Individual Strategy and Social Structure: An Evolutionary Theory of Institutions.

• Author: Mezzetti, Claudio
• Position: Review

By H. Peyton Young. Princeton, NJ: Princeton University Press, 1998; Pp. xiii, 189. $35.00.

In the early 1990s, a series of papers by Peyton Young, Dean Foster, Larry Blume, Michihiro Kandori, George Mailath, and Rafael Rob introduced perturbed Markov processes into evolutionary game theory. This relatively short book by Peyton Young is a survey of the techniques and some of the most interesting substantive results in this area. The book is masterfully written and accommodates different reading levels. It can be read by the nonspecialist who is only interested in understanding the main ideas, or it can be read by the scholar who wishes to learn new tricks to use in his or her own research. This is accomplished by relegating the proofs to appendices, by a judicious use of illuminating examples to develop the theory, and by precise verbal explanation of all the mathematical results.

The book begins with an introductory chapter that overviews the evolutionary approach. Two such approaches are the main departures from traditional economics and game theory. First, equilibrium is not the static outcome of the optimal behavior of agents. Second, agents are not perfectly rational. The evolutionary approach looks at equilibrium as the stable outcome of a dynamic process in which agents are imperfect optimizers that adapt their choices over time in response to the environment. Young convincingly argues that evolutionary models are well suited to explain how social conventions and institutions emerge from the interactions of a large number of different agents. In this respect, a stable convention or institution is like a stable population equilibrium in evolutionary biology.

Chapters 2, 3, and 4 contain the core material of the book. Chapter 2 begins by discussing different varieties of learning behavior and then justifies the adoption of a form of best-reply dynamics as the main object of study. In Young's version of the best-reply dynamics, in each discrete time period n, players are randomly drawn from n populations to play an n-person game. Before choosing an action, a player samples from the history of the game. More precisely, she observes the play in a randomly selected sample of size s from the set of games played in the past m periods, with s [less than] m. The empirical distribution of play from this sample constitutes the player's beliefs about her current opponents' choices. With probability 1 - [Epsilon], the player selects an action that...