Professional Documents
Culture Documents
ENFORCEMENT
This paper presents a rare attempt to quantify the deterrent effect of anticartel policy.
It develops a conceptual framework, which establishes the sort of information necessary
for such quantification. This is then illustrated and calibrated by drawing upon existing
literatures and using evidence from legal cartels to approximate what would be observed
absent policy. Measuring impact by the proportion of all potential harm that is deterred,
our best estimate is two-thirds and, even on conservative assumptions, at least half of
all harms (or seven times the detected harm) is deterred. (JEL H11, K21, L44)
deterrence effects of their work and to assess its We review existing cartel literatures for
macroeconomic impact.”2 guidance on calibrating this framework. On
However, there is a further related issue, which some parts, we can be fairly confident in nar-
is also typically neglected in the evaluation lit- rowing down the possible assumptions: for
erature: what about a second type of unobserved example, previous empirical literatures point
case—those which the CA fails to detect, even to an aggregate probability of detection within
though they involve anticompetitive harm? In the range (0.1, 0.33); and the distribution of
effect, these represent a foregone opportunity, or cartel harm (overcharge) is likely to be positively
even failure of policy; again, this is never quanti- skewed. On other parts, conventional cartel
fied, although the magnitudes involved could also theory provides guidance: notably, on how the
be substantial. deterrence and detection rates are likely to vary
This suggests that a more encompassing across cartels with overcharge. On the other
approach to evaluation would be to ask first hand, we have very little hard evidence on the
“how much potential for anti-competitive harm likely magnitude of the aggregate probability
is out there in an economy?” Then, to assess the of deterrence.
success of policy, further ask “how much of that Inevitably, estimation within this or any other
potential harm is avoided and/or removed by the framework will entail significant uncertainty
presence of anti-cartel laws and the activities and, for this reason, we opt for Monte Carlo
of competition authorities?” The first question experiments, with a strong emphasis on the
is reminiscent of an old literature, provoked by lower bounds of the estimates, focussing more
Harberger (1954), which attempted to quantify on what outcomes can be ruled out, than attach-
the “social costs of monopoly.” Our approach ing undue confidence to specific point estimates.
is undoubtedly ambitious, but the potential Two sets of experiments are presented. The first
value of any practicable methodology to incor- set is based on fairly “general” weak assump-
porate deterrence within policy evaluation is tions; for example, both aggregate deterrence
self-evident—not only for assessing the impact and detection probabilities are allowed to lie
of competition law and its agencies, but also anywhere within wide ranges, and asymmetry
for informing prioritization of resources within within the population distribution is allowed
the authorities. to lie anywhere between perfect homogeneity
We begin by setting out a conceptual frame- and extreme heterogeneity. This plays a largely
work to establish what sort of information would ground-clearing role.
be needed to answer these questions. This is set The second set of experiments is invested with
up as a simple sampling problem in which there more structure in its assumptions, and we explore
is a population of “potential” cartels, all of which how far this narrows down the range of possible
would occur in the absence of anticartel law. A outcomes. Here, assumptions are based on direct
sample is drawn—these are the set of detected observation of the real-world distribution of
cartels—and we derive an estimator for making cartel overcharge gathered originally by Connor
deductions from the sample harm about the mag- (2014). This database is wide ranging, both over
nitudes of the unsampled (deterred or undetected) time and geographically, and although most of
harms. This identifies three types of information the cartels were illegal, a sizeable minority were
required: (a) the aggregate deterrence and detec- legal (occurring in jurisdictions without prohi-
tion probabilities (which determine the aggregate bition, or where exemptions were granted). The
sample size); (b) how these probabilities vary overcharge distribution among these legal cartels
between cartels depending on their overcharge would be observed in the absence of competition
(the sampling differentials); and (c) the extent of law and agencies—the counterfactual population
cartel heterogeneity in overcharge. None of this of potential cartels. This adds more structure not
information is obviously available. The objective only on the population distribution, but also, by
of our work is to demonstrate how one could com- comparing the legal and illegal cartel overcharge
bine existing and published estimates and Monte distributions, Bos et al. (2017) provide guidance
Carlo simulations to retrieve this information. on which types of cartel are less likely to be
observed (detected) once cartels become illegal.
2. “Looking beyond the direct effects of the work of While two important sources of uncertainty
Competition Authorities: Deterrence and Macroeconomic
impact,” Conference hosted by the Netherlands Authority for remain—the aggregates detection and deter-
Consumers and Markets, the UK Competition and Markets rence rates and how far the observed sampling
Authority (CMA) and the EC, September 2015. differentials reflect differences in detection or
DAVIES, MARIUZZO & ORMOSI: ESTIMATING DETERRENCE 3
deterrence—this extra structure does indeed and their findings are used to calibrate our
narrow down the range of feasible outcomes, experiments below.
especially at the lower bound. Our three main While the deterrent effect of cartel law has
results, all based on the most conservative esti- attracted a considerable literature, much of it is
mates, are that: detected harm is only the tip of only of indirect interest for present purposes.3
the iceberg, at least half of all potential harm Among the agencies, the UK and Dutch CAs
is deterred, and at least twice as much harm is have commissioned various research projects.
undetected as is detected. These are survey studies, based on interview-
This paper contributes to, and draws on, ing competition practitioners, lawyers, and com-
various previous literatures. First and most obvi- panies and asking them directly for their opin-
ously, it aims to contribute a missing link to the ions. The most widely cited statistic to emanate
evaluation of the impact of competition policy. from these studies (Deloitte 2007 for the OFT),
This has attracted a voluminous literature, in a is that, for each cartel detected by the CA, there
variety of forms: ex-ante versus ex-post evalu- were at least another five that were deterred
ation; evaluation of individual cases, or broad by competition law.4 This is sometimes referred
areas of policy, or aggregate competition policy. to as the deterrence multiplier. Below, we refer
The quantitative methods used have also varied to “the simple multiplier approach,” which we
widely, including event studies, qualitative sur- employ as a yardstick in our own simulations.
veys, and causal inference studies (see Ormosi Clearly, any estimates based on speculative opin-
2012 for summaries). However, all these works ions in answer to hypothetical questions should
are constrained by the same limitation: driven be treated with considerable caution, and we
by data availability, inferences are made on the do not use any survey estimates directly in our
basis only of the sample of cases in which the own simulations.
CA has intervened, and fail to take account of In the more recent academic literature a num-
unobserved cases, which are deterred or unde- ber of papers are related to our own in that
tected. Of course, this has long been noted by they explore theoretically and/or empirically how
many others, including Werden (2008), Geroski information on detected cartels might be used
(2004, 3), and Baker (2003, 40). More specifi- to draw inferences about other cartels which are
cally within the evaluation literature, our paper unobserved. Miller (2009) develops a model of
is closest to the work of the CAs themselves cartel formation over time and shows empiri-
in evaluating the impact of competition policy cally that the introduction of the 1993 leniency
in aggregate, and/or the broad areas of antitrust program strengthened deterrence. Theoretically,
and merger enforcement (see footnote 1 above). Harrington and Chang (2009) establish the con-
An important theme running through this policy ditions under which information on the number
literature is the use of conservative assumptions of detected cartels, and their durations, can sup-
which give rise to lower bound estimates; given port inferences about how the number of unde-
that one purpose of evaluation is accountability tected cartels is changing. Subsequently, Har-
to stakeholders, this is clearly advisable. We also rington and Wei (2017) focus on the relation-
focus on conservative lower bounds—values of ship between the durations of detected and unde-
deterrence and undetection below which we can tected cartels, and establish empirically the sort
rule out. of selection bias that might be involved. All three
Turning to the more academic literature, papers have made valuable and imaginative con-
empirical studies on the rate of detection provide tributions on how inferences can be drawn about
a useful input into the current paper. The early an inherently unobservable population from a
work of Bryant and Eckard (1991) suggests sample of observed cartels, and this is also our
that in a given year only 13%–17% of cartels
are detected, and Combe, Monnier, and Legal 3. See, for example, the literatures on optimal fines:
(2008) confirm a similar magnitude for Europe Elzinga and Breit (1973), Landes (1983), Kobayashi (2001),
Ginsburg and Wright (2010), Werden, Hammond, and Barnett
using essentially the same method. Ormosi (2012). See also Katsoulacos and Ulph (2013) on whether rev-
(2014) proposes an alternative method that draws enue or profit should be used as the basis for calculating the
an analogy with capture-recapture analysis in optimal fine.
ecological science and finds detection rates to 4. Subsequent studies using similar survey methodolo-
gies confirm that such multipliers are likely to be significant,
be in the same ballpark. Lande and Connor although the precise magnitudes vary considerably: a larger
(2012) provide an exhaustive survey of these follow-up UK survey by London Economics (2011) reports
and other estimates of the rate of detection, much higher (1:28) deterrence rates.
4 ECONOMIC INQUIRY
objective. However, none is designed to quan- (say year) by N. This represents all harmful cases,
tify the magnitude of deterrence, as measured which would occur in the absence of competition
by overcharge: the latter two address nondetec- policy.6 With probability ω some are deterred by
tion rather than deterrence and duration rather the existence of competition law and enforce-
than overcharge, and the first identifies only ment. Within the undeterred subset, some cartels
changes over time, rather than the magnitude, are discovered and prohibited by the CA, but oth-
of harm. ers go undetected. The probability that a cartel is
Finally, two other very recent papers are detected is σ.
closely related to this one. One, Bos et al. (2017) In general, σ and ω can be time variant
is complementary to ours in that it shows, both and interdependent. For example, there may
theoretically and empirically, that cartel deter- be a dynamic causal relation running from
rence will be highest for very low and very high detection, σ, to future deterrence, ω: success
mark-up cartels. This result feeds directly into in detection by the CA might deter some firms
the assumptions employed in this current paper from attempting to contravene the law in the
on the sampling differentials, as will be seen future. Moreover, it is also possible that a more
in Section III.B. The other, Katsoulacos et al. effective CA performs well both in terms of
(2016), shares the same conceptual objective deterrence and detection, therefore ω and σ
as ours, but its approach is very different in at would be contemporaneously correlated as well.
least two respects. First it abstracts from any het- In our experiments below, we allow for the two
erogeneity among cartels, and thereby sidesteps parameters to be either positively correlated
the key issue of selection bias. Second, its main or uncorrelated.
focus is on the rarely observed phenomenon of Suppose the CA detects a sample N S of the N
recidivist cartels.5 cartel population, and estimates the magnitude of
The remainder of the paper is structured as harm they cause, whose value is denoted by H S .
follows. Section II introduces the framework, We refer to H S below synonymously as sample
couched as a simple sampling problem; this or detected harm. It follows that the expected
yields estimators for the unknown population number of detected cartels is:
harm, including deterred and undetected harms,
in terms of sample-detected harms. It identi- (1) N S = (1 − ω) σN,
⏟⏞⏟⏞⏟
fies three building blocks: the aggregate rates ≡λ
of detection and deterrence, the heterogeneity
of cartels within the potential population, and where λ is the proportionate sample size. This
the sampling differentials, which reflect differ- provides two preliminary insights. First, detected
ences in the likelihood of formation and detec- cases are probably a small proportion of the
tion between cartels with different levels of harm. potential population. As an illustrative example,
Section III draws on existing literatures for our if σ = .2 and ω = 0.5, only 10% of all potential
assumptions about these building blocks. The cases are detected. Second, a head count of the
framework is then operationalized in Section number of detected cartels tells us little about the
IV using Monte Carlo experiments to generate efficacy of the CA: detecting 10% of the popu-
estimates of the likely relative magnitudes of lation could indicate a very poor CA, with low
deterred, undetected, and detected harms. Section detection and no deterrence (σ = .1 and ω = 0),
V assesses the likely robustness of our results and or a strong CA, which deters most cartels and
Section VI concludes. detects all those which are undeterred (σ = 1 and
ω = 0.9). Both points are illustrated in the simu-
lations that follow.
II. FRAMEWORK There are similar implications for the magni-
tudes of harm, rather than the number of cases.
A. Preliminary Notation If sample cases are a random draw of the pop-
Denote the total population of all potential car- ulation, then detected harm is related to total
tels in a given economy at a given point in time
6. This counterfactual is an oversimplification to
5. We believe this emphasis is unwarranted because it the extent that some cartels might be deterred by pri-
confuses the fairly frequently observed occurrence of indi- vate enforcement, import liberalization, deregulation, and
vidual firms being involved in more than one cartel, with the other noncompetition policies. Note that we also abstract
very rare possibility of a cartel reforming after having been from the possibility of welfare-enhancing cartels, and
detected. Type 2 errors.
DAVIES, MARIUZZO & ORMOSI: ESTIMATING DETERRENCE 5
C. Selection Bias with Differential Sampling and vice versa. Population asymmetry is captured
Suppose that sampling is random within each by the magnitudes of (pL − hL ) and (hH − pH ), and
segment, but not necessarily across segments, the larger these are, the greater is the impact of
then there is a straightforward estimator for the sampling differentials on population harm.
magnitude of population harm (H).9 If the seg- This intuition can be brought into sharper
ments are sampled in potentially different pro- focus by referring back to the Lorenz curve, in
portions, λi (i = L, M, H), overall proportionate which (hH − pH ) = BC and (pL − hL ) = AD. Thus
sample size of the population is given by: (8) H ⋛ H S ∕λ as (AD∕BC)
NS ( ) ( )
(4) ≡ λ = λL pL + λM pM + λH pH , ⋛ λM − λH ∕ λM − λL .
N
and since pL + pM + pH = 1, we can substitute pM The magnitudes of AD and BC reflect the
and have asymmetry of the distribution, that is, the distance
of the Lorenz curve from the diagonal, but they
( ) ( )
(5) λ = λM + λL − λM pL + λH − λM pH . also depend on the precise form of the underlying
population distribution. ◾
Similarly, proportionate sample harm is:
(6) ( ) ( ) PROPOSITION 1. For a population distribu-
H S ∕H = λM + λL − λM hL + λH − λM hH . tion that generates symmetric Lorenz curves, for
Subtracting Equation (4) from Equation (6), example the lognormal, H will deviate by more
and with simple manipulation, from the simple multiplier (HS /λ) the larger are
the sampling differentials and the more asymmet-
H S ∕λ ric is the population distribution.
(7) H=
(λ −λ ) ( ) .
1 − L λ M pL − hL
Proof. For a symmetric concave Lorenz curve,
(λ −λ ) ( )
+ H λ M hH − pH AD = BC, that is, pL − hL = hH − pH , and
Equation (7) simplifies to:
LEMMA 1. With random sampling, the simple
H S ∕λ
).
multiplier, 1/λ, is an unbiased estimator of total (9) H=
population harm. But with differential sampling, (λ −λ ) (
1 + H λ L hH − pH
population harm also depends on (1) the sam-
pling differentials and (2) population asymmetry. Then H ⋛ H S /λ as λH ⋚ λL ; and for given
(λH − λL ), the absolute difference is increasing
Proof. From inspection, with random sam- with (hH − pH ), where (hH − pH ) is greater for
pling across segments (λL = λM = λH ), more asymmetric distributions.10
H = H S /λ. But, more generally, H ⋛ H S /λ as We assume Lorenz symmetry in the general
(λL − λM )(pL − hL ) − (λH − λM )(hH − pH ) ⋛ 0. experiments below, but merely for ease of illus-
The intuition is fairly obvious. Defining the tration.11 This assumption is not necessary for our
(λ −λ ) (λ −λ ) framework, and is not used in the evidence-based
sampling differentials as L λ M and H λ M ,
then for a given sample proportion λ, and estimate experiments on which our main conclusions are
of sample harm H S , the implied population harm based. ◾
is larger the greater is sampling of the lower tail,
(λL −λM )
λ
, and the smaller is sampling of the upper 10. To avoid confusion note that we refer to two different
types of asymmetry. A distribution is said to be asymmet-
(λH −λM )
tail, λ
, that is, the more that low harm car- ric if it is skewed, in which case the degree of asymmetry
refers to the degree of skewness. A Lorenz curve is asym-
tels are featured disproportionately in samples, metric if it is not symmetric either side of the downward
sloping diagonal. In the current context, all potential popu-
9. With this simple assumption, we effectively approx- lation distributions are likely to be asymmetric, but they may
imate continuous relationships between the sampling rate or may not exhibit asymmetric Lorenz curves. For example,
and case harm with a simple three-step function. Although the lognormal distribution is asymmetric but has symmetric
analytically crude, this is sufficiently flexible for present Lorenz curves.
purposes—the relative magnitudes of the λi can capture 11. For asymmetric Lorenz curves, the Lorenz asymme-
monotonicity or not, and concavity, linearity, or convexity. Of try coefficient (LAC) (Damgaard and Weiner 2000) enters
course, for a given continuous relationship, the magnitudes of as an additional term, where LAC ⋛ 1 as the upper tail con-
the λi are sensitive to how broadly defined are the sizes of the tributes relatively more (less) than the lower tail to overall
different segments (pi ). asymmetry.
DAVIES, MARIUZZO & ORMOSI: ESTIMATING DETERRENCE 7
D. The Estimators for Population Harm and literature for evidence of plausible ranges. There
Deterred and Undetected Harms is a large literature attempting to estimate the
This proposition relates to the relationship cartel detection rate. Lande and Connor (2012)
between population harm and sample harm. We report the results of 25 previous econometric
can interpret it in the present context in two studies, yielding estimates of σ within the range
ways. First, to find total potential population 0.1–0.33. We therefore specify σ as a random
harm in terms of detected and deterred sample variable drawn from a distribution within those
harm, recall that λi = (1 − ωi )σi and Equation (7) supports. For ω, there is much less to go on
becomes: from the previous literature, and this parameter is
allowed to take any possible value between 0 and
H S ∕ (1 − ω) σ 1, except the very extremes. To add a perspec-
(10) H =
(1−ωL )σL −(1−ωM )σM ( ) . tive, note that the midpoints of these two ranges,
1− pL − hL
(1−ω)σ ω = 0.5 and σ = 0.215, imply a deterrence mul-
(1−ωH )σH −(1−ωM )σM ( )
tiplier of approximately 5:1—that is, for every
+ (1−ω)σ
hH − pH
detected case there are five others deterred.12 This
Second, it can also be used to provide an is consistent with the most widely cited estimates
estimator for total undeterred harm (H UR ) if we in the existing literature, from the Deloitte (2007)
treat the detected cases as a sample drawn from survey described earlier, although when interpret-
the population of undeterred cartels. In that case, ing results below, we focus less on midpoints, and
the sample proportion in segment i is λi = σi and more on conservative lower bounds.
population harm refers to the total harm of all
undeterred cartels, H UR . Deterred harm (H DR )
B. The Sampling Differentials
and undetected harm (H UT ) can then be backed
out as simple residuals: For the segment parameters, the ωi and σi , and
thus the sampling differentials, we draw on the
(11) H DR = H − H UR existing theoretical literature to impose qualita-
tive constraints on relative magnitudes. Again,
and this is for both sets of experiments, but later
(12) H UT = H UR − H S . we will impose more structure on the relative
magnitudes of segment parameters using real
Equations (10)–(12) form the basis for the world evidence.
simulations below. For this we require informa- The relative magnitudes of the segment sam-
tion on: pling proportions, the λi , depend on how the prob-
1. the aggregate sampling proportion (λ) and abilities of deterrence and detection vary with
its two constituents: 1 − ω and σ; cartel harm. Here, we employ insights from the
(λM −λL ) existing cartel literature on deterrence and detec-
2. the sampling differentials λ
and tion. For this purpose, we interpret cartel over-
(λM −λH ) charge as synonymous with harm, which is an
λ
, and their constituents (the segments ωi
and σi ); and acceptable approximation given (Bolotova 2009,
3. the asymmetry of the population distribu- 338) the empirical finding of no significant cor-
tion, as represented by (pL − hL ) and (hH − pH ). relation between overcharge and cartel turnover.
We believe the existing literature supports two
To calibrate the parameters, we turn to exist- ordinal assumptions.
ing empirical and theoretical literatures. This
provides some broad guidance on the likely mag- ASSUMPTION 1. The probability of cartel
nitudes of ω and σ, and some ordinal constraints detection is nondecreasing with cartel harm:
on the sampling differentials. σL ≤ σM ≤ σH .
A. The Aggregate Parameters 12. The proportions of detected and deterred cartels are
In the following simulations, the probabilities (1 − ω)σ, and ω, respectively, and if the latter is five times
the former, ω = 5(1 − ω)σ. This is satisfied approximately
of deterrence and detection ω and σ are modeled at the midpoints of our assumed distribution, σ = 0.215 and
as random variables, using the existing empirical ω = 0.5.
8 ECONOMIC INQUIRY
.025
recent theoretical and experimental literatures on
leniency, such as Bigoni et al. (2012), Harrington
.02
and Chang (2009), Fonseca and Normann (2012),
.015
and Jensen and Sørgard (2016), suggest that the
Density
probability of leniency applications increases
.01
with harm. This is because in markets where
cartels cause large harm there is also a larger
.005
incentive to deviate, and less stable cartels are
more likely to apply for leniency. On ex-officio
0
detection, following the Block, Nold, and Sidak 0 10 20 30 40 50 60 70 80 90 100
(1981) model, the prevailing assumption is that Cartel overcharge
the probability of detection increases with the Illegal Legal
TABLE 1
Calibrations: Parameter Assumptions
General Evidence Based
Aggregate deterrence ω ∼ U(0.1, 0.9) ω ∼ U(0.1, 0.9)
Aggregate detection σ ∼ U(0.1, 0.33) σ ∼ U(0.1, 0.33)
Segment detection σL ≤ σM ≤ σH σL ≤ σM ≤ σH
Segment deterrence ωL ≥ ωM ≤ ωH ωL ≥ ωM ≤ ωH
Population proportions pL = pM = pH = 0.333 pL = 0.238, pM = 0.415, pH = 0.347
Population asymmetry Ln(h) ∼ N(μ, s2 ), s ∈{0, 1, 1.5} hL = 0.002, hM = 0.144, hH = 0.854
σ (1−ω )
Sampling differentials λL = Lσ(1−ω)L = 0.5266
σ (1−ω )
λM = Mσ(1−ω)M = 1.5462
σ (1−ω )
λH = Hσ(1−ω)H = 0.6707
Additional Structure. First, assume that the legal The intuition is that higher harm cartels con-
distribution represents a random sample drawn tribute disproportionately to asymmetry, that is
from the population distribution (what would be (hH − pH ) = 0.507 > (pL − hL ) = 0.236 (i.e., the
observed absent policy), then the three segments Lorenz curve is asymmetric) and although rela-
L, M, and H, can be identified as the three parts tively more of the high harm cases are observed
of the distribution delineated by its two points (λL < λH ), this is outweighed by the fact that
of intersection with the illegal distribution (see the unobserved high harm cases are dispropor-
Figure 2). These occur at overcharges of 6.4% tionately harmful. In other words, because the
and 40.0%. In turn, this identifies the propor- most harmful cases are extremely harmful, when
tions of the population and the population harm they are unobserved, merely grossing up an
accounted for by each segment: thus from the estimate of sample (detected) harms using only
legal distribution, in the lower tail pL = 23.8%, aggregate detection and deterrence rates (i.e.,
but hL = 0.2%, and pH = 34.7% and hH = 85.4% assuming all cartels are equally harmful) would
in the upper tail. lead to a 25% underestimate of total population
Second, comparing the masses in the three harm. This finding is of wider general insight
segments between the legal and illegal distribu- into how sample selection bias should caution us
tions provides direct estimates of the sampling when using results on detected cartels to draw
differentials. To see how, note that the low harm inferences about all potential cartels. ◾
segment of the illegal distribution accounts for
σ (1−ω ) RESULT 3. There is a lower bound on the
a proportion, pL Lσ(1−ω)L , of all observed illegal
cartels, while the proportion of the legal sam- aggregate deterrence rate, ω ≥ 0.257.
ple in the low harm segment is pL .21 It follows
Proof. The proof follows by deriving minimum
that the ratio of the illegal to legal proportions
σ (1−ω ) values for each of the three segments ωi and then
in the low segment is Lσ(1−ω)L , that is λL /λ. For aggregating. From Assumption 2 (deterrence is
the other segments λM and λH are derived identi- lowest in the middle segment), and Assumption 1
cally. These are reported in the “Evidence based” (detection is no lower in high harm than medium
column of Table 1, and as can be seen, there is rel- harm cases, σH ≥ σM ), we can back out lower
atively more mass in the middle segment of the values of ωM and ωH . A necessary condition for ω
illegal distribution (1.5462 times the aggregate) to be minimized is ωM = 0; and ωH is minimized
but relatively less in the two tails, especially the σ (1−ω )
when σH is minimized, and since Mσ(1−ω)M =
low harm segment (0.5266). σ (1−ω )
1.546 and Hσ(1−ω)H = 0.671, simple manip-
Three Direct Implications. These estimates do ulation yields ωH = 1–0.671/1.546 = 0.566.
σ σ
not allow us to identify the individual segment In turn, these magnitudes imply: σM = σH =
detection and deterrence parameters—these 1.546 (1 − ω) . To recover minimum ωL , note that
remain underidentified and simulation is still nec- σL (1−ωL ) (1−ω ) ( )
= 0.527, so (1−ω)L = 0.527∕ σL ∕σ .
essary. However, they do provide three important σ(1−ω)
new results even before we turn to simulation. The magnitude of (σL /σ) can be backed out from
the σ aggregation constraint using the above
RESULT 2. Total potential population harm is values for σM /σ and σH /σ. Substituting into the
25% greater than would be predicted by the sim- aggregation constraint for ω yields a pair of
ple multiplier estimate. simultaneous equations yielding a quadratic for
(1 − ωL ) in terms of (1 − ω). The solution has two
Proof. Substituting the above estimates of λi , pi , roots: (ω = 0.353, ωL = 0.660) and (ω = 0.257,
and hi into (7) yields: ωL = 0.256), where the latter is the minimum. ◾
(13) This leads immediately to:
H S ∕ (1 − ω) σ
H= RESULT 4. Deterred harm as a share of total
1 + (1.020) × (0.236) − (0.876) × (0.507)
harm can never be less than 48.4%
= 1.255H S ∕ (1 − ω) σ.
Proof. Evaluating deterred harm at these min-
21. The total number of detected illegal cartels is imum ωi above gives: H DR = ωL hL + ωM hM
Nσ(1 − ω), of which pL NσL (1 − ωL ) are in the low harm + ωH hH = 0.256 × 0.002 + 0 × 0.144 + 0.566 ×
segment. 0.854 = 0.484.
12 ECONOMIC INQUIRY
FIGURE 3
Depicting CA Success. (A) Correlated ω and σ and (B) Uncorrelated ω and σ
A B
1
1
.8
.8
% of potential harm
% of potential harm
.6
.6
.4
.4
.2
.2
0
0
0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1
CA success rate CA success rate
Deterred harm Undetected harm Deterred harm Undetected harm
Detected harm Detected harm
Note: CA success rate is measured by the sum of detection and deterrence and is increasing from 0 to 1 on the horizontal
axis. Correlated ω and σ indicate that a CA strong on detection is also strong on deterrence. Uncorrelated indicates not such
tendency.
many more cases actually occur, and this effect seen, all results are robust, both qualitatively
outweighs the lower rate of detection. and quantitatively.
While this interpretation is offered more as a
thought experiment than a literal description of
V. LIKELY ROBUSTNESS OF RESULTS
the differences between any real-world CAs, it
does drive home two important implications of Our headline result—that at least 50% (and
this paper. First, deterrence is by far the most probably much more) of all potential harm is
important potential impact of anticartel policy, deterred—is the first of its kind, and we hope
and second, estimates of the gains from detected that it will help serve as a useful yardstick for
cartels may be a very misleading indicator of the future work on this topic. Nevertheless, we read-
efficacy of cartel policy. ily acknowledge that some parts of our analysis
Finally, we revisit the assumption underpin- are inevitably speculative, and the purpose of this
ning all the results so far, that ω and σ are per- section is to scrutinize them more closely.
fectly correlated. In policy terms, there are at In essence, our results are driven, theoret-
least three reasons why such a positive correla- ically, by two insights derived from previous
tion should exist. First, firms are more likely to literatures: (1) the chances of detection increase
be deterred if confronted with a CA which has a with overcharge and (2) it is the lowest and high-
strong detection rate. Second, at any point in time, est overcharge cartels which are most likely to
there is likely to be fewer cartels surviving from be deterred. Empirically, we have employed the
previous periods if the CA has been more suc- distribution of overcharges observed historically
cessful in detection. Third, insofar as deterrence for legal distributions and the key assumption is
is determined by not only strong performance in that this approximates the underlying population
detection but also other activities such as com- distribution—what would occur absent cartel
pliance and advocacy, ω and σ are likely to be law and policy. We now return to these key
positively related if a strong CA tends to be better assumptions and also discuss the unavoidable
across the range of activities. uncertainty about the magnitudes of the two key
Nevertheless, we have rerun the experi- aggregate parameters, ω and σ.
ments assuming a zero correlation between
the two parameters (Figure 3B).22 As can be A. The Aggregate Probabilities, ω and σ
It is assumed throughout that σ lies within the
22. This captures the possibility that leniency discounts range of estimates reported in the previous liter-
now weaken the deterrence effects of detection. ature (0.1, 0.33). While this range is relatively
14 ECONOMIC INQUIRY
harm (proxied by overcharge), which would in two stages: how much potential harm can car-
be observed absent cartel policy and therefore tels cause to an economy and how successful are
without deterrence. Nevertheless, this source is CAs in rectifying that harm, either by detection
not without criticism in the wider literature (e.g., or deterrence? The first objective of this paper is
Boyer and Kotchoni 2015) largely concerning methodological—to propose a framework which
the rigor of some of the individual studies from has the potential to answer these questions. It
which Connor compiled the database. Moreover, interprets the cartels detected by a CA as a sam-
and directly relevant for our use of the legal ple drawn from an otherwise unknown popula-
cartel distribution, there are some confounding tion, which also includes deterred cases and cases
factors which complicate any clean compar- which the CA did not detect. This framework
ison between legal and illegal cartels—most identifies three types of information required to
notably in time periods and regional location. quantify the magnitudes of the unknown deterred
However, Bos et al. (2017)—from which our and undetected harms: (1) the aggregate probabil-
estimates derive—have controlled for those ities of deterrence and detection; (2) how these
of these factors which are observable using probabilities vary with cartel harm; and (3) the
propensity score methods. This also shows that heterogeneity in the population among potential
the results are robust to excluding those esti- cartels. This reminds us that the ubiquitous prob-
mates originating from sources, which, by their lem of selection bias is likely to be as important in
nature, were less exposed to academic scrutiny policy evaluation as in most other areas of empiri-
at their time of publication. For these reasons, cal Industrial Organization. There is no reason for
we believe that we have gone as far as possible assuming that those cartels, which are caught and
in this paper, given publicly accessible data at prohibited by CAs are necessarily representative
the time of writing, if one wishes to approx- of other cartels which successfully conceal their
imate the counterfactual—the distribution of existence, or yet other potential cartels, which
harm which would be observed absent cartel are deterred by the existence of the law and the
law/policy—with a distribution of observed CA. On this, the paper underlines a simple but
legal cartels. important wisdom: in many aspects of economic
Notwithstanding this, there may be alternative and social life, strongly positively skewed dis-
approaches to the counterfactual, which sidestep tributions prevail—a small proportion of causes
altogether the need for data on legal cartels. generate a very large proportion of effects. This
Indeed, in the present paper, we have employed is well understood in general and is plausibly
one alternative, in the general experiments of also likely with cartel-induced harm. If so, even
Section IV.A, in which a lognormal is used to small departures from a random sample can lead
depict the underlying distribution of potential car- to potentially large selection bias (we show, as a
tel harm, and exploring sensitivity of results to by-product of our main results, that ignoring car-
varying the degree of asymmetry of that distribu- tel heterogeneity could lead to underestimates of
tion. Using as illustrative the medium asymmetry about 25% for total harm, Result 2 in Section IV).
case reported earlier in Table 2, it can be seen that We calibrate the framework using various
this generated results which are qualitatively sim- results from the existing empirical and theoretical
ilar, if less absolutely pronounced, to our main cartel literatures. Theoretical literature provides
evidence-based results in Table 3. The multipliers the rationale for two assumptions: the probabil-
are broadly similar if smaller, and deterred harm ity of detection increases weakly monotonically
typically dominates undetected harm across the with overcharge, while the probability of deter-
distribution, but its share of total potential harm rence exhibits a U-shaped relationship to over-
could now be as low as one quarter at the 5% con- charge. From the empirical literature we assume
servative lower bound. that the underlying potential population distribu-
tion of overcharge can be approximated by the
VI. CONCLUDING REMARKS
real-world historical distribution of overcharge
for observed legal cartels. If so, a comparison of
This paper raises the ambitious question of the legal overcharge distribution with the distri-
how can we quantify the magnitude of the deter- bution for illegal cartels provides further struc-
rence effect of anticartel enforcement? While it ture on how detection and deterrence vary with
is widely acknowledged that deterrence is proba- overcharge. Nevertheless, the largest gap in our
bly the main effect, this has never been quantified knowledge relates to the magnitude of the aggre-
to date. Our approach is to frame the question gate deterrence probability; here there is little to
16 ECONOMIC INQUIRY
go on from the previous literature, and for that Block, M. K., and J. S. Feinstein. “The Spillover Effect
reason in our simulations we allow this parameter of Antitrust Enforcement.” Review of Economics and
Statistics, 68(1), 1986, 122–31.
to fall within a very wide range of potential val- Bolotova, Y. V. “Cartel Overcharges: An Empirical Analysis.”
ues, and then focus mainly on the lower bounds of Journal of Economic Behavior & Organization, 70(1),
our estimates. In that sense, our headline results 2009, 321–41.
Bos, I., S. W. Davies, J. E. Harrington Jr, and P. L. Ormosi.
are deliberately cautious and conservative. “Does Enforcement Deter Cartels? A Tale of Two
Among the results of our experiments, three Tails.” 2017. Accessed March 15, 2018. https://ssrn
merit emphasis. First, deterrence is overwhelm- .com/abstract=2471425.
Boyer, M., and R. Kotchoni. “How Much Do Cartel[S]
ingly greater than detected cartel harm—our Overcharge?” Review of Industrial Organization, 47(2),
estimates suggest that, at the very least, 50% 2015, 119–53.
of potential harm is deterred. Our typical esti- Brenner, S. “An Empirical Study of the European Corporate
mates are much higher, at two-thirds. Second, Leniency Program.” International Journal of Industrial
Organization, 27(6), 2009, 639–45.
undetected harm is probably much smaller than Bryant, P. G., and E. W. Eckard. “Price Fixing: The Prob-
deterred harm—our upper bound estimate is 40% ability of Getting Caught.” Review of Economics and
of potential harm, but the typical estimate is Statistics, 73(3), 1991, 531–36.
Combe, E., C. Monnier, and R. Legal. “Cartels: The Proba-
less than 25%. Nevertheless, undetected exceeds bility of Getting Caught in the European Union.” BEER
detected harm in most of our experiments (see Paper No. 12, 2008.
Table 3 for example). Third, in aggregate, poten- Competition and Markets Authority. “CMA Impact Assess-
tial harm from the population of cartels which ment 2015–16.” Technical report. London: CMA.
Connor, J. M. “Price-Fixing Overcharges: Revised 3rd Edi-
would occur in the absence of cartel law is a tion.” 2014. Accessed March 15, 2018. http://ssrn.com/
large multiple (at least 13) of the harm actually abstract=2400780.
detected by the CA: detected harm is only the tip Damgaard, C., and J. Weiner. “Describing Inequality in Plant
Size or Fecundity.” Ecology, 81(4), 2000, 1139–42.
of an iceberg. Davies, S., and O. De. “Ringleaders in Larger Number Asym-
Some of the policy implications are self- metric Cartels.” The Economic Journal, 123(572), 2013,
evident. Deterrence is a pivotal arm of cartel pol- F524–44.
Deloitte. “The Deterrent Effect of Competition Enforcement
icy, and it will be largely this which distinguishes by the OFT.” Office of Fair Trading No. OFT962, 2007.
the strong from the weak authority. On the other Elzinga, K. G., and W. Breit. “Antitrust Penalties and Atti-
hand, harm due to undetected cartels is also likely tudes toward Risk: An Economic Analysis.” Harvard
to be considerable, and it can be very misleading Law Review, 86, 1973, 693.
European Commission DG Competition. “2014 Annual Activ-
to measure the success of a CA simply by calcu- ity Report.” Technical Report, DG Competition, 2015.
lating the amount of harm it removes by virtue of Feinberg, R. M. “Antitrust Enforcement and Subsequent Price
cartel busts. Our results imply that a “weak” CA Behavior.” Review of Economics and Statistics, 62(4),
1980, 609–12.
(relatively ineffectual in deterring and detecting) Fonseca, M. A., and H.-T. Normann. “Explicit vs. Tacit
may actually detect more harm than a “strong” Collusion—The Impact of Communication in
CA, simply because its inability to deter leaves Oligopoly Experiments.” European Economic Review,
far more cartels out there to be detected. 56(8), 2012, 1759–72.
Gastwirth, J. L. “The Estimation of the Lorenz Curve and
It is our hope that the current paper provokes Gini Index.” Review of Economics and Statistics, 54(3),
future work in which, for example, alternative 1972, 306–16.
counterfactuals are employed. Within our own Geroski, P. “Is Competition Policy Worth It?” Speech on
the opening of the Centre for Competition Policy
research program, we will work analogously on at the University of East Anglia, 2004. http://www
merger control (for which nondetection is likely .competitioncommission.org.uk.
to be less important), and more generally the Ginsburg, D. H., and J. D. Wright. “Antitrust Sanctions.”
Competition Policy International, 6(2), 2010, 3–39.
scope for using this sort of framework to set the Harberger, A. C. “Monopoly and Resource Allocation.”
relative priorities within CAs. American Economic Review, 44(2), 1954, 77–87.
Harrington, Jr., J. E., and M.-H. Chang. “Modeling the Birth
and Death of Cartels with an Application to Evaluat-
REFERENCES ing Competition Policy.” Journal of the European Eco-
nomic Association, 7(6), 2009, 1400–35.
Aitchison, J., and J. A. Brown. The Lognormal Distribution. . “When can We Expect a Corporate Leniency Pro-
Cambridge University Press, Cambridge, 1957. gram to Result in Fewer Cartels?” Journal of Law and
Baker, J. B. “The Case for Antitrust Enforcement.” Journal Economics, 58(2), 2015, 417–49.
of Economic Perspectives, 17(4), 2003, 27–50. Harrington, Jr., J. E., and Y. Wei. “What can the Duration
Bigoni, M., S.-O. Fridolfsson, C. L. Coq, and G. Spagnolo. of Discovered Cartels Tell Us about the Duration of
“Fines, Leniency and Rewards in Antitrust.” RAND all Cartels?” The Economic Journal, 127(604), 2017,
Journal of Economics, 43(2), 2012, 368–90. 1977–2005.
Block, M. K., F. C. Nold, and J. G. Sidak. “The Deterrent Jensen, S., and L. Sørgard. "Fine Schedule with Heteroge-
Effect of Antitrust Enforcement.” Journal of Political neous Cartels: Are the Wrong Cartels Deterred?" Work-
Economy, 89(3), 1981, 429–45. ing Paper, 2014.
DAVIES, MARIUZZO & ORMOSI: ESTIMATING DETERRENCE 17
Katsoulacos, Y., and D. Ulph. “Antitrust Penalties and Lorenz, M. O. “Methods of Measuring the Concentration
the Implications of Empirical Evidence on Cartel of Wealth.” Publications of the American Statistical
Overcharges.” The Economic Journal, 123(572), 2013, Association, 9(70), 1905, 209–19.
558–81. Miller, N. H. “Strategic Leniency and Cartel Enforcement.”
Katsoulacos, Y., E. Motchenkova, and D. Ulph. “Measur- American Economic Review, 99(3), 2009, 750–68.
ing the Effectiveness of Anti-Cartel Interventions: A Nelson, P., and S. Sun. “Consumer Savings from Merger
Conceptual Framework.” TILEC Discussion Paper No. Enforcement: A Review of the Antitrust Agencies’ Esti-
2016-001, 2016. mates.” Antitrust Law Journal, 69(3), 2002, 921–60.
Kobayashi, B. H. “Antitrust, Agency, and Amnesty: An Eco- Ormosi, P. L. “Evaluating the Impact of Competition Law
nomic Analysis of the Criminal Enforcement of the Enforcement.” OECD Working Paper No. DAF/COMP/
Antitrust Laws.” George Washington Law Review, 715, WP2(2012)5, 2012.
2001, 735–39. . “A Tip of the Iceberg? The Probability of Catching
Lande, R. H., and J. M. Connor. “Cartels as Rational Business Cartels.” Journal of Applied Econometrics, 29(4), 2014,
Strategy: Crime Pays.” Cardozo Law Review, 34(2), 549–66.
2012, 427–90. Werden, G. J. “Assessing the Effect of Antitrust Enforcement
Landes, W. M. “Optimal Sanctions for Antitrust Violations.” in the United States.” De Economist, 156(4), 2008,
University of Chicago Law Review, 50(2), 1983, 433–51.
652. Werden, G. J., S. D. Hammond, and B. A. Barnett. “Deter-
Levenstein, M. C., and V. Y. Suslow. “What Determines rence and Detection of Cartels: Using All the Tools and
Cartel Success?” Journal of Economic Literature, 44(1), Sanctions.” Antitrust Bulletin, 56(2), 2012.
2006, 43–95. Zhou, J. “Evaluating Leniency with Missing Information
London Economics. “The Impact of Competition Interven- on Undetected Cartels: Exploring Time-Varying Policy
tions on Compliance and Deterrence.” Office of Fair Impacts on Cartel Duration.” TILEC Discussion Paper,
Trading No. OFT1391, 2011. 2011-042, 2011.