercise 11.8
a) We have
f (t, x) = f (0, x) +
wt
0
x2 ds +
wt
0
ds B(s, x) = r + tx2 + B(t, x).
b) We have
rt = f (t, 0) = r + B(t, 0) = r.
c) We have
w

T
P (t, T ) = exp
f (t, s)ds
t

w T t
w T t
= exp r(T t) t
s2 ds
B(t, x)dx
0
0

w T t
3
= exp r(T t) t(T t)
B(t, x)dx ,
0
3
t [0, T ].
d) We have
IE
"
w
T t
2 # w
T t w T t
B(t, x)dx
=
IE[B(t, x)B(t, y)]dxdy
0
=t
w T t w T t
= 2t
w T t w y
0
min(x, y)dxdy
xdxdy =
1
t(T t)3 .
3
e) We have

w T t
IE[P (t, T )] = IE exp r(T t) t(T t)3

B(t, x)dx
0
3

2
3
= exp r(T t) t(T t) +
t(T t)3 ,
t [0, T ].
3
6
f) We need to take = 2 /2.
Remark: In order to derive an analog of the HJM absence of arbitrage condition in this stochastic string model, one would have to check whether the
discounted bond price ert P (t, T ) can be a martingale by doing stochastic
calculus with respect to the Brownian sheet B(t, x).
g) We have

w

T
IE exp
rs ds (P (T, S) K)+
0
"
611
N. Privault

+ #
w ST
exp r(S T ) T (S T )3 +
B(T, x)dx K
0
3
m+X

+
IE (xe
K) ,
= erT IE
= erT
"
where x = er(ST ) , m = T (S T )3 /3, and

X=
w ST
0
B(T, x)dx ' N (0, 2 t(T t)3 /3).
Given the relation = 2 /2 this yields

w

T
IE exp
rs ds (P (T, S) K)+
0
= erS
T (S T )3 /12 +
log(er(ST ) /K)
p
T (S T )3 /3
log(er(ST ) /K)
Ke
T (S
+ p
T (S T )3 /3
!
p
log(er(ST ) /K)
= P (0, S) T (S T )3 /12 + p
T (S T )3 /3
rT
KP (0, T )
T )3 /12
T (S T )3 /12 +
log(er(ST ) /K)
p
T (S T )3 /3
!
.
Chapter 12
Exercise 12.1
a) We have

P (t, T2 )
dP (t, T2 )
dP (t, T1 )
d
=
P (t, T2 )
P (t, T1 )
P (t, T1 )
(P (t, T1 ))2
2
(dP (t, T1 ))2
dP (t, T1 ) dP (t, T2 )
+ P (t, T2 )
2
(P (t, T1 )3
(P (t, T1 ))2
rt P (t, T2 )dt + 2 (t)P (t, T2 )dWt
=
P (t, T1 )
rt P (t, T1 )dt + 1 (t)P (t, T1 )dWt
P (t, T2 )
(P (t, T1 ))2
(rt P (t, T1 )dt + 1 (t)P (t, T1 )dWt )2
+P (t, T2 )
(P (t, T1 )3
612
"

(rt P (t, T1 )dt + 1 (t)P (t, T1 )dWt ) (rt P (t, T2 )dt + 2 (t)P (t, T2 )dWt )
(P (t, T1 ))2
P (t, T2 )
P (t, T2 )
= 2 (t)
dWt 1 (t)
dWt
P (t, T1 )
P (t, T1 )
P (t, T2 )
P (t, T2 )
+(1 (t))2
dt 1 (t)2 (t)
dt
P (t, T1 )
P (t, T1 )
P (t, T2 )
P (t, T2 )
1 (t)(2 (t) 1 (t))dt +
(2 (t) 1 (t))dWt
=
P (t, T1 )
P (t, T1 )
P (t, T2 )
(2 (t) 1 (t))(dWt 1 (t)dt)
=
P (t, T1 )
P (t, T2 )
P (t, T2 )
= (2 (t) 1 (t))
dWt = (2 (t) 1 (t))
d Wt ,
P (t, T1 )
P (t, T1 )
t = dWt 1 (t)dt is a standard Brownian motion under the

where dW
T1 -forward measure P.
b) From Question (a) we have
P (T1 , T2 )
P (T1 , T1 )
w

w T1
T1
P (t, T2 )
s 1
=
exp
(2 (s) 1 (s))dW
(2 (s) 1 (s))2 ds
t
P (t, T1 )
2 t

P (t, T2 )
2
exp X v /2 ,
=
P (t, T1 )
2
where
w T1 X is a centered Gaussian random variable with variance v =
Hence by the hint below
(2 (s)1 (s))2 ds, independent of Ft under P.
t
we find
h r T1
i

(K P (T1 , T2 ))+ | Ft
IE e 0 rs ds (K P (T1 , T2 ))+ | Ft = P (t, T1 )IE

P (t, T2 )
= P (t, T1 ) (v/2 + (log(/x))/v)
(v/2 + (log(/x))/v)
P (t, T1 )
= P (t, T1 )(v/2 + (log(/x))/v) P (t, T2 )(v/2 + (log(/x))/v),
P (T1 , T2 ) =
with x = P (t, T2 )/P (t, T1 ).

Exercise 12.2
S is defined from the numeraire Nt := P (t, S) and
a) The forward measure P
this gives
Ft = P (t, S)IE[(
L(T, T, S))+ | Ft ].
b) The LIBOR rate L(t, T, S) is a driftless geometric Brownian motion with
S . Indeed, the LIBOR rate
volatility under the forward measure P
t = Xt /Nt
L(t, T, S) can be written as the forward price L(t, T, S) = X
"
613
N. Privault
where Xt = (P (t, T ) Pr (t, S))/(S T ) and rNt = P (t, S). Since both dist
t
counted bond prices e 0 rs ds P (t, T ) and e 0 rs ds P (t, S) are martingales
under P , the same is true of Xt . Hence L(t, T, S) = Xt /Nt becomes a

S by Proposition 2.1, and computmartingale under the forward measure P
S amounts to removing any dt term in (12.25),
ing its dynamics under P
i.e.
t,
dL(t, T, S) = L(t, T, S)dW
0 t T,
hence L(t, T, S) = L(0, T, S)eWt

S .
Brownian motion under P
c) We find
t/2
t )tR is a standard
, where (W
+
Ft = P (t, S)IE[(
L(T, T, S))+ | Ft ]
2
= P (t, S)IE[(
L(t, T, S)e (T t)/2+(WT Wt ) )+ | Ft ]
t (d+ ))
= P (t, S)((d ) X
= P (t, S)(d ) P (t, S)L(t, T, S)(d+ )
= P (t, S)(d ) (P (t, T ) P (t, S))(d+ )/(S T ),
where em = L(t, T, S)e
(T t)/2
, v 2 = (T t) 2 , and
log(L(t, T, S)/) T t
d+ =
,
+
2
T t
and
d =
log(L(t, T, S)/) T t
2
T t
because L(t, T, S) is a driftless geometric Brownian motion with volatility

S .
Exercise 12.3
a) We take t = 0, T1 = 4, T2 = 5, T3 = 6,
= 1,3 (t) = 0.2, and =
S(t, Ti , Tj ) = 5%. The discount factors are given by P (t, T1 ) = (1 + r)4 ,
P (t, T2 ) = (1 + r)5 , P (t, T3 ) = (1 + r)6 , where r = 5%.
b) By Proposition 12.5 the price of the swaption is
P (t, T1 )P (t, T3 )+ (t, S(t, Ti , Tj )) (t, S(t, Ti , Tj )) (P (t, T2 ) + P (t, T3 )) ,
where d+ and d are given in Proposition 12.5.
Exercise 12.4
a) We have
dP (t, Ti )
= rt dt + ti dBt ,
P (t, Ti )
614
i = 1, 2,
"

and
P (T, Ti ) = P (t, Ti ) exp
w
rs ds +
wT
t
si dBs

1wT i 2
(s ) ds ,
2 t
0 t T Ti , i = 1, 2, hence
log P (T, Ti ) = log P (t, Ti ) +
wT
t
rs ds +
wT
t
si dBs
1wT i 2
( ) ds,
2 t s
0 t T Ti , i = 1, 2, and
1
d log P (t, Ti ) = rt dt + ti dBt (ti )2 dt,
2
i = 1, 2.
In the present model

drt = dBt ,
where (Bt )tR+ is a standard Brownian motion under P, we have
ti = (Ti t),
0 t Ti ,
i = 1, 2.
Letting
dBti = dBt ti dt,
defines a standard Brownian motion under Pi , i = 1, 2, and we have
w

T
1wT 1 2
P (T, T1 )
P (t, T1 )
=
exp
(s1 s2 )dBs
((s ) (s2 )2 )ds
t
P (T, T2 )
P (t, T2 )
2 t
w

T
P (t, T1 )
1wT 1
1
2
2
=
exp
(s s )dBs
(s s2 )2 ds ,
t
P (t, T2 )
2 t
which is an Ft -martingale under P2 and under P1,2 , and
w

T
1wT 1
P (T, T2 )
P (t, T2 )
(s1 s2 )dBs1
(s s2 )2 ds ,
=
exp
t
P (T, T1 )
P (t, T1 )
2 t
which is an Ft -martingale under P1 .
b) We have
1
(log P (t, T2 ) log P (t, T1 ))
T2 T1
1
2
= rt +
((T1 t)3 (T2 t)3 ).
T2 T1 6
f (t, T1 , T2 ) =
c) We have
df (t, T1 , T2 ) =
"
1
d log (P (t, T2 )/P (t, T1 ))
T2 T1
615
N. Privault

1
1
(t2 t1 )dBt ((t2 )2 (t1 )2 )dt
T2 T1
2

1
1
2
1
2
=
(t t )(dBt + t2 dt) ((t2 )2 (t1 )2 )dt
T2 T1
2

1
1
=
(t2 t1 )dBt2 (t2 t1 )2 dt .
T2 T1
2
=
d) We have
1
log (P (T, T2 )/P (T, T1 ))
T2 T1

w
T
1
1
(s2 s1 )dBs ((s2 )2 (s1 )2 )ds
= f (t, T1 , T2 )
t
T2 T1
2
w

T
1
1wT 2
2
1
2
= f (t, T1 , T2 )
(s s )dBs
(s s1 )2 ds
t
T2 T1
2 t

w
T
1
1wT 2
2
1
1
= f (t, T1 , T2 )
(s s1 )2 ds .
(s s )dBs +
t
T2 T1
2 t
f (T, T1 , T2 ) =
Hence f (T, T1 , T2 ) has a Gaussian distribution given Ft with conditional

mean
1wT 2
m = f (t, T1 , T2 ) +
( s1 )2 ds
2 t s
under P2 , resp.
m = f (t, T1 , T2 )
1wT 2
( s1 )2 ds
2 t s
under P1 , and variance

v2 =
wT
1
( 2 s1 )2 ds.
(T2 T1 )2 t s
Hence
i
h r T2

(T2 T1 ) IE e t rs ds (f (T1 , T1 , T2 ) )+ Ft
i
h

= (T2 T1 )P (t, T2 ) IE2 (f (T1 , T1 , T2 ) )+ Ft
i
h

= (T2 T1 )P (t, T2 ) IE2 (m + X )+ Ft

(m)2
v
= (T2 T1 )P (t, T2 ) e 2v2 + (m )((m )/v) .
2
e) We have
L(T, T1 , T2 ) = S(T, T1 , T2 )
616
"

1
P (T, T1 )
1
T2 T1 P (T, T2 )

w

T
1
P (t, T1 )
1wT 1 2
=
exp
(s1 s2 )dBs
((s ) (s2 )2 )ds 1
t
T2 T1 P (t, T2 )
2 t

w

T
1
P (t, T1 )
1wT 1
exp
(s1 s2 )dBs2
(s s2 )2 ds 1
=
t
T2 T1 P (t, T2 )
2 t
w

T
1wT 1
1
P (t, T1 )
1
2
1
(s s )dBs +
(s s2 )2 ds 1 ,
exp
=
t
T2 T1 P (t, T2 )
2 t
=
and by Ito calculus,

1
P (t, T1 )
d
T2 T1
P (t, T2 )

1
1
1
P (t, T1 )
1
=
(t t2 )dBt + (t1 t2 )2 dt ((t1 )2 (t2 )2 )dt
T2 T1 P (t, T2 )
2
2

1
=
+ S(t, T1 , T2 ) (t1 t2 )dBt + t2 (t2 t1 )dt)dt
T2 T1

1
=
+ S(t, T1 , T2 ) (t1 t2 )dBt1 + ((t2 )2 (t1 )2 )dt
T2 T1

1
=
+ S(t, T1 , T2 ) (t1 t2 )dBt2 ,
t [0, T1 ],
T2 T1
dS(t, T1 , T2 ) =
hence
1
T2 T1
+ S(t, T1 , T2 ) is a geometric Brownian motion, with
1
+ S(T, T1 , T2 )
T2 T1

w

T
1wT 1
1
+ S(t, T1 , T2 ) exp
(s1 s2 )dBs2
(s s2 )2 ds ,
=
t
T2 T1
2 t
0 t T T1 .
f) We have
i
h r T2

(T2 T1 ) IE e t rs ds (L(T1 , T1 , T2 ) )+ Ft
i
h r T1

= (T2 T1 ) IE e t rs ds P (T1 , T2 )(L(T1 , T1 , T2 ) )+ Ft

h
i

= P (t, T1 , T2 ) IE1,2 (S(T1 , T1 , T2 ) )+ Ft .
The forward measure P2 is defined by

P (t, T2 ) r t rs ds
dP2
e 0
,
IE
Ft =
dP
P (0, T2 )
0 t T2 ,
and the forward swap measure is defined by

"
617
N. Privault
IE

P (t, T2 ) r t rs ds
dP1,2
,
e 0
Ft =
dP
P (0, T2 )
0 t T1 ,
hence P2 and P1,2 coincide up to time T1 and (Bt2 )t[0,T1 ] is a standard

Brownian motion until time T1 under P2 and under P1,2 , consequently
under P1,2 we have
L(T, T1 , T2 ) = S(T, T1 , T2 )

r
r
T
1
2
2
1
2 2
1 T
1
1
=
+
+ S(t, T1 , T2 ) e t (s s )dBs 2 t (s s ) ds ,
T2 T1
T2 T1
has same law as
1
T2 T1

P (t, T1 ) X 1 Var [X]
e 2
1 ,
P (t, T2 )
where X is a centered Gaussian random variable with variance

w T1
(s1 s2 )2 ds
t
given Ft . Hence
i
h r T2

(T2 T1 ) IE e t rs ds (L(T1 , T1 , T2 ) )+ Ft
= P (t, T1 , T2 )
BS
1
+ S(t, T1 , T2 ),
T2 T1
r T1
t
!
(s1 s2 )2 ds
1
, +
, T1 t .
T1 t
T2 T1
Exercise 12.5
a) We have
L(t, T1 , T2 ) = L(0, T1 , T2 )e
rt
0
1 (s)dWs2 21
rt
0
|1 (s)|2 ds
0 t T1 ,
and L(t, T2 , T3 ) = b. Note that we have P (t, T2 )/P (t, T3 ) = 1 + b hence

P2 = P3 = P1,2 up to time T1 .
b) We use change of numeraire under the forward measure P2 .
c) We have
i
h r T2

IE e t rs ds (L(T1 , T1 , T2 ) )+ Ft

2 (L(T1 , T1 , T2 ) )+ | Ft
= P (t, T2 )IE
i
h
rT
rT
2 (L(t, T1 , T2 )e t 1 1 (s)dWs2 21 t 1 |1 (s)|2 ds )+ | Ft
= P (t, T2 )IE
= P (t, T2 )BS(, L(t, T1 , T2 ), 1 (t), 0, T1 t),
where
618
"

12 (t) =
1 w T1
|1 (s)|2 ds.
T1 t t
d) We have
P (t, T1 )
P (t, T1 )
=
P (t, T1 , T3 )
P (t, T2 ) + P (t, T3 )
P (t, T1 )
1
=
P (t, T2 ) 1 + P (t, T3 )/P (t, T2 )
1 + b
=
(1 + L(t, T1 , T2 )),
(b + 2)
0 t T1 ,
and
P (t, T3 )
P (t, T3 )
=
P (t, T1 , T3 )
P (t, T2 ) + P (t, T3 )
1
=
1 + P (t, T2 )/P (t, T3 )
1 1
=
,
0 t T2 .
2 + b
(16.64)
e) We have
P (t, T1 )
P (t, T3 )
P (t, T1 , T3 ) P (t, T1 , T3 )
1
1 + b
(1 + L(t, T1 , T2 ))
=
(2 + b)
(2 + b)
1
=
(b + (1 + b)L(t, T1 , T2 )),
0 t T2 .
2 + b
S(t, T1 , T3 ) =
We have
1 + b
L(t, T1 , T2 )1 (t)dWt2
2 + b

b
= S(t, T1 , T3 )
1 (t)dWt2
2 + b
dS(t, T1 , T3 ) =
= S(t, T1 , T3 )1,3 (t)dWt2 ,
0 t T2 ,
with

b
1 (t)
S(t, T1 , T3 )(2 + b)

b
= 1
1 (t)
b + (1 + b)L(t, T1 , T2 )
(1 + b)L(t, T1 , T2 )
=
1 (t)
b + (1 + b)L(t, T1 , T2 )
1,3 (t) =
"
619
N. Privault
(1 + b)L(t, T1 , T2 )
1 (t).
(2 + b)S(t, T1 , T3 )
f) The process (W 2 )tR+ is a standard Brownian motion under P2 and

1,3 (S(T1 , T1 , T3 ) )+ | Ft
P (t, T1 , T3 )IE
= P (t, T2 )BS(, S(t, T1 , T2 ),
1,3 (t), 0, T1 t),
where |
1,3 (t)|2 is the approximation of the volatility
1 w T1
1 w T1
|1,3 (s)|2 ds =
T1 t t
T1 t t
(1 + b)L(s, T1 , T2 )
(2 + b)S(s, T1 , T3 )
2
1 (s)ds
obtained by freezing the random component of 1,3 (s) at time t, i.e.

2
1,3
(t) =
1
T1 t
(1 + b)L(t, T1 , T2 )
(2 + b)S(t, T1 , T3 )
2 w
T1
|1 (s)|2 ds.
Exercise 12.6
a) We have
i
h rT
wT
+
IE e t rs ds (P (T, S) ) Ft = VT = V0 +
dVt
0
h
i wt
wt
+
T
= P (0, T ) IET (P (T, S) ) +
s dP (s, T ) +
sS dP (s, S).
0
b) We have
rt

dVt = d e 0 rs ds Vt
= rt e
=
=
rt
rs ds
Vt dt + e
rt
r ds
0 s
dVt
T
rt e 0
(t P (t, T )
rt
+tS P (t, S))dt + e 0 rs ds tT dP (t, T )
tT dP (t, T ) + tS dP (t, S).
rt
rs ds
+ e
rt
0
rs ds S
t dP (t, S)
c) By Itos formula we have

h
i
+
IET (P (T, S) ) |Ft = C(XT , 0, 0)
w t C
(Xs , T s, v(s, T ))dXs
= C(X0 , T, v(0, T )) +
0 x
h
i w t C
+
= IET (P (T, S) ) +
(Xs , T s, v(s, T ))dXs ,
0 x
since the process
h
i
+
t 7 IET (P (T, S) ) |Ft
620
"
d) We have
dVt = d(Vt /P (t, T ))
h
i
+
= d IET (P (T, S) ) |Ft
C
(Xt , T t, v(t, T ))dXt
x
P (t, S) C
(Xt , T t, v(t, T ))(tS tT )dBtT .
=
P (t, T ) x
=
e) We have
dVt = d(P (t, T )Vt )
= P (t, T )dVt + Vt dP (t, T ) + dVt dP (t, T )
C
(Xt , T t, v(t, T ))(tS tT )dBtT + Vt dP (t, T )
x
C
+P (t, S)
(Xt , T t, v(t, T ))(tS tT )tT dt
x
C
= P (t, S)
x
= P (t, S)
f) We have
dVt = d(e
= rt e
rt
0
rs ds
rt
0
Vt )
rs ds
Vt dt + e
rt
0
rs ds
dVt
C
= P (t, S)
x
g) We have
C
dVt = P (t, S)
(Xt , T t, v(t, T ))(tS tT )dBt + Vt dP (t, T )
x
C
=
(Xt , T t, v(t, T ))dP (t, S)
x
P (t, S) C
(Xt , T t, v(t, T ))dP (t, T ) + Vt dP (t, T )

P (t, T ) x

P (t, S) C
= Vt
(Xt , T t, v(t, T )) dP (t, T )
P (t, T ) x
C
+
(Xt , T t, v(t, T ))dP (t, S),
x
hence the hedging strategy (tT , tS )t[0,T ] of the bond option is given by
"
621
N. Privault
P (t, S) C
tT = Vt
(Xt , T t, v(t, T ))
P (t, T ) x
P (t, S) C
= C(Xt , T t, v(t, T ))
(Xt , T t, v(t, T )),
P (t, T ) x
and
tS =
C
(Xt , T t, v(t, T )),
x
t [0, T ].
h) We have
C
(x, , v)
x

x
x
v
1
v
1
=
x
+ log

+ log
x
2
v
v

v
1
x
v
1
x
+ log

+ log
=x
x
2
x
2
v
v

v
1
x
+
+ log
2
v

2

2
x
x
1
1

12
1 v 2 + v
log
2 + v log
e
1
e 2
1
=x
v x
v x
2
2

v
1
x
+
+ log
2
v

log(x/) + v 2 /2
=
.
v
As a consequence we get
P (t, S) C
(Xt , T t, v(t, T ))
tT = C(Xt , T t, v(t, T ))
P (t, T ) x

P (t, S)
(T t)v 2 (t, T )/2 + log Xt
P (t, T )
T tv(t, T )

v(t, T )
1
P (t, S)

+
log
2
v(t, T )
P (t, T )

P (t, S)
log(Xt /) + (T t)v 2 (t, T )/2
P (t, T )
T tv(t, T )

log(Xt /) (T t)v 2 (t, T )/2
=
,
v(t, T ) T t
and
622
"

tS =
C
(Xt , T t, v(t, T )) =
x
log(Xt /) + (T t)v 2 (t, T )/2
v(t, T ) T t

,
t [0, T ], and the hedging strategy is given by

i
h rT
+
VT = IE e t rs ds (P (T, S) ) Ft
wt
wt
= V0 +
sT dP (s, T ) +
sS dP (s, S)
0
0
w t log(X /) (T t)v 2 (t, T )/2
t
= V0
dP (s, T )
0
v(t, T ) T t

wt
log(Xt /) + (T t)v 2 (t, T )/2
+
dP (s, S).
0
v(t, T ) T t
Consequently the bond option can be hedged by shortselling a bond with
maturity T for the amount

log(Xt /) (T t)v 2 (t, T )/2
,
v(t, T ) T t
and by buying a bond with maturity S for the amount

log(Xt /) + (T t)v 2 (t, T )/2
.
v(t, T ) T t
Exercise 12.7
a) Choosing the annuity numeraire Nt = P (Ti , Ti , Tj ) we have
i
h r Ti

IE e t rs ds P (Ti , Ti , Tj )(S(Ti , Ti , Tj ))Ft

i,j P (Ti , Ti , Tj ) (S(Ti , Ti , Tj ))Ft
= Nt IE
NTi
i,j [(S(Ti , Ti , Tj )) | Ft ].
= P (t, Ti , Tj )IE
b) Since S(t, Ti , Tj ) is a forward price for the numeraire P (t, Ti , Tj ), it is a

i,j and we have
martingale under the forward swap measure P
ti,j ,
S(t, Ti , Tj ) = S(t, Ti , Tj )dW
0 t Ti ,
ti,j )tR
(W
+
where
is a standard Brownian motion under the forward swap
i,j .
measure P
c) Given the solution
S(Ti , Ti , Tj ) = S(0, Ti , Tj )eWTi

"
Ti /2
= S(t, Ti , Tj )e(WTi Wt )
(Ti t)/2
623
N. Privault
of (12.28), we find
i,j [(S(Ti , Ti , Tj )) | Ft ]
P (t, Ti , Tj )IE
h
i
T W
t ) 2 (Ti t)/2
i,j S(t, Ti , Tj )e(W
i
= P (t, Ti , Tj )IE
Ft

w
dx
x2 /(2(Ti t))
x 2 (Ti t)/2
p
,
e
= P (t, Ti , Tj )
S(t, Ti , Tj )e
2(Ti t)
T W
t is a centered Gaussian variable with variance Ti t,
because W
i
i,j .
independent of Ft under the forward measure P
d) We find

i,j ( S(Ti , Ti , Tj ))+ | Ft
P (t, Ti , Tj )IE

T W
t) +
i,j ( S(t, Ti , Tj )e2 (Ti t)/2+(W
i
= P (t, Ti , Tj )IE
) | Ft
t (d+ ))
= P (t, Ti , Tj )((d ) X
= P (t, Ti , Tj )(d ) P (t, Ti , Tj )S(t, Ti , Tj )(d+ )
= P (t, Ti , Tj )(d ) (P (t, Ti ) P (t, Tj ))(d+ )/(Tj Ti ),
where em = S(t, Ti , Tj )e
(T t)/2
, v 2 = (T t) 2 , and
log(S(t, Ti , Tj )/) Ti t
d+ =
+
,
2
Ti t
and
d =
log(S(t, Ti , Tj )/) Ti t
2
Ti t
because S(t, Ti , Tj ) is a driftless geometric Brownian motion with volatil i,j .

ity under the forward measure P
Exercise 12.8
a) Apply the Ito formula to d(P (t, T1 )/P (t, T2 ).
b) We have
w

w T1
T1
s 1
LT1 = Lt exp
(s)dB
|(s)|2 ds .
t
2 t
c) We have
i
h
(LT )+ Ft
P (t, T2 )IE
1
r
+
rT
T
Lt e t 1 (s)dBs 21 t 1 |(s)|2 ds Ft
= P (t, T2 )IE
p
= P (t, T2 )BS(, v(t, T1 )/ T1 t, 0, T1 t)
624
"

log(x/K) v(t, T1 )
log(x/K) v(t, T1 )
= P (t, T2 ) Lt
+

.
v(t, T1 )
2
v(t, T1 )
2
d) Integrate the self-financing condition (12.33) between 0 and t.
e) We have

rt
= rt e
=
=
since
rt
rs dsVt
dt + e
rt
r ds
0 s
dVt
rt
r
ds
1
s
rt e 0
t P (t, T1 ) rt e 0 rs ds t2 P (t, T2 )dt
rt
rt
+e 0 rs ds t1 dP (t, T1 ) + e 0 rs ds t2 dP (t, T2 )
t1 dP (t, T1 ) + t2 dP (t, T2 ),
0 t T1 .
rt
dP (t, T1 )
= 1 (t)dt,
P (t, T1 )
dP (t, T2 )
= 2 (t)dt.
P (t, T2 )
f) We apply the Ito formula and the fact that

i
h
(LT )+ Ft
t 7 IE
1
Next we use the fact that
and (Lt )tR+ are both martingales under P.
i
h
(LT )+ Ft ,
Vt = IE
1
and apply the result of Question (f).
g) Apply the Ito rule to Vt = P (t, T2 )Vt using Relation (12.31) and the result
of Question (f).
h) We have
dVt =
C
(Lt , v(t, T1 ))P (t, T1 )(1 (t) 2 (t))dBt + Vt dP (t, T2 )
x
C
(Lt , v(t, T1 ))P (t, T1 )(1 (t) 2 (t))dBt + Vt 2 (t)P (t, T2 )dBt
x
C
= (1 + Lt )
x
C
C
= Lt 1 (t)
(Lt , v(t, T1 ))P (t, T2 )dBt Lt
(Lt , v(t, T1 ))P (t, T2 )2 (t)dBt
x
x
C
+
x

C
C
= Lt
(Lt , v(t, T1 ))P (t, T2 )1 (t)dBt + Vt Lt
(Lt , v(t, T1 )) P (t, T2 )2 (t)dBt
x
x
C
+
(Lt , v(t, T1 ))P (t, T2 )(1 (t) 2 (t))dBt ,
x
=
"
625
N. Privault
hence
dVt = Lt 1 (t)

C
C
(Lt , v(t, T1 ))P (t, T2 )dBt + Vt Lt
(Lt , v(t, T1 )) P (t, T2 )2 (t)dBt
x
x
C
(Lt , v(t, T1 ))P (t, T2 )(1 (t) 2 (t))dBt
x
C
= (P (t, T1 ) P (t, T2 ))1 (t)
(Lt , v(t, T1 ))dBt
x

C
(Lt , v(t, T1 )) dP (t, T2 )
+ Vt Lt
x
C
(Lt , v(t, T1 ))P (t, T2 )(1 (t) 2 (t))dBt
+
x
C
= (1 (t)P (t, T1 ) 2 (t)P (t, T2 ))
(Lt , v(t, T1 ))dBt

x
C
(Lt , v(t, T1 )) dP (t, T2 )
+ Vt Lt
x

C
C
=
(Lt , v(t, T1 ))d(P (t, T1 ) P (t, T2 )) + Vt Lt
(Lt , v(t, T1 )) dP (t, T2 ).
x
x
+
Exercise 12.9
a) We have
S(Ti , Ti , Tj ) = S(t, Ti , Tj ) exp
w
Ti
t
i,j (s)dBsi,j

1 w Ti
|i,j |2 (s)ds .
2 t
b) We have
i
h
+
P (t, Ti , Tj ) IEi,j (S(Ti , Ti , Tj ) ) Ft

+
r
r Ti
2
i,j
1 Ti

= P (t, Ti , Tj ) IEi,j S(t, Ti , Tj )e t i,j (s)dBs 2 t |i,j | (s)ds Ft
p
= P (t, Ti , Tj )BS(, v(t, Ti )/ Ti t, 0, Ti t)
= P (t, Ti , Tj )

log(x/K) v(t, Ti )
log(x/K) v(t, Ti )
S(t, Ti , Tj )
+

,
v(t, Ti )
2
v(t, Ti )
2
where
v 2 (t, Ti ) =
w Ti
t
|i,j |2 (s)ds.
c) Integrate the self-financing condition (12.38) between 0 and t.

d) We have
rt

626
"

rt
rt
dt + e 0 rs ds dVt
j
j
rt
rt
X
X
tk dP (t, Tk )
tk P (t, Tk )dt + e 0 rs ds
= rt e 0 rs ds
= rt e
rs dsVt
k=i
k=i
j
X
tk dP (t, Tk ),
0 t Ti .
k=i
since
dP (t, Tk )
= k (t)dt,
P (t, Tk )
k = i, . . . , j.
e) We apply the Ito formula and the fact that

i
h
+
t 7 IEi,j (S(Ti , Ti , Tj ) ) Ft
and (St )tR+ are both martingales under Pi,j .
f) Use the fact that
i
h
+
Vt = IEi,j (S(Ti , Ti , Tj ) ) Ft ,
and apply the result of Question (e).
g) Apply the Ito rule to Vt = P (t, Ti , Tj )Vt using Relation (12.35) and the
result of Question (f).
h) From the expression (12.37) of the swap rate volatilities we have
C
(St , v(t, Ti ))
x
!
j1
X
(Tk+1 Tk )P (t, Tk+1 )(i (t) k+1 (t)) + P (t, Tj )(i (t) j (t)) dBt
dVt = St
k=i
+Vt dP (t, Ti , Tj )
C
(St , v(t, Ti ))
= St
x
!
j1
X
(Tk+1 Tk )P (t, Tk+1 )(i (t) k+1 (t)) + P (t, Tj )(i (t) j (t)) dBt
k=i
+Vt
j1
X
(Tk+1 Tk )k+1 (t)P (t, Tk+1 )dBt
k=i
j1
= St
X
C
(St , v(t, Ti ))
(Tk+1 Tk )P (t, Tk+1 )(i (t) k+1 (t))dBt
x
k=i
C
+
(St , v(t, Ti ))P (t, Tj )(i (t) j (t))dBt
x
"
627
N. Privault
+Vt
j1
X
k=i
j1
= St i (t)
X
C
(St , v(t, Ti ))
x
k=i
j1
X
C
(St , v(t, Ti ))
St
x
k=i
C
+
x
j1
X
+Vt
k=i
j1
X
C
(St , v(t, Ti ))
x
k=i
X

j1
C
(St , v(t, Ti ))
+ Vt St
x
= St i (t)
k=i
C
+
(St , v(t, Ti ))P (t, Tj )(i (t) j (t))dBt .
x
i) We have
j1
X
C
(St , v(t, Ti ))
x
k=i
X

j1
C
(St , v(t, Ti ))
+ Vt St
x
dVt = St i (t)
k=i
C
+
x
C
= (P (t, Ti ) P (t, Tj ))i (t)
(St , v(t, Ti ))dBt
x

C
+ Vt St
(St , v(t, Ti )) dP (t, Ti , Tj )
x
C
+
x
C
= (i (t)P (t, Ti ) j (t)P (t, Tj ))
(St , v(t, Ti ))dBt

x
C
+ Vt St
(St , v(t, Ti )) dP (t, Ti , Tj )
x
C
=
x
628
"

C
+ Vt St
(St , v(t, Ti )) dP (t, Ti , Tj ).
x
j) We have

C
v
1
x
v
1
x
(x, , v) =
x
+ log
+ log
x
x
2 v
2 v

v
1
x
v
1
x
v
1
x
=x
+ log
+ log
+
+ log
x
2 v
x
2 v
2 v
2
2
1
(v/2+v 1 log(x/)) /2
(v/2+v 1 log(x/)) /2
e
= e
v 2
vx 2

log(x/) v
log(x/) v
+
+
=
+
.
v
2
v
2
k) We have
C
x

C
+ Vt St
(St , v(t, Ti )) dP (t, Ti , Tj )
x

=
+
d(P (t, Ti ) P (t, Tj ))
v(t, Ti )
2

dP (t, Ti , Tj ).
v(t, Ti )
2
dVt =
l) We compare the results of Questions (d) and (k).
Chapter 13
Exercise 13.1 Defaultable bonds.
a) Use the fact that (rt , t )t[0,T ] is a Markov process.
b) Use the tower property (16.25) for the conditional expectation given Ft .
c) We have
rt

d e 0 (rs +s )ds P (t, T )
= (rt + t )e
rt
rt
= (rt + t )e
rt
= (rt + t )e
"
(rs +s )ds
P (t, T )dt + e
rt
(rs +s )ds
(rs +s )ds
0
rt
(rs +s )ds
P (t, T )dt + e
rt
(rs +s )ds
P (t, T )dt + e
dP (t, T )
dF (t, rt , t )
(rs +s )ds F
0
(t, rt , t )drt
x
0
629
N. Privault
2F
1 rt
(t, rt , t )dt + e 0 (rs +s )ds 2 (t, rt , t )12 (t, rt )dt
y
2
x
1 r t (rs +s )ds 2 F
2
(t, rt , t )2 (t, t )dt
+ e 0
2
y 2
2
rt
rt
F
F
+ e 0 (rs +s )ds
(t, rt , t )1 (t, rt )2 (t, t )dt + e 0 (rs +s )ds
(t, rt , t )dt
xy
t
rt
rt
F
F
(1)
(2)
= e 0 (rs +s )ds
(t, rt , t )1 (t, rt )dBt + e 0 (rs +s )ds
(t, rt , t )2 (t, t )dBt
x
y

rt
F
(t, rt , t )1 (t, rt )
+ e 0 (rs +s )ds (rt + t )P (t, T ) +
x
2
F
1 F
1 2F
+
(t, rt , t )2 (t, t ) +
(t, rt , t )12 (t, rt ) +
(t, rt , t )22 (t, t )
2
y
2 x
2 y 2

2F
F
+
(t, rt , t )1 (t, rt )2 (t, t ) +
(t, rt , t ) dt,
xy
t
+ e
rt
0
(rs +s )ds F
hence the bond pricing PDE is

F
(t, x, y)
x
F
1
2F
+ 2 (t, y)
(t, x, y) + 12 (t, x) 2 (t, x, y)
y
2
x
1 2
2F
2F
F
+ 2 (t, y) 2 (t, x, y) + 1 (t, x)2 (t, y)
(t, x, y) +
(t, rt , t ) = 0.
2
y
xy
t
(x + y)F (t, x, y) + 1 (t, x)
d) We have
wt
0

1 (1)
B rt
a t

wt
(1)
Bt
ea(ts) dBs(1)
=
0
a
wt
a(ts)
=
(1 e
)dBs(1) ,
a 0
rs ds =
hence
wT
wT
wt
rs ds =
rs ds
rs ds
t
0
0
wt
wT
a(T s)
=
(1 e
)dBs(1)
(1 ea(ts) )dBs(1)
a 0
a 0

wT
w t a(T s)
=
(e
ea(ts) )dBs(1) +
(ea(T s) 1)dBs(1)
0
t
a
wt
a(T t)
w T a(T s)
a(ts)
(1)
= (e
1) e
dBs
(e
1)dBs(1)
0
a
a t
630
"

1
w T a(T s)
= (ea(T t) 1)rt
(e
1)dBs(1) .
a
a t
The answer for t is similar.
e) As a consequence of the previous question we have
w

wT
T

IE
rs ds +
s dsFt = C(a, t, T )rt + C(b, t, T )t ,
t
and
w
T
wT

s dsFt =
t
t
w
w

T
T

= Var
rs dsFt + Var
s dsFt
t
t
w

wT
T

+2 Cov
Xs ds,
Ys dsFt
Var
rs ds +
2 w T a(T s)
(e
1)2 ds
= 2
a t
w
T a(T s)
+2
(e
1)(eb(T s) 1)ds
ab t
2 w T b(T s)
+ 2
(e
1)2 ds
b t
wT
wT
= 2
C 2 (a, s, T )ds + 2
C(a, s, T )C(b, s, T )ds
t
t
wT
+ 2
C 2 (b, sT )ds,
t
from the Ito isometry.

f) We have

w

wT
T

s ds Ft
P (t, T ) = 1{ >t} IE exp
rs ds
t
t

w
w

T
T

= 1{ >t} exp IE
rs dsFt IE
s dsFt
t
t
w

wT
T
1

rs ds +
s dsFt
Var
exp
t
t
2
= 1{ >t} exp (C(a, t, T )rt C(b, t, T )t )
2w

T
2 w T 2
exp
C 2 (a, s, T )ds +
C (b, s, T )eb(T s) ds
2 t
2 t

wT
exp
C(a, s, T )C(b, s, T )ds .
t
g) This is a direct consequence of the answers to Questions c and f.

"
631
N. Privault
h) The above analysis shows that

w

T

P( > T | Gt ) = 1{ >t} IE exp
s ds Ft
t

2 w T 2
= 1{ >t} exp C(b, t, T )t +
C (b, s, T )ds ,
2 t
for a = 0 and

w

T
2 w T 2

IE exp
rs ds Ft = exp C(a, t, T )rt +
C (a, s, T )ds ,
t
2 t
for b = 0, and this implies

wT
U (t, T ) = exp
C(a, s, T )C(b, s, T )ds
t

= exp
(T t C(a, t, T ) C(b, t, T ) + C(a + b, t, T )) .
ab
i) We have
log P (t, T )
f (t, T ) = 1{ >t}
T

2 2
2
a(T t)
= 1{ >t} rt e
C (a, t, T ) + t eb(T t) C 2 (b, t, T )

2
2
1{ >t} C(a, t, T )C(b, t, T ).
j) We use the relation

w

T

P( > T | Gt ) = 1{ >t} IE exp
s ds Ft
t

2 w T 2
= 1{ >t} exp C(b, t, T )t +
C (b, s, T )ds
2 t
= 1{ >t} e
rT
t
f2 (t,u)du
where f2 (t, T ) is the Vasicek forward rate corresponding to t , i.e.

f2 (t, u) = t eb(ut)
2 2
C (b, t, u).
2
k) In this case we have = 0 and

w

T

P (t, T ) = P( > T | Gt ) IE exp
rs ds Ft ,
t
since U (t, T ) = 0.
632
"

Exercise 13.2
a) Taking (U, V ) = (U, U ), we have
P(U u, V v) = P(U u, U v)
= P(U min(u, v))
= min(u, v)
= CM (u, v),
u, v [0, 1].
b) Taking (U, V ) = (U, 1 U ), we have

P(U u, V v) = P(U u, 1 U v)
= P(U u, U 1 v)
= P(1 v U u)
= 1{01vu1} P(1 v U u)
= 1{0u+v11} (u (1 v))
= (u + v 1)+ ,
u, v [0, 1].
c) We have
C(u, v) = P(U u, V v) P(U u, V 1) P(U u) = u,
u, v [0, 1],
and similarly we find C(u, v) P(U v) = v for all u, v [0, 1], hence
we conclude to (13.18).
d) For fixed v [0, 1] we have
C
C(u + , v) C(u, v)
(u, v) = lim
0
u
P(U u + , V v) P(U u, V v)
= lim
0
P(u U u + , V v)
= lim
0
P (u U u + )
= lim P(V v | u U u + )
0
= P(V v | U = u)
1,
u, v [0, 1], hence
h0 (u) =
C
(u, v) 1 = P(V v | U = u) 1 0,
u
u, v [0, 1], and since h(1) = C(1, v) v = P(V v) v = 0, v [0, 1]

we conclude that h(u) 0, u [0, 1], which shows (13.19).
"
633
N. Privault
Exercise 13.3
From the terminal data of Figure S.16
634
"
1) Fixed recovery rate of 40%

2) Constant default rate for the tenor of the CDS
Using equation 4.3 from the Correlated Default notes, the default rate is solved with a solver (see attached
Excel file). The value 1 is found to be 0.001798747. This gave an implied default probability of 0.001246026
which is very close to that determined by the Bloomberg valuation model at 0.0013 as shown in the first
screen shot.
Fig. S.16: Cashflow data.

we infer STi = 0.10790%, t = Apr 12, 2015, Ti = Mar 20, 2015, = 40%.
Next, from the discount factors of Figure S.17 we solve numerically (13.14)
in the next table:
Date
Delta
Discount Factor Premium Leg Protection Leg
Jun 22, 201 5 0.2611111
0.99952277
0.0002814722 0.0002814708
Sep 21, 2015 0.2527778
0.99827639
0.0002721533
0.000272154
Dec 21, 2015 0.2527778
0.99607821
0.0002715541 0.0002715548
Sum
0.0008251796 0.0008251796
to find the default rate 1 = 0.0017987468 and default probability 0.0012460256,
which is consistent with the value of 0.0013 in Figure S.16.
Chapter 14
Exercise 14.1
a) When t [0, T1 ) the equation reads
dSt = St dt = St dt,
which is solved as St = S0 et , 0 t < T1 . Next, at the first jump time
t = T1 we have
dSt = St St = St dNt = St ,
"
635
N. Privault
Fig. S.17: CDS price data.
which yields St = (1 + )St , hence ST1 = (1 + )ST = S0 (1 + )eT1 .

1
Repeating this procedure over the Nt jump times contained in the interval
[0, t] we get
St = S0 (1 + )Nt et ,
t R+ .
b) When t [0, T1 ) the equation reads
dSt = St dt = St dt,
which is solved as St = S0 et , 0 t < T1 . Next, at the first jump time
t = T1 we have
dSt = St St = dNt = 1,
which yields St = 1 + St , hence ST1 = 1 + ST = 1 + S0 eT1 , and for
1
t [T1 , T2 ) we will find
St = (1 + S0 eT1 )e(tT1 ) ,
t [T1 , T2 ).
More generally, the equation can be solved by letting Yt := et St and

noting that (Yt )tR+ satisfies dYt = et dNt , which has the solution
Yt = Y0 +
wt
0
es dNs ,
t R+ ,
hence in general we have
636
"

St = et S0 +
wt
0
e(ts) dNs ,
t R+ ,
Exercise 14.2
Nt
Y
a) We have Xt = X0
(1 + ) = X0 (1 + )Nt = (1 + )Nt , t R+ .
k=1
b) By stochastic calculus and using the relation dXt = Xt dNt we have

w

wt
t
dSt = d S0 Xt + rXt
Xs1 ds = S0 dXt + d Xt
Xs1 ds
0
0
w
w

w

t
t
t
1
1
= S0 dXt + rXt d
Xs ds +
Xs ds dXt + rdXt d
Xs1 ds
0
0
0
w

t
1
1
1
= S0 dXt + rXt Xt dt +
Xs ds dXt + rdXt (Xt dt)
0
w

wt
t
= S0 dXt + rdt + r
Xs1 ds dXt = rdt + S0 +
Xs1 ds dXt
0
0

wt
1
= rdt + S0 Xt + Xt
Xs ds dNt = rdt + St dNt .
0
c) We have
IE[Xt /Xs ] = IE[(1 + )Nt Ns ] =
(1 + )k P(Nt Ns = k)
k=0
= e(ts)
(1 + )k
k=0
X ((1 + )(t s))k

((t s))k
= e(ts)
k!
k!
k=0
= e(ts) e(1+)(ts) = e(ts) ,
0 s t.
Remarks: We could also let f (t) = IE[Xt ] and take expectation in the
equation dXt = Xt dNt to get f 0 (t) = f (t)dt and f (t) = IE[Xt ] =
f (0)et = et . Note that the relation IE[Xt /Xs ] = IE[Xt ]/ IE[Xs ], which
happens to be true here, is wrong in general.
d) We have

wt
wt
IE[St ] = IE S0 Xt + rXt
Xs1 ds = S0 IE[Xt ] + r IE[Xt /Xs ]ds
0
= S0 e
+r
= S0 et +
wt
(ts)
0
t
r(e
ds = S0 e
1)
,
Exercise 14.3 We have St = S0 ert
Nt
Y
+r
wt
0
ds
t R+ .
(1 + Zk ), t R+ .
k=1
"
637
N. Privault
!2
NT
X
Var [YT ] = IE
Zk IE[YT ]
k=1
IE
n=0
NT
X
!2
Zk t IE[Z1 ]

NT = k P(NT = k)
k=1
!2
n
X
n tn X
=e
IE
Zk t IE[Z1 ]
n!
n=0
k=1
!2
n
n
X
X
n tn X
t
2 2
2
=e
IE
Zk
2t IE[Z1 ]
Zk + t (IE[Z1 ])
n!
n=0
t
k=1
X
n tn
= et
n!
n=0
X
IE 2
Zk Zl +
1k<ln
= et
k=1
n
X
|Zk | 2t IE[Z1 ]
n
X
2 2
Zk + t (IE[Z1 ])
k=1
k=1
X
n tn
n!
n=0
(n(n 1)(IE[Z1 ])2 + n IE[|Z1 |2 ] 2nt(IE[Z1 ])2 + 2 t2 (IE[Z1 ])2 )
X
X
n tn
n tn
+ et IE[|Z1 |2 ]
= et (IE[Z1 ])2
(n
2)!
(n
1)!
n=1
n=2
2et t(IE[Z1 ])2
X
n tn
+ 2 t2 (IE[Z1 ])2 )
(n
1)!
n=1
= t IE[|Z1 |2 ],
or, using the moment generating function of Proposition 14.3,
Var [YT ] = IE[|YT |2 ] (IE[YT ])2
d2
IE[eYT ]|=0 2 t2 (IE[Z1 ])2
=
dw2
= t
|y|2 (dy) = t IE[|Z1 |2 ].
Exercise 14.5
a) We have

1
dSt = + 2 St dt + St dWt + (St St )dNt
2
638
"

1
+ 2 St dt + St dWt + (S0 et+Wt +Yt S0 et+Wt +Yt )dNt
2

1 2
= + St dt + St dWt + (S0 et+Wt +Yt +ZNt et+Wt +Yt )dNt
2

1 2
= + St dt + St dWt + St (eZNt 1)dNt ,
2
=
hence the jumps of St are given by the sequence (eZk 1)k1 .

b) The discounted process ert St satisfies

1
d(ert St ) = ert r + 2 St dt+ert St dWt +ert St (eZNt 1)dNt .
2
such that
Hence by the Girsanov Theorem 14.3, choosing u, ,
1
IE [eZ1 1],
r + 2 = u
2
shows that
IE [eZ1 1]dt)
d(ert St ) = ert St (dWt +udt)+ert St ((eZNt 1)dNt
is a martingale under (Pu,,
).
Exercise 14.6
a) We have
St = S0 et
Nt
Y
(1 + Yk ) = S0 exp t +
k=1
Nt
X
!
t R+ .
Xk
t R+ ,
k=1
b) We have
ert St = S0 exp ( r)t +
Nt
X
!
Xk
k=1
which is a martingale if
0 = r + IE[Yk ] = r + IE[eXk 1] = r + (e
/2
1).
c) We have
er(T t) IE[(ST )+ | St ]
=e
r(T t)
IE S0 exp T +
NT
X
!
Xk
!+

St
k=1
"
639
N. Privault
=e
r(T t)
!+
NT
X
IE St exp (T t) +
Xk

St
k=Nt
NT
X
= er(T t) IE x exp (T t) +
!+
Xk
k=Nt
= er(T t) IE x exp (T t) +
NT
Nt
X
x=St
!+
Xk
k=0
=e
r(T t)
IE

xe
P
(T t)+ n
k=1 Xk
x=St
+
P(NT Nt = n)
x=St
n=0
= e(r+)(T t)
IE

xe(T t)+
Pn
Xk
k=1
+
x=St
n=0
= e(T t)
=e
(T t)
BS(St e(r)(T t)+n
n=0

X
St e(r)(T t)+n
/2
n=0
/2
((T t))n
n!
, r, n 2 /(T t), , T t)
((T t))n
n!
((T t))n
(d+ ) er(T t) (d )
,
n!
with
log(St e(r)(T t)+n
log(St e(r)(T t)+n
/) + r(T t) + n 2 /2
n
log(St /) + (T t) + n 2
=
,
n
d+ =
d =
=
/2
/2
/) + r(T t) n 2 /2
log(St /) + (T t)
,
n
and = r + (1 e
/2
).
Exercise 14.7
a) We have
d(et St ) = et (dNt dt),
hence
et St = S0 +
wt
0
es (dNs d),
and
640
"

St = S0 et +
wt
0
e(ts) (dNs ds),
t R+ .
b) We have
f (t) = IE[St ]
= S0 eat + IE
= S0 e
at
w
wt
0

wt
ea(ts) dNs ea(ts) ds
0
a(ts)
ds
wt
1 eat
= S0 eat + ( )
a
= ( ) + (S0 + ( ))eat ,
a(ts)
ds
t R+ .
c) The process St is a submartingale provided S0 + ( ) < 0.

d) We can write

wT
IE[(ST )] = IE S0 eT +
e(T s) (dNs ds),
0
n
wT
X
X
n w T
1 eT
=
S0 eT +
e(T sk )
0
n! 0
n=0
!
ds1 dsk ,
k=1
t R+ .
Chapter 15
Exercise 15.1
1. We have IE[Nt t] = IE[Nt ]t = tt, hence Nt t is a martingale
if and only if = . Given that
d(ert St ) = ert St (dNt dt),
we conclude that the discounted price process ert St is a martingale if
and only if = .
2. Since we are pricing under the risk neutral measure we take = . Next
we note that
ST = S0 e(r)T (1 + )NT = St e(r)(T t) (1 + )NT Nt ,
0 t T,
hence the price at time t of the option is

er(T t) IE[|ST |2 | Ft ]
= er(T t) IE[|St |2 e2(r)(T t) (1 + )2(NT Nt ) | Ft ]
"
641
N. Privault
= |St |2 e(r2)(T t) IE[(1 + )2(NT Nt ) | Ft ]
= |St |2 e(r2)(T t) IE[(1 + )2(NT Nt ) ]
X
= |St |2 e(r2)(T t)
(1 + )2n P(NT Nt = n)
n=0
X
2 (r2)(T t)
= |St | e
(1 + )2n
n=0
2
= |St |2 e(r2)(T t)+(1+)

2 (r+ 2 )(T t)
= |St | e
((T t))n
n!
(T t)
0 t T.
Exercise 15.2
1. Independently of the choice of a risk-neutral measure Pu,,
we have
rt
rT
er(T t) IEu,,
ST | Ft ] Ker(T t)
[ST K | Ft ] = e IEu,,
[e
= ert ert St Ker(T t)

= St Ker(T t)
= f (t, St ),
for
f (t, x) = x Ker(T t) ,
t, x > 0.
2. Clearly, holding one unit of the risky asset and shorting a (possibly fractional) quantity KerT of the riskless asset will hedge the payoff ST K,
and this hedging strategy is self-financing because it is constant in time.
f
3. Since
(t, x) = 1 we have
x
f
a
(t, St ) +
(f (t, St (1 + a)) f (t, St ))
x
St
t =
2 + a2
(S (1 + a) St )
2 +
St t
=
2 + a2
= 1,
t [0, T ],
2
which coincides with the result of Question 2.

Exercise 15.3
1. We have

1
St = S0 exp t + Bt 2 t (1 + )Nt .
2
2. We have
642
"

1
St = S0 exp ( r)t + Bt 2 t (1 + )Nt ,
2
and
dSt = ( r + )St dt + St (dNt dt) + St dWt ,
hence we need to take
r + = 0,
since the compensated Poisson process (Nt t)tR+ is a martingale.
3. We have
er(T t) E [(ST )+ | St ]
"

+ #
1 2

NT
r(T t)
=e
E
S0 exp T + BT T (1 + )
St
2

+
1 2

= er(T t) E St e(T t)+(BT Bt ) 2 (T t) (1 + )NT Nt St
= er(T t)
P(NT Nt = n)
n=0

St e(T t)+(BT Bt ) 2

St e(r)(T t)+(BT Bt ) 2
= e(T t)
=e
(T t)
(T t)
+

(1 + )n St
X
((T t))n
n!
n=0
= e(r+)(T t)
E
(T t)
+

(1 + )n St
BS(St e(T t) (1 + )n , r, 2 , T t, )
n=0

X
n=0
((T t))n
n!
((T t))n
St e(T t) (1 + )n (d+ ) er(T t) (d )
,
n!
with
log(St e(T t) (1 + )n /) + (r + 2 /2)(T t)
T t
n
log(St (1 + ) /) + (r + 2 /2)(T t)
=
,
T t
d+ =
and
d =
"
log(St e(T t) (1 + )n /) + (r 2 /2)(T t)
T t
643
N. Privault
log(St (1 + )n /) + (r 2 /2)(T t)
.
T t
Exercise 15.4
1. The discounted process St = ert St satisfies the equation
dSt = YNt St dNt , ,
and it is a martingale since the compound Poisson process YNt dNt is
centered with independent increments as IE[Y1 ] = 0.
2. We have
NT
Y
ST = S0 erT
(1 + Yk ),
k=1
hence
erT IE[(ST )] = erT IE S0 erT
NT
Y
!+
(1 + Yk )
k=1
rT
=e
IE S0 e
n=0
rT T
=e
=e
NT
Y
!+
(1 + Yk )

NT = n P(NT = n)
k=1
X
k=0
rT T
rT
X
k=0
IE S0 e
n
Y
rT
!+
(T )
n!
(1 + Yk )
k=1
w1
(T )n w 1
n
1
2 n! 1
S0 e
rT
n
Y
!+
(1 + yk )
dy1 dyn .
k=1
Exercise 15.5
1. We find = where is the intensity of the Poisson process (Nt )tR+ .
2. We have
er(T t) IE[ST | Ft ] = ert IE[erT ST | Ft ] er(T t) IE[ | Ft ]
= ert IE[ert St | Ft ] er(T t)
= St er(T t) ,
since the process (ert St )tR+ is a martingale.
Exercise 15.6
1. We have
dVt = df (t, St )
= rt ert dt + t dSt
644
"

= rt ert dt + t (rSt dt + St (dNt dt))
= rVt dt + t St (dNt dt)
= rf (t, St )dt + t St (dNt dt).
2. We apply the Ito formula with jumps and the martingale property of
t 7 ert f (t, St ) to get
df (t, St ) = rf (t, St )dt
+(f (t, St (1 + )) f (t, St ))dNt (f (t, St (1 + )) f (t, St ))dt,
and we identify the terms in the above formula with those appearing in
(15.16).

Exercise 1 We have
IE[X] =
kP(X = k) = e
k=0
=e
X
k
k
k!
k=0
X
k=1
X k
k
= e
= .
(k 1)!
k!
k=0
Exercise 2 We have
P(eX > c) = P(X > log c) =
=
(log c)/
ey
/2
log c
ey
dy
/(2 2 )
2 2
dy
= 1 ((log c)/) = ((log c)/).

2
Exercise 3
1. If = 0 we have
IE[X] =
xf (x)dx =
1
2 2
y2
1 w
1
=
ye 2 dy =
2
2
by symmetry of the function y 7 ye

w
"
|y|e
y2
2
dy =
lim
A+
wA
|y|e
y2
2
y2
2
x2
xe 22 dx
lim
A+
wA
ye
y2
2
dy = 0,
. Note that we have
dy = 2 lim
A+
wA
0
ye
y2
2
dy
645
N. Privault
= 2 lim
A+

A
y2
A2
e 2
= 2 lim (1 e 2 ) = 2 < ,
A+
y2
hence the function y 7 ye 2 is integrable on R and the above computation of IE[X] is valid. Next, for all R we have
IE[X] =
=
=
=
xf (x)dx =
2 2
w
1
2 2
2 2
1
2 2
(y + )e
y2
2
2
ye
y2
y2
2
2
xe
(x)2
2 2
dx
dy
dy +
e 22 dy =
2 2
y2
e 22 dy
f (y)dy = P(X R) = .
646
"
Index
-algebra, 498
Internal Ratings-Based formula, 439
absence of arbitrage, 15, 342
abstract Bayes formula, 316
adapted process, 83, 409
admissible portfolio strategy, 109
affine model, 341
American forward contract, 305
American option
finite expiration, 294
perpetual, 282
annuity num
eraire, 388, 399
annuity numeraire, 413
arbitrage, 13
absence of, 15
continuous time, 109
discrete time, 33
opportunity, 14
arbitrage price, 47, 140
Asian option, 189, 244
attainable, 19, 24, 47, 111
backward induction, 52, 55
Barrier forward contract, 264
down-and-in, 265, 570
down-and-out, 265, 571
up-and-in, 265, 566
up-and-out, 265, 568
Barrier options, 199
down-and-in, 202
down-and-out, 202, 209, 211, 217
up-and-in, 202
up-and-out, 202, 207, 208, 212
barrier options, 188
Basel II, 439
Bernoulli distribution, 509

BGM model, 370
binary option, 151
binomial distribution, 509
Black
(1976) formula, 387
caplet formula, 385
Black-Scholes
formula, 114, 124, 141, 326, 371
PDE, 112, 114, 123, 127, 214, 217, 489
with jumps, 476
Black-Scholes calibration, 162
bond
defaultable, 412
option, 384
pricing PDE, 343, 368, 630
zero-coupon, 341, 342
bridge model, 376
Brownian bridge, 100
Brownian motion, 73
call option, 5
call/put duality, 327
cap, 387
cap pricing, 387
caplet, 385
pricing, 385
Cauchy distribution, 507
CEV model, 341
change of measure, 137
change of num
eraire, 315, 327
characteristic function, 520
Chasles relation, 87
CIR model, 340
CIR process, 374
Clark-Ocone formula, 60, 241
complete market, 20, 24, 139
647
N. Privault
complete space, 82
completeness
discrete time, 41
Compound Poisson martingale, 468
compound Poisson process, 449, 473
conditional
expectation, 512, 517
probability, 502
conditional expectation, 35
conditional survival probability, 407
conditioning, 502
contingent claim, 18, 33, 41, 47
attainable, 19, 24, 111
continuous-time limit, 67
continuous-time model, 105
copulas, 442
correlation problem, 366
coupon
bond, 342
rate, 347
coupon bond, 413
Courtadon model, 340
credit default swap, 413, 424
CRR model, 42
Dothan model, 341, 348

drifted Brownian motion, 135
Dupire PDE, 169
default rate, 409

defaultable bonds, 439
Delta, 113, 117, 126, 146148
hedging, 144, 332, 333
density
function, 505
marginal, 508
digital option, 151
discounted asset prices, 32
discrete distribution, 509
distribution
Bernoulli, 509
binomial, 509
Cauchy, 507
discrete, 509
exponential, 506
gamma, 507
Gaussian, 506
geometric, 509
lognormal, 250, 507
marginal, 513
negative binomial, 510
Pascal, 510
Poisson, 510
stationary, 180
uniform, 506
dividends, 129, 152
Doob-Meyer decomposition, 302
failure rate, 408

Fatous lemma, 132, 276
Feller condition, 340
filtration, 74, 269
enlargement of, 412
finite differences
explicit method, 488, 490
implicit method, 489, 491
floorlet, 387
foreign exchange, 323
foreign exchange option, 325
forward
contract, 69, 114, 127, 547
measure, 382
rate, 350
swap rate, 353
forward contract, 69, 114, 127, 547
American, 305, 584
forward rate, 350
spot, 351, 363, 385
swap, 353
forward start options, 149
fugazi (the), 160
fundamental theorem
discrete time, 41, 42
efficient market hypothesis, 40

elasticicity of diffusion, 341
enlargement of filtration, 412
entitlement ratio, 6, 116, 119, 164166
equivalent probability measure, 17, 22,
137
Euler discretization, 493
event, 498
exchange options, 328
exercise price, 5
exotic option, 33, 53
exotic options, 185
discrete time, 59
expectation, 510
conditional, 512, 517
exponential
distribution, 409, 506
model, 478
exponential distribution, 448
exponential Vasicek model, 340
Gamma, 148
648
"

gamma distribution, 507
gamma process, 457
Garman-Kohlagen formula, 325
Gaussian
cumulative distribution function, 68
distribution, 115, 506
random variable, 521
Geman-Yor method, 250
generating function, 100, 520
geometric
distribution, 509
geometric Brownian motion, 93
Girsanov theorem, 137, 138, 320
for jump processes, 464, 473
Greeks, 128, 129, 146, 152
heat equation, 120, 487
heat map, 195
hedging, 20, 51, 55, 59, 142
hedging by change of num
eraire, 330
hedging strategy, 143
hedging with jumps, 481
historical
probability measure, 136
volatility, 182
historical volatility, 159
hitting probability, 278
hitting time, 273
HJM
condition, 358
model, 356, 411
Ho-Lee model, 341
Hull-White model, 341, 357
implied volatility, 160
independence, 502, 505, 506, 509, 512,
517, 519, 521
independent increments, 133, 466, 467
indicator function, 504
instantaneous forward rate, 352
Interest rate model
affine, 341
Constant Elasticity of Variance, 341
Courtadon, 340
Cox Ingersoll Ross, 340
Dothan, 341, 348
exponential Vasicek, 340
Ho-Lee, 341
Hull-White, 341
Vasicek, 339
inverse Gaussian process, 457, 459
IRB formula, 439
It
o
isometry, 79, 81, 84, 453
"
process, 90, 91, 112

stochastic integral, 79, 84, 131
It
o formula, 90, 150
with jumps, 456
It
o table, 91
with jumps, 456
Jensen inequality, 552
jump-diffusion process, 473
key lemma, 410
L
evy process, 457
Laplace transform, 520
law of total expectation, 513
law of total probability, 502, 513
Leibniz integral rule, 358
LIBOR
model, 354
rate, 354
swap rate, 355, 391, 393
Lipschitz function, 328
local volatility, 166, 490
lognormal
approximation, 250
distribution, 507
long forward contract, 484, 485
lookback option, 218
call, 231
put, 188, 223, 225
marginal
density, 508
distribution, 513
Margrabe formula, 329
market completeness, 20, 24, 41
Markov property, 328, 332
martingale, 35, 131, 270
compound Poisson, 468
discrete time, 38
method, 139
Poisson, 466, 467
submartingale, 270
supermartingale, 270
transform, 39, 131
maturity, 5
maximum of Brownian motion, 190
mean hitting time, 281
mean reversion, 339
Merton model, 479
Milshtein discretization, 494
Minkowski inequality, 82
moment generating function, 520, 638
Musiela notation, 356
649
N. Privault
negative binomial distribution, 510
negative inverse Gaussian process, 459
Nelson-Siegel, 364
nonlocal operator, 477
num
eraire, 110, 313
num
eraire invariance, 330
optimal stopping, 294
option
on average, 188
on extrema, 186
writer, 20
optional
sampling, 274
stopping, 274
Ornstein-Uhlenbeck process, 339
Partial integro-differential equation, 477
partition, 502, 513
Pascal distribution, 510
path dependent options, 59
path integral, 55, 317
payoff function, 6, 185
PDE
Black-Scholes, 112, 114, 123
integro-differential, 477
variational, 297
PIDE, 477
Poisson
compound martingale, 449, 473
distribution, 510
process, 443
Poisson process, 409
portfolio, 12, 30
portfolio strategy, 106
admissible, 109, 111
power option, 149
predictable process, 38, 50, 453
predictable representation, 142, 145
pricing, 47, 53
with jumps, 474
probability
conditional, 502
density function, 505
distribution, 505
measure, 501
equivalent, 17, 22
space, 497
process
gamma, 457
inverse Gaussian, 457
predictable, 38, 50, 453
stable, 457
stopped, 273
variance gamma, 457

put option, 5
random variable, 503
rate
default, 409
forward, 350
forward swap, 353
instantaneous forward, 352
LIBOR, 354
LIBOR swap, 355, 391, 393
recovery rate, 412, 423
reflexion principle, 190
replication, 20
risk-neutral measure, 16, 473
risk-neutral measures
discrete time, 40
riskless asset, 68, 105
running maximum, 187
self-financing portfolio, 330, 332
continuous time, 105, 107, 481
discrete time, 31
short-selling, 12, 24, 118
spot forward rate, 351, 363, 385
square-integrable
functions, 78
random variables, 81
stable process, 457, 458
stationary distribution, 180
stochastic
calculus, 88
default, 409
differential equations, 96
integral, 49, 78, 83
with jumps, 452
process, 29
stopped process, 273, 274
stopping time, 272, 409
theorem, 274
Stratonovich integral, 543
strike price, 5, 19
string model, 377
strong Markov property, 448
submartingale, 270
super-hedging, 20
supermartingale, 270
survival probability, 407
Svensson parametrization, 364
swap, 353
measure, 315, 389, 400, 402
swaption, 390
650
"

telescoping sum, 356
tenor structure, 315, 353, 381, 413, 423
Theta, 128, 129, 152
tower property, 38, 39, 50, 51, 55, 85,
133, 145, 343, 513, 519, 527, 629
triangle inequality, 82
two-factor model, 368
vega notional, 183

volatility
historical, 159, 182
implied, 160
local, 166, 490
smile, 162
surface, 161
uniform distribution, 506
warrant, 6, 116
vanilla option, 33, 53

variance, 515
variance gamma process, 457, 458
variational PDE, 297
Vasicek model, 339
"
yield, 351, 385

curve, 363
zero-coupon bond, 341, 342
651
N. Privault
652
"
Author index
Aas, K. 439
Achdou, Y. 170
Feller, W. 340
Folland, G.B. 74
Follmer, H 11, 17, 20, 41, 42, 68
Fouque, J.-P. 178
Fouque, J.P. 159, 172, 177
Barrieu, P. 250
Bermin, H. 241
Bjork, T. 27, 365, 366
Bosq, D. 445
Bosq, S. 445
Brace, A. 370
Breeden, D.T. 168
Brigo, D. 345, 369, 606
Carr, P. 249, 250
Castellacci, G. 425, 442
Chan, C.M. 35
Chen, R.R. 413
Cheng, X. 413
Cont, R. 452, 457, 463, 468
Cox, J.C. 42, 340
Dana, R.A. 233
Dash, J. 317
de Chavez J., Ruiz 59
Dellacherie, C. 411
Devore, J.L. 497
Di Nunno, G. 59, 145
Doob, J.L. 270, 274, 302
Dothan, L.U. 341, 348
Dudley, R.M. 82
Duffie, D. 412
Dufresne, D. 250
Dupire, B. 169
El Karoui, N. 315, 331
El Khatib, Y. 241, 244
Elliott, R.J. 411, 412
Eriksson, J. 35, 215
Fabozzi, F. 413
"
Garman, M.B. 325

Gatarek, D. 370
Gatheral, J. 177, 183
Geman, H. 247, 250, 315, 331, 576
Gerber, H.U. 309
Gibson, M. 428
Glasserman, P. 493
Grasselli, M.R. 426
Guo, X. 410, 412
Hagan, P.S. 173
Harrison, J.M. 41, 42, 111
Heath, D. 358
Heston, S.L. 172, 178
Hiriart-Urruty, J.-B. 18
Huang, J.Z. 413
Hull, J. 357, 428
Hurd, T.-R. 426
Ikeda, N. 86
Ingersoll, J.E. 340
Jacod, J. 412, 497
Jamshidian, F. 331, 384
Jarrow, R. 358, 410, 412
Jeanblanc, M. 233, 411, 412, 484
Jeulin, Th. 412
Jones, S. 423
Kallenberg, O. 520
Kemna, A.G.Z. 247
Kim, Y.-J. 385
Kohlhagen, S.W. 325
653

N. Privault
Korn, E. 493
Korn, R. 493
Kreps, D.M. 41, 42
Kroisandt, G. 493
Kumar, D. 173
Lamberton, D. 59, 255
Lando, D. 409, 412
Lapeyre, B. 255
Lee, S. 423
Lemarechal, C. 18
Lesniewski, A.S. 173
Levy, E. 250, 252
Li, D.X. 422, 432
Litzenberger, R.h. 168
Liu, B. 413
Longstaff, F.A. 299, 301
Maisonneuve, B. 411
Margrabe, W. 329
Matsumoto, H. 246
Menn, C. 410, 412
Mercurio, F. 345, 369, 606
Merton, R. 330
Merton, R. C. 428, 439
Merton, R.C. 426
Meyer, P.A. 302, 411
Mikosch, T. 543
Morton, A. 358
Musiela, M. 370
Revuz, D. 74
Rochet, J.-C. 315, 331
Rogers, C. 257
Ross, S.A. 42, 340
Rouault, A. 250
Rubinstein, M. 42
Rudin, W. 79, 81
Salmon, F. 422
Santa-Clara, P. 377
Sato, K. 468
Schied, A. 11, 17, 20, 41, 42, 59, 68
Schoenmakers, J. 394
Schroder, M. 249, 250
Schwartz, E.S. 299, 301
Shi, Z. 257
Shiryaev, A.N. 111
Shiu, E.S.W. 309
Shreve, S. 198, 207, 222, 261, 286,
307, 336, 565
Singleton, K. 412
Sircar, K.R. 159, 172, 177
Sircar, R. 178
Slna, K. 172, 177, 178
Sornette, D. 377
Steele, J.M. 295
Tankov, P. 452, 457, 463, 468
Teng, T.-R. 331, 387, 394
Turnbull, S.M. 250
Nguyen, H.T. 445

Norris, J.R. 448
Uy, W.I. 350
ksendal, B. 59, 145
Vasicek, O. 339, 345, 428, 439

Vorst, A.C.F. 247
Papanicolaou, A. 172
Papanicolaou, G. 159, 172, 177, 178
Persson, J. 35, 215
Pintoux, C. 349, 350
Pironneau, O. 170
Pitman, J. 497
Pliska, S.R. 111
Proske, F. 59, 145
Protter, P. 90, 97, 138, 145, 320,
329, 331, 343, 344, 411, 497
Rebonato, R. 173
Wakeman, L. 250
Watanabe, S. 86
White, A. 357, 428
Widder, D.V. 120
Williams, D. 59
Wong, H.Y. 35
Woodward, D.E. 173
Yor, M. 74, 245, 247, 250, 411, 412,
576
Yu, J.D. 252
654
"
References
1. Y. Achdou and O. Pironneau. Computational methods for option pricing, volume 30 of Frontiers in Applied Mathematics. Society for Industrial and Applied
Mathematics (SIAM), Philadelphia, PA, 2005. (Cited on page 170).
2. P. Barrieu, A. Rouault, and M. Yor. A study of the Hartman-Watson distribution motivated by numerical problems related to the pricing of Asian options.
J. Appl. Probab., 41(4):10491058, 2004. (Cited on page 250).
3. H. Bermin. Essays on Lookback options: a Malliavin calculus approach. PhD
thesis, Lund University, 1998. (Cited on page 241).
4. T. Bj
ork. Arbitrage Theory in Continuous Time, volume 121 of Oxford Finance.
Oxford University Press, 2004. (Cited on page 27).
5. T. Bj
ork. On the geometry of interest rate models. In Paris-Princeton Lectures
on Mathematical Finance 2003, volume 1847 of Lecture Notes in Math., pages
133215. Springer, Berlin, 2004. (Cited on pages 365 and 366).
6. D. Bosq and H.T. Nguyen. A Course in Stochastic Processes: Stochastic Models
and Statistical Inference. Mathematical and Statistical Methods. Kluwer, 1996.
(Cited on page 445).
7. A. Brace, D. Gatarek, and M. Musiela. The market model of interest rate
dynamics. Math. Finance, 7(2):127155, 1997. (Cited on page 370).
8. D.T. Breeden and R.h. Litzenberger. Prices of state-contingent claims implicit
in option prices. Journal of Business, 51:621651, 1978. (Cited on page 168).
9. D. Brigo and F. Mercurio. Interest rate modelstheory and practice. Springer
Finance. Springer-Verlag, Berlin, second edition, 2006. (Cited on pages 345,
369, and 606).
10. P. Carr and M. Schr
oder. Bessel processes, the integral of geometric Brownian
motion, and Asian options. Theory Probab. Appl., 48(3):400425, 2004. (Cited
on pages 249 and 250).
11. G. Castellacci.
Bootstrapping Credit Curves from CDS Spread
Curves.
Available at SSRN: http://ssrn.com/abstract=2042177 or
http://dx.doi.org/10.2139/ssrn.2042177, 2008. (Cited on pages 425 and 442).
12. R.-R. Chen, X. Cheng, F.J. Fabozzi, and B. Liu. An explicit, multi-factor credit
default swap pricing model with correlated factors. Journal of Financial and
Quantitative Analysis, 43(1):123160, 2008. (Cited on page 413).
13. R.-R. Chen and J.-Z. Huang. Credit spread bounds and their implications
for credit risk modeling. Working paper, Rutgers University and Penn State
University, 2001. (Cited on page 413).
14. M. ClarkS(Director). Trillion Dollar Bet. PBS Nova Documentaries, 2000.
(Cited on page 3).
655
N. Privault
15. R. Cont and P. Tankov. Financial modelling with jump processes. Chapman &
Hall/CRC Financial Mathematics Series. Chapman & Hall/CRC, Boca Raton,
FL, 2004. (Cited on pages 452, 457, 463, and 468).
16. J.C. Cox, J.E. Ingersoll, and S.A. Ross. A theory of the term structure of interest
rates. Econometrica, 53:385407, 1985. (Cited on page 340).
17. J.C. Cox, S.A. Ross, and M. Rubinstein. Option pricing: A simplified approach.
Journal of Financial Economics, 7:87106, 1979. (Cited on page 42).
18. R.-A. Dana and M. Jeanblanc. Financial markets in continuous time. Springer
Finance. Springer-Verlag, Berlin, 2007. Corrected Second Printing. (Cited on
page 233).
19. J.W. Dash. Quantitative finance and risk management. World Scientific Publishing Co. Inc., River Edge, NJ, 2004. (Cited on page 317).
20. C. Dellacherie, B. Maisonneuve, and P.A. Meyer. Probabilit
es et Potentiel, volume 5. Hermann, 1992. (Cited on page 411).
21. J. L. Devore. Probability and Statistics for Engineering and the Sciences.
Duxbury Press, sixth edition, 2003. (Cited on page 497).
22. J.L. Doob. Stochastic processes. John Wiley & Sons Inc., New York, 1953.
23. J.L. Doob. Classical potential theory and its probabilistic counterpart. SpringerVerlag, Berlin, 1984. (Cited on page 270).
24. L.U. Dothan. On the term structure of interest rates. Jour. of Fin. Ec., 6:5969,
1978. (Cited on pages 341 and 348).
25. R.M. Dudley. Real analysis and probability, volume 74 of Cambridge Studies in
Advanced Mathematics. Cambridge University Press, Cambridge, 2002. Revised
reprint of the 1989 original. (Cited on page 82).
26. D. Duffie and K.J. Singleton. Credit risk. Princeton Series in Finance. Princeton
University Press, Princeton, NJ, 2003. (Cited on page 412).
27. D. Dufresne. Laguerre series for Asian and other options. Math. Finance,
10(4):407428, 2000. (Cited on page 250).
28. B. Dupire. Pricing with a smile. Risk Magazine, 7(1):1820, 1994. (Cited on
page 169).
29. R.J. Elliott and M. Jeanblanc. Incomplete markets with jumps and informed
agents. Math. Methods Oper. Res., 50(3):475492, 1999. (Cited on page 412).
30. R.J. Elliott, M. Jeanblanc, and M. Yor. On models of default risk. Math.
Finance, 10(2):179195, 2000. (Cited on page 411).
31. J. Eriksson and J. Persson. Pricing turbo warrants. Preprint, 2006. (Cited on
pages 35 and 215).
32. W. Feller. Two singular diffusion problems. Ann. of Math. (2), 54:173182,
1951. (Cited on page 340).
33. G. B. Folland. Real analysis. Pure and Applied Mathematics (New York). John
Wiley & Sons Inc., New York, second edition, 1999. (Cited on page 74).
34. H. F
ollmer and A. Schied. Stochastic finance, volume 27 of de Gruyter Studies
in Mathematics. Walter de Gruyter & Co., Berlin, 2004. (Cited on pages 11,
17, 20, 41, 42, 59, and 68).
35. J.P. Fouque, G. Papanicolaou, and K.R. Sircar. Derivatives in financial markets
with stochastic volatility. Cambridge University Press, Cambridge, 2000. (Cited
on page 159).
36. J.P. Fouque, G. Papanicolaou, K.R. Sircar, and K. Slna. Multiscale Stochastic
Volatility for Equity, Interest Rate Derivatives, and Credit Derivatives. Cambridge University Press, Cambridge, 2011. (Cited on pages 172, 177, 178, 180,
and 181).
37. M.B. Garman and S.W. Kohlhagen. Foreign currency option values. J. International Money and Finance, 2:231237, 1983. (Cited on page 325).
38. J. Gatheral. The Volatility Surface: A Practitioners Guide. Wiley, 2006. (Cited
656
"

39. H. Geman, N. El Karoui, and J.-C. Rochet. Changes of num
eraire, changes of
probability measure and option pricing. J. Appl. Probab., 32(2):443458, 1995.
(Cited on pages 315 and 331).
40. H. Geman and M. Yor. Bessel processes, Asian options and perpetuities. Math.
Finance, 3:349375, 1993. (Cited on pages 247, 250, and 576).
41. H.U. Gerber and E.S.W. Shiu. Martingale approach to pricing perpetual American options on two stocks. Math. Finance, 6(3):303322, 1996. (Cited on
page 309).
42. M. Gibson. Understanding the risk of synthetic CDOs. Fedreal Reserve Board
Working paper, 2004. (Cited on page 428).
43. P. Glasserman. Monte Carlo methods in financial engineering, volume 53 of
Applications of Mathematics (New York). Springer-Verlag, New York, 2004.
Stochastic Modelling and Applied Probability. (Cited on page 493).
44. M.R. Grasselli and T.-R. Hurd. Credit risk modeling. Lecture notes, McMaster
University, January 3, 2010. (Cited on page 426).
45. X. Guo, R. Jarrow, and C. Menn. A note on Landos formula and conditional
independence. Preprint, 2007. (Cited on pages 410 and 412).
46. P.S. Hagan, D. Kumar, A.S. Lesniewski, and D.E. Woodward. Managing smile
risk. Wilmott Magazine, pages 84108, 2002. (Cited on page 173).
47. J.M. Harrison and D.M. Kreps. Martingales and arbitrage in multiperiod securities markets. Journal of Economic Theory, 20:341408, 1979. (Cited on
pages 41 and 42).
48. J.M. Harrison and S.R. Pliska. Martingales and stochastic integrals in the theory
of continuous trading. Stochastic Process. Appl., 11:215260, 1981. (Cited on
page 111).
49. D. Heath, R. Jarrow, and A. Morton. Bond pricing and the term structure of
interest rates: a new methodology. Econometrica, 60:77105, 1992. (Cited on
page 358).
50. S.L. Heston. A closed-form solution for options with stochastic volatility with
applications to bond and currency options. The Review of Financial Studies,
6(2):327343, 1993. (Cited on pages 172 and 178).
51. J.-B. Hiriart-Urruty and C. Lemar
echal. Fundamentals of convex analysis.
Grundlehren Text Editions. Springer-Verlag, Berlin, 2001. (Cited on page 18).
52. J. Hull and A. White. Pricing interest rate derivative securities. The Review of
Financial Studies, 3:537592, 1990. (Cited on page 357).
53. J. Hull and A. White. Valuation of a CDO and an n-th to default CDS without Monte Carlo simulation. Journal of Derivatives, 12:823, 2004. (Cited on
page 428).
54. N. Ikeda and S. Watanabe. Stochastic Differential Equations and Diffusion
Processes. North-Holland, 1989. (Cited on page 86).
55. J. Jacod. Grossissement initial, hypoth`
ese (H) et th
eor`
eme de Girsanov. In
Th. Jeulin and M. Yor, editors, Grossissements de filtrations: exemples et applications, volume 1118 of Lecture Notes in Mathematics, pages 614. SpringerVerlag, Berlin, 1985. (Cited on page 412).
56. J. Jacod and P. Protter. Probability essentials. Springer-Verlag, Berlin, 2000.
57. F. Jamshidian. An exact bond option formula. The Journal of Finance,
XLIV(1):205209, 1989. (Cited on page 384).
58. F. Jamshidian. Sorting out swaptions. Risk Magazine, 9(3):5960, 1996. (Cited
on page 331).
59. M. Jeanblanc and N. Privault. A complete market model with Poisson and
Brownian components. In R. Dalang, M. Dozzi, and F. Russo, editors, Seminar on Stochastic Analysis, Random Fields and Applications (Ascona, 1999),
volume 52 of Progress in Probability, pages 189204. Birkh
auser, Basel, 2002.
"
657
N. Privault
60. M. Jeanblanc, M. Yor, and M. Chesney. Mathematical methods for financial
markets. Springer Finance. Springer-Verlag London Ltd., London, 2009.
61. Th. Jeulin. Semi-martingales et grossissement dune filtration, volume 833 of
Lecture Notes in Mathematics. Springer Verlag, 1980. (Cited on page 412).
62. S. Jones. The formula that felled Wall Street. Financial Times Magazine, April
24 2009. (Cited on page 423).
63. K.Aas. The basel II IRB approach for credit portfolios: A survey. Note
No SAMBA/33/05, Norwegian Computing Center, October 2005. (Cited on
page 439).
64. O. Kallenberg. Foundations of Modern Probability. Probability and its Applications. Springer-Verlag, New York, second edition, 2002. (Cited on page 520).
65. A.G.Z. Kemna and A.C.F. Vorst. A pricing method for options based on average
asset values. Journal of Banking and Finance, 14:113129, 1990. (Cited on
page 247).
66. Y. El Khatib. Contributions to the study of discontinuous markets via the
Malliavin calculus. PhD thesis, Universit
e de La Rochelle, 2003. (Cited on
page 244).
67. Y. El Khatib and N. Privault. Computations of replicating portfolios in complete
markets driven by normal martingales. Applicationes Mathematicae, 30:147
172, 2003. (Cited on page 241).
68. Y.-J. Kim. Option pricing under stochastic interest rates: An empirical investigation. Asia-Pacific Financial Markets, 9:2344, 2002. (Cited on page 385).
69. R. Korn, E. Korn, and G. Kroisandt. Monte Carlo methods and models in
finance and insurance. Chapman & Hall/CRC Financial Mathematics Series.
CRC Press, Boca Raton, FL, 2010. (Cited on page 493).
70. D. Lamberton and B. Lapeyre. Introduction to stochastic calculus applied to
finance. Chapman & Hall, London, 1996. (Cited on pages 59 and 255).
71. D. Lando. On Cox processes and credit risky securities. Review of Derivative
Research, 2:99120, 1998. (Cited on pages 409 and 412).
72. S. Lee. Formula from hell. Forbes.com August 8 2009. (Cited on page 423).
73. E. Levy. Pricing European average rate currency options. Journal of International Money and Finance, 11:474491, 1992. (Cited on pages 250 and 252).
74. D.X. Li. On default correlation - a copula function approach. The Journal of
Fixed Income, 9(4):4354, 2000. (Cited on pages 422 and 432).
75. F.A. Longstaff and E.S. Schwartz. Valuing american options by simulation: a
simple least-squares approach. Review of Financial Studies, 14:113147, 2001.
76. W. Margrabe. The value of an option to exchange one asset for another. The
Journal of Finance, XXXIII(1):177186, 1978. (Cited on page 329).
77. H. Matsumoto and M. Yor. Exponential functionals of Brownian motion. I.
Probability laws at fixed time. Probab. Surv., 2:312347 (electronic), 2005.
78. R. Merton. Theory of rational option pricing. Bell Journal of Economics,
4(1):141183, 1973. (Cited on page 330).
79. R. C. Merton. On the pricing of corporate debt: The risk structure of interest
rates. Journal of Finance, 29:449470, 1974. (Cited on pages 426, 428, and 439).
80. T. Mikosch. Elementary stochastic calculuswith finance in view, volume 6 of
Advanced Series on Statistical Science & Applied Probability. World Scientific
Publishing Co. Inc., River Edge, NJ, 1998. (Cited on page 543).
81. J.R. Norris. Markov Chains, volume 2 of Cambridge Series in Statistical
and Probabilistic Mathematics. Cambridge University Press, Cambridge, 1998.
Reprint of 1997 original. (Cited on page 448).
82. G. Di Nunno, B. ksendal, and F. Proske. Malliavin Calculus for L
evy Processes
with Applications to Finance. Universitext. Springer-Verlag, Berlin, 2009. (Cited
658
"

83. Basel Committee on Banking Supervision. An explanatory note on the basel II
IRB risk weight functions. July 2005. (Cited on page 439).
84. A. Papanicolaou and R. Sircar. A regime-switching Heston model for VIX and
S&P 500 implied volatilities. Quant. Finance, 14(10):18111827, 2014. (Cited
on page 172).
85. C. Pintoux and N. Privault. A direct solution to the Fokker-Planck equation
for exponential Brownian functionals. Analysis and Applications, 8(3):287304,
86. C. Pintoux and N. Privault. The Dothan pricing model revisited. Math. Finance,
21:355363, 2011. (Cited on pages 349 and 350).
87. J. Pitman. Probability. Springer, 1999. (Cited on page 497).
88. N. Privault. Stochastic analysis of Bernoulli processes. Probab. Surv., 5:435483
(electronic), 2008. (Cited on page 59).
89. N. Privault. Stochastic analysis in discrete and continuous settings with normal
martingales, volume 1982 of Lecture Notes in Mathematics. Springer-Verlag,
Berlin, 2009. (Cited on pages 59, 60, 61, 82, 86, 132, 145, and 403).
90. N. Privault. An Elementary Introduction to Stochastic Interest Rate Modeling
(Second Edition). Advanced Series on Statistical Science & Applied Probability,
16. World Scientific Publishing Co., Singapore, 2012. (Cited on pages vi, 357,
365, 366, 367, 369, and 385).
91. N. Privault. Understanding Markov Chains. Springer Undergraduate Mathematics Series. Springer, 2013. (Cited on page 448).
92. N. Privault and T.-R. Teng. Risk-neutral hedging in bond markets. Risk and
Decision Analysis, 3:201209, 2012. (Cited on pages 331, 387, and 394).
93. N. Privault and W.T. Uy. Monte Carlo computation of the Laplace transform of
exponential Brownian functionals. Methodol. Comput. Appl. Probab., 15(3):511
524, 2013. (Cited on page 350).
94. N. Privault and J.D. Yu. Stratified approximations for the pricing of options
on average. to appear in Journal of Computational Finance, 2015. (Cited on
page 252).
95. P. Protter. A partial introduction to financial asset pricing theory. Stochastic
Process. Appl., 91(2):169203, 2001. (Cited on pages 145 and 331).
96. P. Protter. Stochastic integration and differential equations, volume 21 of
Stochastic Modelling and Applied Probability. Springer-Verlag, Berlin, second
edition, 2004. (Cited on pages 90, 97, 138, 145, 320, 329, 343, 344, and 411).
97. R. Rebonato. The SABR/LIBOR Market Model Pricing, Calibration and Hedging for ComplexInterest-Rate Derivatives. John Wiley & Sons, 2009. (Cited on
page 173).
98. D. Revuz and M. Yor. Continuous Martingales and Brownian Motion. SpringerVerlag, 1994. (Cited on page 74).
99. L.C.G. Rogers and Z. Shi. The value of an Asian option. J. Appl. Probab.,
32(4):10771088, 1995. (Cited on page 257).
100. M. Rubinstein. Pay now, choose later. Risk Magazine, 4:1313, 1991. (Cited on
page 149).
101. W. Rudin. Real and Complex Analysis. Mc graw-Hill, 1974. (Cited on pages 79,
81, and 82).
102. J. Ruiz de Ch
avez. Predictable representation of the binomial process and
application to options in finance. In XXXIII National Congress of the Mexican Mathematical Society (Spanish) (Saltillo, 2000), volume 29 of Aportaciones
Mat. Comun., pages 223230. Soc. Mat. Mexicana, M
exico, 2001. (Cited on
page 59).
103. F. Salmon. Recipe for disaster: the formula that killed Wall Street. Wired
Magazine, 23 February 2009. (Cited on page 422).
"
659
N. Privault
104. P. Santa-Clara and D. Sornette. The dynamics of the forward interest rate curve
with stochastic string shocks. The Review of Financial Studies, 14(1):149185,
105. K. Sato. L
evy processes and infinitely divisible distributions, volume 68 of Cambridge Studies in Advanced Mathematics. Cambridge University Press, Cambridge, 1999. (Cited on page 468).
106. J. Schoenmakers. Robust LIBOR modelling and pricing of derivative products.
Chapman & Hall/CRC Financial Mathematics Series. Chapman & Hall/CRC,
Boca Raton, FL, 2005. (Cited on page 394).
107. M. ScorseseS(Director). The Wolf of Wall Street. Paramount Pictures, 2013.
108. A.N. Shiryaev. Essentials of stochastic finance. World Scientific Publishing Co.
Inc., River Edge, NJ, 1999. (Cited on page 111).
109. S.E. Shreve. Stochastic calculus for finance. II. Springer Finance. SpringerVerlag, New York, 2004. Continuous-time models. (Cited on pages 198, 207,
222, 261, 286, 307, 336, and 565).
110. J.M. Steele. Stochastic Calculus and Financial Applications, volume 45 of Applications of Mathematics. Springer-Verlag, New York, 2001. (Cited on page 295).
111. S. Turnbull and L. Wakeman. A quick algorithm for pricing European average
options. Journal of Financial and Quantitative Analysis, 26:377389, 1992.
112. O. Vasi
cek. Loan portfolio value. Risk Magazine, pages 160162, 01 Dec 2002.
113. O. Vasi
cek. An equilibrium characterisation of the term structure. Journal of
Financial Economics, 5:177188, 1977. (Cited on pages 339 and 345).
114. O. Vasi
cek. Probability of loss on loan portfolio. KMV Corporation Technical
Report, 1987. (Cited on page 428).
115. D.V. Widder. The heat equation. Academic Press, New York, 1975. Pure and
Applied Mathematics, Vol. 67. (Cited on page 120).
116. D. Williams. Probability with martingales. Cambridge Mathematical Textbooks.
Cambridge University Press, Cambridge, 1991. (Cited on page 59).
117. H.Y. Wong and C.M. Chan. Turbo warrants under stochastic volatility. Quant.
Finance, 8(7):739751, 2008. (Cited on page 35).
118. M. Yor. Grossissements de filtration et absolue continuit
e de noyaux. In Th.
Jeulin and M. Yor, editors, Grossissements de filtrations: exemples et applications, volume 1118 of Lecture Notes in Mathematics, pages 614. SpringerVerlag, Berlin, 1985. (Cited on page 412).
119. M. Yor. On some exponential functionals of Brownian motion. Adv. in Appl.
Probab., 24(3):509531, 1992. (Cited on page 245).
660
"
"
661
N. Privault
This book is an introduction to the stochastic calculus and PDE approaches to the pricing and hedging of financial derivatives, including vanilla
options and exotic options. The presentation is done both in discrete and
continuous-time financial models, with an emphasis on the complementarity between algebraic and probabilistic methods. It also covers the pricing
of some interest rate derivatives, American options, exotic options such as
barrier, lookback and Asian options, and stochastic models with compound
Poisson jumps. The text is accompanied with a number of figures and simulations, and includes 20 examples based on actual market data.
662
"

Stochastic Finance

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Stochastic Finance

Uploaded by

Copyright:

Available Formats

Nicolas Privault

Notes on Stochastic Finance

This version: April 25, 2013

Notes on Stochastic Finance

Notes on Stochastic Finance

Assets, Portfolios and Arbitrage . . . . . . . . . . . . . . . . . . . . . . . . . .

Pricing and hedging in discrete time . . . . . . . . . . . . . . . . . . . . . .

Brownian Motion and Stochastic Calculus . . . . . . . . . . . . . . . .

The Black-Scholes PDE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Martingale Approach to Pricing and Hedging . . . . . . . . . . . . .

Notes on Stochastic Finance

11 Forward Rate Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

12 Pricing of Interest Rate Derivatives . . . . . . . . . . . . . . . . . . . . . .

13 Credit Risk, CDSs and CDOs . . . . . . . . . . . . . . . . . . . . . . . . . . . .

15 Pricing and Hedging in Jump Models . . . . . . . . . . . . . . . . . . . . .

16 Basic Numerical Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Appendix: Background on Probability Theory . . . . . . . . . . . . . . . .

This version: April 25, 2013

Notes on Stochastic Finance

As if a whole new world was laid out before me. . . . . . . . . . . . . .

Illustration of the self-financing condition (2.3). . . . . . . . . . . . . . . . . .

Sample paths of a one-dimensional Brownian motion. . . . . . . . . . . . .

Illustration of the self-financing condition (5.2). . . . . . . . . . . . . . . . . .

Drifted Brownian path. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

Gamma of a European option. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Brownian motion Bt and its supremum Xt . . . . . . . . . . . . . . . . . . . . 187

This version: April 25, 2013

Notes on Stochastic Finance

Graph of the underlying asset price. . . . . . . . . . . . . . . . . . . . . . . . . . .

Drifted Brownian path. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

11.1 Graph of t 7 rt in the Vasicek model. . . . . . . . . . . . . . . . . . . . . . . .

This version: April 25, 2013

13.3 Function x 7 1 (x) + ( r) T t/ for > r, = r, and

14.1 Sample path of a Poisson process (Nt )tR+ . . . . . . . . . . . . . . . . . . . . .

16.1 Divergence of the explicit finite difference method. . . . . . . . . . . . . . . . 491

Notes on Stochastic Finance

Animated figures (work in Acrobat reader).

Modern mathematical finance and quantitative analysis require a strong

Albert Einstein, Uber

Notes on Stochastic Finance

Notes on Stochastic Finance

European Call and Put Options

Fig. 0.3: Sample price processes simulated by a geometric Brownian motion.

Notes on Stochastic Finance

Fig. 0.4: Infogrames stock price curve.

An illustration - pricing and hedging in a binary model

The following issues can be addressed:

at time t = 1, one has to:

3. invest those $2 + $2 = $4 into the purchase of = 1 unit of stock valued

Notes on Stochastic Finance

IE[P ] = $3 P(S1 = $5) = $2

P(S1 = $5) + P(S1 = $2) = 1.

from which we can infer the probabilities

Assets, Portfolios and Arbitrage

1.1 Definitions and Formalism

= (0) , (1) , . . . , (d)

denotes the prices (i) > 0 at time t = 0 of d + 1 assets numbered

1.2 Portfolio Allocation and Short-Selling

Assets, Portfolios and Arbitrage

= (i) S (i) (1 + r) (i)

or S (1 + r) per unit of stock invested, which is positive provided

[start from 0 or even with a debt]

iii) P( S > 0) > 0.

[a profit is made with non-zero probability]