You are on page 1of 5

Management 402 Logs and Exponentials

Fall 2012

In class it has been convenient to make use of the logarithmic transformation of regression data. The purpose of this note is to brush up on the salient facts about logarithms. We give only a brief sketch, concentrating only on the facts needed to understand lecture. Lets start with some basic facts about powers. Powers Starting from the beginning, for any number x: x0 = 1, x1 = x, and x2 = (x)(x). Continuing in this fashion, for any positive integer, n, xn = x(xn1 ) = x(x)...(x) (which amounts to saying that xn = x multiplied by itself n times). Negative powers are dened as follows: 1 xn = n x 1/2 root. Thus x = x. We will also need some familiar algebraic

The 1/nth power is the nth properties of powers:

xa xb = xa+b (xa )b = xab Logs For a positive number, a, the logarithm base a of x is the power to which we must raise a to get x. Thus the logarithm base 2 of 8 is 3, written in terms of symbols: log2 (8) = 3. To check note that 23 = 2(2)(2) = 8. Exponentiation and logarithm are inverses in the sense that aloga (x) = x The relationship between logs and exponentiation gives us some very useful properties of logs:

loga (a) = 1, 1

loga (xy ) = loga (x) + loga (y ), and loga (xz ) = z loga (x). The log function is concave and increasing, but is not dened for x 01 . As x approaches 0 from above, loga (x) decreases without limit. Suppose we know loga (x), how do we nd logb (x) where b = a? We know that x = aloga (x) . Hence logb (x) = logb (aloga (x) ) Hence logb (x) = loga (x) logb (a). Translating from logarithms base a to logarithms base b requires only knowing logb (a), it also says that logarithm base b of x is proportional to logarithm base a of x. The constant e and natural logs The irrational number e is often used as the base when taking logarithms. When we take logs base e we call these natural logs2 . The constant e is equal3 . to 2.718.... and it does not have a nite decimal representation (it shares this property with the other popular irrational number, ). We use the symbol ln(x) to denote loge (x). The inverse function corresponding to natural log is the exponential function, ex . Since they are inverses we have eln(x) = x and ln(ex ) = x ln(e) = x.
1

More precisely, a real logarithm of a non-positive number is not dened, in the same sense that we say

the square root function is not dened for negative real numbers. Just as with square roots, we may remedy this problem by expanding our view point to encompass complex numbers (numbers involving the imaginary number i). A treatment of complex logarithms is beyond the scope of this note. 2 The term natural log seems to be an oxymoron to most students. Taking logs doesnt seem particularly natural to begin with, taking logs to an irrational base seems even less so. 3 In a course on advance calculus, we would obtain e as a limit: ex = lim
n

1+

x n

For example, setting x = 1 and taking n = 5000 gives (1 + 1/5000)5000 = 2.7180101 whereas e to seven decimal places is 2.7182818. Modifying the example above by taking n = 5000, and x = 0.05, we have (1 + 0.05/5000)5000 e0.05 1.0512708336 1.05127109638 2.628 107 .

The following gure graphs, ln(x), ex , and ex .. 4 3 2 1 0 -3 -2 -1 -1 -2 -3 -4 Continuous Compounding Suppose that you have an asset with an intial value of $1 and which increases in value at a rate of 10% per year compounded annually. The value of the asset at then end of a year will be $1.11 . In 2 years the asset will be worth $1.12 . Suppose, however that we compound the growth semi-annually, the formula for the value after one year would be $(1 +
0.1 2 ) 2

ln(x) exp(x) exp(-x)

= 1.052

and after 2 years is $1.054 . If we were to divide the year into n periods and compound each
.1 2n years growth in n equal installments the value of the asset would be $(1 + ( 0n )) . The

table below gives the value after two years for dierent compounding intervals

Compounding interval Annual Semi-Annual Monthly Weekly Monthly Continuous $(1 +

Return in 2 years $(1 + 0.1)2 = 1.12 = 1.21


0.1 2(2) ) = 1.054 = 1.2155 2 .1 12(2) $(1 + 0 ) = 1.008324 = 1.2204 12 .1 52(2) $(1 + 0 ) = 1.0019104 = 1.2212 52 0.1 365(2) $(1 + 365 ) = 1.0003730 = 1.22137 .1(2)

$e

= 1.221403

To generalize this a bit, suppose that the asset experiences compound growth at rate r per year compounded n times per year (equally spaced), at the end of the rst year the 3

value of the asset would be $(1 + (r/n))n , at the end of t years it will be worth $(1 + (r/n))nt . As the number of compounding periods (n) increases without limit we have 1+ r n
nt

ert

At the end of t years, compounded continuously at rate r, the asset will be worth ert . In a similar fashion, an asset which depreciates at rate r compounded continuously, has a value of $ert after t years. Note that when we write ert or ert , r is the rate of appreciation or depreciation per year and t is the number of years, the time scale for r and t must match. The Rule of 69 Once we have in hand the idea that the function ert represents the value of an asset undergoing constant continuous growth rate of r for a length of time t, we can construct a rule of thumb for how long it takes for that asset to double in value. Our hypothetical asset starts with value 1, i.e. er0 = 1. How long until it reaches value 2? This is pretty simple to work out. We wish to nd the value of t that makes 2 = ert Taking natural logs on both sides yields ln(2) = rt Hence t = ln(2)/r. Now the rule of thumb: ln(2) 0.693147 . . ., round this to 0.69. Our asset doubles every .69/r time units. To make the arithmetic easier, write r as a percentage, and multiply 0.69 by 10, then t 69/(percentage growth rate). If I have an asset that is growing at a continuously compounded annual rate of 10%, it will double (approximately) every 69/10 = 6.9 years. To check: e(0.1)(6.9) = 1.993716 2. For slower compounding rates (monthly, annual, etc.) the rule of 69 provides an approximate lower bound on the doubling time. Logarithmic Scales Many natural phenomena are most conveniently described on a logarithmic scale. The Richter scale for earthquake intensity is roughly: R = log10 (A/A0 ) 4

where A is the magnitude of the the deection of the needle on a seismograph (which measures amount of ground movement), and A0 is a base level of movement. Suppose now that we experience two earthquakes, one a magnitude 3 and the next a magnitude 4. How does the amount of movement change between the two? 3 = log10 (A1 /A0 ) 103 = (A1 /A0 ) A1 = A0 103 and 4 = log10 (A2 /A0 ) 104 = (A2 /A0 ) A2 = A0 104 so A2 = 10A1 . The second earthquake created 10 times the shake as the rst. Sound intensity is also typically measured on a logarithmic scale. The magnitude of a sound of intensity X in decibels is d = 10 log10 (X/X0 ) where X0 is a reference sound level close to the threshold of human hearing. Setting X = X0 yields d0 = 10log10 (X0 /X0 ) = 10 log10 (1) = 0. A signal at the threshold of perception (detectable by a young person with normal hearing) is a 0 dB signal. A rock concert will often produce a 120 dB signal. To nd the sound intensity for a 120 dB signal solve 120 = 10 log10 (X/A0 ) or X = 1012 A0 or one trillion times stronger! The amazing thing is that we can hear anything at a rock concert. It speaks to the marvelous dynamic range of our sense if hearing. We also typically measure the frequency (or pitch) of sound on a logarithmic scale. The standard diatonic scale comprises 12 semitones or half-steps in an octave4 . The frequency of a pitch doubles with each octave, every 12 semi-tones yields a doubling of the pitch. In fact, one could, without cringing too much, call this the logarithmic scale.

The standard western diatonic has seven steps in an octave (do, re, mi, fa,so, la, ti, do). However, these steps are not equally spaced in frequency, so it is closer to our notion of logarithmic scale to count semi-tones, or half-steps.

You might also like