In my class we recently discussed a paper by Higgins and Rodriguez (2006)—published in the Journal of Financial Economics—that contains an important lesson for researchers who want to apply the difference-in-differences (DiD) method in competition analysis and merger control.
“We find evidence consistent with the proposition that deteriorating R&D productivity could be the motivation underlying the acquisition of research-intensive firms.”
DiD estimators try to solve the problem of causal inference by comparing a treatment and control group over time. In competition analysis, this technique is often applied to compare a group of firms that engage in mergers and acquisitions (treatment group), with another group of firms that don’t (control group). The goal is to see whether the acquiring firms are able to charge higher prices due to lower competition in the aftermath of a merger.
To get the intuition behind DiD, have a look at the figure below. First, we need to check the difference in average price levels between both groups before the acquisition (difference 1). Second, we assess the difference in prices after the acquisition (difference 2). If difference 2 is larger than difference 1, we have solid evidence that acquiring firms are able to charge higher prices because of the anticompetitive effects of the merger. In this case, an antitrust authority such as the FTC or the European Commission would probably want to intervene to protect consumer interests.
By now the name should also make sense. The causal effect of acquisitions is estimated by looking at the difference (or change) between difference 1 and difference 2:
Effect of acquisitions = difference 2 – difference 1
The great advantage of DiD estimators is that treatment and control groups don’t have to be entirely similar, like with other statistical methods. That means, it’s no problem if there is a positive difference 1 before the acquisition. This milder requirement is great in applied work, because we know that acquirers are often very different from the general population of firms (larger, greater financial resources, etc). However, the causal interpretation of DiD estimators hinges crucially on the assumption that nothing else of importance changes in the meantime. Everything on top of difference 1 is attributed to the effect of the treatment. Therefore, simultaneous changes in firms’ prices due to other influence factors need to be ruled out. We also say that the difference between treatment and control group needs to be “time-invariant” or “constant over time”.
This brings me back to the quote from above. Higgins and Rodriguez show that firms in the pharmaceutical industry make use of acquisitions predominantly in times when their own product pipeline is deteriorating. I.e., they buy up other firms when their own drug portfolio starts to become less profitable and no new blockbuster drugs are in sight. Hence, without their acquisitions, they would probably be forced to charge lower prices some time soon. If their M&A strategy plays out though, they can add a bunch of drugs with healthy sales forecasts to their existing portfolio and, at the same time, get rid of some annoying competitors to fight the price decline.
If that’s the case, however, the time-invariance assumption mentioned earlier would be violated and our DiD estimates might be biased. Then, even if there was an anticompetitive effect of mergers and acquisitions, we wouldn’t be able to detect it, because only firms with a declining sales and R&D trajectory engage in them. All we would find, would be price levels staying the same for acquiring firms compared to firms from the control group. Consequently, competition authorities would remain inactive although they should better stop those acquisitions for the good of consumers.
In pharma—the industry Higgins and Rodriguez study—all these things are nicely observable since products are well-defined and their sales forecasts are quite predictable. Drugs are usually patent-protected, which gives firms monopoly rights for a certain period of time, after which sales drop sharply due to competition from generic drugs. Moreover, future drug developments are easy to trace, because firms need to apply for FDA approval way in advance. But if similar things are going on in other industries too—and it’s hard to tell—we should be cautious with interpreting difference-in-differences estimates in competition policy analysis.