Within antitrust there appear to be two distinct trends regarding the use of sophisticated economic models such as merger simulations. Increasingly, economists at enforcement agencies and private economic consulting firms have been employing such models, and there is some evidence that courts have relied upon them. Similarly, the European Commission has used merger simulations in matters involving mobile telephony and industrial commodities, among other industries. On the other hand, there is also a strong dissident movement that abjures economics and would reject the use of such models and replace them with various populist rules. Some observers see the roots of this dissident movement in the failure of economic models to predict the aforementioned financial crisis of 2007–2008. Certainly merger outcomes do not seem always to be what has been predicted ex ante. Moreover, it is not unusual for both economic experts on opposing sides of antitrust matters to point to economic models in drawing their conclusions.
While Kay and King’s primary concerns are with the use of models in the macroeconomic and monetary theory worlds from which the authors hail, their theses concerning the use of economic models under uncertainty would seem to hold lessons for the practice of antitrust as well.
Kay and King contend that economic models fail to perform as advertised largely because they too often ignore the critical distinction between risk and uncertainty or, perhaps more accurately, because such models are simply incapable of accurately modeling the latter. Kay and King’s overall conclusions can be broadly summarized as follows:
Models as often applied neglect the distinction between uncertainty and risk and assume that uncertainty can be modeled in the same probabilistic framework as risk.
Modeling economic processes and outcomes can be challenging since the processes underlying them are often not “stationary” in the sense that they are not governed by unchanging physical laws.
The optimization model of human behavior that underlies most economic models can be too simplistic and can lead researchers astray if taken too literally.
Drawing on the insights of economists such as Frank Knight and John Maynard Keynes, the economics profession has long recognized that risk and uncertainty are not understood synonymously. To both Knight and Keynes, risk applies to situations where, although the outcome is not known, all possible outcomes and the probability associated with each are (such as the flip of a fair coin). Uncertainty refers to situations where the odds or even the possible outcomes themselves are unknown. Thus, while all future events are in a certain sense unknowable, for many we have tools to estimate their probability based on past events that are very similar to possible future ones or it is reasonable to use logical inference to determine those odds (as in the flip of a coin). Uncertainty arises when such tools are not applicable to future events.
Kay and King make a further distinction between uncertainty that is resolvable and uncertainty that is not. The latter they label radical uncertainty. Resolvable uncertainty is uncertainty which can be resolved with more information (e.g., I am uncertain as to the capital of Pennsylvania, but I can remove the uncertainty by obtaining more information). Radical uncertainty is a situation where there is no suitable means of estimating the probabilities before the outcome is known. An extreme example of radical uncertainty might be something like the discovery of the New World. Prior to its discovery (and even for some years after) residents of the Old World could not even conceive of such an event let alone attach probabilities to it before it occurred.
Kay and King note that uncertainty is a function not only of ignorance, but also of the fact that the underlying processes are not stationary. Models are useful for predicting states of the world that are within the contours of what has already happened and where there is good reason to believe that the world is relatively stable. In situations where the underlying processes are stationary, and one knows with reasonable precision the possible future states and their likely probabilities, models can be extremely powerful, such as when NASA is able to launch an interplanetary mission that predicts where a spacecraft will be a number of years into the future. When it comes to predicting situations that are outside previous and recent experiences (e.g., a great recession, the 9/11 attacks, a pandemic) models are often of little avail. Models cannot predict unique and radically uncertain events. This results from both ignorance and the non-stationarity of the processes.
Kay and King also argue that human behavior should not be characterized by individuals optimizing among a number of choices. They contend that in many situations, individuals don’t lay out an array of choices and then optimize among them, but rather use narrative and contextual reasoning to manage uncertainty. For example, they point to work done by the American psychologist Gary Klein, who has concluded that on the battlefield and in various emergency situations decision makers search for the first workable solution they can find rather than the optimal one. While the authors are critical of the economist’s standard model of optimizing behavior, they are equally critical of behavioral economists’ assertions that human beings are often irrational. Rather, they use an evolutionary perspective to argue that certain behaviors that may appear irrational from an individualistic perspective are rational when seen in terms of group survival. For example, while altruism and certain forms of loss aversion may appear problematic from an individualistic optimizing perspective, they are entirely “rational” from an evolutionary and group perspective.
Despite their criticisms, Kay and King’s analysis is not a call to eschew the use of economic models entirely. Rather, it is a call to realize what models can and cannot realistically be expected to do, and an offer of suggestions for supplementing models with other forms of reasoning to improve the decision-making process. Indeed, the authors’ analysis of economic models reads at times as if it came from a standard introductory economics course. Thus, the authors note that models are useful for identifying critical factors and for giving some sense of how these factors have interacted in the past and might interact in the present or future where the underlying processes are relatively stationary. The authors also take the uncontroversial, but sometimes ignored, position that models are simply self-contained worlds, and the correlation between these models and the real world is one of analogy rather than to provide tight predictions.
As noted above, Kay and King are primarily concerned with the use of models in the macroeconomic and monetary theory worlds rather than the microeconomic approach relevant to industrial organization and antitrust. Indeed, the closest reference to antitrust in the book is a digression by the authors noting that the French mathematician, Joseph Bertrand, namesake of one of the most widely used models in antitrust, was critical of using probabilistic reasoning in cases of uncertainty. Nevertheless, their book provides lessons for antitrust practitioners as well.
Some of Kay and King’s criticisms of the use of models will appear relatively quotidian, both to those working in the antitrust field and to relatively sophisticated readers in general, although they do offer several useful reminders of how models have been misused. For example, Kay and King spend a chapter discussing how models are often used to justify choices made by policy makers ex post rather than to guide those choices in the first place. It is not difficult to draw an analogy to antitrust here. Thus, it would be uncontroversial to most adjudicators that models created in the ordinary course of business (analogous to models used to guide policy) are more probative than models that have been created as part of an undertaking’s advocacy (analogous to models used to justify policy choices ex post). They also note that, at best, the variables used in models are to varying extents imperfectly measured. At worst, they are simply made up.
At first blush, it is not obvious that economic models employed in antitrust suffer from concerns over radical uncertainty and non-stationarity in the same way as do models seeking to predict the effects of a pandemic or even macroeconomic policy. Antitrust models are generally called upon to predict what would happen in the relatively near future assuming most factors do not change (other than the practice or transaction under consideration). Nevertheless, Kay and King’s analysis does suggest a need for caution. The greater the extent that the “but for” world represents a significant structural break from the “as is” world, the more questionable the predictions of a model are likely to be.
For example, assume one is considering a merger that results in a large change in current market structure as measured through conventional concentration measures such as the Hirshman-Herfindahl Index (HHI). The analyst might be tempted to argue that this is of little practical consequence, since even if the model does not precisely calculate the size of a price increase, surely it could be counted on to get the direction of the effect right. However, if the cognizable efficiencies from such a merger are also potentially large and outside the range of experience, they are equally unpredictable. Under such circumstances, modesty regarding the power of such a model to predict net effects would appear apt. Modesty would also appear to be called for in other contexts where structural breaks are involved, such as when innovation is at issue or in cases involving considerations of nascent or potential entry. This is not to say that admittedly imperfect models cannot assist decision makers in a world of significant uncertainty. Nevertheless, one should be wary of assuming that economic models are well-positioned to provide tight and accurate estimates of effects in such cases.
In offering suggestions for how to move beyond simply modeling and thereby improve our ability to make good decisions, Kay and King make use of the distinction between deductive, inductive and abductive reasoning originated by the American pragmatist philosopher Charles Sanders Pierce. Deductive reasoning reaches logical conclusions from stated premises. Inductive reasoning seeks to make generalizations from observations. Abductive reasoning seeks to provide the best explanation for a single event. Another term for abductive reasoning that might be familiar to antitrust practitioners is case study.
Indeed, the legal framework within which virtually all antitrust matters are ultimately adjudicated essentially relies heavily on abductive reasoning. It involves a search for the “best explanation”—a persuasive narrative account of events related to the case of which economic models are only one input. In civil proceedings, that narrative must be better than any alternative narrative (i.e., “more likely than not”). In criminal proceedings, the narrative must be sufficiently compelling that no materially different account of events could be seriously entertained (i.e., “beyond a reasonable doubt”).
To fulfill the mission of enabling decision makers to make better decisions under uncertainty, Kay and King argue that all forms of reasoning are required. Further, they argue economics has a role to play in all aspects of the analysis––narrative, data, and models. Neither policy makers nor economists themselves should limit the role of economics and the economist to setting up and testing models alone. Economists can aid the narrative analysis by providing framing. This involves identifying critical factors and how these factors have interacted in the past and by assessing how they might interact in the future. Not every useful insight is capable of being quantified, and economists need to be cognizant of that fact. That said, economic analysis needn’t abandon modeling (or the use of mathematics or econometrics) in helping to inform decisions. Each type of reasoning has its place; none are likely to be sufficient by themselves, yet all can be usefully employed in concert with one another.
All in all, this book is a useful and accessible read for most antitrust practitioners. It offers both provocative arguments regarding models, as well as more mundane thoughts that every practitioner should be reminded of. Usefully, the authors present particularly lucid discussions of several concepts central to understanding decision making under conditions of risk and uncertainty to which lawyers and other policy makers may have been exposed but have not had carefully explained to them. (One such concept is Bayesian reasoning.) Overall, greater adherence to the authors’ general theses regarding the role and limitations of economic modeling applied to antitrust may serve to counter the criticisms of those who would prefer that economics have little or no role in antitrust at all.