تصمیم گیری رفتاری، پیش بینی، نظریه بازی و نقش بازی

Green’s finding that the outcome of role-play provides forecasts that are superior to those of game theorists highlights some of the unrealistic assumptions used in traditional game theory. In this commentary I discuss how elements studied in the behavioral decision literature impact the manner in which people behave in conflict situations studied by Green, and in the spectrum auction conducted in the United States. The main behavioral elements discussed are loss aversion, myopia, and the winner’s curse. Green (2002) presents compelling evidence that when it comes to forecasting the outcomes of real world conflicts, role-play can be more effective than the forecasts made by game theorists. Yet most of the commentaries on Green’s paper urge caution when interpreting his findings to imply that game theory should be abandoned as a forecasting tool. For example, Bolton (2002) points out that it is now common practice in business school negotiation classes to use role-play and game theory together. He suggests that in the future the most useful lessons will come from work that combines game theory with experimental economics, along the lines described in Erev, Roth, Slonim, and Barron (2002). Although I concur with Bolton’s general conclusion, I would argue that game theory’s value as a forecasting tool will continue to be limited until game theorists revamp its structure along behavioral lines. Traditional game theory assumes that players are fully rational in respect to preferences (expected utility), judgments, and strategic choices. Yet the behavioral decision literature documents that preferences routinely violate the axioms of expected utility, and that judgments exhibit systematic biases and errors. In making the case for the explicit incorporation of behavioral concepts into game theory, I draw on two sets of experiences. The first is my own experience from a setting described by Bolton, teaching a graduate business school course that makes dual use of game theory and role-play. The second is the experience of the spectrum auction being conducted by the US government.

What is the practical relevance of game theory for the kind of day-to-day forecasting situations faced by practitioners, such as the individual conflicts studied by Armstrong (2001) and Green (2002)? It seems to me that Green is essentially correct, that in a head-to-head contest with role-play, traditional game theory comes out a distinct second. What gives role-play its edge over game theory? In my opinion, the answer is that role-play outcomes emerge from the actual interaction of real human beings, whereas game theoretic outcomes emerge from the theoretical interaction of idealized human beings. The behavior of real human beings reflects emotion and errors in judgment, as embodied within psychological phenomena such as loss aversion, myopia, and winner’s curse. In contrast, the idealized human beings modeled by game theory are rational in respect to their preferences, judgments, and choice of strategies. The development of behavioral game theory is well underway (see Thaler, 1988 and Rabin, 1994; and especially Camerer, 2001). The introduction of learning, as discussed by Erev et al. (2002) is a small, but important step. The predictive power of game theory will improve as the underlying assumptions and equilibrium concepts are made more realistic. Indeed role-play exercises and experiments should help game theorists to develop well-structured behaviorally based models. McAfee and McMillan (1996) state that “the real value of theory is in developing intuition” rather than applying “complicated models that try to capture a lot of reality” (p. 172). It seems safe to say that from the perspective of the FCC, game theory was a powerful tool to have used in designing the US spectrum auction. The auction form they used provided a decent mechanism for bidders to aggregate licenses, and it raised more revenue than was forecast. At the same time, McAfee and McMillan make a stronger assertion, stating that the role of theory “is to show how people behave in various circumstances” (p. 172). Of course, this is exactly the forecasting issue studied by Green. On this dimension, the evidence suggests game theory has had limited success. Winning bidders in the spectrum auction were afflicted by the winner’s curse. And the C block auction produced a rather untidy outcome that is still unfolding. The results from the studies by Green and Armstrong suggest that in its current form, game theory has had limited success in forecasting “how people behave in various circumstances”.