www.ccp.uea.ac.uk Three Types of Inefficiency in Strategic Offers: Empirical Identification from Merger Remedy Settlements Luke Garrod, Bruce Lyons and Andrei Medvedev
Mar 28, 2015
www.ccp.uea.ac.uk
Three Types of Inefficiency in Strategic Offers: Empirical Identification from Merger Remedy
Settlements
Luke Garrod, Bruce Lyons and Andrei Medvedev
The institutions of EC remedy agreement provide a natural experiment to test theory of strategic offers
INTRODUCTION
If both parties are rational and have complete information, mutually beneficial agreement will…
• definitely be reached and• be reached immediately
Incomplete information can explain some delay…• screening for other’s type• signalling of own type
Evidence from…• experiments • labour bargaining
EC remedy agreements involve…• bargaining strategy from firms• more passive competition agency
The institutions of EC remedy agreement provide a natural experiment to test theory of strategic offers
INTRODUCTION
Two parties - Competition agency and merging firms
Discrete ‘rounds’ (2-phase investigation)• More information gathered in phase II• Allows early (phase I) or late agreement (phase II)… or no agreement
Legally specified… • Time limits to each phase (i.e. limited evidence gathering)• Order of who can make offers and who can accept/reject• Agency decision must be based on evidence
Three types of error…
• Type 1 – remedy too stringent• Type 2 – remedy insufficient to prevent market power• Type 3 – remedy agreed late
We present a theoretical model based on Lyons and Medvedev (2007) which allows us to predict which types of errors will occur and when
EMPIRICAL PREDICTIONS
Delay to Phase II more likely if:• Complex or imprecise merger appraisal (high σ1)• Delay is relatively less costly to the firms (K/π)
Model does not predict any effect of:• Obvious harm of the merger (αT)
Remedies in phase I:• Too stringent (Type 1 error) if issues are simple and/or delay is costly to firms
• Insufficient (Type 2 error) if issues are complex and/or delay is not costly
Remedies in phase II are:• Late (Type 3 error) if issues are simple and/or delay is costly to firms
We collected data from mergers with remedies agreed for horizontal aspects in phase I and phase II during the period 1999-2006
THE VARIABLES
variable exp sign description
ComplexityNumber of markets low ? with [x-35)% market share
uncertshare + with [35-45]% market sharehigh ? with 45+% market share
Coordinated effects collusion + =1 if coordinated effects considered; 0 otherwise
Transatlantic help EEAUS - =1 if 1 firm from US & 1 from EEA; 0 otherwiseUSonly ? =1 if firms from US; 0 otherwisenonUS base =1 if neither firm from US; 0 otherwise
ExperienceIndustrial indexp1990 - # mergers with 2-digit NACE classification beforeCoordinated effects coordexp1990 - # mergers with coordinated concern before
Cost of delayRemedy how much? concernprc + (# concern markets)/(# markets)
Does size matter?mean market shares s1s2 ? for concern markets
si ? increment for concern markets
phase I phase II
variable obs mean std dev min max obs mean std dev min max
markets 92 12.60 15.41 1 75.0 38 18.03 27.87 1 142.0high barriers % 92 30.80 35.20 0 100.0 38 57.40 39.10 0 100.0low barriers % 92 6.60 18.00 0 92.9 38 4.00 11.20 0 50.0
concern 92 5.44 7.83 1 38.0 38 10.97 20.08 1 113.0high barriers % 92 49.30 45.50 0 100.0 38 66.40 41.30 0 100.0low barriers % 92 0.00 0.00 0 0.0 38 0.00 0.00 0 0.0
Phase II documents tend to be more complete and detailed compared to those produced for phase I decisions. Without the correct attention, this could lead to bias.
POTENTIAL BIAS
The bias in the dataset can be highlighted by the entry barriers variable
Another bias: only cursory analysis for simple markets in phase II documents…
… filter used to eliminate markets not usually discussed in phase II documents
At low levels, the market share filter removes mostly markets that do not cause competition concerns
MARKETS REMOVED
If there is no bias we would expect the market share filter to removed a proportion of (1340/2084=) 0.643 phase I markets from the dataset
PROPORTION REMOVED
Sample size is n = 130… 30 other mergers are not included in the dataset due to lack of data or because they raised predominantly vertical issues
DESCRIPTIVE STATISTICS
Variable Obs Mean Std Dev Min Max
phase2 130 0.292 0.457 0 1 nonus 130 0.754 0.432 0 1 usonly 130 0.162 0.369 0 1 eeaus 130 0.085 0.279 0 1 combrev2006 84 31649 34009 2613 193941
markets 130 14.185 19.907 1 142 prodmark 130 6.485 8.098 1 54 geomark 130 2.419 2.617 1 17
concern 130 7.054 12.859 1 113 coord 130 0.300 1.111 0 9 noncoord 130 6.754 12.865 0 113
collusion 130 0.200 0.402 0 1 coord concern 130 0.115 0.321 0 1
Sample size is n = 130… 30 other mergers are not included in the dataset due to lack of data or because they raised predominantly vertical issues
DESCRIPTIVE STATISTICS
Variable Obs Mean Std Dev Min Max
s1s2 130 60.972 16.743 28.33 100 si 130 18.388 9.555 1 50 sr 122 18.623 11.537 0 60
ns1ns2 103 36.171 9.957 20 70 nsi 103 10.215 4.830 2.5 30 nsr 86 26.275 9.980 5 60
Probit Analysis shows what factors affect the likelihood that a merger will fail to be agreed in phase I
PROBIT ANALYSIS
phase2=1 exp sign
constant -1.2955 *low -0.0282uncertshare + 0.0694 *high 0.0189collusion + 0.8170 *coordexp1990 - -0.0805 *indexp1990 - -0.0568 ***concernprc + 0.0190 ***s1s2 0.0109si -0.0312 *usonly 0.9013eeaus - -1.0439 **
Pseudo R-squared 0.2764
coefficient
Probit Analysis shows what factors affect the likelihood that a merger will fail to be agreed in phase I
FURTHER SPECIFICATION
phase2=1 exp sign
constant -1.6622 **low -0.0395uncertshare + 0.1376 ***high -0.0007collusion + 0.7647coordexp1990 - -0.4770 **indexp1990 - -0.0866 ***coordconprc + 0.1241 ***noncoordprc + 0.0212 ***s1s2 0.0193lowsi 0.0111bigsi -0.0491 ***sdbigsi 0.0608 *usonly 1.2751 *eeaus - -1.5107 ***
Pseudo R-squared 0.3956
coefficient
Probit Analysis shows what factors affect the likelihood that a merger will fail to be agreed in phase I
MARGINAL EFFECTS
phase2=1 exp sign marginal effects
constant -1.3952 *low -0.0268uncertshare + 0.0801 ** 0.0235high 0.0030collusion† + 0.8438 * 0.2865coordexp1990 - -0.0803 * -0.0236indexp1990 - -0.0621 *** -0.0182concernprc + 0.0205 *** 0.0060s1s2 0.0103lowsi -0.0035bigsi -0.0384 ** -0.0113sdbigsi 0.0555 * 0.0163usonly 0.9022eeaus† - -1.1205 ** -0.2313
Pseudo R-squared 0.2960
coefficient
Our results are not dependent upon the level of the market share filter, as the results are relatively robust for a number of different filters (including 0%)
SENSITIVITY ANALYSIS
phase2=1 exp sign 15% 20% 25%
constant -1.1147 -1.2459 * -1.3952 * -1.3953 *low -0.0099 -0.0226 -0.0268 -0.0491uncertshare + 0.0655 * 0.0804 * 0.0801 ** 0.0860 **high 0.0061 0.0031 0.0030 0.0037collusion + 0.8441 * 0.9087 ** 0.8438 * 0.8761 **coordexp1990 - -0.0845 * -0.0888 * -0.0803 * -0.0772 *indexp1990 - -0.0630 *** -0.0636 *** -0.0621 *** -0.0611 ***concernprc + 0.0215 *** 0.0220 *** 0.0205 *** 0.0186 ***s1s2 0.0086 0.0086 0.0103 0.0109lowsi -0.0132 -0.0139 -0.0035 -0.0004bigsi -0.0412 ** -0.0396 ** -0.0384 ** -0.0389 **sdbigsi 0.0532 0.0553 * 0.0555 * 0.0584 *usonly 0.8196 0.8839 0.9022 0.8914eeaus - -1.1252 ** -1.1490 ** -1.1205 ** -1.0457 **
Pseudo R-squared 0.3093 0.3159 0.2960 0.2868
0%
The theory suggests Type 1 errors are likely occur when probability of phase II is low and Type 2 are likely to occur when probability of phase II is high
TYPE 1 AND 2 ERRORS
All phase II mergers are Type 3 errors but those with a low probability of phase II are likely to be explained by poor strategic play as opposed to complexity and delay cost
TYPE 3 ERRORS
We present a theoretical model based on Lyons and Medvedev (2007) which allows us to predict which types of errors will occur and when
CONCLUSIONS
Delay in reaching agreement arises when: • competition issues are complex and • delay is costly to the firms
Theory suggests:Remedies in phase I:
• Too stringent (Type 1 error) if issues are simple and/or delay is costly to firms• Insufficient (Type 2 error) if issues are complex and/or delay is not costly
Remedies in phase II are:• Late (Type 3 error) if issues are simple and/or delay is costly to firms
Empirics back up the theory and suggest:• Type 1 errors are more common than Type 2 errors• Type 3 errors also occur from poor strategic actions from players
Here are the 'top five' mergers for each error type that, according to our analysis, are most likely to be errors
TOP FIVES
Merger Predicted probability of phase 2
TYPE 1 M.3558 CYTEC / UCB - SURFACE SPECIALTIES 6.14E-07M.3593 APOLLO / BAKELITE 0.0000576M.3544 BAYER HEALTHCARE / ROCHE (OTC BUSINESS) 0.0002739M.2854 RAG / DEGUSSA 0.0016818M.1932 BASF / AMERICAN CYANAMID (AHP) 0.0031749
TYPE 2 M.1795 VODAFONE AIRTOUCH / MANNESMANN 0.7607729M.3770 LUFTHANSA / SWISS 0.6541686M.1571 NEW HOLLAND / CASE 0.6272017M.3235 TEIJIN / ZEON / JV 0.6139716M.3225 ALCAN / PECHINEY (II) 0.5641928
TYPE 3 M.3916 T-MOBILE AUSTRIA / TELE.RING 0.0301744M.2060 BOSCH / REXROTH 0.1079204M.1673 VEBA / VIAG 0.1168239M.2972 DSM / ROCHE VITAMINS 0.1559581M.1845 AOL / TIME WARNER 0.1822359