Skip to main content
SpringerLink
Log in

Fifty is the New Forty: EU Merger Policy Permits Higher Market Shares After the 2004 Reform

  • Published:
Review of Industrial Organization Aims and scope Submit manuscript

Abstract

I analyze empirically all of the European Commission’s decisions regarding “unilateral effects” aspects of horizontal mergers before and after the 2004 reform, which introduced the “significant impediment to effective competition” test in merger policy. I find that, after the reform, the Commission did not change its stance toward mergers to monopoly or quasi-monopoly (almost always challenged) and mergers in un-concentrated markets (almost never). The new test produced more frequent challenges when the combined entity is not the largest firm, but these cases remain rare. The Commission’s stance toward mergers that fall between these polar opposites appears to have been tougher pre-reform ceteris paribus.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Log in via an institution

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Institutional subscriptions

Fig. 1

Similar content being viewed by others

Notes

  1. EC 139/2004 has applied since May 1, 2004, when it replaced EC 4064/89 (enacted on Sep. 1, 1990).

  2. Guidelines on the assessment of horizontal mergers under the Council Regulation on the control of concentrations between undertakings (2004/C 31/03), Official Journal of the EC (OJ), Feb. 5, 2004.

  3. See: European Commission Press Release IP/03/1027 (Brussels, July 16 2003).

  4. EC 139/2004, Recital 29, and 2004 HMG, Section VII.

  5. See Vickers (2004).

  6. I do not study instances when the EC had ‘collective dominance’ concerns: The merger could enable collusion between the merging parties and certain remaining rivals.

  7. See EC 139/2004, Article 1.

  8. The EC may conditionally clear the concentration at the end of Phase I by accepting the commitments the merging firms offered early on, without the need for an in-depth market investigation.

  9. See: http://ec.europa.eu/competition/mergers/cases/.

  10. An antitrust market consists of “products and/or services which are regarded as interchangeable or substitutable by the consumer, by reason of the products’ characteristics, their prices and their intended use” and comprises the area “in which the conditions of competition are sufficiently homogeneous and which can be distinguished from neighboring geographic areas,” see: Commission Regulation (EC) No 802/2004 of 21 April 2004 implementing EC 139/2004 (OJ L 133, 30.4.2004, p. 1).

  11. In rare cases, the EC identifies relevant markets where there is neither horizontal nor vertical overlap; see the discussion of ‘conglomerate’ markets below (Sect. 3.4).

  12. Case M.1672—Volvo/Scania is one of these relatively rarer cases.

  13. See: Market Share Ranges in Non-Confidential Versions of Merger Decisions, EC-Directorate-General for Competition (December 11, 2008).

  14. I do not cover papers that focus on normative issues; e.g., Duso et al. (2007, 2013) use stock-market data to evaluate EC merger policy from a welfare standpoint without relying on EC-reported market shares.

  15. Phase II decisions for the EC (166 observations) and ‘second request’ cases for the FTC (109).

  16. The paper explains that observed differences between the EC and the FTC can be due to policy effects (one agency being ‘tougher’ than the other) and case-mix effects (one agency facing more problematic cases). Regression results are used to decompose observed differences between policy and case-mix effects.

  17. The dataset covers all published decisions, in any language. Decisions that were taken under the ‘simplified procedure’ that was introduced in 2000 to deal with mergers that do not raise concerns because of no/very limited overlap between the merging firms’ activities are excluded because they do not contain any information on the markets where the limited overlaps (if any) occur.

  18. On December 14, 2013, the EC raised the threshold from 15 to 20%: Now horizontal mergers falling short of a combined 20% can qualify for a simplified procedure.

  19. For instance, conglomerate issues arise when the merging parties produce different goods (e.g., different kinds of alcoholic beverages, or different types of household appliances) that belong to separate antitrust markets, but the EC may be concerned about the merged entity strengthening its product portfolio.

  20. I exclude observations when no overlap occurs because the issues in potential competition or portfolio/conglomerate effects cases are different, and hence their inclusion could be inappropriate. I am indebted to Dr. Coate for pointing this fact to me. Including these observations would not affect the empirical estimates since there are very few of them as compared to the 13,818 observations in my dataset.

  21. For instance, the EC may leave open whether the geographic market for titanium dioxide is EU-wide or world-wide. In this case, the dataset contains both alternative market definitions.

  22. The log transformation fits the data better than regressions in levels. Most previous studies use logs of concentration variables too, thus enabling comparisons across studies.

  23. Hsiao (1983) shows that regressing a continuous dependent variable on ranges’ midpoints produces consistent estimates if the unobserved explanatory variable is uniformly distributed over the range.

  24. This assumption is consistent with how the EC defines the range (see footnote 13).

  25. Sometimes the EC provides useful additional information; e.g., it may report that two firms each having [10–20%] market share merge, and the delta HHI falls between 250 and 350. This implies that the midpoints of the ranges (15%) cannot be the expected shares: 2 × 15 × 15 = 450, i.e., outside the 250–350 range.

  26. Since all variables (including expected concentration level) are based on what the EC states, one could argue that, to some degree, they all reflect the EC’s subjective views, or its mistakes (e.g., when defining relevant markets, etc.). I recognize this issue, which this study shares with all other papers in this literature.

  27. See the methodology section for a brief discussion of the omitted variable bias in latent variable models.

  28. All studies with a 0/1 entry dummy variable that were reviewed in Sect. 2 find positive and significant coefficients in all specifications, often at the 0.01% confidence level, as do studies that code the EC’s qualitative assessment of entry barriers as a variable that takes values of 0,1, 2, or 3 (absent, low, medium, high).

  29. Including an ordinal assessment (absent, low, medium, high) as a variable (0, 1, 2, or 3) would make this issue disappear. The issue would reappear if a set of (three) dummy variables were used instead.

  30. In principle, shares/HHI too can be interacted with ‘subjective’ dummy variables when the EC concludes that high shares “are not indicative of a dominant position.” See for instance M.1930, Ahlstrom/Andritz, ¶66.

  31. This was true until the end of 2013. The EC has now raised the threshold to 30%.

  32. Guidelines on the assessment of non-horizontal mergers under the Council Regulation on the control of concentrations between undertakings (2008/C 265/07), OJ, October 18, 2008 (¶¶ 91–92).

  33. Consumer good is 1 for over-the-counter (OTC) drugs; for prescription drugs and other medical products Consumer good is 0. The EC normally defines separate markets for OTC and prescription drugs due to legal, marketing and distribution differences; e.g., consumers (not doctors) choose OTC drugs, whose price public healthcare systems do not cover.

  34. Hence, e-books and hardcopies, or DVDs and music/video downloads have the same dummy variable value. Regression results remain largely the same if these pairs of products are assigned different dummy variables.

  35. Nomenclature statistique des activités économiques dans la Communauté européenne. Details on the level of disaggregation that are associated with these dummies are available upon request.

  36. See M.3354-Sanofi-Synthelabo/Aventis.

  37. The EC typically defines the geographic markets for pharmaceuticals as national markets; it is thus not unusual that a given prescription drug calls for reviews of dozens of markets: At the end of 2013, there were 28 EU countries. The framework is designed to review many markets efficiently and thoroughly. See ten Have et al. (2016) for a discussion of this framework.

  38. I define “European countries” to include EU countries (between 1991 and 2014, their number increased from 15 to 28), and European Economic Area and European Free Trade Association countries (Switzerland, Norway, and Iceland are notable European but non-EU countries).

  39. Exceptions occur for ranges such as [45–55%], or when ranges are reported for each merging firm but not for the merged entity (e.g., a 20-point wide range for the combined entity). These cases represent less than 15% of the 13,818 observations. Observations whose expected market share is equal to the buckets’ bounds (e.g., an expected share of 40% from a [45–55%] range) are assigned randomly to either the upper or lower bucket.

  40. The conclusions with regard to the reduction in the challenge rate would not change when considering only the 85% of the observations that can be assigned to a bucket with certainty. Moreover, the differences are larger when excluding markets from unconditional authorization decisions (about half of the total).

  41. Under CEV, the observed explanatory variable is correlated with the ME, and extra steps are needed (i.e., instrumental variables) to estimate the model. When the dependent variable is continuous, Wooldridge (2002) notes that using an observed variable that is measured with error instead of the true unobserved value only increases the regression error’s variance but does not violate any of the standard ordinary least squares assumptions—provided that the ME is uncorrelated with the observed value that is used in the regression.

  42. I also estimate separate regressions where post-merger HHI and delta HHI take the place of M and D.

  43. The additivity of the composite error follows from the linearity of the latent variable model.

  44. A linear combination of bounded measurement errors cannot be normally distributed over [− ∞, + ∞]; at most, it can be distributed as a truncated normal. Yet, the composite error could still be approximately normal, which justifies the choice of a heteroskedastic over a homoskedastic parametric approach.

  45. The magnitudes are scaled by the irretrievable variance of the error term. Observed differences (or lack thereof) could be entirely due to the error. Coefficient ratios do not depend on the unknown scaling factor.

  46. The pre-reform means of the industry dummy variables do not differ significantly from the post-reform means; the overall mean is thus representative of the average industry in both periods.

  47. The marginal effect for a given explanatory variable is a function of all estimated coefficients, hence the estimated coefficient and the associated marginal effect are not guaranteed to be both insignificant or significant at a given confidence level. I report significance levels that are associated with marginal effects, and highlight instances when discrepancies arise. On this issue, see Greene (2009).

  48. The consumer dummy variables can in part proxy for these factors because countervailing buyers’ power or customer sophistication factors do not typically play a major role in these consumer-oriented markets.

  49. While ten Have et al. (2016) argue that the EC’s framework for assessing pharmaceutical cases appears highly focused on preventing Type II errors (mistakenly allowing anticompetitive mergers), in practice the EC share/HHIs thresholds are not closer to 0 (blocking all mergers would minimize Type II errors).

  50. The first three bins are based on ¶¶ 19–20 of the 2004 HMG: The EC “is unlikely to identify horizontal competition concerns in a market with a post-merger HHI below 1000. […] The Commission is also unlikely to identify horizontal competition concerns in a merger with a post-merger HHI between 1000 and 2000 and a delta below 250, or a merger with a post-merger HHI above 2000 and a delta below 150, except where special circumstances […] are present”.

  51. In the example, the delta HHI and the merging firms’ contribution to the pre-merger HHI would be underestimated too. However, the midpoint approach will yield the right expected delta HHI if certain circumstances hold.

  52. The distribution of the largest merging parties’ market share is not uniform, however.

  53. The same conclusions about the midpoint approach’s shortcomings apply if the EC does not state that the named firms’ shares sum to 100, or if there is a residual group of firms that account for a share in a given range.

  54. That is, 4-tuplets close to the upper bound in each of the four dimensions.

  55. When the EC does not name any individual rival, the smallest firm for the calculation of the residual is the combined merged entity.

  56. In this example, the information would affect the midpoint via a reduction of the ‘high addition to the HHI’ component, while leaving the ‘low addition’ unchanged. In other cases, both or just the ‘low’ addition may vary.

  57. See for instance M.5890—Heinzel/ Europapier.

  58. Case M.2547 Bayer/Aventis Crop Science.

  59. Typically, the study relies on tens of millions of draws.

  60. Based on all draws, not just the first 150.

  61. This market is thus quite different from the rest of the dataset: In the whole dataset, 50% of the 13,818 observations have an expected “Other” that is below 26.61%; 25% of the observations have an expected “Other” that is below 5.85%.

References

  • Bergman, M. A., Coate, M. B., Jakobsson, M., & Ulrick, S. W. (2010). Comparing merger policies in the European Union and the United States. Review of Industrial Organization, 36(4), 305–331.

    Article  Google Scholar 

  • Bergman, M. A., Coate, M. B., Mai, A., & Ulrick, S. W. (2016). Does merger policy converge after the 2004 European Union reforms? Retrieved June 25, 2018, from SSRN https://ssrn.com/abstract=2844402.

  • Bergman, M. A., Jakobsson, M., & Razo, C. (2005). An econometric analysis of the European Commission’s merger decisions. International Journal of Industrial Organization, 23, 717–737.

    Article  Google Scholar 

  • Duso, T., Gugler, K., & Szücs, F. (2013). An empirical assessment of the 2004 EU merger policy reform. The Economic Journal, 123, F596–F619.

    Article  Google Scholar 

  • Duso, T., Neven, D. J., & Röller, L. H. (2007). The political economy of European merger control: Evidence using stock market data. Journal of Law and Economics, 50(3), 455–489.

    Article  Google Scholar 

  • Fernández, B. M., Iraj, H., & Jegers, M. (2008). The implementation of the European Commission’s merger regulation 2004: An empirical analysis. Journal of Competition Law and Economics, 4(3), 791–809.

    Article  Google Scholar 

  • Greene, W. (2009). Discrete choice modeling. In T. Mills & K. Patterson (Eds.), The handbook of econometrics (Vol. 2, pp. 473–556)., Applied econometrics London: Palgrave.

    Chapter  Google Scholar 

  • Harvey, A. C. (1976). Estimating regression models with multiplicative heteroskedasticity. Econometrica, 44(3), 461–465.

    Article  Google Scholar 

  • Hsiao, C. (1983). Regression analysis with a categorized explanatory variable. In S. Karlin, T. Amemiya, & L. Goodman (Eds.), Studies in econometrics, time series and multivariate statistics (pp. 93–129). New York: Academic Press.

    Chapter  Google Scholar 

  • Mai, A. (2016). Is EU merger policy less stringent after its 2004 reform? PESO Working Paper, School of Social Sciences, Södertörn University, 1. Retrieved June 25, 2018, from https://www.diva-portal.org/smash/get/diva2:900189/FULLTEXT01.pdf.

  • Monti, M. (2004), February. Convergence in EU-US antitrust policy regarding mergers and acquisitions: an EU perspective. In Paper presented at the UCLA Law First Annual Institute on US an EU Antitrust Aspects of Mergers and Acquisitions, Los Angeles.

  • ten Have, F., Truijens Martinez, J., & Demertzi, E. (2016). The European Commission’s pharmaceutical merger control practice: an overview of the state of play. World Competition, 39(1), 85–117.

    Google Scholar 

  • Vickers, J. (2004), February. Merger policy in Europe: retrospect and prospect. In Paper presented at the CRA conference on current topics in merger and antitrust enforcement, Washington DC.

  • Wooldridge, J. M. (2002). Econometric analysis of cross section and panel data. Cambridge, MA: The MIT Press.

    Google Scholar 

Download references

Author information

Authors and Affiliations

  1. Takoma Park, MD, 20912, USA

    Federico Mini

Authors
  1. Federico Mini
    View author publications

    You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to Federico Mini.

Appendix: Methodology to Compute Variables Measured with Error

Appendix: Methodology to Compute Variables Measured with Error

Assume that the EC decision reports that the acquiring party will have a [85–95%] share after adding the target firm’s [20–30%] share.

Based on this information alone, it would be right to expect the merged party to have 90% and the target 25%, although it would be wrong to expect that the merged party’s contribution to the post-merger HHI is 902: Doing so underestimates the merged firms’ contribution to the post-merger HHI by (width of the range)2/12 = 8.33 points.Footnote 51

In this case, the midpoint approach yields the expected values because the whole symmetrical quadrilateral that represents the admissible {combined share, share increment} pairs lies below the 100 threshold (a line in a 2-dimensional space). The midpoint approach is a useful shortcut to compute expected values because each random variable has a symmetric marginal distribution that is centered on the midpoint.Footnote 52

Assume now that the EC reported that this is a 4-to-3 merger where the merged firm would face a rival with [5–15%] and another rival with [0–5%];Footnote 53 see Table 7.

Table 7 Shortcomings of the midpoint approach
Full size table

Now the sum of the upper bounds is above 100: The 4-dimensional set of points representing the individual ranges gets its top cornerFootnote 54 ‘cut off as’ inadmissible because it is above a 3-dimensional hyperplane (the set of 4-tuplets that sum to 100).

The midpoint shortcut no longer works, since the cut-off affects the individual expected values in ways that cannot be reliably replicated by reducing all midpoints either proportionally (Table 7, col.) or by a fixed amount (col. 7). The domains of the marginal distributions are not necessarily the whole ranges; the distributions are no longer necessarily symmetric, which makes the adjustments to the midpoint that is needed to match the true expected values vary in ways that are hard to replicate: cols. 6 and 7.

The values in column 2 are computed by drawing many vectors of market shares {m1; m2; r1; r2} from a uniform distribution in R4, and then dropping draws that do not meet all of the restrictions in the published decision.

This ‘numerical approach’ makes it possible to use all of the information the EC reports: When the EC reports that a firm is larger/smaller than another among the N firms that operate in a market, the corresponding non-admissible N-tuplets will be dropped. The same reasoning applies to ranges for the post-, pre-merger, and delta HHI, etc.: These restrictions can be modelled as inequalities that identify inadmissible N-tuplets.

Restrictions that correspond to equality constraints—e.g., the merged firm’s share is equal to two times the largest competitors’ share; the delta HHI is equal to 500—reduce the random multivariate vector’s dimension from N to N−1: One variable can be expressed as a function of other(s).

The numerical approach also allows the computation of HHIs that are based on market shares that sum to 100 in all cases—even when the sum of the ranges’ upper bounds for the named firms is less than 100, and/or the EC reports that “others” account for “the rest”—possibly reporting a range for their total share (say, [0–10%]).

In such cases, for each single draw, I compute the addition to the HHI that arises from the residual share as the midpoint of two opposite alternatives: (1) the residual market share is held by identical very small firms, each with a market share of 0.01% (so that the addition is less than one HHI point); and (2) the residual market share is held by identical firms that each have a market share as large as the smallest individual rival.Footnote 55 When the EC reports information that requires deviating from this method—say, “other firms have 5% or less each”—that information is also taken into account.Footnote 56

Once the distributions of the random variable of interest have been numerically simulated, it is straightforward to compute the expected combined market share (or the expected logarithm thereof) and the expected increase in market share, as well as their standard deviation (and the covariance between combined market share and increase in market share). The same applies to HHIs and their logs.

I define the increase in market share as the combined market share minus the share of the largest merging party. Also, whenever the lower bound of a merging party’s market share range is 0% (say, [0–5%]), it is changed to [1–5%]. The change from 0 to 1% is made to be consistent with EC practice: The EC typically does not report information on horizontally affected markets when the addition of market shares is de minimis, often specifying that this means that the smaller merging party has “less than 1%.”Footnote 57

The figures below visually summarize the method that has been described above with reference to the French market for products to control fleas on small companion animals.Footnote 58 Merial [50–60] bought Bayer [0, 10]; Nestlé [10, 20] was the largest rival; and smaller unnamed firms had the remaining [30–40].

Figure 2 shows the first 150 draws from a multivariate uniform distribution in R3 (‘Other’ is the complement to 100 of the sum of the three named firms).Footnote 59 The figure highlights that draws such that Merial, Bayer, and Nestlé shares are close to the lower bound of the respective ranges are more likely, because their sum cannot exceed 70% since other firms have at least 30%.

Fig. 2
figure 2

Draws for the French market for flea control products for cats and dogs

Full size image

Figure 3 shows the implied distributionsFootnote 60 of for the random variables whose expected values are used in the Share regression. The expected Other market share is 32.50%.Footnote 61 The figures of the implied distributions for post-merger HHI and delta HHI are available upon request.

Fig. 3
figure 3

French market for flea control products for cats and dogs, Share and Ln (Share) distributions

Full size image
Table 8 Regression results
Full size table

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Mini, F. Fifty is the New Forty: EU Merger Policy Permits Higher Market Shares After the 2004 Reform. Rev Ind Organ 53, 535–561 (2018). https://doi.org/10.1007/s11151-018-9645-9

Download citation

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s11151-018-9645-9

Keywords

Search

Navigation