Evolutionary models in laboratory and auditing games.

Abstract

Traditional game theory assumes that each player in a strategic interaction is able to predict (and optimize against) the strategies of the other players. This dissertation investigates evolutionary game theory (EGT) as an alternative method of analyzing strategic interactions. The central assumption of EGT is that players increase the frequency of strategies that would have been relatively successful in previous rounds. Chapter 1 describes the predictions of EGT when little is known of how players respond to past outcomes. The analysis determines that for a variety of response dynamics, players' strategies will always converge to strict dominance solvable Nash equilibria, and sometimes converge to strict Nash equilibria, but will never converge to mixed strategy equilibria. Experiments in Chapter 2 use these results to demonstrate that an evolutionary process is the most likely cause of the equilibrium behavior that arises in laboratory games. Chapter 3 uses EGT to incorporate the "anchoring and adjustment" heuristic into a strategic auditing game, in which an auditee chooses whether or not to misrepresent an account balance, while an auditor chooses whether or not to accept the reported balance. The analysis reveals the possibility of cyclical behavior, in which periods of frequent auditee misrepresentation are followed by frequent auditor rejection, which leads to less frequent auditee misrepresentation, which in turn leads to infrequent auditor rejection. The amplitude of such cycles is shown to be mitigated by an increase in the informativeness of the data collected by the auditor. Experiments in Chapter 4 provide support for the analysis, by demonstrating the existence of cycles in a laboratory setting.