Ex-post Evaluation of Mergers in the Supermarket Industry

We restrict our analysis to cities where both firms involved in the merger were already present before the event. Hence, a merger should always affect market concentration. Table 2 provides evidence for this change in market concentration for the mergers that we evaluate.

Another widely used approach to predict the impact of horizontal mergers is structural merger simulations (Hausman et al. 1994; Nevo 2000).

In the first group we find work in household appliances (Ashenfelter et al. 2013), retail consumer products (Ashenfelter and Hosken 2010), and newspapers (Chandra and Collard-Wexler 2009). In the latter, studied industries include gasoline (Hastings 2004), hospitals (Romano and Balan 2011), supermarkets (Huang and Stiegert 2009, Allain et al. 2017) and bookstores (Aguzzoni et al. 2016).

A different branch of the literature looks at stock market reactions. One example is Hosken and Simpson (2001), who infer merger effects indirectly through stock returns on the announcement day. They find very small stock effects, suggesting that no anti-competitive effects were expected. In general, stock returns seem to be unreliable guides to the effects of mergers (Mcafee and Williams 1988; Kwoka and Gu 2015).

A paper that is closely related to ours is Allain et al. (2017), but with some key differences: they examine the price effects of mergers in the supermarket industry, but consider one particular event in France, whereas we study multiple events that occurred in different years in the United States. This allows us to estimate a distribution of merger effects and assess how merger and city characteristics affect the exact location of the effect within that distribution.

We define “short term” as the 6 months that follow an event and “long term” as everything after that.

The Federal Trade Commission (FTC) did not object to any supermarket merger in the period that is studied in this paper. The absence of an average price effect might be evidence that any ex-ante screening mechanism for these mergers by the FTC found no reason for concerns.

A few papers look beyond the immediate price effect. That is the case for quality in healthcare (Romano and Balan 2011) and for short- versus long-term effects in railroad mergers (Winston et al. 2011). Berry and Waldfogel (2001) look at concentration events in the radio broadcasting sector and find that those increase variety, which is consistent with our findings. George (2007) examine the variety effects of newspaper groups following acquisitions. Her conclusion is similar to ours, but our variety results are within store/firm while her analysis is within a group of firms (with the same owner). Davis (2010) evaluates the effects of mergers on the frequency and depth of sales promotions.

Table 2 provides evidence for a higher market concentration after the mergers that we evaluated, and anecdotal evidence suggests that the acquiring chain usually closes some stores of the acquirer. Unfortunately we do not have detailed information on the number of closures in our sample, even though newspaper articles report the closure of some of the stores that took part in our sample of mergers.

For example, these economies of scale or scope can happen because of lower procurement, marketing and distribution costs (e.g., stores can share the same warehouse and get cost savings through streamlining product distribution functions) and/or the number of consumers who visit each store gets larger (Kaufman 2000b).

By private-label brands, we mean store brands, which is consistent with the way that our dataset defines them.

UPP is a first-order approximation to a full merger simulation, which indicates, in a simple formula, if a merger will likely result in higher or lower prices. It trades-off efficiency gains with the reduction in competition.

The information of the mergers was collected from the website Supermarket News (supermarketnews.com).

According to the Horizontal Merger Guidelines, a merger raises concerns if the HHI increase is larger than 200 (and post-merger HHI is larger than 2500).

Our working assumption is that the effect of mergers in this category is a good prediction of what happens in the entirety of products that are sold in a typical store.

Nielsen defines supermarkets as all grocery stores (using the SIC code) with over $2 million in annual sales.

We later test our results using alternative definitions of the control group.

The reason for this concern is potential zone pricing: Stores in markets that are very close to event cities may be in the same zone and suffer an effect. Control stores in other states may have very different price levels , but that is fine because our estimator only captures differential changes that are caused by the event

The Universal Product Code (UPC) consists of twelve numerical digits, which are uniquely assigned to each product. The UPC is the barcode mainly used for tracking products in stores.

It may be the case that this quarterly price measure overstates the actual mean price during that quarter if positive sales are more frequent with lower prices. Yet there is no reason to believe that this will affect our estimated coefficients as only differential effects are relevant

According to Basker and Noel (2013), “the FTC has often used a three- or five-mile radius around each supermarket to denote the store’s catchment area”

FTC File No. 001 0057, https://​www.​ftc.​gov/​news-events/​press-releases/​2000/​06/​ftc-seek-injunction-block-kroger-co-purchase-winn-dixie.

The average value of the population in treated and control cities is not very different, but this is only because some of the cities in the control group have a very large population. If instead we examine the median values of the population, we have some important differences: the median population in the control cities is 66,900, whereas in the treated cities the median population is 173,404 before the merger and 169,241 after the merger.

We allow for some differential time patterns in the specification as well.

In order to settle on this specification we ran a series of Chow tests across different groups of stores. To be precise, we tested for differences in time patterns before the events took place. We could not reject that acquirer stores in event markets and in non-event markets had the same time pattern (p value of 0.877 for variety and 0.391 for prices). Similarly, we cannot reject that Other stores in event markets and in non-event markets had the same time pattern for variety (p value of 0.637). However, we do reject similar patterns for Other stores in the price regression. For that reason, in the online appendix we estimate a different specification that also allows for different time patterns for the two groups. The results are qualitatively similar.

In specification (3) we control by “store address” fixed effects instead of the richer “store address × chain” fixed effects that we use in the remaining columns. That is necessary to identify a separate “Target” effect. Therefore, one should interpret the estimated target effect with caution. There can be a bias if, for example, acquirer chains have systematically lower prices than the chains that are being acquired.

In the online appendix we report regression results by product category, separately for private label and national brands.

We also estimated our baseline regressions using regular prices (after removing sales) instead of average prices. We do not report those results because of space constraints, but they show a similar picture. The main difference is that the negative coefficient on regular prices for acquirer stores becomes significant. This is consistent with the explanation provided in Trindade (2011)

One related concern is the difference between the effects of mergers in general and mergers “at the enforcement margin”. The latter group consists of those events in markets where there is less competition ex-ante and, therefore, are more likely to be scrutinized by the antitrust agencies. Since we do not observe sales for every store in each city (but only those that have an agreement with Nielsen), we cannot directly measure the effects for that group of stores. When, in the next subsection, we examine the heterogeneity of the results, we estimate the effect of mergers in cities with high HHI before the merger. That variable can be thought of as a proxy for the likelihood that a merger will be investigated by the relevant antitrust agency. The list of FTC actions that are reported in the Online Appendix further suggests that most of the mergers in this industry are not “at the enforcement margin” and therefore we cannot draw policy conclusions about our results.

The results are available in the Online Appendix.

The first step of the procedure was to geocode the address of every store in our sample to obtain latitude and longitude. Then, we computed the distance between stores in an event city and those in the control group using the Haversine formula. Stores belonging to the same chain of a treated store and within a 100 mile band distance from treated stores were dropped.

Davis (2010) finds evidence of economies of scale and improved bargaining power after large national and regional supermarket mergers. Our results suggest that economies of scale and improved bargaining power can also occur in a merger in a small city with effects on concentration and their effects can manifest on variables other than prices.

Despite the fact that these mergers are local, efficiency gains are to be expected. For example, Holmes (2011) shows that a big national chain—Wal-Mart in his example—has a multiplicity of distribution centers, each serving a few stores. Two different chains that merge will likely start using only one of its distribution centers and drop the least efficient.