Is Game Theory of Any Value for the Historical Analysis of Institutions?

MARCH 4, 2008

I was much much struck by generally pessimistic tone of Gregory Clark’s lengthy review in the JEL’s September issue of Avner Greif’s Institutions and the Path to the Modern Economy. These comments have wider implications for the application of economic tools (especially game theory) to the analysis of historical outcomes, particularly in relation to institutions, and I have therefore thought it worth excerpting from the review here (at some length).

When You Have More Variables than Data You Aren’t Explaining Anything

As noted, Greif defines an institution as a self-reinforcing set of behaviors. Greif pioneered in applying game theory to historical institutional analysis and his 1993 study of the Maghribi traders remains a classic of this still modest genre. This was certainly an exciting development for economists. For the first time, [sic] seemingly grounded the explanation of informal institutions in optimizing individual rational behavior. Behaviors that would seem to the layman to be based on blind irrational custom could be shown to be consistent with individual optimization. Given the incredible intellectual elaboration of game theory, and its meager harvest in terms of actual economic applications, the finding was welcome to both game theorists and to economic historians. [ed: a tough but fair assessment …] The Maghribi study also allowed for the possibilities of institutional change resulting just from changes in parameters. Since the equilibrium depended on certain parameter values, changes in transportation costs or observability could terminate the old equilibrium and lead to a new one. The 1993 article seemed to point to new micro foundations for institutions that would ground them in individual maximizing behavior.

But this book is almost certainly not what many economists who welcomed the 1993 article expected as the generalization of its ideas. Some indeed will be shocked by, and perhaps hostile to, the path Greif has taken. Were economists of a more literary bent, the word apostasy would be on their lips. In a search for generality, Greif concludes that such a set of limited rational actor assumptions is not constraining enough to describe real-world institutions. For a start, “multiple equilibria usually exist in the repeated situations central to institutional analysis” (p. 125). There have to be more constraints on the structure of the interaction to explain the equilibrium. These constraints include “cognitive norms” (p. 128) as well as “the social and normative foundation of behavior” (p. 143). Issues such as “losses of esteem,” “norms,” “fairness,” or “social exchange” have to be introduced. Also such social and normative behavior is “situationally contingent” (p. 144). [ed: and we now have so many parameters we could probably explain anything …]

Greif posits this as just an extension and elaboration of the original individualistic rational-actor game theoretic ideas. Once we are compelled to admit, however, into the explanatory apparatus almost the entire sociological zoo of ill defined and unmeasurable constructs, we lose all explanatory power. Explanatory power requires few objects and small degrees of freedom. Greif notes that “a useful feature of game theory is that it allows us to study all intertransactional linkages—economic, coercive, social and normative—simultaneously” (p. 147). But he does not seem to appreciate the price of this generality in terms of testability. All we are left with is the idea that people operating within institutions act as they do because, given the cognitive, intellectual, cultural, and normative constraints they face, their actions seem to them as being the best available. But, in an informal sense, we knew that already. Without any consideration of the ins and outs of game theory, we can appreciate that any lasting institution likely constitutes some set of self-reinforcing behaviors. Yanomamo males, for example, engaged in recurrent raids against other bands aimed at capturing women and revenging previous raids (Napoleon A. Chagnon 1983). This was clearly an institution in the sense of Greif and must be maintained by some kind of self-reinforcing set of behaviors. But we knew that, even if we had never studied game theory. So what insights have we gained from page after page of elaboration on the idea of equilibria and the elements that enter into them (pp. 124-53)? If we were able to reduce all such social equilibria to a game theory equilibrium of purely self interested rational individuals interacting with common knowledge that would be a radical, novel, and testable theory. This book denies that possibility, but without providing any alternative that has empirical content. [pp. 735-736, emphasis added]

The Problem of Too Many Equilibria (in Dynamic Games with Beliefs)

… Greif here starts from the basis that we will never be able to predict institutional structure from exogenous features of the situation—including institutional history. … Given the many potential stable equilibria in each institutional context, the outcomes are inherently unknowable. After the attention given to elaborating the theory of institutional stability and dynamics in the preceding 350 pages, this conclusion comes as something of a surprise. The structure and tone of the previous discussion is that of laying the groundwork for a theory of institutions. The reader now learns that the extended theory encompasses a perhaps uncountable number of possible institutional equilibria, so that there can be no advance prediction.

Just as deductive methods cannot succeed, Greif asserts also that inductive generalization about institutional forms will also fail to reveal any patterns. This is because unobservable elements of the situation—beliefs and norms—are crucial to the determination of the outcome. The same observable elements will be associated with radically different institutional equilibria. … [p.737]

Case Studies (and Historical Anecdote) Aren’t Economics (or Economic History

[The empirical approach recommended by Greif] As conducted in the book, [this] is essentially the method of “analytical narratives” popularized by Greif and Robert Bates, Margaret Levi, Jean-Laurent Rosenthal, and Weingast. An analytical narrative consists of matching institutional detail to a formal, or more often informal, interpretation of the situation as some kind of rational choice equilibrium, interpreted in the broad sense above (Bates et al. 1998). It is not clear how this is distinguished from such things as Harvard Business School case studies. As applied by Greif and his colleagues, an “analytical narrative” seems to be just an interpretation of an institution in terms of a loosely defined equilibrium. This is fine as an approach to generating hypotheses, but as an endpoint of analysis, as it generally is in the book, it offers little conviction. [p. 737]

In Conclusion: There Isn’t Much of a Future

… Greif intends in his book to develop at least the outline of a new, micro grounded theory of institutions. Stating, explaining, and elaborating this theory takes 503 densely written pages, including a primer on game theory. By the end, however, this reviewer, to the contrary, read it mostly as a demonstration of the impossibility of a systematic account of institutions along the lines he proposes. The efflorescence of concepts, combined with the constriction of possible empirical tests, makes … prediction and testing impossible. And this shows in the case studies conducted in the book. Each institution in his formulation has to be analyzed in its full idiosyncracy, aided by the expert judgment of the investigator as to the social and epistemological context. But, as we saw in the case of the Podesteria, that kind of analysis, even in the hands of a careful enquirer like Greif, is fraught with the danger of conflating conjecture and fact. Kant’s Prolegomena to any Future Metaphysics as a Science never led to his proposed science of metaphysics. Unfortunately Greif’s Prolegomena to a future institutional theory similarly serves mainly to indicate the barriers to a science of institutions.