Quantitative Techniques for Competition and Antitrust Analysis_8 docx

7.1. Quantifying Damages of a Cartel 373 Q P, c S' S P Inelastic P Elastic Demand Elastic Demand Inelastic P 0 Figure 7.8. Pass-on with elastic and inelastic demand. For a formal demonstration, consider a price-taking firm, solving max q pq  C.qIc/; where C.qIc/ represents the total cost function and c represents a cost driver. From this problem we derive the firm supply function q D s.p; c/ and from that in turn, given N identical firms, we derive an industry supply curve S.pIc/, increasing in p and decreasing in c. We may now define a function F.pIc/ Á D.p/  S.pIc/ D 0; which implicitly defines the equilibrium price as a function of our cost driver, c. We can then apply the Implicit Function Theorem to get an expression for the pass-through @p=@c. Specifically, totally differentiating gives @D.p/ @p D @S.pIc/ @p C @S.pIc/ @c @c @p ; which in turn suggests that when the downstream market is perfectly competitive the pass-on effect can be expressed as @p @c D @S.pIc/ @c  @D.p/ @p  @S.pIc/ @p à D   @ ln S.pIc/ @c Ã  @ ln D.p/ @p C @ ln S.pIc/ @p à ; where the latter equality follows by noting first that D.p/ D S.pIc/, second that for any nonzero differentiable function f.p/we can write @ ln f.p/ @p D 1 f.p/ @f .p/ @p 374 7. Damage Estimation and thirdly by multiplying top and bottom by minus one. Finally, note that (i) the demand elasticity is negative while the supplyelasticityis positive so that the denom- inator will be positive and (ii) supply will decline as costs increase so that the numer- ator is also positive, making the ratio positive so that equilibrium prices increase with cost, @p=@c > 0. Furthermore, we conclude that the pass-on depends on both the demand and supply elasticities as well as on the cost elasticity of supply. Both elastic demand or elastic supply make the denominator large and hence reduce the pass-through down toward zero. Similarly, and intuitively, when the cost elasticity of supply is small so that costs tend not to impact on ability to supply the downstream good, the rate of pass-through will be small. Verboven and Van Dijk (2007) derive the analytical formulas for the pass-on rate under perfectly competitive markets and under markets with oligopolistic competi- tion. Furthermore, they evaluate the relative importance of the pass-on and output effects for a variety of settings. They note that the pass-on effect should be applied and the amount of the overcharge discounted by this effect when the claimant oper- ates in a fully competitive setting. But when the claimant’s industry—the down- stream industry—is less competitive, the output effect and the loss of sales volume by the claimant starts mitigating the effect of the pass-on on the claimants profits. The output effect should in such cases limit the discount in the damages granted by a pass-on defense. Their paper provides analytical expressions for the total discount to be applied to the overcharge of the cartel, taking into account both the pass-on and the output effects. Cournot competition in quantities the pricing function has the form P.Q/CP 0 .Q/q D mc; which under firm symmetry implies P.Q/CP 0 .Q/ Q N D mc or 1 C 1 NÁ.Q/ D mc P.Q/ ; where Á is the price elasticity of demand, Á.Q/ Á @ ln P.Q/ @Q D 1 P.Q/ @P .Q/ @Q : This equation defines implicitly our equilibrium output F.Q;mc/ Á 1 C 1 NÁ.Q/  mc P.Q/ D 0; 7.1. Quantifying Damages of a Cartel 375 which in turn means, given the demand equation, we can calculate the implied level of prices since the inverse demand curve describes p D P.Q/. First, recall that if p D P .Q.mc//, then @p @mc D @P .Q/ @Q @Q.mc/ @mc ; where the former is just a property of the inverse demand equation and the latter we can calculate by applying the implicit function theorem to the Cournot equilibrium equation: F.Q;mc/ D 0: Noting that 13 @F .Q; mc/ @Q Á 1 N.Á.Q// 2 @Á.Q/ @Q C mc P.Q/ 2 @P .Q/ @Q and @F .Q; mc/ @mc Á 1 P.Q/ ; we can apply the implicit function theorem by noting that @Q.mc/ @mc D  @F .Q; mc/ @mc à@F .Q; mc/ @Q à 1 and hence, given the results above, that @p @mc D @P .Q/ @Q @Q.mc/ @mc D @P .Q/ @Q  1 P.Q/ Ã  1 N.Á.Q// 2 @Á.Q/ @Q C mc .P .Q// 2 @P .Q/ @Q à : Rearranging gives @p @mc D @P .Q/ @Q @Q.mc/ @mc D @ ln P.Q/ @Q  1 N.Á.Q// @ ln Á.Q/ @Q  mc P.Q/ @ ln P.Q/ @Q à 1 () so that canceling terms gives @p @mc D  Q N @ ln Á.Q/ @Q  mc P.Q/ 1 Q à 1 ; where Á is the price elasticity of demand. Note that in the Cournot model the sensi- tivity of the price elasticity of demand to the output level affects the pass-through. The expression does not allow us to predict whether the pass-on under Cournot will be lower or higher than under perfect competition. 376 7. Damage Estimation 7.1.4 Timing the Cartel An area we have not yet considered is the timing of the cartel. We need to understand the time when the cartel was active since damages will accrue over that period. In fact, getting the periodsufficiently approximately correct maybe at least as important for the final damages number as pinning down exactly what the difference between collusive and competitive prices would have been in any given time period. In addition, most methodologies rely at least to some extent on pre-cartel or post- cartel data to extract information about the competitive scenario and it is therefore rather important that the data deemed to be the result of competition are in fact genuinely the result of competition, or something very close to it. Most commonly investigators use direct data from company executives to time the cartel: notes from diaries, records of meetings, emails referring to meetings or exchange of information, and memos describing pricing schemes. All these are the best sources for timing the cartel, as well as proving it existed in the first place. Because they are generally simple and less controversial pieces of evidence, it is by far the preferred source of information. Such information may be obtained from raiding company offices or executives’home addresses.Alternatively, it may emerge from the now widespread use of leniency programs, where leniency (particularly for second and subsequent leniency applications in a given case) can sometimes be conditional on providing evidence about the workings of a cartel. However, if there is not enough hard documentary evidence to time the cartel precisely, investigators may want to consider looking for a structural break in the data. The idea is to look for a change in the competition regime prevailing in the industry and the intuition is that we expect changes in conduct to be associated with otherwise unexplained changes in the levels of prices and/or quantities being sold. One way to do this is to specify dummy variables that allow for multiple possible starting and finishing dates. For instance, one might run the following regression: p t D x 0 t ˇ C˛ 1 D April 06 to May 06 t C ˛ 2 D June 06 to July 06 t C " t : This specification nests two timing options with two different starting dates. If ˛ 1 D 0 and ˛ 2 >0, then the starting date of the cartel is June 2006. If ˛ 1 D ˛ 2 >0, then the starting date of the cartel is April 2006. One can undertake a similar exercise for the end dates of the cartel but end dates are often trickier to pin down than start dates. Reversion to competition can be a gradual process and is not always marked by a discrete event such as a meeting among executives. Cartels often collapse little by little due to cheating, entry, a diversion of interests, or due to scrutiny by a competition authority. One may observe prices falling with several attempts to re-establish coordination having some limited success. Documenting and incorporating these data into the analysis may not be straightforward. 7.2. Quantifying Damages in Abuse of Dominant Position Cases 377 Additionally, there are reasons to think that the cartel may be replaced by a competition regime that is not necessarily genuine competition. The fact that a cartel had explicitly solved the problem of agreeing what it meant to be colluding meant that the first of Stigler’s conditions for tacit collusion may be satisfied, namely agreement (see the discussion in chapter 6). There are numerous indications that tacit collusion may be more likely after periods of explicit collusion and examples that are widely cited include those which followed the breakdown of the electrical cartels in the late 1950s. 14 Alternatively, firms in the previously cartelized industry which are being exposed to damage claims may sometimes have an incentive to price above the noncollusive level in the post-cartel period in order to minimize the size of their penalty (Harrington 2003). Finally, it is worth noting that the focus on claims made by downstream firms in the discussion of cartel damages reflects, in part, a legal reality, at least in Europe. The fact is that groups of final consumers often find it very difficult to coordinate together to generate a successful damages claim. Legal fees in a damages case can be substantial, even if a regulator has already put together a civil case establishing there was a cartel, while each consumer’s damage may be small. For example, in the football shirt case in the United Kingdom (JJB Sports) that consumer organization Which? took to the Competition Appeals Tribunal on behalf of consumers, each consumer was awarded £20 in damages from the company. However, since in the United Kingdom this kind of private action requires consumers to opt into the group of consumers that were represented byWhich?, only approximately1,000 consumers were expected to receive £20 each in compensation while almost one million shirts were estimated to have been affected by the cartel. 15 The possibility for a limited form of U.S style class-action suits, where groups of consumers would need to opt out of an action rather than opt into it, is under consideration in a number of European jurisdictions. 16 7.2 Quantifying Damages in Abuse of Dominant Position Cases Damages are mostly explicitly calculated for cartel infringements. However, monop- olization cases (or in EU language abuse of a dominant position cases) may also harm the process of competition and ultimately consumers. Because the tradition of 14 Specifically, the General Electric–Westinghouse case provides an example where it was subsequently alleged that tacit collusion replaced the explicit collusion of the late 1950s (see Porter 1980). 15 See, for example, “Thousands of football fans win ‘rip-off’ replica shirt refunds” (http://business. timesonline.co.uk/tol/business/law/article3159958.ece). The other aspect of the incentive to take such cases on behalf of consumers is the allocation of costs. If a case is won by a consumer organization, it can seek its costs; however, this “loser-pays” principle puts a considerable risk of a large downside on consumer organizations if the court decides that a claim for damages is without merit. 16 See www.oft.gov.uk/news/press/2007/63-07 for the United Kingdom and http://ec.europa.eu/comm/ competition/antitrust/actionsdamages/index.html for the European Commission’s consultation on private actions. 378 7. Damage Estimation private litigation is not yet fully developed in Europe, there are not many examples of calculated damages for individual misconduct unrelated to price fixing. This sec- tion only briefly introduces the topic and draws from Hall and Lazear (1994) and the Ashurst (2004) study for the European Commission. 7.2.1 Lost Profits Abuses of dominant position hurt consumers directly through exploitative abuses (high prices)but additional harm toconsumers often also occurs because competition has been impaired in some way. For example, rivals have been prevented from operating in the market either entirely or perhaps their scale of operation has been reduced. In either case, we will say they have suffered from an exclusionary abuse. Who, if anybody at all, is entitled to claim damages is a matter of law and differs by jurisdiction. The calculation of damages arising from an abuse of dominant position is a fairly uncommon activity for competition authorities, far rarer than damage calculations are for cartels. One reason may be that the damage inflicted by a dominant firm on customers and the extra profits generated by the abusive conduct can be very difficult to calculate whenever there is a significant element of exclusionary abuse. Indeed, there are few well-understood methodologies for evaluating the damage caused by exclusionary abuses, although a simulation model could be used in principle. By its very nature estimating what competition would have been like with additional firms active is a very difficult exercise. The quantification of the additional profits generated by the abuse, on the other hand, may be of interest if the authority wants to assess the incentives that firms face for engaging in abusive behavior of some kind. The methods presented here could also be used for such a purpose. When the injured party is a rival and not a customer, the damage calculation is even less straightforward. Typically, damages will be expressed as the additional profits that would have been obtained if the abuse had never taken place. The counterfactual is more difficult to establish than the effect on consumers since it will involve the performance of a particular firm if it had faced different conditions on the market. While our current generation of simulation models might be used to incorporate individual abusive conduct and to produce comparative static results of outcomes with andwithout the conduct, the data required toundertake suchan exercise robustly would quite possibly rarely be available. The design of a counterfactual and the quantification of the profit differential with and without the conduct is the most essential and also the trickiest part of such a damage estimation exercise. There are, however, other empirical issues that will also be relevant. For example, if plaintiffs can recover interest from their past losses, there will have to be a calculation of the present value of past damages. Similarly, future losses due to irreparable damage will have to be divided by a suitable discount rate in order to be expressed in net present value. The choice of the 7.2. Quantifying Damages in Abuse of Dominant Position Cases 379 interest rate and the discount factor theoretically appropriate will take into account the characteristics of the firm and the risk of the investment. While such general statements are widely acknowledged to be standard practice, they are not the same as stating the right number for any given context. Doing so with any confidence would require a substantial endeavor. Finally, the timing of injury may not coincide with timing of the infringement since injury can extend beyond the infringement and the claimant may not have been directly affected by the abuse since it took place. 7.2.2 Valuation of Lost Profits The quantification of lost profits due to an abuse of dominant position by another firm may well mostly involve using accounting data and accounting concepts to construct the profitability that would have occurred in the counterfactual world where no abuse took place. One approach is to base the damage calculations on the claimant company’s earnings: the damages will be the discounted estimated change in the cash flow. The cash flow is defined as the firm’s earnings actually received minus the costs actually incurred. The calculation of cash flow would exclude depreciation since the cost of depreciation is not actually paid. Assumptions must be made about how costs would have changed with different output and revenues. The calculation of “but for” cash flow will have to be carefully based on information about the company situation before the injury and its likely prospects on the market. The latter, in particular, means that a sufficiently deep knowledge of the firm and industry is required for such an exercise, and/or at least a willingness to make reasonable assumptions. A second approach to evaluating lost profits is to use a market-based approach. Damages could be estimated by calculating the loss of sales due to injury and multiplying that by the stock market valuation of a similar company as a multiple of its sales. If a similar company’s stock price implies a valuation of double the sales revenues, the damage to lost sales will be double. This approach eliminates the need to discount the loss in profits over time but the calculation of the loss in sales raises the same issues as the calculation of the “but for” cash-flow or the “but for” scenario in general.A related assets-based approach would calculate the damages as the change in the book value of assets before and after the infringement. Of course, for such an approach to be a sensible one, the analyst must be confident that the change in asset valuation is a consequence of the abuse and reflects the value of damage. Each of these techniques has advantages and disadvantages and they all raise the challenge of constructing a credible “but for” world. Case handlers may have to draw on the knowledge and industry expertise of an array of professionals such as indus- try experts, accountants, and strategy managers in order to construct a reasonable estimate of such damages. 380 7. Damage Estimation 7.3 Conclusions  Cartels increase prices and diminish output causing both a loss in total welfare and also a transfer of welfare from customers to producers. Profits go up and consumer surplus will generally go down under a cartel relative to a competitive market.  The total harm caused by a cartel to its customers consists of a direct effect on the customers who buy from the cartel in the form of an increase in prices and also an indirect effect due to the restriction in output on those customers who decide not to buy from a cartel given its high prices. If the cartel sells an input to downstream firms who then sell on to final consumers, damages to the downstream firm may be mitigated by the downstream firm’s ability to pass on the increase in its costs to final consumers.  In practice, cartel damages are often approximated by the direct damage or the total amount of the overcharge to the customers. This is the increase in price times the actual quantity sold during the cartel period.  Quantifying the damages will require estimating the price that would have prevailed absent the cartel. When market conditions do not vary greatly, this can be done by looking at historical time series and taking the price of the com- petitive periods as the benchmark competitive price during the cartel period. If market conditions do vary over time, one may nonetheless be able to use a regression framework to predict the “but for” prices during the cartel period. Structural simulations of the market are also possible but require reasonable assumptions on the nature of demand and the type of competition that would prevail absent the cartel.  Using the trend in the prices of a similar product to infer the price in the cartelized market is also possible, assuming such a benchmark is available. Applying a “reasonable” margin to the cost of the cartelized industry during the cartel can also provide a “but for” price when such “reasonable” margin can be inferred from the industry history or other benchmark markets.  Timing the cartel is a necessary part of damage estimation. It is best done using documentary evidence but evidence of unexplained structural breaks in the pricing patterns can sometimes also provide useful guidance.  The treatment of the pass-on effect in the calculation of damages depends on the legal framework. The extent of the pass-on will depend on the sensitivity of the firm’s supply function to the change in costs and also on the demand and supply elasticities that it faces. When the output effect is very large, so that a downstream firm’s profits suffer as they lose the margins that would have been earned on competitive volumes, the ability to pass on cost increases may not successfully mitigate the damage suffered by the downstream firm. 7.3. Conclusions 381  In addition to the difficulties in cartel cases, the exercise of quantifying dam- ages in cases of abuse of dominant position (attempted monopolization) is further complicated by the difficulty in defining the “but for” world. Dynamic and strategic elements which are difficult to incorporate might be particularly relevant in such settings. For example, suppose a claim for damages were made following the EU’s case against Microsoft for abuse of dominance. To evaluate the damages suffered by rival firms, we may need to take a view on the counterfactual evolution of the computer industry—by any measure a nontrivial task. 8 Merger Simulation Simulating markets in order to predict the unilateral effect of mergers on prices has seen considerable growth in popularity since the method was refined during the 1990s in a series of papers including the famous papers by Farrell and Shapiro (1990), Werden and Froeb (1993b), and Hausman et al. (1994). Such exercises, called merger simulations, are used for two purposes. First, they can serve as a screening device. In that case a standard model is usually taken as an admittedly very rough approximation to the world withthe expectation that the merger simulated with that model provides at least as good a screen as the use of market shares or concentration indices alone and hence is a complementary assessment tool to these simple methods. The second purpose of merger simulation involves building a more substantial model with the explicit aim of providing a realistic basis for a “best guess” prediction of the likely effects of a merger. Although merger simulation is now familiar to most antitrust economists and has been applied in a number of investigated cases, authorities remain cautious in the use of the results of these simulations as evidence. One important reason is that most authorities’ decisions are subject to review by judges and the courts have not universally embraced merger simulation as solid probative material. In turn, the reason for judicial concern is that merger simulation models are based on important structural assumptions regarding the nature of consumer demand, the nature of firm behavior, and the structure of costs. Evaluating whether a simulation model is likely to be accurate therefore implies determining the appropriateness of those assumptions. Unfortunately, there is usually considerable uncertainty regarding the price-setting mechanism in the market, the nature of demand, and the nature of costs. Yet a model builder must make explicit assumptions about each of these important elements of a merger simulation model. The alternative empirical approach is to try to use “natural experiments.” In some cases natural experiments will allow an empirical evaluation with fewer explicit assumptions. We discussed this important approach in detail in chapter 4. Such an approach is, however, not always either available or convincing. As a result, many investigations use a mixture of theoretical arguments, quantitative indicators, and qualitative descriptions of industry features to decide whether a merger will lead to a substantial lessening of competition (SLC) causing prices to rise. Such [...]... and margin behavior must be consistent with the reality of the industry It is therefore vital to take the time to refine and check the model sufficiently before proceeding to the merger forecasting exercise Methods to check the validity of simulation models include both the use of “in-sample” and “out-of-sample” predictions Consider, for example, a differentiated products Bertrand model On the one hand,... estimate suitable demand and cost models and embed those into a suitable framework describing both firms’ individual motivations and the nature of their interaction There is usually also a computation problem, since we must solve for best responses and then equilibrium For example, in a pricing game, we must solve for the firms’ pricing equations under the different models of oligopolistic competition being... simulations and it is therefore not surprising that in confrontational judicial or regulatory settings, arguments will often revolve around the adequacy of the demand specification There are several standard demand functions that are commonly used to describe consumer preferences and each of them will have implications for the prediction of the effect of a merger For an in-depth discussion of the main techniques, ... demand and cost sides of the model and then inputting them into the pricing/quantity/advertising equations, or one can attempt to estimate the demand, cost, and pricing/quantity/advertising equation parameters together The right approach may depend on the data and in particular on the reliability of the supply side of the model and the equilibrium assumption since these provide a great deal of information... about demand-side parameters, if the model is correct If it is not correct, then imposing the supply side of the model and the equilibrium concept being used will bias the estimates of the demand-side parameters This observation can also potentially form the basis of a Durbin–Wu–Hausman style test of the pricing equations For example, in the Bertrand model we can estimate the pricing and demand models... the demand system the marginal costs can potentially be retrieved from the equilibrium price equation That is, instead of solving the J first-order conditions for equilibrium prices given demand functions and marginal cost functions, instead we may solve the J first-order conditions for the J marginal costs, one for each product We do so by assuming that we know about the shape of the demand system and. .. equations as known values, and then solve for the J remaining unknowns, namely the marginal costs, mcj , j D 1; : : : ; J (For an algebraic description of this process using linear demand curves, see section 8.3.2.4.) Typically, we will therefore use the pricing equations in two ways First, using pre-merger data on prices and the model of demand we will solve the pricing equations for the J marginal costs... merger involves calculating the best response functions for both the pre-merger and post-merger scenarios and solving for the corresponding equilibrium prices and quantities In the Cournot model, if firms are symmetric in costs, the only difference between the pre- and post-merger scenarios will be the total number of firms operating in the market and so this is the variable that will need to be adjusted... curves directly from industry cost information if this is possible However, sometimes, given the pricing equations, the market prices and demand parameters, marginal costs can be inferred In those cases, the accuracy of the cost estimate will be hugely dependent on both the demand estimates and the model of competition being the “right” model It is vital then to perform appropriate reality checks to... equilibrium for models involving perfect information, and Bayesian Nash equilibrium for models involving imperfect information.12 11 Note that in a particular sense “behavioral economics” is misnamed All economics is behavioral but most neoclassical economics makes the behavioral assumption that consumers maximize utility and that firms maximize profits As an aside, it is striking that when talking to competition . 373 Q P, c S' S P Inelastic P Elastic Demand Elastic Demand Inelastic P 0 Figure 7 .8. Pass-on with elastic and inelastic demand. For a formal demonstration, consider a price-taking firm,. a claim for damages is without merit. 16 See www.oft.gov.uk/news/press/2007/63-07 for the United Kingdom and http://ec.europa.eu/comm/ competition/ antitrust/ actionsdamages/index.html for the. estimation may be possible and good practice for econometric and regression analysis applies. If there are insufficient data or indeed insufficient time available for estimation and the model is being