Detecting Cartels Joe Harrington Johns Hopkins University - PowerPoint PPT Presentation

slide1 n.
Skip this Video
Loading SlideShow in 5 Seconds..
Detecting Cartels Joe Harrington Johns Hopkins University PowerPoint Presentation
Download Presentation
Detecting Cartels Joe Harrington Johns Hopkins University

play fullscreen
1 / 29
Download Presentation
Detecting Cartels Joe Harrington Johns Hopkins University
Download Presentation

Detecting Cartels Joe Harrington Johns Hopkins University

- - - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript

    1. Detecting Cartels Joe Harrington (Johns Hopkins University) Advances in the Economics of Competition Law June 2005

    3. What is it that we are trying to detect? Two forms of collusion Explicit collusion - firms coordinate through direct communication. Tacit collusion - firms coordinate through some mutual understanding and without the aid of direct communication. Target is explicit collusion. Stages in the process Screening - identifying markets where collusion is suspected. Verification - systematically distinguishing between collusion and competition. Prosecution - developing economic evidence to determine guilt.

    4. Empirical methods for finding collusion Structural: Identifying industry traits conducive to collusion. Behavioral: Identifying collusive behavioral patterns. Focus is on behavioral methods. Questions to be addressed What are current methods for identifying collusion? How easily can a cartel pass a test for collusion? Can the antitrust authority play a more active role in screening industries for cartels?

    5. Overview of behavioral empirical methods Is behavior inconsistent with competition? Is there a structural break in behavior? Does the behavior of suspected colluding firms differ from that of competitive firms? Does a collusive model fit the data better than a competitive model? Methods vary according to the type of data that is required the need for prior information about collusion reduced form or structural estimation methods

    6. Is Behavior Inconsistent with Competition? Approach Specify a class of competitive models and identifies properties of equilibrium. Do these properties hold? If yes, then there is no evidence of collusion. If no, then this could be due to misspecification of cost and demand assumptions behavioral assumptions (collusion) Is the manner in which results differ from competition consistent with some model of collusion?

    7. Procurement auctions for highway maintenance projects (Bajari and Ye, 2003) Key maintained hypothesis: After conditioning on publicly available information, bidders' costs are independent. Property of competitive solution Firms' bids are independent. Test: Is the unexplained part of one firm's bid independent of the unexplained part of another firm's bid? Implementation Estimate a pricing equation for each firm. Calculate the residuals of each firm's bid function, eit. Test the hypothesis that the coefficient of correlation for eit and ejt is zero.

    8. Inferences Suppose the test of independence fails. The model is then misspecified and collusion is one possible source of misspecification. Is the failure of the test consistent with some model of collusion? Consider a collusive scheme in which cartel members, who are not designated to bid to win, submit phantom bids. Suppose the test doesn't fail. Firms could still be colluding as the cartel members can pass the test by appropriately scaling their "competitive bids". Issues Doesn't test for collusion. Omitted variables and misspecification are serious problems.

    9. Is There a Structural Break in Behavior? (Collusive) sources of structural breaks Formation of a cartel. Demise of a cartel. Approach: Finding a breakpoint Search without prior knowledge. Candidate breakpoints Events conducive to cartel formation. Events that make collusion more effective. Example: Trade associations Trade associations are used as a cover for cartel meetings and have been created for that express purpose. The Amino Acid Manufacturers International Association was formed by members of the lysine cartel.

    10. Test: Is there a break in the relationship among firms' prices around the time of the creation of the association? The Oklahoma Highway Department only started receiving identical bids at procurement auctions some time after the Asphalt Refiners Association was formed. Warning One might expect structural change even if firms are not colluding. The formation of an association could lead to enhanced correlation of firms' prices because it promotes the exchange of information which homogenizes firms' beliefs. But would it result in higher average prices? Important to consider the various implications of a trade association and identify those which are unique to collusion. Issues Doesn't test for collusion. There can be various sources of structural change.

    11. Does the Behavior of Suspected Colluding Firms Differ from that of Unsuspected Firms? Approach Estimate a model of (suspected) colluders' behavior. Estimate a model of competitive firms' behavior. Compare them. Common implementation Estimate reduced form price equations - regressing price on cost and demand shifters - for colluding firms and non-colluding firms. Test: Are the estimated coefficients statistically different? Reality check Do the non-colluding firms act in a manner consistent with a competitive model? Do the colluding firms act in a manner consistent with some model of collusion?

    12. Sources of competitive benchmark. Unsuspected firms in the same market. Firms in a comparable market where collusion is not suspected. Suspected firms during a time when they are not thought to have been colluding. Before cartel formation or after cartel collapse. Breakdown of collusion during the cartel regime.

    13. Procurement auctions for school milk (Porter and Zona, 1999) A market is a school district where each district awards an annual contract for the supply of school milk. Empirical model A reduced form model of a firm's bid level is estimated for all competitive firms plus a suspected firm. Independent variables: distance between district and plant, district enrollment, etc. Estimation 1: Slope coefficients are constrained to be the same for the competitive firms and the suspected firm. Estimation 2: Slope coefficients are allowed to differ for the suspected firm. Null hypothesis is that the coefficients are the same. Likelihood ratio test is used to evaluate the null hypothesis. The exercise is conducted for each of the suspected cartel members.

    14. Example Cost shifter: distance between the processing plant and the school district. A firms bid should be increasing in distance. Competitive firms' bids were found to be increasing in distance. Bids of all three suspected colluding firms were less sensitive to distance and two of them had their bids decreasing in distance.

    15. Collusive story Collusion will be effective only in those districts/markets for which non-colluding firms are neither numerous nor have a significant cost advantage (such as being the closest processors). Colluding firms may then be submitting higher bids in districts for which they have a distance advantage - so collusion works - and, in more distant markets, are forced by competition to submit lower bids (in spite of the higher transportation costs).

    16. Issues Finding a competitive benchmark. This may require prior information as to which firms may be colluding, in which markets there may be collusion, and over what time there may have been collusion. It is not useful with all-inclusive global cartels such as the vitamins cartel. Endogeneity of the competitive benchmark. Why aren't some firms suspected of colluding? Will the empirical analysis pick up all relevant differences? Why is there suspected collusion in one market and not in another comparable one? Suppose there is tacit collusion in the unsuspected market? The "competitive" benchmark is then not so competitive after all. Do explicitly colluding firms behave differently than tacitly colluding firms?

    17. Does a Collusive Model Fit the Data Better than a Competitive Model? Approach Specify a structural model of collusion and of competition in which firms' prices depend on cost and demand shifters. Estimate the model(s) and see which fits the data better. The competitive benchmark is estimated rather than identified ex ante. Wheat auctions in India (Banerji and Meenakshi, 2004) Industry structure: three large buyers (with a total market share of about 45%). Competitive model Three largest buyers draw valuations from different distributions than that of the remaining buyers (all who are assumed to have the same distribution). The winning bid is the second-order statistic over the valuations of three large buyers and the small buyers.

    18. Collusive model Bid rotation in which the three buyers randomly decide on the buyer to participate in a particular auction. The winning bid is the second-order statistic over the valuations of one large buyer and the small buyers. Collusive model fits the data better than the competitive model. Advantages One does not require prior information about who may or may not be colluding. Applicable even if the cartel is all-inclusive (all firms and all markets) and data is only available during the cartel regime.

    19. Disadvantage - misspecification bias. Usual concern with structural estimation. Misspecification is apt to be a more serious concern for the collusive model. There are many more collusive models than competitive models and, for each collusive model, there can be many equilibria. Milk cartels (Pesendorfer, 2000) In the Florida school milk cartel, firms used side payments. Market shares fluctuated over time because contracts could go to the most efficient firm with the other firms receiving transfers as compensation. In the Texas school milk cartel, there was no evidence of side payments and market shares were stable.

    20. Beating a Test for Collusion Can a cartel easily beat these tests? Some tests are easy to beat. Consider the test for independence of bids (Porter and Zona, 1993; Bajari and Ye, 2003). Firms' bids are independent under the competitive model and lack of independence is take as evidence consistent with collusion. This test can be beat by the cartel members appropriately scaling their "competitive bids".

    21. Some tests are costly to beat. Consider the finding that bids for school milk contracts are decreasing in distance (between the processor and the district). An unconstrained cartel finds it optimal to bid high in nearby collusive markets but bid low in more distant competitive markets. Cartel members could avoid failing this test by making their bids increasing in distance but it would lower profit. In choosing their bids, a smart cartel would trade-off cartel profit with the probability of detection. It may reduce the power of a test but not eliminate it entirely. Identifying a structural break. Collusion must mean a change in the price-generating process; in principle one should be able to pick up a break by monitoring the average price change. The cartel can reduce the power of this test by manipulating price changes but it foregoes profit in doing so

    22. Reasons that tests may have power even with smart cartels. Cartel foregoes profit from circumventing them. The need to maintain cartel stability may constrain them from appearing "competitive." In an auction, the cartel may need to cluster bids to avoid cheating (Marshall and Marx, 2004). The cartel may need to lower prices during times of strong demand (Rotemberg and Saloner, 1986). With imperfect monitoring, periodic reversion to lower prices may be required to maintain collusion but the resulting structural break could trigger detection (Green and Porter, 1984).

    23. A More Active Role for the Antitrust Authorities: Screening for Price-Fixing Screening of other forms of illegal activity Tax avoidance Insider trading in securities markets Credit card fraud Screening refers to a cost-effective method for identifying industries whose behavior is sufficiently suggestive of collusion so as to warrant verification. Criteria for systematic and ubiquitous screening. The screen should be based on easily available data. The screen should be routinizable so that it can be conducted with minimal human input. The screen should be costly for the cartel to beat.

    24. Data To be practical, screening must rely on easily available data which, in many cases, will mean price data. In some instances, quantity and some cost or demand shifters may also be easily available. Consider a product with a primary input which trades on commodity markets; for example, raw sugar used in the production of refined sugar. If cartel members manufacture in one country and sell in another - such as with the vitamins cartel - then exchange rate fluctuations are a cost shifter.

    25. Methods Search for collusive markers Price Lower price variance under collusion. Higher correlation in bids (at an auction) under collusion. Negative correlation in price and quantity (in response to demand fluctuations). Market Share Market share has more intertemporal structure when firms collude. Market share is more stable under collusion. Market share is negatively correlated over time under collusion.

    26. Price variance of frozen perch (Abrantes-Metz, Froeb, Geweke, and Taylor, 2005).

    27. Search for structural breaks in the stochastic process producing prices or some other measure of firm behavior. Example: average price - bid-ask spreads in Nasdaq (Christie and Schultz, 1999)

    28. What is a reasonable objective for an activist cartel detection policy? Trying to reject competition in favor of collusion is probably not feasible. Most collusive patterns can be made consistent with competition by appropriately choosing the demand and cost structure. Lack of theoretical knowledge and high quality data makes verification difficult. Identifying suspicious behavioral patterns is probably feasible. Analogy is with detection of credit card fraud. It involves looking for structural breaks and, more generally, statistical anomalies. Though we do not have as good data, we may have better theories.

    29. Coniurationes Caveant! (Cartels Beware)