Draft March 6, 2013
Determinants of U.S. Antitrust Fines of Corporate Participants of Global Cartels*
John M. Connor Professor Emeritus, Purdue University, West Lafayette, Indiana
[email protected]
Douglas J. Miller Department of Economics, University of Missouri, Columbia, MO 65211-‐‑6040, USA
*Earlier versions were presented at the 7th International Industrial Organization Conference, Boston, April 3-‐‑5, 2009 and an address at the 11th annual meeting of the American Antitrust Institute, Washington DC, June, 2010. The authors thank three anonymous referees of this Review, Roger King, and other discussants for their comments.
1 Electronic copy available at: http://ssrn.com/abstract=2229300
Abstract
For criminal violations of the Sherman Act, although guided by federal sentencing guidelines, U.S. Department of Justice has great latitude in recommending corporate cartel fines to the federal courts, and its recommendations are nearly always determinative. In this paper, we analyze the determinants of variation in size of criminal fines imposed by the Antitrust Division of the DOJ on 118 corporate participants of hard-‐‑core global cartels. Our behavioral model provides the first direct test of the optimal deterrence theory of antitrust crimes. Regressions are fitted to a sample of the corporations that participated in international cartels and that were fined between 1996 and March 2010. The predictive power of the optimal-‐‑deterrence model is quite good. We find that U.S. corporate cartel fines are strongly directly related to economic injuries from collusion. However, U.S. fines do not conform to the ¢Ȃȱpredictions about the probability of detection and conviction of clandestine cartels. We also find that fines complement other antitrust penalties: the ȱȱȱȱȱȱȂȱȱȱȱȱ and private damages paid. Key words: antitrust, Sherman Act, DOJ, Antitrust Division, cartel, collusion, price-‐‑ fixing, optimal deterrence, fines, penalties. JEL codes: L41, L44, L65, L11, L13, N60, K21, K14
2 Electronic copy available at: http://ssrn.com/abstract=2229300
INTRODUCTION ȱȱȱȱȱǯǯȱȱȱ ȱǻȱȃȱȄȱȱ ȃ ȄǼȱȱȱoldest antitrust authority in the world. It has been prosecuting price fixing with increasing severity for more than a century. Together with its sister competition authority in the European Union, the Competition Directorate of the European Commission (EC), the two comprise the most powerful and influential government agencies for detecting and punishing cartels. However, until the mid 1990s obtaining evidence on well hidden international cartel activity was very difficult for prosecutors. However, the number of cartels detected and fines imposed rose considerably after workable amnesty programs were introduced. During 1990-‐‑2009, these two authorities imposed $25.3 billion in fines on 1200 companies for overt international price-‐‑fixing violations. In its published cartel-‐‑infringement decisions, the EC provides detailed descriptions of ȱ ȱ ¢Ȃȱ ¢ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ published Cartel Fining Guidelines. Similarly, the DOJ follows the U.S. Sentencing Guidelines (USSGs) as a starting point for sentencing companies guilty of hard-‐‑core price fixing. However, the USSGs are not as precise as those of the EC, partly because the Guidelines suggest a range of appropriate fines rather than a point fine. Moreover, virtually every corporate defendant receives ȱȱȱȃȄȱȱȱ in fines that are below ȱ ȱ ȱ ȱ Ȃȱ ǯ Guilty plea negotiations between DOJ prosecutors and defendants over fine discounts are confidential, and ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ Ȃȱ cooperation. Consequently, for criminal violations of the U.S. Sherman Act, the DOJ has considerable discretion in recommending corporate cartel fines to a federal judge. Whether the cartel-‐‑sentencing decisions of the DOJ are capable of prediction is an open question. The Antitrust ȱȱȱȱȱȱȱȱȂȱȱ-‐‑fixing prohibitions. It states that its criminal penalty procedures adhere to the law-‐‑and-‐‑ 3
economics principles of optimal deterrence Posner [1]. Indeed, when the U.S. Sentencing Guidelines for federal crimes by organizations were being debated in the late 1970s, the DOJ was instrumental in making optimal deterrence the explicit foundation of the Guidelines, Cohen [2], Connor and Lande [3]. Fines for criminal antitrust violations were to be formulated on the basis of the illegal gains or economic damages generated by the offending entity; if the damages could not be estimated in practice, then formulas were devised to compute proxies for economic damages.1 Optimal deterrence is also served by considering the probability of detection of particular cartels; fines should be higher if this probability is low. Thus, both the starting point for sentencing cartelists and the internal negotiations for plea agreements are governed by optimal deterrence principles. Implementing optimally deterring fines is subject to constraints on DOJ decision-‐‑ making. The new presidential administration ushered in by the 1992 election brought a new commitment to prosecuting large international cartels. In addition, resource constraints may have prompted the DOJ to focus its enforcement efforts on industries especially susceptible to cartelization. The DOJ found that evidence was insufficient to prove the beginning dates of cartels with long durations; in such cases proxy measures of harm will understate the appropriate fine. The economic theory of crime has received little empirical verification. In particular, a review of the empirical law-‐‑and-‐‑economics literature finds very few studies that quantitatively estimate the variation in corporate criminal fines and no such studies for cartel fines. This paper provides a novel test of the predictive power of optimal deterrence principles underlying the enforcement activities directed at an important corporate crime. 1 A discussion of the role of optimal deterrence in the U.S. Sentencing Guidelines for cartel violations is given in Connor and Lande [3]. Briefly, the authors of the USSGs deemed that an overcharge of 10% of affected sales was a reasonable rebuttable presumption for most cartels, and then doubled that figure to arrive at a base fine that would take into account the overcharge, the dead-‐weight loss, and the need for deterrence.
4
Objective This paper analyzes the determinants of variation in cartel sanctions imposed on more than 100 corporate participants of global cartels by the Antitrust Division of the Department of Justice from 1995 to December 2008. The models to be tested primarily draw upon testable propositions suggested by optimal deterrence theory. However, we also augment the specification of the models and hypotheses by considering the stated policies and historical sentencing practices of the DOJ and the federal judiciary. These latter facȱ ¢ȱ ȱ ȱ ȱ ȱ ȱ Ȃȱ ¢ȱ ȱ ȱ ¢ȱ optimal fines. Importance of the Topic Greater understanding of the determinants of cartel fines is important for policy and disciplinary reasons. First, tȱ Ȃȱ policies and procedures are often held up as an exemplary, highly successful paradigm for the scores of antitrust authorities that have developed active anticartel programs in the past two decades. Now that it has accumulated a substantial record of enforcement, a retrospective analysis is feasible. Second, of interest to the law-‐‑and-‐‑economics discipline is the extent to which DOJ sentencing practices conform to the tenets of the optimal deterrence theory of crime, now the dominant basis for antitrust law enforcement. The one empirical study assessing the adherence of corporate sentencing included few antitrust convictions in its data set. Third, DOJ officials often emphasize the idiosyncratic features of sentencing, going so far as to deny the predictability of negotiated fines in advance of plea bargaining. If so, this raises doubts about the transparency and proportionality of cartel fines. Organization The rest of this paper is organized as follows. First, we examine optimal deterrence theory for testable hypotheses. Second, we describe the U.S. statutes and methods ¢ȱ ȱ ȱ ȱ ¢ȱ Ȃȱ ǯǯȱ ȱ ¢ǯȱ ǰȱ ȱ ¢ȱ ȱ ȱ empirical literature on corporate cartel fines. Fourth, we discuss the data sources and
5
sample. Fifth, we lay out our behavioral model, variables and hypotheses. Finally, we explain and discuss the results of the regression analysis. THEORY AND PRACTICE OF SETTING CORPORATE PENALTIES ȱ ȱ ȱ ȱ ȱ ȱ ȱ Ȃȱ -‐‑fine recommendations is the law-‐‑ and-‐‑economics theory optimal deterrence. Although the theory provides general guidelines, U.S. laws and prosecutorial practices also influence the imposition of fines. For example, the USSGs incorporate some culpability factors that are difficult for outsiders to observe or quantify. Moreover, the outcome of nearly all contemporary cartel cases takes place in the context of plea bargaining, which is an unobservable process subject to a great deal of case-‐‑by-‐‑case variation in outcome. Optimal Deterrence Theory Although Bentham and other classical economists wrote about the economic rationality of crime, modern interest dates from a seminal paper by Becker, Ehrlich [4], Becker [5]. This approach assumes that offenders respond rationally to incentives. They are utility maximizers who optimally allocate their time among competing legal and illegal activities. The decision to engage in crime is related to the expected marginal benefits of alternative activities, the perceived probability of apprehension and conviction, and the expected marginal penalties imposed for various crimes. The dual of utility maximization by a decision maker evaluating a crime is minimization of social costs of detection, conviction, and monitoring or incarceration. These costs can be private (antitrust compliance training, legal defense costs, etc.) or public (policing markets, supporting prosecutors and the judicial system, operating penal systems). For a survey of the economics of crime and formal proofs of these propositions, see Garoupa [6] or Polinsky and Shavell [7, 8].
6
In the context of cartels, optimal deterrence theory is couched in terms of the expectations of the founders and managers of cartels. Individual expectations about cartel penalties are formed on the basis of Information from historical experience -‐‑-‐‑ that of the firm itself, its legal advisors, and of other firms that were defendants in comparable price-‐‑fixing litigation. The expected size of expected monetary penalties affects both the probability of detection and the rate of cartel formation. If expected fines are low, the incentive for applying for leniency is low, cartel defections slow, and the likelihood of detection is lowered. Therefore, increasing penalties will make cartels more fragile and increase detection rates. Assuming that the benefits of overt collusion derive from exogenous market characteristics, up to some point higher penalties efficiently discourage the formation of an optimal number of cartels. Specifically, the decision to form a new cartel or to enter an existing cartel is positively related to the gain (anticipated additional monopoly profits) or harm (damages to victims) and is negatively related to the expected probability of detection (p) and expected severity of penalties (E(F)). The perceived probability of detection by a criminal is not directly observable, but prosecutors may have sufficient information to develop some notion of the extent to which a cartel attempted to remain clandestine and thereby elevate the imposed penalties.2 Penalties for corporations include fines, private monetary settlements, legal defense costs, debarment, and loss of reputation. 3 Although the Division seldom bases its fine recommendations on gain or harm, guidelines have developed proxies for them. Because expected fines are formed by actual fines, models too may be framed by observed fines. What is available is ex post rather than the ideal 2
Cooperation after amnesty or during plea bargaining requires defendants to divulge evidence of destruction of meeting agendas, minutes, travel records, or other cover-‐up conduct. Since at least the mid 1990s, virtually all cartel defendants have had to offer full cooperation as a condition in their plea agreements. 3 ĞĐŬĞƌ͛Ɛϱoriginal model assumed that a crime was committed by a single utility maximizer, which could well describe an owner-‐managed small business. However, most modern cartels are populated by large businesses with a cadre of professional top managers who do not have financial control of the company. If principal-‐agent problems exist, optimal corporate sanctions may still exist under some situations, Cohen [2]. For example, in many ĐŽƵŶƚƌŝĞƐ͕ĞŵƉůŽLJĞƌƐŵĂLJƉĂLJĨŽƌĂŶĞŵƉůŽLJĞĞ͛ƐƉĞƌƐŽŶĂůĨŝŶĞ͘hŶĚĞƌŽƚŚĞƌĐŽŶĚŝƚŝŽŶs, a combination of corporate monetary penalties and executive incarceration may be optimal.
7
ex ante penalties. Actual penalties are a good surrogate for expected penalties if criminals are risk-‐‑neutral. In its simplest form, an optimal fine is F*=HARM/p. One additional principle of optimal deterrence is that all types of monetary (or monetary-‐‑equivalent) sanctions are fungible. Thus, fines will be rationally lower when prosecutors or judges have knowledge or expectations that a defendant will pay extra-‐‑ jurisdictional fine or civil penalties (OTHPEN).4 The Division has instituted a policy of substituting more and longer prison sentences for ever larger corporate fines. In sum, the first-‐‑order condition for an optimal criminal sanction is: F*=HARM/p Ȯ OTHPEN. [1] ȱ Ȃȱȱ¢ȱ The DOJ has been widely extolled for its energetic campaign against cartels that took off in the mid 1990s Klawiter [9]; Connor 2007c[10].5 It has adopted more aggressive investigatory techniques, increased the severity of corporate and individual sanctions, and instituted cooperation with many antitrust authorities around the world. Perhaps ȱ ǰȱ ȱ Ȃȱ ȱ ȱ ȱ ȱ ret cartels improved because of the revised 1993 Corporate Leniency Program Spratling and Arp [11], Hammond [12]. The idea that a qualified leniency applicant should receive a 100% reduction in its potential cartel fine is well grounded in economics. Game-‐‑¢ȱ ȱ ȱ ȱ Ȃȱ 4
The majority of private civil suits are resolved well after criminal fines are imposed, but these suits tend to be filed within a month or two from the time a formal investigation begins or the first guilty pleas are made public. In addition, counsel for private plaintiffs often inform prosecutors of evidence in their possession. Thus, prosecutors typically have concurrent knowledge of actual or planned private suits and claimed damages. If global cartels were prosecuted by a global antitrust authority, then OTHPEN would always be zero. 5 In international opinion surveys of antitrust enforcement, the DOJ almost always ranks at or near the top in the admiration of antitrust lawyers (Hoj 2007).
8
Dilemma show that under a wide array of conditions full leniency reduces cartel stability (i.e., it induces members of functioning cartels to defect by confessing to antitrust authorities) (Aubert et al. [13], Spagnolo [14]). ȱ Ȃȱ ȱȱ¢ȱȱȱ ¢ȱȱȱȱ only the first member of a cartel that applies and that meets a few criteria. 6 Because applicants know those conditions in advance, acceptance by the DOJ into the Program is not discretionary Ȯ ȱ ȱ ȃȄȱ ȱ ȱ ȱ ȱ ǯȱ ȱ ¢ȱ Program has resulted in grants of immunity (i.e., full leniency) from federal government fines to scores of cartelists and their officers. ȱ Ȃȱȱ¢ȱȱȱ to have worked. A sophisticated game-‐‑theoretic analysis finds that the U.S. program increased detection of cartels by about 60% after 1993 (Miller 2009[15]. For the remaining members of a cartel, the DOJ typically resolves criminal matters through confidential guilty-‐‑plea negotiations, which incorporate partial leniency7 discounts of less than 100% OECD [16], ICN [17]. The signed guilty plea agreement, when approved by a federal court, is tantamount to a criminal prosecution. Only in rare instances do corporate plea negotiations break down, followed by an indictment and trial.8 We discuss partial leniency in more detail below, but it is apparent that the effectiveness of full leniency programs depends on substantial expected fines and balanced partial leniency policies. Immunity from prosecution will be attractive to 6 dŚĞŵĂŝŶĐƌŝƚĞƌŝĂĂƌĞƚŚĂƚƚŚĞĂƉƉůŝĐĂŶƚŵƵƐƚŶŽƚďĞƚŚĞŝŶŝƚŝĂƚŽƌŽƌ͞ƌŝŶŐůĞĂĚĞƌ͟ŽĨƚŚĞĐĂƌƚĞůĂŶĚƚŚĂƚƚŚĞ application must either be made before the DOJ has begun an investigation of the cartel (Type A Leniency) or before the DOJ has sufficient information to sustain a conviction (Type B Leniency). See http://www.usdoj.gov/atr/public/guidelines/0091.htm. 7
K:ŽĨĨŝĐŝĂůƐŽĨƚĞŶƐƉĞĂŬĂďŽƵƚ͞ĚŽǁŶǁĂƌĚĚĞƉĂƌƚƵƌĞƐ͟Žƌ͞ƌĞǁĂƌĚƐ͟ĨŽƌƐĞĐŽŶĚ-‐in cooperating firms. We prefer ƚŚĞƚĞƌŵƉĂƌƚŝĂůůĞŶŝĞŶĐLJďĞĐĂƵƐĞ͞ƌĞǁĂƌĚƐ͟ĐĂŶŝŶĐůƵĚĞŵĂŶLJďĞŶĞĨŝƚƐŽƚŚĞƌƚŚĂŶĐŽŽƉĞƌĂƚŝŽŶĚŝƐĐŽƵŶƚƐ͕ƐƵĐŚĂƐ shaving time from the true conspiracy period, reducing the scope of products known to have been cartelized, keeping the number of counts to a smaller number than the maximum possible, or reducing the number of ĞŵƉůŽLJĞĞƐƚŽďĞ͞ĐĂƌǀĞĚŽƵƚ͟ĨŽƌƉƌŽƐĞĐƵƚŝŽŶ͕,ĂŵŵŽŶĚϮϬ͘ 8 The Division does prosecute a few cartel managers at trial each year, but excluding four or five very small family-‐ operated firms, only one corporation has been convicted at trial for price fixing since 1994 ʹ DƌƐ͘ĂŝƌĚ͛ƐĂŬĞƌLJŝŶ 1996 (Connor 2007a). In a 2001 trial that convicted Mitsubishi Corp., the issue was not price fixing per se but rather whether it had liability for a joint venture that had admittedly fixed the prices of graphite electrodes [http://www.justice.gov/atr/cases/indx216.htm].
9
cartelists ¢ȱȱȱȱȱȱȱȱ ȱȱȃȱȱȱȱȄȱȱȱ painfully higher penalties. The present study examines cartel fines for plea agreements rewarding partial leniency. Our sample excludes companies that were granted immunity under the Leniency Program because the decision criteria are quite different from those used to impose fines and because the identity of amnestied firms is not known with certainty. U.S. Sentencing Laws: Fines in Theory There are two statutes governing the setting of criminal antitrust fines by U.S. courts Connor and Lande [3]. First, in 1987 the U.S. Sentencing Guidelines for Organizations (USSGs) first became law. They specify the calculation of a range of fines within which the courts, upon the recommendation of the DOJ, were required to impose a specific corporate fine. In January ŘŖŖśȱȱȱȂȱȱȱBooker rendered the use of the USSGs advisory rather than mandatory, but it is evident that federal prosecutors and judges continue to be guided by them. Second, beginning in the mid-‐‑1990s the DOJ realized that when hard-‐‑core price fixing became a felony crime in 1974, courts were ȱ ȱ ȱ ȱ ¢ȱ ȱ ȃȱ ȱ ǰȄȱ ¢ǰȱ ŗŞȱ ȗřśŝŗǯȱ ȱ courts are instructed to apply whichever statute results in the largest fine. In most cartel guilty plea agreements, both the USSGs and the alternative fine provision are cited as the legal bases of the negotiated fine Connor 2008a[18]. One difference between the two fining methods is that fines imposed under the authority of the USSGs are subject to an absolute statutory limit, whereas there is no such limit under the alternative fining method. The statutory cap from July 1990 to July 2004 was a corporate fine of $10 million. For illegal conduct occurring after July 2004, the Antitrust Criminal Penalty Enhancement and Reform Act raised the cap to $100 million. Courts do not challenge corporate fine recommendations under the statuary
10
cap. But if the DOJ decides to recommend a fine above the statutory limit, it must appeal to the alternative fine statute.9 The mechanics of applying the USSGs require three steps and can seem rather complicated. First, on the assumption that the typical cartel achieves a 10% collusive mark-‐‑up,10 that percentage is doubled and multiplied by the ¢Ȃȱ affected commerce; this is termed the base fine. Second, the DOJ computes a total culpability score, which is the sum of a base score plus aggravating factors11 and minus mitigating factors.12 Third, the base fine is multiplied by two culpability multipliers to yield a fine range. The total score is converted into a culpability multiplier range that can start from as low as 0.75 to as high as 4.0. The top multiplier is always double the bottom multiplier. Thus, U.S. cartel defendants face fines that range within 15% to 80% of their affected sales. The second method under the alternative felony statute simply doubles the economic harm inflicted on direct purchasers by each defendant. There are no culpability adjustments; rather, the overcharge alone summarizes the degree of culpability. Consequently, the recommended fine is a single number, not a range. Proving ȱȱȃ¢ȱȱȱǰȄȱ ȱȱȱȱȱ¢ȱ 2005, would be challenging to prosecutors in a trial setting, so what the DOJ has done since then has been to negotiate by mutual agreement with some defendants an 9
The first breach in the $10-‐million cap occurred in August 1995 when Norwegian manufacturer Dyno-‐Nobel was fined, after agreeing to plead guilty, slightly more than $10 million for its role in the Explosives cartel (Connor 2007a:7). Since then dozens of fines above $10 million have been imposed on corporate cartel members. Moreover, since 1999 several fines above $100 million have been approved. 10 That this assumption may be too low for the typical cartel, see Connor (2007b). 11 dŚĞŵŽƐƚĐŽŵŵŽŶŽŶĞƐŝŶĐĂƌƚĞůĐĂƐĞƐĂƌĞ;ϭͿ͞ŝŶǀŽůǀĞŵĞŶƚŝŶŽƌƚŽůĞƌĂŶĐĞŽĨ͟ƚŚĞĐƌŝŵĞďLJƚŽƉŵĂŶĂŐĞƌƐ͕ǁŝƚŚ culpability rising with the size of the company and (2) recidivism within the past ten years. These cause the base fines to rise to a 40% to 80% range. 12 They are ;ϭͿĂŶĞĨĨĞĐƚŝǀĞŝŶƚĞƌŶĂůĐŽŵƉůŝĂŶĐĞŽƌĞƚŚŝĐƐƉƌŽŐƌĂŵĂŶĚ;ϮͿ͞ƐĞůĨƌĞƉŽƌƚŝŶŐ͕ĐŽŽƉĞƌĂƚŝŽŶ͕ĂŶĚ ĂĐĐĞƉƚĂŶĐĞŽĨƌĞƐƉŽŶƐŝďŝůŝƚLJ͟ĨŽƌƚŚĞĐƌŝŵĞ͘DŽƐƚĐĂƌƚĞůists are awarded small mitigating points, so a typical Guidelines range might be 30% to 60% of affected sales.
11
overcharge figure that will serve as the basis of the fine calculation. The defendant agrees not to contest this negotiated overcharge figure. Plea Bargaining and Cooperation Discounts: Fines in Practice
There is a trade-‐‑off between the conservation of constrained prosecutorial resources and the size of discounts offered on corporate cartel fines. On the one hand, more rapid acceptance of guilty pleas can be induced by offering relatively large discounts from recommended cartel fines OECD [16]. Large discounts will permit an antitrust authority to pursue more cases that involve difficult proof of guilt. Deterrence is improved. On the other hand, a deep discounting policy will lead to lower expected fines, fewer amnesty applications (i.e., fewer cartel detections), and a greater number of cartel formations. Deterrence is hobbled. The DOJ has a long-‐‑standing practice of negotiating ȃdownward departuresȄ from the mandatory or suggested Guidelines ranges in order to persuade alleged violators to plead guilty. Plea bargaining in criminal cases with prosecutors is a long-‐‑established practice in most common-‐‑law nations, and in the United States the vast majority of criminal cases are resolved by means of a deal in which a guilty plea is obtained in return for a promise of a reduced sentence Fisher [19].13 Nearly all of the hundreds of cartel convictions in the United States have been secured through guilty pleas Hammond [20]. 13
Kobayashi (1992) develops a formal game-‐theoretic model that has assumptions that describe cartel plea bargaining. That is, it incorporates simultaneous plea-‐bargaining between a prosecutor (who is maximizing total penalties) and several defendants that have been detected engaging in a single crime and allows one defendant to offer inculpatory information about other defendants to the prosecutor. The penalty facing any defendant is exogenous. This model predicts that the size of penalties from plea bargaining is positively related to the ĚĞĨĞŶĚĂŶƚ͛Ɛ ;ϭͿ ex ante chance of conviction and (2) value of information for convicting the remaining co-‐ conspirators.
12
Little has been published about the frequency of granting downward departures, the size of the discounts, how the administrative standards14 are employed in practice, or the effects of fine discounting on cartel deterrence.15 Discounts for cooperation are normally granted after one ȱ ȱ ȱ ȱ ȱ ȱ ¢ȱ ȱ ȱ Ȃȱ Corporate Leniency Program.16 Although plea negotiations are largely a black box, published guidelines do exist. Arriving at a mutually satisfactory discount is governed by procedures contained in the DO Ȃȱ Grand Jury Manual DOJ 1991[21]. When a firm requests to begin negotiations for a criminal guilty plea, the starting point for the DOJ is customarily the minimum fine in the Guidelines range. The upper point in the range is double the lower end. That is, without special circumstances, a typical defendant is granted a downward departure of 50% from the maximum liability under the Guidelines even before negotiations begin.17 Sȱȱȱ ȱȱȱȱ Ȃȱ range gives defendants the benefit of the doubt. Under the USSGs a court may, upon the recommendation of prosecutors, depart below ȱ Ȃȱȱȱȱ¢ȱȱȃdzȱȱȱȱȱ ȱ ȱ ȱ ȱ £ȱ ȱ ȱ ȱ ȱ dzȄȱ ǻUSSC 2005:§8C4.1(a)[22]). The major form of cooperation is divulging of secret information ȱ ȱ Ȃȱ ȱ ǯȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ ¢ȱ ȱ
14
The guidelines for entering into plea-‐bargaining negotiations are found in DOJ [21]. These guidelines exist to ͙͞ĞŶƐƵƌĞƚŚĂƚƉůĞĂĂŐƌĞĞŵĞŶƚƐĞŶƚĞƌĞĚŝŶƚŽďLJĨĞĚĞƌĂůƉƌŽƐĞĐƵƚŽƌƐĚŽŶŽƚďĂƌŐĂŝŶ ĂǁĂLJũƵƐƚŝĐĞ͕͟,ĂŵŵŽŶĚϮϬ. 15 Cooter and RubinfĞůĚ;ϭϵϴϵ͗ϭϬϴϮͿŶŽƚĞƚŚĂƚƚŚĞ͙͞ĚĞĐŝƐŝŽŶƚŽĂƐƐĞƌƚĂůĞŐĂůĐůĂŝŵŝƐĚŝĨĨŝĐƵůƚƚŽŝŶǀĞƐƚŝŐĂƚĞ ĞŵƉŝƌŝĐĂůůLJ͙͟ďĞĐĂƵƐĞŽĨƚŚĞĂďƐĞŶĐĞŽĨƉƵďůŝĐƌĞĐŽƌĚƐ͘/ŶĚĞĞĚƚŚĞir survey cites only one empirical study. 16 When cartel members formally apply for leniency, counsel representing the firm brings a proffer letter to the DOJ outlining what it has to offer by way of information on the illegal activity. When the letter is submitted, the ĂƉƉůŝĐĂŶƚƌĞĐĞŝǀĞƐĂ͞ŵĂƌŬĞƌ͟ƚŚĂƚĞƐƐĞŶƚŝĂůůLJŝŶĨŽƌŵƐƚŚĞĂƉƉůŝĐĂŶƚŽĨŝƚƐƉůĂce in the queue. The first applicant that fully qualifies receives amnesty (or immunity). The next applicant is called second-‐in, the next third-‐in, and so forth. These latter applicants are eligible for partial leniency. 17 The two exceptions are when a defendant is not qualified for full amnesty because it was a ringleader or it failed ƚŽƌĞǀĞĂůŝƚƐƉĂƌƚŝĐŝƉĂƚŝŽŶŝŶĂƐĞĐŽŶĚĐĂƌƚĞů͕ǁŚŝĐŚƚŚĞŶĐĂƵƐĞƐƚŚĞK:ƚŽĐŽŶƐŝĚĞƌŝŵƉŽƐŝŶŐĂ͞WĞŶĂůƚLJWůƵƐ͟ĨŝŶĞ that is near the GuidelŝŶĞ͛ƐŵĂdžŝŵƵŵ͘ůůĚĞĨĞŶĚĂŶƚƐĚĞƉŽƐĞĚďLJƚŚĞK:ĂƌĞĂƐŬĞĚƚŚĞ͞KŵŶŝďƵƐYƵĞƐƚŝŽŶ͟ƚŚĂƚ requires revealing possible additional cartels.
13
member of a cartel and divulged about other members.18 The specific types of cooperation expected from firms that have admitted their guilt comprises a rather short list (Spratling 1999: 4-‐‑9[23]: x
Produce all information, wherever located, that the DOJ requests
x
Permit all relevant information to be shared with foreign authorities
x
Secure the cooperation of all employees for interviews or testimony
x
Immediate cessation of collusion
Clearly the major benefit to prosecutors of cooperation is the ability to assemble testimony (eyewitness accounts of meetings and communications among the conspirators), written documents (memoranda of meetings, spreadsheets, or scorecards), other indisputable records of contacts (telephone logs, travel receipts, and the like), and electronic recordings of cartel activity Ȯ all of which would add up to an airtight case against all the defendants should the case go to trial. In a criminal antitrust system, prosecutors have many reasons to prefer resolving convictions through guilty-‐‑ plea agreements. Defendants give up their rights to trial and to appeals. Trials can take years of preparation and months of courtroom time for several prosecutors. Trial losses embarrass the DOJ. Often pȱ ȱ Ȃȱ -‐‑economic teams that are many times larger, better financed, and more experienced.19 When a second-‐‑in firm offers to cooperate, the proffered assistance may be quite valuable, but is more likely to be partially duplicative of what the first firm has already offered. Nevertheless, even duplicative information available from additional witnesses may be valuable to prosecutors. The cooperation value for information offered by third-‐‑ in and successive firms declines. 18
Cooperation discounts do not apply to self-‐ŝŶĐƌŝŵŝŶĂƚŝŶŐĨĂĐƚƐ͘͞^ĞůĨ-‐reporting, cooperation, and acceptance of respŽŶƐŝďŝůŝƚLJ͟ĂƌĞŵŝƚŝŐĂƚŝŶŐĐŝƌĐƵŵƐƚĂŶĐĞƐƌĞǁĂƌĚĞĚďLJƌĞĚƵĐƚŝŽŶƐŝŶƚŚĞĚĞĨĞŶĚĂŶƚ͛ƐĐƵůƉĂďŝůŝƚLJƐĐŽƌĞƵŶĚĞƌƚŚĞ Guidelines. 19 /ŶůĂƚĞϭϵϵϰ͕ƚŚĞK:͛ƐƉƌŽƐĞĐƵƚŝŽŶŽĨ'ĞŶĞƌĂůůĞĐƚƌŝĐĂŶĚĞĞĞƌƐŽŶƐŽůŝĚĂƚĞĚĨŽƌƉƌŝĐĞĨŝdžŝŶŐŝŶƚŚĞŐůŽďĂů market for industrial diamonds was dismissed after the prosecution presented its case. Commentary cited the overwhelming legal resources of GeneƌĂůůĞĐƚƌŝĐƌĞůĂƚŝǀĞƚŽƚŚĞ'ŽǀĞƌŶŵĞŶƚ͛ƐƌĞƐŽƵƌĐĞƐĂƐĂŵĂũŽƌĨĂĐƚŽƌŝŶƚŚŝƐ defeat for the DOJ (Connor 2007a: 75).
14
ȱ Ȃȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ ¢ȱ accepted after brief, pro forma judicial hearings. Indeed, the most common type of sentencing agreement is binding upon the court OECD [16]. That is, judges rarely question the negotiated plea agreements, and defendants know this.20 PREVIOUS EMPIRICAL STUDIES Surveys of the general crime-‐‑and-‐‑punishment literature note the paucity of quantitative empirical studies of the determinants of punishment; they cite the lack of appropriate data and methodological problems as reasons for so few studies Ehrlich 1987: 723[4]. CamȂ [24] survey includes several empirical studies of crime in the Beckerian tradition, but none examined antitrust enforcement. Levitt and Miles [25] argue that valid economic studies of crime first appeared only in the mid-‐‑1990s. Our review of the empirical law-‐‑and-‐‑economics literature finds only three studies that may be considered antecedents for the present paper in the sense that they test the validity of optimal deterrence theory. The first quantitatively estimates the variation in criminal fines for companies convicted for a wide range of U.S. federal crimes: fraud, tax evasion, environmental crimes, and the like Cohen [2]. Antitrust violations comprise a small share of Ȃ sample.21 A second study of contemporary international price-‐‑fixing violations examines the variation in penalties on the cartels as a whole, not the corporate members of those cartels, Bolotova and Connor [26].
20
Connor (2007a) notes only one case in which a supervising judge challenged the DOJ. A DOJ-‐authored report confirms that only one instance of judicial refusal occurred between 1997 and 2007 (OECD [16]. 21 In Cohen [2],Table 2, only 8.2% of the 961 sampled companies were convicted for antitrust violations. Parker and Atkins (1999) might be another antecedent study, except for the fact that they explicitly eliminate antitrust violations from their data set.
15
Cohen [2] examines variation in the size of U.S. corporate criminal penalties for a full sample of up to 961 companies criminally convicted during 1984 -‐‑1990 and for a subsample of 285 observations with estimates of harm available. He specifies a Tobit model with 21 independent variables, of which three relate to optimal deterrence theory. His major conclusions are: (1) the monetary value of harm to the market has the expected positive effect on both fines and total penalties; 22 (2) when a judge was aware of additional civil penalties, the fine was significantly reduced; and (3) the size and severity of penalties was unrelated to two proxies for the probability of detection (ibid. pp. 404-‐‑406). Although the Cohen study is a seminal piece with several fascinating results, it is of limited value as an antecedent for this paper because only 8% of his full sample consists of antitrust violators; moreover, the subsample test contains no antitrust violators. Bolotova and Connor [26]analyze the determinants of variation in total monetary penalties imposed on international cartels that were sanctioned between 1990 and 2005. Unlike Cohen [2], they examine antitrust sanctions worldwide for whole cartels, not their corporate members. Bolotova anȱȂȱȱǽŘǾȱȱȱDZ Sanctioni ƽȱ΅ƸȱΆȘ(Overchargei) Ƹȱȱ·ȱȘi ƸȱΔȘi + Όi ƸȱΉi . [2] Sanctioni is specified as either fines only or total penalties (fines and compensation recovered by private parties) for iƽȱ ŗǰȱ dzŚȱ . The explanatory variables in both specifications are: dollar or percentage damages to buyers (Overchargei), the volume of sales affected by the cartel (AffectedSalesi), cartel duration (Durationi), a set of control variables Ό that capture other differences among defendants or jurisdictions, and an error term (Ήi). 22 Total penalties include U.S. government fines, restitution, and civil and administrative settlements. Cohen does not contemplate penalties imposed outside the United States. The dependant variables, harm and other sanctions, were converted to natural logs, which results in coefficients that are elasticities. The elasticity of harm on fines was 0.41 and on all penalties 0.71.
16
ȱ ȱ Ȃȱ ȱ ȱ ȱ ŗŗŘȱ ȱ ȱ ȱ ȱ śŜȱ international cartels by antitrust authorities or courts during 1991 -‐‑ 2005. All variables measured in dollars are highly positively skewed. While all the cartels were fined by at least one authority, in many jurisdictions fines or penalties were zero. The model is estimated by the Tobit ML procedure, and six determinants explained 64% of the variation in government fines and 33% of the variation in total penalties. That is, cartel fines are considerably more predictable than are private settlements. Bolotova and Connor [26]:Table 4, find that affected sales is positively related to both fines and total penalties, as expected. The only surprising result concerns the role of overcharges: the dollar overcharge is negatively related to both fines and total sanctions, and the percentage overcharge is unrelated to penalties. These overcharge findings are contrary to optimal deterrence principles ȱ ȱ ȱ ȱ Ȃȱ ǽŘǾȱ findings. In the present study, we hope to resolve these inconsistent results.
SAMPLING AND DATA DESCRIPTION The sample of convicted global cartelists is drawn from an original data set, Private International Cartels (PIC). PIC identifies and collects information on the members, market characteristics, penalties, and other legal-‐‑economic dimensions of all international cartels discovered by any antitrust authority since January 1990. The members are the companies and their executives that were identified by prosecutors as participants in illegal hard-‐‑core price-‐‑fixing schemes. Every cartel has members headquartered in two or more nations; global cartels are international cartels that operated in two or more continents. For a large proportion of the cartels we identified ȱȱȱȱȱȱ ȱȱȱǻȃȱ ȄǼǯȱ¢ǰȱ ȱȱ large minority of the cartels, PIC contains market price effects (the buyerȂ overcharge) of these cartels.23 23 For details of data collection methods, see Connor (2010). 17
The sample employed in this paper comprises 124 companies in 39 global cartels penalized by the United States Government for price fixing from 1996 to March 2010.24 ȱ ȱ ȱ ȱ ¢ȱ ȱ ȃȱ -‐‑ǰȄȱ ȱ ¢ȱ that Levitt and Miles [25] identify as responsible for progress in economic studies of crime. The first global cartel fined is Lysine and the most recent Marine Hose. All companies were convicted through guilty-‐‑plea agreements that conferred partial leniency. The sample excludes about 30 companies in these cartels that were apparently granted immunity from criminal prosecution by the DOJ.25 A summary of the sample is given in Table 1. EMPIRICAL MODELS AND HYPOTHESES Empirical Models
Following equation [1] above, the dependent variable in our empirical model is the amount (in current U.S. dollars) of fines imposed on a corporate cartel participant by the United States (USF). To develop an econometric model of USF based on optimal deterrence theory, we must adapt equation [1] above to reflect the available data for the explanatory variablesǯȱ ȱ ǰȱ ȱ ȱ ȱ ¢ȱ ȱ ȱ ȱ Ȃȱ ȱ and is only a partial measure of the market injuries. However, if our model corrects for variation in the elasticity of demand, the unobserved deadweight loss will be proportional to the measurable overcharge, and HARM is a relevant proxy for the deadweight loss resulting from the cartel actions. Data limitations compel us to find a proxy for HARM, though a surrogate variable may still have good predictive value because damages are rarely employed by the DOJ to calculate recommended fines. 24
The DOJ attempted to convict a few companies for global price fixing but lost at trial. Only four companies were dropped from the sample because of incomplete data for a key variable, U.S. affected sales (ASUS). 25 Nor does the sample encompass about 200 companies that participated in convicted global cartels with affected sales in the United States that were not punished or given immunity by the DOJ; the vast majority was convicted by other antitrust authorities. Reasons for lack of punishment by the DOJ may include: inadequate evidence should the suspect demand a jury trial, low affected sales, large previous or anticipated monetary penalties by other parties, the statute of limitations, and inadequate DOJ resources to investigate or prosecute certain cartels.
18
Accordingly, we follow Bolotova and Connor [26] and substitute the firm-‐‑specific U.S. affected sales (ASUS) as a proxy for HARM. Recall that the Sentencing Guidelines require that the DOJ use ASUS to begin formulating a recommended fine. Almost all of the whole-‐‑cartel affected sales were adapted from DOJ press releases or speeches and 40% of the data on firm-‐‑specific U.S. affected sales came directly from posted DOJ sentencing agreements; the rest of estimates of ASUS were based on market shares taken from various reliable sources. To verify that ASUS is a good proxy for HARM, we computed the correlation between the market-‐‑level observations of AS and HARM for a subset of 26 cartels in our data set for which both variables are available, and the sample correlation is very strongly positive (r=+0.98). To see whether the DOJ pays some attention to injuries outside the United States, we also include a variable that represents market-‐‑level affected sales in the rest of the world (ASROW). To account for variation in the elasticity of demand, we introduce variables that capture differences in market supply and demand characteristics through a fixed-‐‑effects approach. Second, p is the probability of detection and conviction of various cartels (thus, 1/p is the difficulty of detection and conviction). Of course, 1/p cannot be directly measured and must be represented by appropriate proxy variables. Some of the proxies relate to the organizational features of the cartel, and there are many suggestions from cartel theory, Jacquemin and Slade [27], Grout and Sonderegger [28]. For example, cartels with many small firms are more likely to be unstable than those with few firms Carlton and Perloff [29]; Davies and Olkzak [30], which suggests that inequality of size among firms in a cartel makes punishment of defectors more likely than a cartel of more symmetric participants. Also, bid rigging may be harder to discover than price-‐‑ or quantity-‐‑fixing collusion; markets where cartels sell to many buyers tend to facilitate long-‐‑lived conspiracies; cartels with a large share of members from Europe or Asia may find clandestine agreements more compatible with their business cultures than firms from North America; and cartels with a large fringe are expected to be easier to detect than cartels with near monopolies over supply.
19
Third, OTHPEN is the actual size of monetary penalties imposed on the firm by other antitrust authorities or on thȱ Ȃȱ ȱ . The largest component of OTHPEN is recoveries in private suits; fines on executives are negligible. Equation [1] suggests that optimal deterrence can be achieved by the sum total of several types of monetary penalties or non-‐‑monetary punishments that have monetary equivalency.26 That is, expected fines imposed on these global cartels by other antitrust authorities (OTHF) are, dollar for dollar, perfect substitutes for USF. Furthermore, the expected cost of private settlements (PVT) also deters cartel formation. Whether the number and severity of individual penalties affects USF is an open question, though deterrence theory clearly suggests that they ought to substitute for USF, especially if the culpable individuals are top managers. Thus, OTHPEN may be represented by the sum of alternative penalties (OTHF and PVT) that ought to explain variation in USF, especially because U.S. prosecutors tend to have good notions of the size of all future penalties at the time a plea is being negotiated.27 In the case of global cartels, the DOJ is nearly always the first to impose a penalty. Canada typically follows by six months and the EU by three years; private settlements also usually follow DOJ fines within two to five years. These lags may lead to measurement problems with OTHPEN.28 Finally, as noted above, the monetary variables exhibited very high degrees of skewness. Following a similar decision by Cohen [2], we specify our econometric model based on equation [1] to be logarithmic in the key monetary variables (e.g., USF, ASUS, and ASROW) and linear in the non-‐‑monetary control variables. Although we initially tried logarithmic transforms of the OTHF and PVT variables, we found that a quadratic specification (with the variables divided by 100 for proper scaling) was supported by the model diagnostic tests. Accordingly, we used linear and quadratic terms based on 26
The most obvious example is prison sentences on executives of cartel members; the monetary equivalent is the sum of money an individual would pay to secure his freedom and reverse the stigma of conviction. Legal defense costs (including defeŶĚĂŶƚƐ͛ŵĂŶĂŐĞƌŝĂůƚŝŵĞͿ͕ǁŚŝĐŚĂƌĞƌĂƌĞůLJƌĞǀĞĂůĞĚ͕ĂƌĞƉĞŶĂůƚŝĞƐ͘ŽƌƉŽƌĂƚĞĚĞďĂƌŵĞŶƚĂŶĚ reputational loss also have monetary equivalents. None of these are easy to measure. 27 DOJ prosecutors resolve corporate and executive penalties simultaneously, receive information on private suits ĚŝƌĞĐƚůLJĨƌŽŵƉůĂŝŶƚŝĨĨƐ͛ĐŽƵŶƐĞů͕ĂŶĚŚĂǀĞĨƌĞƋƵĞŶƚĐŽŵŵƵŶŝĐĂƚŝŽŶƐǁŝƚŚŶŽŶ-‐U.S. prosecutors. 28 ĐƚƵĂůƉĞŶĂůƚŝĞƐĂƌĞƵƐĞĚĂƐĂƉƌŽdžLJĨŽƌK:ƉƌŽƐĞĐƵƚŽƌƐ͛ĞdžƉĞĐƚĂƚŝŽŶƐĨŽƌŽƚŚĞƌƉĞŶĂůƚŝĞƐ͘Kd,WEŝƐĂĐĐƵƌĂƚĞ for the companies fined during 1996-‐2003 but is understated for many fines imposed during 2003-‐2008 because antitrust authorities abroad had not acted nor had private suits been settled as of march 2010.
20
the sum of OTHF and PVT to form our proxy for OTHPEN in the final version of the empirical model. The final form of the econometric model is: ǻǼȱƽȱ΅ȱƸȱΆ1ȉLN(ASUSǼȱƸȱΆ2ȉLN(ASROWǼȱƸȱ·ȉǻŗȦǼȱ ƸȱΈ1ȉOTHPEN + Έ2ȉOTHPEN2 ƸȱΏȉȱƸȱΉ ,
[3]
where CONTROLS is a vector of variables that capture variation in the degree of culpability of a defendant (not already reflected in ASUS or ASROW), demand and supply ȱ ȱ ȱ Ȃȱ ȱ ȱ ȱ ȱ ȱ -‐‑price elasticity of demand, and lȱ ȱ ȱ ȱ ȱ ȱ Ȃȱ ¢ȱ ȱ ȱ ǯ 29 Industry dummies will crudely capture structural differences in demand. Duration of cartels, if independent of HARM or AS, is a factor that increases culpability under the USSGs. Bid rigging also increases USSG culpability scores. Rigging tenders offered by government agencies may be treated with a different severity by the DOJ than when private firms are buyers. Hypotheses for Particular Proxy Variables
First, as predicted by the economic theory of crime, optimal cartel sanctions are a ȱ ȱ ȱ ȱ Ȃȱ ȱ ǻȱ US and ASROW as proxies). Therefore, we hypothesize30 ȱ Ά1>0 and Ά2>0 in equation [3]. Furthermore, the magnitudes of the estimated coefficienȱȱΆ1 and Ά2 are meaningful. If USF performs only a compensatory function (is sub-‐‑ǼǰȱȱΆ1+ Ά2 is less than one. If U.S. fines ¢ȱǰȱȱΆ1+ Ά2 is one; finally, if the fines are over-‐‑deterring, the magnitude ȱΆ1+ Ά2 is greater than one. We also expect that ASUS will be more strongly related to USF than ASROW because the USSGs require the base fine to be computed using AS US. Further, as noted above, ASUS is a firm-‐‑specific proxy for HARM while ASROW is a market-‐‑level observation of affected sales that takes the same value for all firms in the same cartel. 29
Inability to pay sets an upper limit on fines. We have no measures on inability to pay.
30
These are alternative hypotheses.
21
Second, we expect that optimal fines to be inversely related to p, the probability of detection and conviction of a cartel. Given that we use a set of proxy variables to represent ǰȱȱȱȱ· is a vector of parameters. The sign of an element of ·ȱ ȱ be positive when the associated proxy variable indicates that the chances of cartel detection are low, the costs of detection are high for buyers or prosecutors, or difficulties of prosecution are high. Four variables are designed to capture variation in cartel structures that proxy the chances a cartel will be discovered by buyers or antitrust authorities. Bid rigging (BIDRIG) is an aggravating factor under the USSGs, perhaps because it is harder to detect and because cartels find it easier to monitor cheating under open-‐‑record laws.31 Moreover, governments are alleged to be inept at detecting collusion compared with procurement specialists with firms in the private sector. 32 Similarly, GOVTBUYS may be interpreted as a factor that raises the cost of prosecution of bid-‐‑rigging cartels, because investigations of such cartels place the burden of proof on prosecutors to establish restitution (damages calculations) that not needed for prosecution of most classic cartels. BIDRIG and GOVTBUYS are correlated conceptually and empirically, so at least one of them should be positively related to the optimal USF. When a cartel has a dominant member (LEADER = 1), cartels are likely to be more stable and harder for the authorities to catch. Thus, when these three determinants take on high values, optimal fines are higher and these elements of ·ȱ ȱȱǯȱȱȱ other hand, large-‐‑membership cartels (N is high) are expected to more discoverable because they are more fragile, i.e., more likely to foster whistle-‐‑blowers, which suggests a negative sign for this element of ·ǯ
31
The U.S. Sentencing Guidelines impose higher fines for bid-‐rigging schemes because they were believed to generate systematically higher overcharges. For a discussion of this issue, see Connor and Lande [3]. 32 There is a large body of writings in the branch of economics known as Public Choice that critically examines the assumption of neutrality of politicians and civil servants that is common in the economics of taxation and spending ;dƵůůŽĐŬϭϵϴϳͿ͘dƵůůŽĐŬƌĞĨĞƌƐƚŽƚŚŝƐƚŽƉŝĐĂƐƚŚĞ͞dŚĞŽƌLJŽĨƵƌĞĂƵĐƌĂĐLJ͘͟dŚĞŵĂŝŶŚLJƉŽƚŚĞƐĞƐĂƌĞƚŚĂƚĐŝǀŝů servants cannot always be counted on to reflect the priorities of their duly elected managers, and that they make decisions that serve their self interests (job security, promotion, aggrandizement of authority, and perks). Similarly, government procurement agencies may become captives of rent-‐seeking by firms subject to antitrust enforcement.
22
Other relevant proxy variables affect p through the costs and difficulties of prosecution after detection. For example, when a cartel has a record of conducting protracted plea negotiations, optimally deterring fines will ȱȱǻ·ȱ ȱȱǼǯȱ To illustrate, consider a possible proxy for cover-‐‑up: the length of time the DOJ took to investigate a case (PROBE)33; a lengthy probe may well signal that the defendants had destroyed most of the evidence needed to convict them or that defendants were stubbornly adversarial in plea negotiations. Given that plea negotiations are intended to be labor-‐‑ saving substitutes for trials, it is reasonable for prosecutors to impose higher penalties on firms that were particularly uncooperative during negotiations. Alternatively, we considered PROBEDUM, which is a dummy variable that equals one if the firm delayed the progress of the investigation (e.g., by destroying evidence), as a means to partly mitigate the potential measurement error in PROBE. Thus, either PROBE or PROBEDUM is expected to be positively related to USF. On the other hand, MANYBUYERS assists discovery because it is representative of the number of potential tips that may be generated, which in general reduces the costs of conviction. For this reason, MANYBUYERS is the one determinant with dual effects; the net effect on USF depends on which if any of these effects dominates. 34 In sum, USF is hypothesized to be inversely related to N, but directly related to PROBE, PROBEDUM, LEADER, BIDRIG, and GOVTBUYS. The expected effect of MANYBUYERS is ambiguous. Third, optimal deterrence regards all monetary penalties as fungible and, thus, we expect USF to decline as OTHPEN rises. That is, to the extent to which DOJ prosecutors are cognizant of investigations that have a likelihood of resulting in additional fines or prior fines imposed by other antitrust authorities on the company (OTHF), USF will be lower; similarly, high expected future private penalties (PVT) will lower the optimal 33
PROBE has significant measurement errors caused by the secrecy that surrounds DOJ investigations, whether internal to the Division or through a grand jury. In a minority of cases an investigation is revealed on the same day that the first cartel indictment is announced. More commonly, especially in global cartel cases, the start of an investigation becomes public when corporations reveal that subpoenas are served, when prosecutors exercise search warrants, or cooperating foreign antitrust authorities conduct simultaneous raids with the DOJ. 34 The BIDRIG and GOVTBUYS variables are also somewhat interrelated with MANYBUYERS because much bid rigging is directed at tenders issued by government agencies, which are monopsonies for particular contract proposals (i.e., low value of MANYBUYERS). If this is correct, the optimal USF might decline when MANYBUYERS is high.
23
USF. A related suggestion was explored by Cohen [2], who finds some evidence35 that ȱȃdz ȱȱȱȱȱȱȱȱ¢ȱȱȱ substitutes for purposes ȱ ǯȄȱ Besides other monetary penalties on the company, unless leniency is granted, the DOJ imposes sanctions on cartel managers. Executives are typically fined very little (the median fine is $100,000), but incarceration of 12 months (the 1990-‐‑2008 average) could have considerable opportunity cost for high-‐‑level executives Connor [31]. Thus, we also considered the variable PRISON, which is a continuous variable that equals the months of prison time that the Ȃ executives were sentenced. Theory suggests that PRISON is a substitute for USF. Fourth, we also include industry-‐‑specific dummy variables (fixed effects). Although we cannot assign expected signs to their coefficients, the industry pattern of discovered cartels over wide swaths of history is to find cartels in markets for industrial intermediate materials with high barriers to entry; the organic chemicals industry is an exemplar. A history of cooperative conduct may foster cartelization. Industries populated by firms that were in recent years36 subject to government price regulations also seem likely to support more stable cartels; recently deregulated industries -‐‑-‐‑ such as airlines, surface freight transportation, telecommunications, insurance, and banking Ȯ have had a history of passive and cooperative pricing conjectures that may carry over into a deregulated industry regime. Therefore, we examine the effects of cartels formed in the chemical (CHEM) and service (SERVICE) industries on USF relative to the reference group of all other industries (most of them non-‐‑chemical manufacturing). We also included dummy variables to indicate the headquarters location for European (EUR) and Asian (ASIA) firms. We have no hypotheses for these variables. ȱ ¢ȱ ȱ ȱ Ȃȱ ȱ ǻǼȱ ¢ȱ ȱ ȱ £ȱ ȱ ǯȱ ǰȱ Connor [31] noted signs of increasing intolerance of international cartels over time in statements of 35
The probability of a corporate employee being convicted along with his employer increases with the amount of harm, and this relationship is stronger when the firm is large and not closely held. 36 Deregulation of most of these sectors began in the United States in 1979 and was mostly complete by 1990. However, we are examining global cartels, and in much of the rest of the world, deregulation was contemporaneous with our sample period (1990-‐2010).
24
DOJ officials, and this suggests that T will have a positive sign. On the other hand, the period 1996-‐‑2010 spans two presidential administrations, and there is some evidence from DOJ and FBI workload statistics that investigative resources shrank and anti-‐‑cartel enforcement slackened somewhat in 2001-‐‑2005 compared to 1996-‐‑2000 reference period and recovered thereafter Connor [32]. If in fact there was a reduced anti-‐‑cartel commitment in 2001-‐‑2009, the sign of the coefficient of T will be negative. Thus, the sign on T is ambiguous. Alternatively, we replaced T with BUSH1 and BUSH2, which are dummy variables that equal one the first Bush term (2001-‐‑2005) and the second Bush term (2005-‐‑2009). The reference period for BUSH1 and BUSH2 is the first part of the sample period (1996-‐‑2000), which roughly covers the second Clinton administration. Cartel DURATION is hypothesized to be positively related to the size of cartel sanctions. As noted above in the discussion of cartel discounting, plea bargains often include a concession to a defendant on the dates of its collusion. In other cases, the DOJ shortens the cartel span because it lacks documentary or testimonial evidence on the beginning stages of a lengthy cartel. Often, subsequent convictions, particularly in private rights of action, find U.S. courts approving a settlement based on a significantly longer conspiracy period than is revealed by DOJ plea agreements. As a result, HARM or AS may be understated when calculating USF. Based on evidence from model specification tests, we included the natural logarithm of DURATION in the final version of Model [3].
The complete set of independent or explanatory variables considered for use in Model [3] and their associated hypotheses are summarized in Table 2.
25
ESTIMATION RESULTS
We estimated Model [3] by ordinary least squares (OLS) based on the full set of explanatory variables described in Tables 1 and 2. 37 Based on these preliminary results, we condensed two sets of explanatory variables that exhibited strong pairwise correlations and may have led to potentially harmful collinearity in the fitted model. First, three dummy variables (GOVTBUYS, BIDRIG, and SERVICE) were nearly coincident; the pairwise correlations among these variables ranged from 0.44 to 0.86. For this reason, we only included the BIDRIG variable in the model because it largely encompasses the cases represented by the other two variables and because it is an aggravating factor in the Guidelines. Second, we found that the quadratic term for OTHPEN (i.e., the squared value of OTHF plus PVT) was statistically significant. Third, we identified other explanatory variables that exhibited limited statistical significance (LEADER, TIME, PROBE, EXECS, ASIA, SERVICE, and ASROW) and estimated the refined model by excluding these explanatory variables. The OLS parameter estimates, t statistics, and associated p-‐‑values for the final version of Model [3] are presented in Table 3. The 13 independent variables that remain explain 76.5% of the variation in the natural log of USF. This degree of goodness of fit is quite satisfactory given the highly disaggregated nature of our data.38 ȱȱȱȱǰȱ ȱȱ¢Ȃs RESET procedure to test for the presence of unspecified nonlinearities and also conducted the Breusch-‐‑Pagan-‐‑ Godfrey (BPG) and White tests for heteroskedasticity, Wooldridge [33]. The test result for the RESET procedure reported at the bottom of Table 3 shows that the null 37
There were a substantial number of missing observations for LEADER and PVT that were missing, and these values were recoded as zeros. The problem with PVT is that it is understated for cartels fined in the last five or six years, because settlements typically take years to be resolved after a fine is imposed or the parties wish to keep them confidential. 38 The fitted model reported by Bolotova and Connor [26],Table 4, explains 64% of the variation in more aggregated total cartel fines, albeit with fewer independent variables.
26
hypothesis (correctly specified functional form) was not rejected at the 10% level. However, the test results for the White and BPG tests provide mixed evidence for heteroskedasticity. Therefore, we computed heteroskedastic-‐‑robust t test statistics based on the White estimator, and these show that our interpretations of statistical significance are unchanged when one accounts for the presence of heteroskedasticity in the data. TABLE 3 HERE The signs and significance of the independent variables yield several interesting conclusions. First, the marginal effect of LN(ASUS) is significantly positive, as expected. The double-‐‑ȱȱȱ ȱȱȱȱ ȱȱ ȱΆ 1 is an elasticity. Consequently, our model predicts that the dollar value of U.S. cartel fines increases by roughly 5.9% as firm-‐‑specific U.S. affected sales (ASUS) increases by 10%. Further, the fact that the marginal effect for LN(ASROW) is insignificant implies that U.S. prosecutors pay no heed to welfare effects outside the United States. The US-‐‑specific affected sales elasticity (0.59) is also significantly less than one at the 1% level, which strongly implies that the imposed fines are only partially compensatory in their function.39 From an ex post perspective, deterrence is not being served by U.S. fines alone. Second, for the proxy variables related to the odds of detecting cartels, the hypotheses are not supported by the estimation results. In particular, LEADER, GOVTBUYS, and BIDRIG were not significant. Out of the 22 bid-‐‑rigging observations, 18 were fines imposed on firms that had primarily engaged in bid-‐‑rigging against the U.S. military. ǰȱ Ȃ Theory of Bureaucracy is not supported, and the policy conclusion is that bid-‐‑rigging conduct is not in practice an aggravating factor in setting U.S. global cartel fines.40 While the DOJ is expected to impose higher fines against noncooperative defendants, we find that PROBE in insignificant and that PROBEDUM is significantly negative. ȱ ȱ ȱ ¢ȱ ȱ ȱ ȱ Ȃȱ ǯȱ 39
ŽŚĞŶ͛ƐϮƌĞŐƌĞƐƐŝŽŶĂŶĂůLJƐŝƐŽĨĐŽƌƉŽƌĂƚĞĐƌŝŵŝŶĂůĨŝŶĞƐŝŶƚŚĞĞĂƌůLJϭϵϴϬƐĐŽŵƉƵƚĞƐĂŶĞůĂƐƚŝĐŝƚLJŽĨϬ͘ϰϭ (Table 5). In another model that includes restitution and all other federal monetary penalties, the elasticity is 0.71. 40 Bid rigging premia may be applied to fines involving localized conspiracies, which are excluded from our sample.
27
Rather, long-‐‑lasting probes may signal that prosecutors judged their evidence relatively weak. It is also possible that PROBEDUM=0 reflects the secrecy of an investigation; in this case, the negative sign on PROBEDUM means that when news about a DOJ investigation leaks, the expected USF is lower.41 MANYBUYERS has an ambiguous expected sign, and we find ȱ ȱ ȱ ¢Ȃȱ ¢ȱ insignificant. Inconsistent with expectations, we find that companies in cartels with one more member than other cartels (an increase in the number of firms N) are predicted to incur roughly 12% higher U.S. fines, and this estimated coefficient is statistically significant at the 1% level. We find it puzzling that the DOJ should treat defendants in well populated cartels more severely. Third, we find that the best formulation for OTHPEN is quadratic. The impact of other penalties on USF is increasing at a decreasing rate. Further, the estimated point at which the marginal impact of OTHPEN on USF would begin to decline is at $516 million, which is higher than nearly all of the observed OTHPEN values in our sample. Thus, the estimated marginal impact of OTHPEN on USF is effectively positive, and DOJ prosecutors are not following the principles of optimal deterrence for the observed range of other penalties. We also note that the potential measurement error in the OTHPEN values (explained in our data discussion) is not a likely cause of this outcome. In general, measurement errors generate attenuation bias in the OLS estimator such that the expected value of the estimator is smaller in absolute value, but the expected sign of the estimator is unchanged. That the result for the PRISON coefficient is significantly positive has a similar interpretation as OTHPEN. Both corporate and individual U.S. penalties are complements to U.S. corporate fines, rather than substitutes as optimal deterrence theory posits.42
41
It is noteworthy that 47 corporate observations (38% of the total) had zero values for PROBE. This is an impossible number. In effect, when PROBE = 0 this captures those cases for which a grand jury operated in complete secrecy, i.e., its existence was only revealed to the press on the day the first defendant pled guilty. When PROBEDUM=1, either a very public raid occurred, a defendant revealed receiving a subpoena, or the existence of a Grand Jury leaked. There is one observation that may be an outlier; the Industrial Diamonds case dragged on for more than 10 years because the remaining duopolist (DeBeers of South Africa) was outside the reach of U.S. law; De Beers had a modest fine imposed. 42 In our sample 22 companies had one or more executive sentenced to prison; the median term was 11 months. However, PRISON may be an inadequate proxy for the opportunity cost of individual prosecutions. The opportunity
28
Fourth, four of the five control variables remaining in the final model are statistically significant. We did not have prior expectations for the sign on the coefficient for the chemical industry dummy variable (CHEM), nevertheless we are a bit surprised that defendants in this collusion-‐‑prone industry received a statistically significant reduction of roughly 113% in USF relative to other types of firms.43 Similarly, although we found no evidence of a general linear time trend, the final model establishes large effects from the two dummy variables that represent anti-‐‑cartel enforcement during the George W. Bush administrations. The dummy coefficient for BUSH1 is significantly negative at the 10% level, and the estimated magnitude of this coefficient suggests that the USF values decreased in 2001-‐‑2004 by roughly 50% (relative to the latter Clinton administration). The dummy variable for BUSH2 (2005-‐‑2009) had an even greater effect of negative 172% Figure 1). The estimates should be robust because half of our sample is drawn from each Presidential administration (Table 2), and a detailed analysis of the data indicates the likely reasons for these large estimates. Although the mean and median values of USF did decline during the BUSH1 period relative to the Clinton administration fines, the mean and median values of USF during BUSH2 were modestly higher than the Clinton values. However, ASUS increased sharply after 2001, and the large negative coefficient estimates for the BUSH1 and BUSH2 dummy variables indicate that the observed fines underperformed relative to expectations during 2001-‐‑2009 and were lower than the conditional expected fines (i.e., given the characteristics of the cartels and their members, such as affected sales) (Figure 2). The final significant control variable is a dummy variable EUR, which takes a value of one when tȱ Ȃȱ ȱ ȱ ȱ ȱ ȱ ȱ ǯȱ Ȃȱ ȱ ȱ ¢ȱ positive, and its value suggests that European firms were fined roughly 41% more than Asian and North American companies. Rather than representing a discriminatory effect, we suspect that European firms as a group have some undetected culpability factor not accounted for in the model.44 Finally, we find that the marginal effect associated with the natural logarithm of DURATION is positive but insignificant at the cost of incarceration is unknown, but may be rather high. Another problem is that about one-‐fourth of all indicted executives in international cartels abroad with little chance of being extradited to the United States. 43 This odd result cannot be explained by ASUS because the US affected sales of chemical manufacturers in the sample is 30% as high as the size of the remaining cartelists. 44 European firms, for example, tend to be high on lists of cartel recidivists (Connor and Helmers 2006).
29
10% level. The estimated coefficient suggests that USF increases by about 1.8% given for each 10% increase in cartel duration. The weakness of DURATION suggests that cartels with longer duration receive higher USF through the harm caused rather than through any independent adjustment. Figures 1 & 2 HERE DISCUSSION In general, the predictability of criminal sanctions is held to be an element of judicial efficacy and fairness. Under concepts of optimal deterrence of crime, potential law-‐‑ breakers are presumed to be able to predict with some degree of certainty the material benefits and costs of criminal conduct ex ante. That is, rational criminal decisions are based on the assumption that reasonably accurate expectations about probable penalties can be formed at the time a participant is weighing the benefits and costs of initiating the illegal conduct. Moreover, one of the guiding principles of sanctioning illegal behavior is that of proportionality, i.e., among comparable defendants the fine should fit the crime. Optimally deterring penalties are inherently proportional (to the harm caused). An analysis of the sources of variation in cartel fines is also useful in understanding and appraising how antitrust enforcement is implemented in practice. The variables that measure predictions drawn from optimal deterrence theory of crime are only partially successful in predicting variation in fines on corporations convicted of global price fixing in the United States. The dollar value of fines imposed is strongly positively related to the proxy for the economic injuries imposed on U.S. buyers. However, the impacts of the variables representing the probability of antitrust detection and conviction do not conform at all ȱȱ ¢Ȃȱ . There is no evidence that the DOJ fines bid-‐‑rigging schemes more heavily than conventional price-‐‑fixing cartels. Intra-‐‑cartel asymmetry and the numerosity of buyers are likewise unrelated to cartel fines. The effects of the number of corporate members of the cartel and prior public information of the existence a DOJ investigation have signs contrary to theoretical predictions. Another element of optimal deterrence theory that does not 30
hold up well is the idea that other antitrust penalties are good substitutes for U.S. corporate fines in deterring cartel conduct. Rather, we find evidence that the DOJ piles on higher fines when sentencing the cartel managers to heavier prison sentences and when other antitrust monetary penalties rise. Among the control factors tested, three are noteworthy. Ceteris paribus, U.S. cartel fines during both Bush administrations were significantly lower than those imposed in the Clinton administration. Guilty firms in the chemicals sector were treated more leniently. And we find that European violators paid heavier fines than companies from other continents. Given the mixed levels of disaggregation of the data employed in this study (i.e., some variables are firm-‐‑specific, some cartel-‐‑specific), the overall fit of the models is quite good. Nevertheless, because model estimation was potentially affected by harmful collinearity and measurement limitations, we found it difficult to include some other reasonable determinants of U.S. cartel fines. Factors such as an inability to pay,45 defections from the cartel to seek amnesty, and recidivism are omitted from our model. Further experimentation with alternative measures of possibly substitute penalties may be productive. For example, one could examine whether the size or timing of corporate fines of particular authorities (Canada, EU, etc.) might provide more explanatory power than the geographically aggregated penalties that we employed. Also, our measure of ȱ ¡Ȃȱ ȱ -‐‑-‐‑ the number of months of prison sentenced -‐‑-‐‑ could conceivably be replaced by more appropriate monetary measures of the opportunity cost of such sentences. Another obvious extension would be to develop a more complex model that takes into account the possibly interrelated decisions of the DOJ, the European Commission, and settlements in private antitrust suits. 45 A traditional ƌĞĂƐŽŶĨŽƌĚŝƐĐŽƵŶƚŝŶŐĐĂƌƚĞůĨŝŶĞƐĂƌŝƐĞƐĨƌŽŵĂĚĞĨĞŶĚĂŶƚ͛ƐŝŶĂďŝůŝƚLJƚŽƉĂLJ͘ĞĐĂƵƐĞŵŽƐƚĐĂƌƚĞůƐ arise in concentrated industries, the exit of even one company can raise industry concentration. Thus, prosecutors are loath to propose and courts are unůŝŬĞůLJƚŽĂĐĐĞƉƚ ĨŝŶĞƐŚŝŐŚ ĞŶŽƵŐŚƚŽĐĂƵƐĞĂ ĚĞĨĞŶĚĂŶƚ͛ƐďĂŶŬƌƵƉƚĐLJ͘/Ŷ ĂĚĚŝƚŝŽŶ͕ĨŝŶĞƐƚŚĂƚĂƌĞƚŽŽůĂƌŐĞŵĂLJŝŵƉĂŝƌĂĚĞĨĞŶĚĂŶƚ͛ƐĂďŝůŝƚLJƚŽĐŽŶƚƌŝďƵƚĞƚŽĚĂŵĂŐĞƐƉĂLJŵĞŶƚƐŝŶ related private suits. However, one empirical study suggests that financial principles rarely find imposed fines high enough ƚŽĞŶĚĂŶŐĞƌĂĨŝƌŵ͛ƐƐƵƌǀŝǀĂů;ƌĂLJĐƌĂĨƚet al. 1997).
31
REFERENCES [1] R.A. Posner, Antitrust Law: Second Edition. Chicago: University of Chicago Press (2001). [2 ] M.A. Cohen, Theories of Punishment and Empirical Trends in Corporate Criminal Sanctions. Managerial and Decision Economics 17 (1996): 399-‐‑411. [3] J.M. Connor, R. H. Lande. How High Do Cartels Raise Prices? Implications for Optimal Cartel Fines, Tulane Law Review 80 (December 2005): 513-‐‑570. [4] I. Ehrlich, Crime and Punishment, pp. 721-‐‑723 in The New Palgrave Dictionary of Economics: Vol. 1, John Eatwell et al. (editors). London: Macmillan (1987). [5] G.S. Becker, Crime and Punishment: An Economic Approach. Journal of Political Economy 76 (1968): 169-‐‑217. [7] A.M. Polinsky, S. Shavell. The Optimal Tradeoff Between the Probability and Magnitude of Fines. American Economic Review 69 (1979)
880-‐‑91.
[8] J.M. Connor,The Economic Theory of Public Enforcement of the Law. Journal of Economic Literature 38 (March 2000): 45-‐‑76. [9] D. Klawiter, After the Deluge: The Powerful Effect of Substantial Criminal Fines, Imprisonment, and Other Penalties in the Age of International Criminal Enforcement. George Washington Law Review 69 (2001): 745-‐‑765. [10] J.M. Connor, Effectiveness of Sanctions on Modern International Cartels. Journal of Industry, Competition, and Trade (2007c). [11] G.R. Spratling, D. J. Arp. The Status of International Cartel Enforcement Activity in the U.S. and Around the World, address at the American Bar Association Section of Antitrust Law, Fall Forum, Washington, DC (November 16, 2005). 32
[12] S.D. Hammond, The Evolution of Criminal Antitrust Enforcement over the Last Two Decades, address at the 24th annual National institute of White Collar Crime, Miami, Florida, Feb. 25, 2010 [13] Aubert, Cecile, P. Rey, W. E. Kovacic, The Impact of Leniency and Whistle-‐‑Blowing Programs on Cartels. International Journal of Industrial Organization 24 (November 2006): 1241-‐‑66. [14] G. Spagnolo, Leniency and Whistleblowers in Antitrust, prepared for P. Buccirossi (Ed.), Handbook of Antitrust Economics. Cambridge, Mass.: MIT Press (2007). [http://www.cepr.org/meets/wkcn/6/6641/papers/spagnolo.pdf] [15] N.H. Miller, Strategic Leniency and Cartel Enforcement. American Economic Review (June 2009). [16] OECD. Plea Bargaining/Settlement of Cartel Cases (DAF/COMP(2007)38). Paris: Organization of Economic Co-‐‑Operation and Development (January 22, 2008). [available at http://www.oecd.org/dataoecd/12/36/40080239.pdf] [17] ICN. Cartel Settlements: Report to the ICN Annual Conference. Kyoto, Japan (April 2008). [http://www.icn-‐‑kyoto.org/documents/materials/Cartel_WG_1.pdf] [18] J.M. Connor, A Critique of Cartel Fine Discounting by the U.S. Department of Justice: SSRN Working Paper (revised April 24, 2008a). [available at SSRN: http://ssrn.com/abstract=977772] [19] G. Fisher, ȱȂȱDZȱȱ ¢ȱȱȱȱȱǯ Palo Alto: Stanford University Press (2003). [20] S.D. Hammond, The U.S. Model of Negotiated Plea Agreements: A Good Deal with Benefits for All, address before the OECD Competition Committee Working Party No. 3. Paris, France (October 17, 2006). [21] DOJ (Antitrust Division). ȱ ¢ȱDZȱȱşȱȄȱȄǯȱ Washington, DC (November 1991). [http://www.usdoj.gov/atr/public/guidelines/207144.pdf]. [22] USSC. 2005 Federal Sentencing Guideline Manual. Washington, DC: U.S. Sentencing Commission (November 1, 2005). [http://www.ussc.gov/2005guid/tabcon05_1.htm]
33
[23] G.R. Spratling, Negotiating the Waters of International Cartel Prosecutions: Antitrust Division Policies Relating To Plea Agreements In International Cases, speech at the 13th annual National Institute of White Collar Crime, San Francisco, California (March 4, 1999). [24] S. Cameron, The Economics of Crime Deterrence: A Survey of Theory and Evidence. Kyklos 41 (1988): 301-‐‑323. [25] S.D. Levitt, T. J. Miles, Economic Contributions to the Understanding of Crime. Annual Review of Law and Social Science 2 (2006): 147-‐‑164. [26] Y. Bolotova, J.M. Connor, Cartel Sanctions: An Empirical Analysis, 6th International Industrial Organization Conference, Alexandria, Virginia, May 16-‐‑17, 2008. [with Yuliya Bolotova] [27] A. Jacquemin, M. E. Slade, Cartels, Collusion, and Horizontal Merger, in Richard Schmalensee and Robert D. Willig (editors.), Handbook of Industrial Organization, Volume 1. Amsterdam: Elsevier (1989) [28] P.A. Grout, S. Sonderegger, Predicting Cartels, Office of Fair Trading Discussion Paper (OFT 773). London: Office of Fair Trading (March 2005). [29] D.W. Carlton, J.M. Perloff, Modern Industrial Organization: Fourth Edition. Boston: Pearson (2005). [30] S. Davies, M.Olczak, Tacit versus Overt Collusion Firm Asymmetries and DZȱȂȱȱǵ Competition Policy International 4 (2008): 175-‐‑200. [31] J.M. Connor, Anti-‐‑Cartel Enforcement by the DOJ: An Appraisal. Competition Law Review Vol. 5, Issue 1 (December 2008b). [32] J.M. Connor, Cartels and Antitrust Portrayed: Individual Penalties: SSRN Working Paper (March 2009). [http://ssrn.com/abstract=1372854] [33] J. Wooldridge, Introductory Econometrics: A Modern Approach (4th edition), Mason, OH: South-‐‑Western Cengage Learning (2009). FOOTNOTE REFERENCES
34
J.M. Connor, Global Price Fixing: 2nd Updated and Revised Edition: Studies in Industrial Organization No. 26. Heidelberg, Germany: Springer (2007a). J.M. Connor, Price-‐‑Fixing Overcharges: Legal and Economic Evidence, Chapter 4 in John B. Kirkwood (editor), Volume 23 of Research in Law and Economics. Oxford, Amsterdam and San Diego: Elsevier (2007b). J.M. Connor, Price-‐‑Fixing Overcharges: Legal and Economic Evidence: Second Edition: SSRN Working Paper (2010). [http://ssrn.com] J.M.Connor, C.G. Helmers, Statistics on Modern Private International Cartels: Working Paper #06-‐‑11. West Lafayette, Indiana: Purdue University (November 2006). [http://papers.ssrn.com/sol3/papers.cfm?abstract_id=944039], [http://www.agecon.purdue.edu/working_papers/workingpaper.connor.11.10.06. pdf] R.D. Cooter, D. L. Rubinfeld, Economic Analysis of Legal Disputes and Their Resolution. Journal of Economic Literature 27 (1989): 1067-‐‑1097. C. Craycraft, J.L. Craycraft, J. C. Gallo, ȱȱȱȱȂȱ¢ȱȱ¢ǯȱ Review of Industrial Organization 12 (1997): 171-‐‑183 J.Hoj, Competition Law and Policy Indicators for the OECD Countries: Economics Department Working Paper No. 586 (ECO/WKP(2007)28. Paris: Organization of Economic Co-‐‑ Operation and Development (August 8, 2007). [http://www.olis.oecd.org/olis/2007doc.nsf/LinkTo/NT00002ED6/$FILE/JT032308 25.PDF] B.H. Kobayashi, ȱ ȱȱDZȱȱ¡ȱȱȃȄȱ Plea Bargains. RAND Journal of Economics 23 (Winter 1992): 507-‐‑517. F.S. Parker, R. A. Atkins, Did the Corporate Criminal Sentencing Guidelines Matter? Some Preliminary Empirical Observations. Journal of Law and Economics 42 (1999): 423 453. G. Tullock, Public Choice, in pp. 1040-‐‑1044 of The New Palgrave Dictionary of Economics: Vol. 3, edited by John Eatwell, et al. London: Macmillan (1987). 35
Table 1. U.S. Corporate Global Cartel Sanctions: Descriptive Statistics. Variable Units Dependent variable: USF $US million Basis for Damages: HARMUS* $US million HARMWORLD* $US million ASUS $US million ASROW $US million Detection probability: GOVTBUYS dummy BIDRIG dummy PROBE years PROBEDUM dummy LEADER** dummy MANYBUYERS dummy N (cartelists) Number Other Penalties: OTHF $US million PVT $US million OTHPEN $US million EXECS number PRISON months Controls: TIME (T) years CLINTON dummy BUSH1 dummy BUSH2 dummy NO AM dummy EUR dummy ASIA dummy CHEM dummy
Mean 42.4 1591 15276 401.1 50398 0.14 0.18 1.03 0.62 0.79 0.73 9.7 25.5 54.5 80.0 4.3 3.7 12.2 0.40 0.23 0.36 0.23 0.48 0.29 0.48
Median 11.0 87.1 494 82.5 1395 0 0 0.78 1 1 1 7 8.25 6.45 20.1 3.0 0 12 0 0 0 0 0 0 0
Std. Dev. 70.2 3419 42053 1029 155,140 0.35 0.38 1.43 0.49 0.41 0.44 6.8 44.0 128.9 145.3 5.0 12.7 4.0 0.49 0.43 0.48 0.42 0.50 0.46 0.50
Minimum 0.01 0.0 0.0 0.5 0.3 0 0 0 0 0 0 2 0 0 0 0 0 6 0 0 0 0 0 0 0
Maximum 400.0 10,290 221,167 8,275 943,000 1 1 10.4 1 1 1 25 254.3 794.9 893.5 19 99 18 1 1 1 1 1 1 1 36
SERVICE DURATION
dummy months
0.23 85.6
0 71
0.43 65.6
0 3
1 365
Source: Global Cartel Fines spreadsheet dated 4/25/2010. *= Only 92 observations with available data; ** = only 66 observations with available data.
Table 2. Definitions and Expected Signs of Variables Explaining Variation in USF Explanatory Variable Cartel Injuries:
Definition
U.S. overcharges attained by cartel, a continuous variable measured in $US million (if not available, ASUS substituted) World overcharges attained by cartel, a continuous variable HARMWORLD measured in $US million (if not available, ASROW substituted) ASUS The volume of U.S. affected sales for each firm in the cartel, a continuous variable measured in current $US million ASROW The volume of non-‐U.S. market sales affected by cartel, a continuous variable measured in current $US million Factors that affect detection probability a: PROBE Length of time from the date a formal investigation is launched (if known) to the date the first member of the cartel is fined by the DOJ, a proxy for the effort required to convict, measured in years PROBEDUM =1 if the PROBE was greater than zero; if zero, the DOJ and Grand Jury investigation was successfully kept secret until the first member of the cartel was indicted and pled guilty. BIDRIG =1 if bid-‐rigging cartel, a form of collusion easier to hide from buyers (and an aggravating factor under the Sentencing Guidelines) GOVTBUYS =1 if the U.S. government is a principal buyer of the product LEADER =1 if intra-‐cartel market shares are unequally distributed, i.e., cartel has a leader with at least a 30% cartel production share and usually 40%+; may make cartels more stable (harder to catch) because dominant firms are highly credible sources of punishment of defection in cartels. N Number of sellers in cartel; increases the probability of defection, effective cheating, and the likelihood of applying for amnesty. MANYBUYERS =1 if cartel sells to more than 100 buyers; with dispersed buyers, individual transactions are more costly for a cartel to monitor, which makes cheating harder to detect for the other members of the cartel. On the other hand, a large number of buyers implies that ƚŚĞƌĞĂƌĞŵĂŶLJƉŽƐƐŝďůĞƚŝƉƐƚĞƌƐ͕ǁŚŝĐŚƌĞĚƵĐĞƐƉƌŽƐĞĐƵƚŽƌƐ͛ĐŽƐƚƐŽĨ detection. Other Substitute Penalties:
Expected Sign +
HARMUS
+ + +
+
+
+ +
+
-‐ +/-‐*
37
OTHF PVT
OTHPEN EXECS
PRISON Controls: TIME (T)
CLINTON BUSH1 BUSH2 NO AM EUR ASIA CHEM SERVICE DURATION
Size of fines imposed on the company by a government antitrust authority outside the United States measured in current $US million Size of company settlements in private suits by direct or indirect purchasers in the United States or Canada measured in current $US million OTHPEN = OTHF + PVT and is also measured in current $US million Number of executives charged or sanctioned by the DOJ for criminal price fixing in the same cartel, a proxy for the severity of fines or ŝŵƉƌŝƐŽŶŵĞŶƚŽĨƚŚĞĐŽŵƉĂŶLJ͛ƐĐĂƌƚĞůŵĂŶĂŐĞƌƐ Months of prison time assigned to cartel executives as sentences
-‐
T may proxy greater overall severity of fines over time, measured by the last two digits of the year after 1990 in which the first member of the cartel was fined by the DOJ =1 if the fine was imposed during 1996-‐1999 =1 if the fine was imposed during 2001-‐2004 =1 if the fine was imposed during 2005-‐2009 =1 if firm is US or Canadian =1 if firm is headquartered in Europe =1 if firm is headquartered in Asia or Oceania =1 if a chemical product market =1 if transportation or other service-‐sector market Cartel duration, longest dates proven by any antitrust authority measured in months
+
-‐ -‐ -‐ -‐
reference
-‐ -‐ reference
+ + +/-‐* +/-‐* +
a) When a factor represents a small probability of detection (hard to catch) or large probability of difficult or costly conviction, the optimal USF will be high (i.e., positive coefficient), and vice-‐versa when detection is easy or the effort needed to convict is low. * The estimated coefficient is expected to have either a positive or a negative sign, depending on which force predominates.
38
Table 3. OLS Estimation Results for Model [3] with Unadjusted and Heteroskedastic-‐ Robust (White) Standard Error Estimates OLS Unadjusted Heteroskedastic-‐robust t statistic p-‐value t statistic p-‐value Estimates Intercept
-‐1.2876
-‐1.72
0.089
-‐1.96
0.5911
9.18
0.000
10.42
Detection probability: BIDRIG -‐0.1649 PROBEDUM -‐0.5050 N (cartelists) 0.1162 MANYBUYERS 0.4804
-‐0.43 -‐1.73 3.58 1.25
0.669 0.086 0.001 0.216
-‐0.44 -‐1.84 3.27 1.39
Other Penalties: OTHPEN OTHPEN2 PRISON
0.6268 -‐0.0607 0.0183
3.29 -‐2.40 2.32
0.001 0.018 0.022
3.89 -‐3.01 3.97
Controls: BUSH1 BUSH2 CHEM EUR LN(DURATION)
-‐0.4964 -‐1.7243 -‐1.1331 0.4088 0.1789
-‐1.74 -‐4.41 -‐2.33 1.83 1.01
0.085 0.000 0.021 0.070 0.317
-‐1.79 -‐2.51 -‐1.95 1.80 1.00
Diagnostic Statistics: RESET F stat (nonlinearity) RMSE 1.0286 White stat (heteroskedastic) R2 statistic 0.7648 BPG stat (heteroskedastic)
0.296 120.4 18.6
Proxy for Harm: LN(ASUS)
0.052 0.000 0.662 0.069 0.001 0.166 0.000 0.003 0.000 0.076 0.014 0.054 0.075 0.318 0.828 0.018 0.137
39
Fig. 1.Predicted Fines by President 120
CLINTON
100 80
BUSH1
60 40 20 0 -‐20
96
97
98
99
00
01
02
03
04
05
06
07
08
-‐40 BUSH2
-‐60 -‐80 -‐100 Index
40
Fig. 2. US Affected Sales per Convicted Member of Global Cartels 2000.00 1800.00 1600.00
1400.00 1200.00
1000.00 800.00
AS per firm
600.00 400.00 200.00 0.00
41