Dr. Mike Walker Vice President, CRA International, London mwalker@crai.co.uk

1. Merger simulation models: useful or just plain dangerousGlobal Competition Law Centre Seminar21 October 2005 Dr. Mike Walker Vice President, CRA International, London mwalker@crai.co.uk Professor of Economics, Loughborough University m.walker@lboro.ac.uk

2. Outline of presentation • Background to use of merger simulation models • Problems with merger simulation • Is there a role for such models in merger control? • Conclusions on merger simulation models October 2005

3. Why is Europe beginning to use merger simulation models?  • Traditional standard for judging a merger was whether it “created or enhanced a dominant position” • Dominance has been primarily a question of market shares • Post-merger share of more than 40% • Post-merger share of leading 3 firms more than 70% • This puts the emphasis on market definition • The new merger standard is whether a merger is likely to lead to a “significant impediment to effective competition” • Thought to be equivalent to US standard of “substantial lessening of competition”, which focuses on consumer harm • So tests that directly measure post-merger price increases are now more relevant than in the past October 2005

4. What is a merger simulation model? • Model of competition in an industry that allows the effect on prices of increased concentration to be measured directly • Typically based on Bertrand differentiated products model • Necessary inputs are pre-mergerelasticities, marginal costs, prices • Calibrate on current competitive outcome, then allow two or more firms to merge • Loss of inter-firm competition reflected in higher prices post-merger, subject of merger efficiencies October 2005

5. Volvo Scania Post-merger total Sweden 45 46 91 Finland 34 31 65 Denmark 29 30 59 Ireland 22 27 49 Norway 38 32 70 Example: Volvo/Scania merger • Competitive concern was in heavy trucks in Nordic countries Market Shares (%) Source: Para. 65 of the Commission’s Decision October 2005

6. Volvo/Scania Competitors Rigid Tractor Rigid Tractor Denmark 11.55 8.17 0.26 0.19 Finland 10.03 7.83 0.39 0.24 Ireland 10.87 7.36 0.21 0.30 Norway 13.17 8.63 0.32 0.28 Sweden 22.34 12.64 0.47 0.32 Example: Volvo/Scania merger (2) Using a nested logit model estimated the following post-merger price rises, based on list prices for 1997 and 1998 for 16 EEA countries October 2005

7. Problems with merger simulations – elasticity issues • Merger simulation requires • Estimate of current own-price and cross-price elasticities • Assumption about how elasticities vary as prices rise • Predicted post-merger price rises can vary greatly depending on elasticity assumptions • Crooke et al considered four alternative functional forms • Log-linear, linear, logit and AIDS • Using Monte Carlo analysis, found that on average predicted post-merger price rises were • Three times larger for log-linear vs. linear • Two times larger for AIDS vs. linear • 50% larger for logit vs.linear October 2005

8. -1.5 0.09 0.03 0.16 Firm 1 Firm 2 Firm 3 Firm 4 0.5 -1.33 0.06 0.23 Firm 1 0.61 0.22 -1.81 0.3 Firm 2 0.47 0.12 0.04 -1.33 Firm 3 Firm 4 Problems with merger simulations – elasticity issues (2) Relatively small inaccuracies in elasticity estimates can have significant effects on predicted post-merger price rises Example • 4 firms pre-merger • Firms 2 and 3 merging • Market shares of 63%, 16%, 5% and 15% respectively • Elasticities as follows: October 2005

9. Linear AIDS Log-linear Firm 1 0.1 0.9 0.0 Firm 2 1.6 12.5 12.9 Firm 3 4.3 17.1 28.2 Firm 4 0.2 1.5 0.0 Problems with merger simulations – elasticity issues (3) • Predicted post-merger price rises under three alternative functional forms are October 2005

10. ii Before After % Change Linear AIDS Linear AIDS Linear AIDS Firm 2 +10% 1.6 12.5 1.5 8.1 -6 -35 Firm 3 +10% 4.3 17.1 3.9 13.0 -9 -24 Firm 2 -10% 1.6 12.5 1.8 24.6 13 97 Firm 3 -10% 4.3 17.1 4.8 24.5 12 43 Problems with merger simulations – elasticity issues (4) • Predicted post-merger price rises as own-price elasticities vary by +/- 10% October 2005

11. Checking the facts fit the model (1) • If a model is going to be convincing to a regulatory authority, then the assumptions of the model must fit the facts of the industry • Are marginal cost estimates and elasticity estimates consistent with the underlying model of competition? • If not, is this because • Our estimates are poor? • The true underlying model of competition is not as assumed (i.e. how plausible is the assumption of a Nash equilibrium?) • Simulation models typically used in oligopoly situations where competition may well be “soft” October 2005

12. Checking the facts fit the model (2) • For instance, study used in Volvo/Scania predicted Lerner indices of 0.35–0.56, but industry experts argued the correct figure was 0.3 • Is the 0.3 an accounting fiction? • Do manufacturers “discount” trucks in order to make follow-on sales of spare parts? • Is it due to the use of list prices? • Can the model explain the past? • Shifts in market shares? • Concord Boat v. Brunswick Corp. October 2005

13. At best, only half the story • Common claim is that merger simulation allows one to dispense with market definition and competitive effects analysis and instead go straight to the “answer” • But merger simulations omit important factors • Barriers to entry and expansion • Buyer power • Potential for post-merger co-ordination • So still need to carry out the competitive effects analysis October 2005

14.  Retail data, manufacturer mergers • Price and quantity data is usually retail data, but the merger is often of manufacturers • Retail elasticities will in general differ from wholesale elasticities • W = Rpw • pwusually does not equal 1 • Wholesale elasticity usually less than the retail elasticity, but can be more • Froeb, Tschantz and Werden [2002] show that pass-through can be complete or non-existent, even in a simple game, depending on nature of contracts between retailers and manufacturers October 2005

15. Does non-price competition matter? • Merger simulation models usually omit non-price issues • But often carried out in branded goods industries, where we know non-price issues are important • e.g. promotions, competition for shelf-space • Simulation models focus on the internalisation of cannibalism • Is this the most important effect? • Empirical question October 2005

16. Does non-price competition matter? (cont) • The Marketing Professor’s view • “Bertrand is a posted price model. …You post your price. You change it if sales were not consistent with your expectations about demand. There’s nothing about trying to get business away from your competitors. There’s nothing about positioning your product differently, doing any of the sort of things which real world marketing is all about. I think those things are likely to be very important, but it’s the empirical issue. The issue is really, does internalisation of cannibalisation – which is all that drives Bertrand simulation models – dominate all the other stuff which is really very important?”(Scheffman, 2004) October 2005

17. The positive story? • Cost efficiencies • A rough cut • Part of the information matrix • Divestment analysis October 2005

18. Cost efficiencies • Can use a simulation model to take account of the effect of marginal cost efficiencies resulting from a merger • Still suffers from problems to do with elasticities • Can calculate compensating marginal cost efficiencies • Suffers to a lesser extent from elasticity issues • The efficiency defence is still weak in Europe • Professor Röller has downplayed their usefulness • Must be “merger specific” October 2005

19. Cost efficiencies (2) Variation in Own Price Elasticity Critical MC Reduction (%) Firm 2 Firm 3 Firm 2 Firm 3 0% 0% 13.0 19.1 10% 0% 9.2 17.3 0% 10% 11.8 15.5 10% 10% 8.4 14.1 -10% 0% 21.7 21.1 0% -10% 14.4 24.6 -10% -10% 24.1 27.3 October 2005

20. A rough cut • One defence of merger simulations is that they provide a useful first cut at the merger appraisal • Weak, given elasticity issues raised above • Provide an upper-bound using a log-linear approach • Only if assume no increase in co-ordinated behaviour • Any number is better than no number? • Not generally a respectable argument October 2005

21. Part of the information matrix • If the results of a merger simulation are consistent with the rest of the market analysis, then this provides support to the market analysis • Limited claim: how much value added does the simulation provide? • In the Philip Morris/Papastratos decision the Commission wrote • “The parties have provided the results of a merger simulation that shows that on average the market price increase post-merger would be minimal. The simulation model assumes that the merging parties’ products compete in different segments, or in other words, that the degree of substitutability between their products is low. The market investigation has confirmed the market segmentation. The results of the simulation confirm that the present merger would not lead to significant price increase in the Greek cigarette market.” (para. 32) • What did the simulation add? • Putting the results of the competitive effects analysis into a formal model can sharpen the rigour of analytical thinking October 2005

22. Divestment analysis • A good model of the industry can allow the effect of divestments to be simulated • In particular, can allow alternative divestment packages to be ranked in terms of their likely competitive effect • Regulatory authorities will still also talk to third parties, but the simulation can save time on proposals that are unlikely to be acceptable October 2005

23. Conclusions • Merger simulation models are dangerous unless handled with great care • At the moment, the grand claims often made for merger simulation models cannot be defended • They do not allow the investigator to avoid the competitive effects analysis • The precision they offer is more apparent than real • They omit competitively important factors • Merger simulation models currently have a much more limited role to play than usually claimed • Need to be bespoke models tailored to the particular industry facts • They may increase one’s confidence in the results of the competitive effects analysis • They may help to put efficiency claims into context • They can aid the divestment analysis October 2005