Meta-Stability of Interacting Adaptive Agents

Abstract

The adaptive process can be considered as being driven by two fundamental forces: exploitation and exploration. While the explorative process may be deterministic, the resultant effect may be stochastic. Stochastic effects may also exist in the expoitative process. This thesis considers the effects of stochastic fluctuations inherent in the adaptive process on the behavioural dynamics of a population of interacting agents. It is hypothesied that in such systems, one or more attractors in the population space exist; and that transitions between these attractors can occur; either as a result of internal shocks (sampling fluctuations) or external shocks (environmental changes). It is further postulated that such transitions in the (microscopic) population space may be observable as phase transitions in the behaviour of macroscopic observables. A simple model of a stock market, driven by asexual reproduction (selection plus mutation) is put forward as a testbed. A statistical dynamics analysis of the behaviour of this market is then developed. Fixed points in the space of agent behaviours are located, and market dynamics are compared to the analytic predictions. Additionally, an analysis of the relative importance of internal shocks(sampling fluctuations) and external shocks( the stock dividend sequence) across varying population size is presented.