An Artificial Adaptive Speculative Stock Market

The model presented in this paper considers the interaction among traders acting in a financial market. Each agent bases her evaluation of the asset on an (agent-specific) interpretation of an (partially agent-specific) information set. The evaluation process of an agent is simulated by a neural network, which can be interpreted as a nonlinear way to process the available information. It is shown in the paper that, given our assumptions on the structure of the market, each agent learns from her experience and updates her “model of the world” according to the distance between the actual average opinion of the market and her forecast of the same variable. We study the long-run distribution of the behavior of the market, and show the importance that history may have in determining the long-run equilibrium. We also study the performances of various agents in terms of wealth, and show the importance of learning ability. The market shows long periods of calm followed by sudden burst of volatility, when the agents revise their average opinion.