Feedback complexity, bounded rationality, and market dynamics

In a set of experimental markets, we investigate how the dynamic structure of the market as well as its price-setting institutions affect market performance and stability. We compare the outcomes to two alternative hypotheses: The standard neoclassical assumption of optimality and rational expectations, and a behavioral hypothesis based on previous studies of humans in dynamic decision making tasks, where subjects frequently ignore critical elements of the feedback structure in which they operate. We consider the implications of such misperceptions for the process of market adjustment, as well as the ability of market forces and financial incentives to mitigate these effects. We find that feedback structure has a strong effect on performance relative to optimal, as the markets showed large fluctuations in prices and quantities. The only condition where performance (but not the dynamics) are relatively unaffected is where computer-mediated automatic price-clearing eliminates cumulative imbalances. The observed behavior is consistent with individual misperceptions of feedback, thus demonstrating that markets moderate but do not eliminate the negative impact of misperceptions of feedback. We analyze the decisions of individual subjects by fitting them to simple equations and then used the estimated equations in a dynamic simulation model of the complete market. The simulations reproduce the most salient features of the dynamic behavior, and variations between markets can be related to differences in certain parameter values of the decision functions that can be interpreted as reflecting the degree of misperceptions of feedback. In this manner, the analysis constitutes a link between observation of individual decision making behavior and the theory of aggregate market outcomes. An examination of decision timing data and verbal protocols provide evidence that increasing complexity of the decision-making task leads subjects to simplify their task by ignoring certain aspects, particularly strategic interactions between firms, and revert to simpler, more reactive behavior. Moreover, subjects tend to attribute observed oscillations and other systematic behavior patterns to exogenous forces rather than their own interaction with the system. This latter phenomenon may have important implications for the ability of agents to learn over time to improve their behavior. Draft August 20, 1996 2