ECONOMICS COMMITTEE NEWSLETTER

Transcription

1 ECONOMICS COMMITTEE NEWSLETTER Contents Welcome... 2 Call for Articles... 3 Economics Committee Programming at the Section of Antitrust Law, Spring Meeting, Washington, DC, April 5-8, Recent Economics Committee Programming... 5 Counterintuitive Signs in Reduced Form Price Regressions... 7 The Theory of Contestable Markets and Its Legacy in Antitrust Practice...20 Antitrust Enforcement in Two-Sided Markets...29 Economics Committee Leadership...38 Contact Information...39 Volume 16, Number 1 1 winter 2016

2 Welcome It is our pleasure to welcome you to the winter 2016 edition of the Economics Committee Newsletter. The newsletter aims to provide a forum where Section of Antitrust Law and Economics Committee members can share their views on topics related to the relationship of antitrust law and economics worldwide. In this edition of the newsletter, we include three articles by economists and practitioners in the field. Yonghong An, Michael A. Williams and Wei Zhao consider price regressions and provide a checklist for practitioners in evaluating the applicability of interpretations of signs of estimated coefficients based on simultaneous equation, multicollinearity, measurement error and indirect effect of independent variables. José Alberro and Rainer Schwabe examine the theory of contestable markets and its legacy in antitrust practice. Juliette Caminade, Rebecca Kirk Fair, Federico Mantovanelli and David Toniatti analyze antitrust enforcement in two-sided markets. Whatever your background, these articles will provide valuable insights and perspectives. The newsletter is intended to provoke discussion. As a result, the opinions expressed are only those of the authors and not necessarily those of the American Bar Association, the Section of Antitrust Law, the Economics Committee or its subcommittees or any other individuals or entities. We hope that you enjoy the newsletter! Kind Regards Matthew Hall (McGuireWoods LLP) and Cani Fernández (Cuatrecasas Gonçalves Pereira SLP), Co-editors Volume 16, Number 1 2 winter 2016

3 Call for Articles We are always looking for articles for future issues of the Economics Committee Newsletter. If you have an article or an idea for an article regarding the current or potential future use of economics in analyzing issues of antitrust law worldwide, please share it with us. Contact Matthew Hall at or Cani Fernández for more information. at Volume 16, Number 1 3 winter 2016

4 Economics Committee Programming at the Section of Antitrust Law, Spring Meeting, Washington, DC, April 5-8, 2016 GUPPI-BASED HARBOR: SAFE OR PERILOUS? Wednesday, April 6, 2016, 1:45 3:15 pm Presented by the Economics, Federal Civil Enforcement and Mergers & Acquisitions In the recent Dollar store merger, the FTC commissioners disagreed on the proper application of a GUPPI based approach to analyzing mergers. How effective is the GUPPI as a tool/screen for predicting competitive effects and/or establishing safe harbor? How does the GUPPI measure up to an HHI-based safe harbor? CHAIR: Gary P. ZANFAGNA, Chief Antitrust Counsel, Honeywell International, Inc., New York, NY MODERATOR: Joanna TSAI, Charles River Associates, Washington, DC PANELISTS: Paul T. DENIS, Dechert LLP, Washington, DC Joseph FARRELL, University of California Berkeley, Berkeley, CA Henry SU, Attorney Advisor, Office of Chairwoman Edith Ramirez, Federal Trade Commission, Washington, DC Joshua D. WRIGHT, George Mason University School of Law, Arlington, VA FUNDAMENTALS - ECONOMICS Wednesday, April 6, 2016, 3:30 5:00 pm Presented by the Economics Committee Understanding basic economic principles is critical to antitrust analysis. Four economists will teach basic economic principles related to antitrust analysis, including market definition, market power, competitive effects and econometrics. The session should provide an introduction to concepts and techniques that appear repeatedly in the economic analysis of antitrust issues. CHAIR: Donald K. STOCKDALE JR, Bates White Economic Consulting, Washington, DC MODERATOR: Michael KHEYFETS, Edgeworth Economics, Washington, DC PANELISTS: Debra J. ARON, Navigant Economics, Chicago, IL Keith J. BRAND, Economist, Federal Trade Commission, Washington, DC Nicholas HILL, Assistant Section Chief, U.S. Department of Justice, Antitrust Division, Washington, DC Paola VALENTI, Columbia University, New York, NY Volume 16, Number 1 4 winter 2016

5 PRESENTING ECONOMIC EVIDENCE IN MERGER TRIALS Wednesday, April 6, 2016, 3:30 5:00 pm Presented by the Economics, Mergers & Acquisitions and Trial Practice Committees The outcome of merger trials can turn on the resolution of conflicting economic testimony and whether that testimony is intelligible and consistent with a recognized theory of harm and the record facts. The Sysco and H&R Block trials are two examples. What is a proper SSNIP test (with and without a data rich record)? What do price discrimination markets look like? A debate will ensue. CHAIR & MODERATOR: Ian T. SIMMONS, O Melveny & Myers LLP, Washington, DC PANELISTS: Timothy F. BRESNAHAN, Stanford University, Department of Economics, Palo Alto, CA Deborah L. FEINSTEIN, Director, Bureau of Competition, Federal Trade Commission, Washington, DC Mark A. ISRAEL, Compass Lexecon, Washington, DC Daniel M. WALL, Latham & Watkins LLP, San Francisco, CA PRICE WAR: RECONCILING CONFLICTING NATIONAL PRICING RESTRAINTS Thursday, April 7, 2016, 8:30 10:00 am Presented by the Economics, Distribution & Financing and Pricing Conduct Committees Economic laws do not stop at geographic borders, but the US, Canada, EU, Asia, and other areas vary in how they treat pricing restraints such as conditioned discounts, Internet sales and national account restrictions, MFN clauses, RPM, MAP, slotting allowances, and agency/consignment arrangements. The panel will examine how companies can develop effective, yet legal, global distribution policies. CHAIR & MODERATOR: Deena Jo SCHNEIDER, Schnader Harrison Segal & Lewis LLP, Philadelphia, PA PANELISTS: Joy K. FUYUNO, Director of Competition Law for Asia, Microsoft Corporation, Singapore Yong HUANG, University of International Business and Economics, Beijing Robert E. KWINTER, Blakes Cassels & Graydon LLP, Toronto Jorge PADILLA, Compass Lexecon, Madrid Recent Economics Committee Programming Transcripts of Economics Committee programmes are available on its website and on Section of Antitrust Law Connect. The most recent programmes are listed below: Volume 16, Number 1 5 winter 2016

6 A Session with Professor Massimo Motta, Chief Competition Economist, European Commission, DG Competition, January 11, 2016 Fundamentals of Antitrust Economics Series: Market Definition, December 14, 2015 Implications of Sixth Circuit s Collins Inkjet Corp. v. Eastman Kodak Co. Decision, December 10, 2015 Francine Lafontaine, Director of FTC s Bureau of Economics, Speaking on the Economics of Anticompetitive Practices, September 10, 2015 Nancy Rose, Deputy AAG for Economic Analysis, Speaking on Bargaining Leverage and Competitive Effects, June 25, 2015 Volume 16, Number 1 6 winter 2016

7 Counterintuitive Signs in Reduced Form Price Regressions I. Introduction Yonghong An 1, Michael A. Williams 2 and Wei Zhao 3 Reduced form price regressions are widely used in price-fixing cases for quantifying damages and in merger investigations for estimating likely price effects. In some cases, the estimated coefficients of one or more independent variables may have signs inconsistent with economic theory. For example, basic economic theory shows that an increase in demand, all else equal, leads to an increase in price. Similarly, an increase in cost, all else equal, leads to an increase in price. In a major price-fixing case, plaintiffs expert s reduced-form regression model showed that an increase in the values of variables measuring demand led to lower prices, and an increase in the value of a cost index also led to lower prices. The defense s expert concluded, on this basis, that the regression model was fundamentally flawed. Plaintiffs expert responded by arguing that the field of econometrics recognizes that the interaction of supply and demand effects in a reduced-form regression model can yield such counterintuitive signs even though the underlying supply and demand effects conform to economic theory. 4 We discuss how to interpret such counterintuitive signs and whether they cast doubt on the reduced-form model. II. Measuring Antitrust Damages and Estimating Merger Effects on Prices In assessing damages in price-fixing cases, a central economic question is the following: given that defendants colluded in the conspiracy period, how much have prices been elevated above what they would have been but for the collusion? To answer this question, we need to estimate but-for prices, i.e., the prices that buyers would have paid had there been no collusion, and compare those prices 1 Assistant Professor of Economics, Texas A&M University, y.an@tamu.edu. 2 Director, Competition Economics LLC, mwilliams@c-econ.com. 3 Consultant, Competition Economics LLC, wzhao@c-econ.com. 4 See, e.g., ABA Section of Antitrust Law, Proving Antitrust Damages (ABA Publishing, 2010); and Jonathan Baker and Daniel Rubinfeld, Empirical Methods in Antitrust Litigation: Review and Critique, American Law and Economics Review 1 (1999): Volume 16, Number 1 7 winter 2016

8 with the actual prices. In antitrust practice, the estimated but-for prices are generally obtained from a reduced form price regression. 5 In the standard framework, supply and demand are two separate functions that jointly determine price and quantity. As discussed below, in econometric terms these functions can be specified separately in two structural equations. Structural equations may contain both exogenous and endogenous variables. An exogenous variable is uncorrelated with the error term in the regression. This is important because the classical linear regression model assumes that the explanatory variables in a regression are uncorrelated with the error term. 6 An endogenous variable is correlated with the error term in the regression. Endogeneity gives rise to biased estimates of the regression parameters, as discussed below. One way of handling such endogeneity problems is to combine the structural equations into a single reduced form equation that contains only exogenous variables. Reduced-form regressions are often used in antitrust because they can yield unbiased estimates of but-for prices. A reduced form price regression is typically estimated using one of two approaches: the dummy variable approach or the forecasting approach. 7 Both approaches regress price on demand-side and supply-side control variables. The dummy variable approach utilizes observations both from the cartel period and a non-collusive, benchmark period and includes a dummy variable for the cartel period. The forecasting approach uses only observations from the benchmark period. For the dummy variable approach, the but-for price is obtained by setting the cartel dummy to zero for the cartel period and calculating the prices predicted by the regression estimates. For the forecasting approach, estimated coefficients are used to predict but-for prices in the cartel period. 5 See, e.g., Kai Hüschelrath, Kathrin Müller, and Tobias Veith, Concrete Shoes for Competition: the Effect of the German Cement Cartel on Market Price, Journal of Competition Law and Economics 9, no. 1 (2013): ; and Niels Frank and Phillip Schliffke, P. (2013), The Post-Cartel Equilibrium Puzzle in the German Cement Market: A Reply to Hüschelrath, Müller, and Veith, Journal of Competition Law and Economics 9, no. 2 (2013): See e.g., Peter Kennedy, A Guide to Econometrics, 6 th ed. (Oxford, UK: Wiley-Blackwell, 2008). 7 For a discussion of these two approaches, see Justin McCrary and Daniel Rubinfeld, Measuring Benchmark Damages in Antitrust Litigation, Journal of Econometric Methods, 3, no. 1 (2014): 63-67; see also Paul Godek, Time-Series Models for Estimating Economic Damages in Antitrust (and Other) Litigation: The Relative Merits of Predictive versus Dummy-Variable Approaches, CPI Antitrust Journal 1 (2011):1-7; and Halbert White, Robert Marshall, and Pauline Kennedy, The Measurement of Economic Damages in Antitrust Civil Litigation, Antitrust Section Economic Committee Newsletter (Spring 2006). Volume 16, Number 1 8 winter 2016

9 In merger cases, price regressions may be used in evaluating the effect of a merger on prices. 8 Prices are regressed on market structure variables affected by the merger as well as demand and supply control variables. For example, in predicting the price effects of a merger between two retail chains, each local market can be characterized by number and type of chains and the number of stores each chain owns. These market structure variables would change following the merger. In addition, local markets also differ in terms of demographic variables that affect local demand and supply conditions. Store-level prices can be regressed on these market structure variables and demographic variables. With the estimated coefficients, counterfactual prices can be calculated that compare prices with and without the merger. In both the damage analysis for price-fixing cases and merger analysis, a common criticism raised in litigation practice is that the estimated but-for prices are not reliable because one or more of the estimated coefficients in the reduced form price regression have signs inconsistent with economic theory. We discuss under what circumstances such a criticism is valid or invalid. III. Counterintuitive Signs and But-for Price Prediction A. Criteria for Accurate Predictions A primary criterion used to judge the accuracy of predictions from a regression model is whether the predicted values are biased. A predictor is unbiased if the expected prediction error equals zero. In a linear price regression, if the estimated coefficients are unbiased then the prediction based on those coefficients is unbiased as well. Any factors that affect only the standard error of an estimate but not its expectation do no harm to the prediction according to the unbiasedness criterion. A second criterion regarding the accuracy of a predictor is its variance, which measures the dispersion of predicted values. We can combine these two criteria, i.e., the bias and variance of the predicted value, by calculating the mean squared prediction error, which equals the variance of the predicted value plus the square of its bias. According to this criterion, we could have a poor prediction even though it is unbiased because the variance of the predicted value could be large. In a linear regression, the variance of predicted values is related to the 8 See, e.g., Orley Ashenfelter, David Ashmore, Jonathan Baker, Suzanne Gleason, and Daniel Hosken, Empirical Methods in Merger Analysis: Econometric Analysis of Pricing in FTC v. Staples, International Journal of the Economics of Business 13, no. 2 (2006): Volume 16, Number 1 9 winter 2016

10 variance of the estimated coefficients. 9 This suggests that we need to be careful with the prediction when the counterintuitive signs are caused by large variances of the estimated coefficients. Even with unbiased price prediction, the counterintuitive signs could be signals of large variances and low accuracy of predictions. We discuss several common interpretations of counterintuitive signs. For each interpretation, we address two questions. First, would that interpretation actually explain a counterintuitive sign, thus potentially justifying the reducedform regression despite the counterintuitive sign, and under what circumstances would such an interpretation hold? Second, would the same interpretation lead to the conclusion that the predicted prices from the reduced-form regression are biased? In evaluating these alternative interpretations, we produce a checklist for valid interpretations of counterintuitive signs. B. Simultaneous Equations Interpretation The expectation of the sign of a coefficient for a given control variable is based on the relationship between price and the control variable as predicted by economic theory. Demand and cost shifters enter the demand function and the supply function, respectively. Observed prices, which are equilibrium prices, are solved by equating demand and supply. For example, suppose (1) both the demand and supply functions are linear in price and (2) there are no common variables other than price in the two functions. We can then derive the empirical, structural form of the price regression model as follows: Demand function: (1) Supply function:, (2) where (or ) is the vector of demand (or cost) shifters. Equating (1) and (2) and rearranging terms gives: (3) 9 Even if the variance of an estimated coefficient is low, such that it is highly statistically significant, the value of the coefficient may not be economically significant. See, e.g., Stephen Ziliak and Deirdre McCloskey, The Cult of Statistical Significance: How the Standard Error Costs Us Jobs, Justice, and Lives (University of Michigan Press, 2007). Volume 16, Number 1 10 winter 2016

11 Economic theory provides predictions on the signs of coefficients in Equations (1) and (2), while the reduced-form regression model estimates the coefficients and in Equation (3) (the equation representing the market equilibrium). The left and right panels of Figure 1 show, respectively, the effects of an increase in demand and supply on price. Variables that increase demand in Equation (1) will increase price in the reduced form Equation (3). Conversely, variables that increase supply in Equation (2) will decrease price in the reduced form Equation (3). Intuitively, this means that the signs of coefficients of the demand shifters in the structural demand Equation (1) should have the same signs as the coefficients on in the reduced form Equation (3). Conversely, the signs of coefficients of the cost shifters in the structural supply Equation (2) should have the opposite signs as the coefficients on in the reduced form Equation (3). Figure 1: Price Effects of Increases in Demand and Supply From Equation (3), the signs of the coefficients of the demand control variables depend on both and ( ). In general, a higher price leads to decreased demand, so, and a higher price leads to increased supply, so. Together, these results imply that that (. This implies that the coefficients on in Equation (3),, should have the same sign as the coefficients on in Equation (1),. Thus, in this simple case, the sign of the coefficient in the reduced-form regression model is predictable (and testable) given the economic theory. Suppose that economic theory suggests that in Equation (1), i.e., demand increases in. As illustrated in the left panel of Figure 1, when increases in shift the demand to the right, the equilibrium price also increases, which corresponds to. Volume 16, Number 1 11 winter 2016

12 Similarly, from Equation (3), the signs of the coefficients of the supply control variables depend on both and ( ). As discussed above, (. This implies that the coefficients on in Equation (3),, should have the opposite sign as the coefficients on in Equation (2),. Thus, in this simple case, the sign of the coefficient in the reduced-form regression model is predictable (and testable) given the economic theory. Suppose that economic theory suggests that in Equation (2), i.e., supply increases in. As illustrated in the right panel of Figure 1, when increases in shift the supply to the right, the equilibrium price decreases, which corresponds to. When there is a variable that affects both demand and supply, 10 the sign of the coefficient on that variable in the reduced-form regression may be undetermined. Let denote the common variable(s) and rewrite the structural demand and supply functions as Demand function: (4) Supply function:, (5) Then the reduced form, equilibrium price equation is (6) Equation (6) shows that the sign of coefficient on the common variable is determined by. In some cases, economic theory cannot provide a prediction for the sign of. However, if the theory suggests that and should have opposite signs, or the theory predicts that and should have the same sign, but one coefficient should be much larger than the other, then we have a prediction of the sign of coefficient on. In theory, both the demand and the supply functions may not necessarily be linear or even approximately linear. For example, the demand function based on the nested logit discrete choice model is highly nonlinear. 11 However, 10 For example, if the product in question is wheat, then the price of barley affects both wheat demand and wheat supply. 11 See, e.g., Steven Berry, Estimating Discrete-Choice Models of Product Differentiation, RAND Journal of Economics 25, No. 2 (1994): ; see also Aviv Nevo, A Practitioner s Guide to Estimation of Volume 16, Number 1 12 winter 2016

13 nonlinearity does not change our argument above and might not create any distortions in signs of coefficients if (a) the demand and supply functions are well behaved 12 and (b) the relationship between the dependent variable and the independent variable is monotonic around the equilibrium. A well-behaved theoretical model, linear or nonlinear, should produce an equilibrium price function consistent with economic theory. IV. Problems in Estimation Counterintuitive signs in reduced-form regressions can be caused by several factors, including multicollinearity, measurement error, failing to consider indirect effects of independent variables on the dependent variable, as well as misspecification of the model. 13 Thus, counterintuitive results can be caused by a variety of sources. We analyze whether a counterintuitive sign can be explained from an econometric perspective, thus potentially justifying the reduced-form regression despite the counterintuitive sign. If such an econometric explanation does not exist, counterintuitive signs may indicate fundamental problems that render the empirical results from the reduced-form regression unreliable. A. Multicollinearity Interpretation Particularly in price regressions, multicollinearity can cause a discrepancy between the signs of the estimated coefficients and their theoretical predictions. Suppose that two demand shifters, say GDP and unemployment, are highly correlated. The problem is that the majority of the variation in GDP and unemployment is common, and common variation is not used in estimating the coefficients on GDP or unemployment. 14 Only variation specific to GDP (or unemployment), which is likely to be limited given the multicollinearity, is used in estimating the coefficient. This leads to two problems regarding the sign of coefficients. First, if only common variation captures the theoretical link between the regressors and the dependent variable, then this theoretical link is not reflected in the coefficients. For example, common variation in GDP and unemployment is supposed to capture variation in overall demand, and we expect Random-Coefficients Logit Models of Demand. Journal of Economics & Management Strategy 9, no. 4 (2000): In general, well-behaved demand and supply functions are respectively downward and upward sloping. See, e.g., Hal Varian, Intermediate Microeconomics: A Modern Approach (8th Edition), (W.W. Norton & Company, 2010). 13 For example, the reduced form price regression could be specified as linear, but the structural model could be nonlinear. In addition, the structural model may not be specified correctly relative to the underlying true model. 14 The common variation is still used in predicting the prices. Volume 16, Number 1 13 winter 2016

14 a higher price for higher overall demand. Since this common variation is not reflected in the coefficients, we might find a negative coefficient on either GDP or unemployment, or both. Second, since little variation is unique to each regressor, the coefficients would be estimated with large standard errors. Regarding the example of GDP and unemployment, the result is that the estimated coefficients can be negative due to large standard errors, while the underlying true coefficients are positive. In addition, if estimated coefficients of the multicollinear variables are statistically significant, in which case the multicollinearity problem is not severe, the second argument regarding insufficient unique variation is unlikely to be relevant. However, the first argument, i.e., that the theoretical link is only reflected by common variation, might still apply. Thus, even if the coefficients are statistically significant (and there is no measurement error or misspecification of the model), we still might find counterintuitive signs of coefficients. Thus, if control variables are multicollinear, a reduced-form regression may still be consistent with economic theory despite the presence of counterintuitive signs. Finally, note that this discussion is based on the unbiasedness criterion for price predictions. When the large variances in estimated coefficients of variables with multicollinearity lead to large variances in price prediction, the accuracy of the prediction is in question according to the mean squared prediction error criterion. B. Measurement Error Interpretation Two types of measurement error may appear when we estimate but-for prices. First, price (i.e., the dependent variable) may be measured with error. Namely,, where, are respectively the observed prices, the true prices, and the error. We assume the error is random and independent from the true prices. For the simplicity of exposition, we assume that the prices are only affected by a single regressor : (7). (8) The main consequence of the presence of measurement error in the dependent variable is that those errors inflate the standard errors of the Volume 16, Number 1 14 winter 2016

15 regression coefficient estimates. 15 Hence, the estimated coefficient may have the opposite sign of the underlying true value due to a large standard error. Thus, if some estimated coefficients contradict economic theory, measurement error in the dependent variable is one possible explanation. However, since measurement error in the dependent variable only produces counterintuitive signs through large standard errors, it is unlikely that the problem is caused by measurement error in prices if the counterintuitive coefficient is statistically significant. In other words, if the dependent variable is measured with error, a reduced-form regression may still be consistent with economic theory despite the presence of counterintuitive signs, but likely only if the coefficients of the variables with counterintuitive signs are statistically insignificant. The second type of measurement error is that an independent variable is mismeasured as. We assume that the error is independent from the true variable. Suppose another independent variable is, and the price regression is where Z has no measurement error. The consequence of measurement error in is that both and could be biased and incorrect signs may occur. Counterintuitive estimates can be due to measurement error as illustrated in this case. Unfortunately, the prediction in such a case would be biased because of the biased coefficient. In summary, if we are confident attributing the counterintuitive signs of coefficients to measurement error in an independent variable, then it is very likely the corresponding prediction is not reliable. C. Indirect Effect Interpretation 16 A counterintuitive sign on a coefficient might be pseudo counterintuitive if the researcher fails to understand the difference between the total effects and the ceteris paribus effect of that variable. By pseudo counterintuitive, we mean that the estimated sign is different from the expectation not because the estimation is in error, but because the researcher s expectation is in error. That is, the researcher has ignored indirect effects. For example, suppose price,, depends on (9) 15 If the measurement errors have nonzero means, they will also shift the intercept even when the true slope is zero. 16 Strictly speaking, the indirect effect also can be understood as a simultaneous equations problem. For the purpose of studying reduced form price regressions, we differentiate the analysis of indirect effects from the simultaneous equation system of demand and supply functions discussed in Section III.B. Volume 16, Number 1 15 winter 2016

16 control variables and, and affects both directly and indirectly through. Economic theory predicts that the total effect of on is positive. Then one might incorrectly argue that a negative estimated coefficient on is counterintuitive. But this conclusion would fail to take into account the fact that the coefficient on captures only the direct effect, which might have an opposite sign from the total effect. However, not all indirect effects would lead to unpredictable signs of reduced-form coefficients. Consider a price regression on a grocery store competition. For simplicity, ignore other variables and potential endogeneity issues and suppose the true model is that price depends linearly on the number of grocery stores in the market and population, and the number of grocery stores in the market depends linearly on population: P (10). (11) In addition, suppose the theory predicts that,, and the total effect of population on price is positive: P (12) Then if estimation of Equation (10) gives a negative estimate of, can we argue that it is reconcilable with the theory because of the indirect effect through the number of stores? The answer is no. To see this, we can substitute in Equation (11) into Equation (10) and rearrange the terms: P (13) Comparing Equation (12) with Equation (13), we have. Since,, and, we must have. Thus, an estimate of negative cannot be explained using the indirect effect argument. If there is no other good explanation as discussed in previous sections, then this suggests a modeling or data problem and cast doubts on the price prediction from the reduced-form regression. Volume 16, Number 1 16 winter 2016

17 V. Summary: A Checklist for Practitioners Problems in Estimation To sum up our findings, we provide a checklist for antitrust practitioners. When a counterintuitive sign is found in a price regression, one can go through this list for all valid interpretations that preserve the credibility of the price prediction. If none of the interpretations on our list apply, then the counterintuitive sign is an indicator of modeling or data problems that invalidate the price predictions from the reduced-form regression. When an estimated coefficient has a counterintuitive sign, the first thing to check is if it is statistically significant. A coefficient might be statistically insignificant because of weak dependence, not enough variation in the variable, heteroskedasticity, as well as other factors. In such cases, the estimated coefficient may be positive when the expectation is negative or vice versa due to large standard errors. However, large standard error alone will not bias the predicted price. When the variable in question has ambiguous or limited economic influence on prices, an insignificant coefficient with a counterintuitive sign generally raises less concern. Commonly used explanations for counterintuitive signs of estimated coefficients include simultaneous equations, multicollinearity, measurement error, and indirect effects of independent variables. Table 1 summarizes the applicability and implications of these different potential interpretations. 17 Lastly, whenever the check is based on statistical significance of the coefficient in question, the economist should make sure that the calculated standard errors account for data issues such as heteroskedasticity and autocorrelation. For example, if heteroskedasticity is present but ignored, then the calculated standard error is too small. The estimated regression may lead one to conclude that an estimated coefficient is statistically significant when it is not and rule out potential explanations of counterintuitive signs such as measurement error in the dependent variable. 17 Note that omitted variable bias may also cause counterintuitive signs. However, omitted variable bias generally leads to biased predictions and, therefore, cannot be used in justifying counterintuitive signs of estimated coefficients. Spurious correlation also can contribute to counterintuitive signs. Nevertheless, in practice causality between control variables is clear in most cases, and it is less likely that correlation is misinterpreted as causality. Thus, the concern of spurious correlation is less severe. Volume 16, Number 1 17 winter 2016

18 TABLE 1: APPLICABILITY OF COMMONLY USED INTERPRETATIONS FOR COUNTERINTUITIVE SIGNS Possible Interpretations Simultaneous equations with no common controls in demand and supply functions Simultaneous equations with common controls in demand and supply functions Multicollinearity Measurement error in the dependent variable Measurement error in an independent variable Indirect Effect Applicability and Implication Signs are predictable by economic theory. Counterintuitive and statistically significant coefficients suggest potential modeling or data problems. Signs are not predictable if, in theory, the common variable affects supply and demand in the same direction. Estimated coefficients likely to be statistically insignificant. However, if the theoretical link between price and the multicollinear variables is only reflected by common variation, then signs are not predictable. Counterintuitive and statistically significant coefficients suggest potential modeling or data problems. May be the reason for counterintuitive and statistically significant coefficients. In addition, the prediction in such a case would be biased. The economist needs to understand whether the theoretical prediction relates to the total effect or the direct effect of an independent variable. Whether indirect effects would lead to unpredictable signs is case-specific and can be studied, as for example, in Section IV.C. Volume 16, Number 1 18 winter 2016

19 VI. Conclusions Coefficients that are statistically and economically significant, and that have economically counterintuitive signs (particularly on variables that affect only demand or only supply but not both) generally indicate substantial problems that call into question the reliability of the underlying model. When the price regression model is linear, or at least well-behaved, the counterintuitive signs of coefficients are not likely explainable by the simultaneous equations argument. For price prediction to be valid, measurement errors in the dependent variable can be used to explain a counterintuitive sign only if the corresponding coefficient is statistically insignificant. Measurement errors in control variables may be the cause of a counterintuitive sign, but such measurement errors also would lead to biased price predictions. When there is no multicollinearity, the counterintuitive signs could be an indicator of model misspecification or poor quality of data, which cast doubt on predicted, but-for prices based on the reduced-form regression model. However, with multicollinearity, counterintuitive signs may arise even when the reduced-form model is valid and, thus, yields unbiased predicted but-for prices. In this case, the researcher as a check could drop one or more of the collinear control variables to ameliorate the multicollinearity issue and see if the counterintuitive sign(s) persist. If so, the reduced-form regression may well be misspecified, casting doubt on the resulting predicted but-for prices. Lastly, counterintuitive signs may result from a misunderstanding of the difference between the total effect and the ceteris paribus effect. The researcher should study carefully the direct and indirect effects predicted by economic theory when relying on this argument for counterintuitive signs of reduced-form coefficients. Volume 16, Number 1 19 winter 2016

20 The Theory of Contestable Markets and Its Legacy in Antitrust Practice 1 José Alberro 2 and Rainer Schwabe 3 Lawyers and economists working in antitrust are likely to encounter the Theory of Contestable Markets (henceforth TCM ) at some point in their professional lives. In this article we provide a brief, non-technical explanation of TCM and we chronicle its rise to prominence, its decline in popularity within mainstream Economics, and its enduring influence on the practice of antitrust. I. What Is a Contestable Market? In 1982, industrial organization economists William Baumol, John Panzar, and Robert Willig published a book titled Contestable Markets and the Theory of Industry Structure. 4 They proposed an innovative framework that sought to generalize the concept of perfect competition, 5 describing market entry and exit conditions that would induce socially optimal behavior even in oligopolistic or monopolistic markets. This stark separation between market structure (i.e., concentration) and market outcomes (i.e., prices and quantities sold) was a challenge to the Structure-Conduct-Performance paradigm, which had dominated the profession since the seminal work of Joe Bain in the 1950s. 6 As such, Baumol and his coauthors proclaimed their theory to be an uprising in the theory of industry structure and a new theory of industrial organization. 7 Much of the Economics profession and many practitioners seemed to agree with this assessment. As noted by William Shepherd just two years after the publication of 1 The views expressed in this article are solely those of the authors, who are responsible for the content, and do not necessarily reflect the views of Cornerstone Research. 2 Senior Advisor, Cornerstone Research. 3 Associate, Cornerstone Research. 4 William Baumol, John Panzar, and Robert Willig, Contestable Markets and the Theory of Industry Structure (San Diego, CA: Harcourt Brace Jovanovich, Revised Edition, 1988). We cite the 1988 revised edition of the book; the book was originally published in The book developed a unified framework for TCM, building on articles published by the authors over the preceding dozen years. These articles are listed in the book s Copyrights and Acknowledgments section, Baumol et al., at William Baumol, Contestable Markets: An Uprising in the Theory of Industry Structure, American Economic Review 72, no. 1 (1982): 1 15, at 2. 6 Joe Bain, A Note on Pricing in Monopoly and Oligopoly, American Economic Review 39, no. 2 (1949): ; Joe Bain, Workable Competition in Oligopoly: Theoretical Considerations and Some Empirical Evidence, American Economic Review 40, no. 2 (1950): 35 47; Joe Bain, Relation of Profit Rate to Industry Concentration: American Manufacturing, , Quarterly Journal of Economics 65, no. 3 (1951): ; Joe Bain, Industrial Organization (New York, NY: John Wiley & Sons Inc., 1968). 7 Baumol, supra note 5; Baumol et al., supra note 4, at vi. Volume 16, Number 1 20 winter 2016

21 the Baumol et al. book, the topic [was] being installed not only in the literature, but also in graduate curricula and in policy debates about antitrust and regulation. 8 According to Baumol et al., a contestable market is characterized by the following conditions: 9 1. Free Entry: All firms, including those that do not currently produce in the market, have access to the same production and distribution technologies. Furthermore, there are no regulatory barriers to entry. All firms (new or existing) that wish to offer products to the market are on a level playing field with respect to other firms in the market in terms of consumer perception there are no branding advantages. In summary, there are no barriers to entry as George Stigler defined them: A cost of producing which must be borne by a firm which seeks to enter an industry but is not borne by firms already in the industry Free Exit: There are no impediments to firms leaving the market. When a firm exits the market, it can recover the depreciated value of its market-specific investments; i.e., there are no sunk costs. 3. Sticky Short-Term Prices: Incumbent firms cannot adjust their prices immediately. This may be due to contract provisions or other practical impediments. 4. Price-Sensitive Consumers: Consumers immediately react to differences in price, purchasing the least costly alternative. This corresponds to the consumer behavior assumed in the widely used Bertrand-Nash model of oligopolistic behavior. Together, these assumptions imply that hit-and-run entry, a key implication of TCM, is possible. Hit-and-run entry takes place when a firm enters the market, sells its products at a price below the prevailing price in that market but above the competitive level and then exits before incumbent firms can react. 8 See William Shepherd, Contestability vs. Competition, American Economic Review 74, no. 4 (1984): , at The description is based on Baumol, supra note 5, at 3, and Avinash Dixit, Recent Developments in Oligopoly Theory, American Economic Review 72, no. 2 (1982): 12 17, at 15. An alternative treatment is provided by Jean Tirole, The Theory of Industrial Organization (Boston, MA: MIT Press, 1988), at George Stigler, The Organization of Industry (Chicago, IL: University of Chicago Press, 1983), at 3. Volume 16, Number 1 21 winter 2016

22 When hit-and-run entry is possible, incumbent firms will not price their products above the competitive level; if they do, a hit-and-run entrant will take advantage of the situation by capturing all of the sales in the market, leaving incumbent firms empty-handed. A. Airline Markets as the Prototypical Contestable Markets Proponents of TCM pointed to airline routes between city-pairs as real world examples of contestable markets. Baumol et al. provide a characteristically persuasive explanation: Consider two towns between which the demand for travel is only sufficient to support one flight a day. This is a natural monopoly market. And yet, because airline equipment (virtually capital on wings ) is so very freely mobile, entry into the market can be fully reversible. In principle, faced with a profitable opportunity in such a market, an entrant need merely fly his airplane into the airport, undercut the incumbent s price, and fly the route profitably. Then, should the incumbent respond with a sufficient price reduction, the entrepreneur need only fly his airplane away to take advantage of some other lucrative option even if he only returns his rented aircraft or resells it in the well-functioning secondary aircraft market. Thus, it is highly plausible that air travel provides real examples of contestable markets. 11 Subsequent research, however, suggested that airline routes were not perfectly contestable. As Baumol and Willig note in their 1986 retrospective on TCM, post-deregulation experience in the airlines industry has revealed several elements of the structure of supply that conflict significantly with the conditions necessary for the pure theory of contestability to apply without modification. 12 Several econometric studies found a statistical relationship between profits and market structure in airline markets, contradicting the predictions of TCM. Moreover, when entry occurred, established firms reduced their fares in response, violating the sticky price condition described above. 13 Other structural factors such as constraints in the supply of airport facilities and particular airplane 11 Baumol et al., supra note 4, at William Baumol and Robert Willig, Contestability: Developments since the Book, Oxford Economic Papers 38 (1986): 9 36, at Indeed, airline markets are sometimes cited as examples of predatory pricing, where incumbents lower prices below cost in order to drive entrants out of the market. See, for instance, the description of Northwest s response to Spirit Airlines entry to the Detroit to Philadelphia market in Luis Cabral, Introduction to Industrial Organization (Cambridge, MA: MIT Press, 2000), at 275. Volume 16, Number 1 22 winter 2016

23 models, as well as union-negotiated wage contracts, further departed from the basic assumptions of TCM. However, researchers found that potential competition constrains price in these markets, providing support for TCM s notion that potential competition can be a competitive constraint on market participants. 14 B. Criticisms of Contestable Markets Theory Criticisms of TCM followed close on the heels of Baumol et al. s book. 15 Some economists faulted the authors of TCM for overstating their contribution and underemphasizing the influence of the work of Bain, Stigler, and others on barriers to entry, as well as other contemporary theories of imperfect competition such as limit pricing. 16 More relevant, however, were claims that TCM s assumptions lack generality and, therefore, that the theory was empirically implausible. One powerful criticism of TCM centered on the apparent disconnect between the model s assumptions on the flexibility of prices relative to capacity. For hit-and-run entry to be feasible, a firm must be able to bring new products to the market before incumbent firms can adjust their prices in response. However, as Avinash Dixit noted, [t]he traditional presumption in industrial organization is the opposite, that is, that prices can be changed more quickly than sunk capacity. 17 Ultimately, these arguments together with the failure to establish a consistent body of empirical evidence to support TCM, as noted above for the case of airline markets, resulted in most academic economists putting TCM aside. Church and Ware s widely cited textbook on Industrial Organization sums up current attitudes toward TCM, stating that [t]he theory of contestability has 14 Steven A. Morrison and Clifford Winston Empirical Implications and Tests of the Contestability Hypothesis, Journal of Law and Economics 30, no. 1 (1987): See Martin L. Weitzman, Contestable Markets: An Uprising in the Theory of Industry Structure: Comment, American Economic Review 73, no. 3 (1983): ; Marius Schwartz and Robert Reynolds, Contestable Markets: An Uprising in the Theory of Industry Structure: Comment, American Economic Review 73, no. 3 (1983): ; Shepherd, supra note 8; Michael Spence, Contestable Markets and the Theory of Industry Structure: A Review Article, Journal of Economic Literature 21, no. 3 (1983): See Shepherd, supra note 8, at 572, 574, noting that TCM was offered with unusual self-praise, and that [i]n 1956, Joe Bain added the concept of entry barriers, as an element that could modify internal market power. The logical possibility that free entry could neutralize internal market power was known from the outset, but it was set aside as an extreme case. See also Dixit, supra note Dixit, supra note 9, at 16. Volume 16, Number 1 23 winter 2016

24 provoked considerable controversy over its logical possibility, robustness, and empirical relevance. 18 II. Contestable Markets Theory through the Lens of the DOJ/FTC Merger Guidelines The influence of TCM on antitrust practice over the years can be traced through the lens of the DOJ/FTC Merger Guidelines. In particular, the concept of potential entry has played a significant role in the Merger Guidelines since The 1982 and 1984 Merger Guidelines introduced criteria consistent, but perhaps not directly inspired by, TCM; the 1992 Guidelines reflected the main concepts of TCM; and the 2010 Guidelines reflect the profession s turn away from TCM. The 1982 Guidelines gave entry a central role in merger enforcement for the first time, stating that [i]f entry into a market is so easy that existing competitors could not succeed in raising price for any significant period of time, the Department is unlikely to challenge mergers in that market. 20 In effect, potential entrants whose participation in the market was judged to be sufficiently likely and timely following a 5 percent increase in price were considered to be part of the market and assigned market shares. 21 While this view owes at least as much to the work of Bain, Stigler, and others on the importance of barriers to entry as it does to TCM, contemporary commentators noted its congruence with TCM, and the Guidelines were often interpreted in light of TCM. 22 The Jeffrey Church and Roger Ware, Industrial Organization: A Strategic Approach (New York, NY: McGraw-Hill, 2000), at 509. See also Rhys Evenden and Alan Williams, Contestability: The Debate and Industry Policy, Economic Analysis and Policy 30, no. 1 (2000): 75 90, at 76 ( Contestability theory no longer holds widespread support amongst academic economists in the field of microeconomic policy because the assumptions have come to be regarded as implausible as a matter of logic or empirical evidence. ). 19 For an account of the doctrine of potential competition through the years, see Darren Bush and Salvatore Massa, Rethinking the Potential Competition Doctrine, Wisconsin Law Review (2004): U.S. Department of Justice, Merger Guidelines (1982), at 15, available at 21 This was the expression of the concept of supply-side or production substitution in market definition. 22 Eleanor M. Fox, The New Merger Guidelines a Blueprint for Microeconomic Analysis, Antitrust Bulletin 27 (1982): , at 541 ( The categories of Ease of Entry and Other Factors are very important.... Even if the market is dominated by a single firm, it is possible that entry conditions make the leading firm s market contestable by firms that can enter and leave without incurring unrecoupable costs (e.g., airlines in an unregulated environment, seeking new routes). ); Janusz Ordover and Robert Willig, The 1982 Department of Justice Merger Guidelines: An Economic Assessment, California Law Review 71, no. 2 (1983): , at 563 ( This view squares with a generally accepted notion that in a market unprotected by entry barriers, incumbent firms cannot exercise market power. In such markets, which we term contestable markets, the threat of entry is an absolute constraint on incumbents market power. ); Richard Schmalensee, Horizontal Merger Policy: Problems and Changes, Journal of Economic Perspectives 1, no. 2 (1987): 41 54, at 51 ( The position of the Guidelines on the extreme case of contestable markets is theoretically correct: If entry into a market is so easy Volume 16, Number 1 24 winter 2016