A common practice in merger screening is to estimate and consider market shares and concentration. However, as stated in the United States Department of Justice (“DOJ”) and Federal Trade Commission (“FTC”) Horizontal Merger Guidelines (“Guidelines”), “[m]arket shares may not fully reflect the competitive significance of firms in the market or the impact of a merger.” For example, standard measures of market concentration do not account for cost efficiencies that may occur as the result of a proposed merger. Thus, it is important to consider whether market concentration measures, such as the Herfindahl Hirschman Index (“HHI”), are good indicators for assessing whether a proposed merger is likely to harm competition in the presence of merger-specific efficiencies.
Nathan H. Miller, Marc Remer, Conor Ryan and Gloria Sheu (“Upward Pricing Pressure as a Predictor of Merger Price Effects,” International Journal of Industrial Organization, 2017) test the accuracy of UPP and HHI as merger screening tools. The authors use a large-scale Monte Carlo experiment with a variety of merger scenarios — generated by randomly drawing market shares and using varying demand substitution patterns. The authors find that the change in HHI, in many cases, can be a good indicator of likelihood of price increases due to a merger.
I extend on this approach to test the accuracy of the HHI as a merger screening tool in the presence of merger-specific efficiencies. Following Miller et al, I also use Monte Carlo simulations and classify mergers into five potential categories based on the post-merger HHI and the change in HHI. The five categories are 1) post-merger HHI greater than 2500 and change in HHI greater than 200; 2) post-merger HHI greater than 2500 and change in HHI greater than 100, but less than or equal to 200; 3) post-merger HHI greater than 1500 and less than 2500, and change in HHI greater than 100; 4) post-merger HHI less than or equal to 1500; and 5) change in HHI less than 100. These categories reflect the likelihood a merger will be investigated (categories one through three are likely to raise possible competitive concerns, while categories four and five are unlikely to raise competitive concerns).
For the Monte Carlo simulations, I use four standard functional forms for the demand side. These are Logit demand, Log-Linear demand, Linear demand and Almost Ideal demand. I also assume a market structure for each industry, which may contain four, six, or eight firms competing with differentiated products. I also consider efficiencies generated using two different cost formulations, Generalized Leontief and Quadratic functional forms. The Generalized Leontief cost formulation will generate higher merger-specific efficiencies than those generated through the Quadratic cost formulation, which allows for a comparison between mergers that are likely to generate more substantial merger efficiencies and those that are not. For each scenario, I draw 3,000 mergers, for a total of 72,000 merger simulations.
I find that the more substantial and significant the merger efficiencies are, the less likely it is that classifications using the post-merger HHI and the change in HHI will be accurate indicators of post-merger price increases. For example, consider the Monte Carlo simulations for mergers that fall under category one (with a post-merger HHI exceeding 2500 and a change in HHI exceeding 200) and that are in industries with four firms. For these simulations, I find that the probability of a five percent price increase ranges from approximately 30 percent to 58 percent, depending on the demand system, for mergers under a Generalized Leontief cost structure, while for mergers under a Quadratic cost structure, it ranges from approximately 61 percent to 90 percent. I also considered the probability of a ten percent price increase and find that the probability of a ten percent price increase ranges from approximately 17 percent to 54 percent for mergers under a Generalized Leontief cost structure and from approximately 21 percent to 80 percent for mergers under a Quadratic cost structure.
Additionally, any screening tool, including the HHI, is not a perfect predictor of post-merger price increases, and this can lead to false positives and false negatives — known as type I and type II errors. I estimate the likelihood of these type I and type II errors for the five merger categories, using each demand system and each functional form of cost efficiencies. I consider two thresholds, a price increase of five percent and a price increase of ten percent, as indicating that a merger is anticompetitive. For example, I consider the following combination: the merger is considered anticompetitive if it results in a price increase of more than five percent and the merger belongs to category one (post-merger HHI exceeding 2500 and change in HHI exceeding 200). Monte Carlo simulations for this combination indicate that the type I and type II errors increase with greater merger-specific efficiencies. I find that the sum of the type I and type II errors range from 40 percent to 48 percent for this combination, in industries with four firms, for mergers under a Generalized Leontief cost structure, while for mergers under a Quadratic cost structure, the sum of the type I and type II errors range from 24 percent to 33 percent.
The DOJ and FTC, following their Guidelines, use both the post-merger HHI and the change in HHI to classify mergers into those that are more likely to raise competitive concerns and those that are not. However, these classifications may not be strong predictors of whether a proposed merger will harm competition when there are substantial merger-specific efficiencies. There is a need to better understand merger screening models whenever efficiencies are involved and recognize that typical methods (such as market share and concentration) often do not reflect market dynamics. Structural-based models and estimates of unilateral effects are additional tools that may improve merger screening.