Randomness, imitation or reason explain agents' behaviour into an artificial stock market?

Abstract

This paper shows that the effects of the interaction between simple artificial agents in a well defined economic environments, such as an artificial stock market, are useful to understand microeconomic events. We employ here the connectionist cross-target (CT) method [4] to build artificial neural subjects that make (i) guesses about their own actions and (ii) guesses about the effects of those actions. Each subject, learning and acting, develops the coherence between the two types of guesses. Our artificial interacting agents buy and sell shares, following cyclical behaviour (buying when the price rises and selling when the price diminishes) or developing risk aversion or true anti cyclical strategies. The last complex strategy can emerge from imitation between agents and randomness instead of reason.