In this paper we describe a simple model of adaptive agents of different types, represented by Learning Classifier Systems (LCS), which make investment decisions about a risk free bond and a risky asset under a well defined stock market environment. Our main aim is to explore the degree of reliability that artificially intelligent agents can have when applied to real life economic problems. We do this by evaluating whether an LCS is able to represent competent traders in a real market scenario in which daily stock prices and dividends are given to the agents exogenously, so permitting us to focus on the dynamics and evolution of the behaviour of these evolving traders without having to be concerned about how their actions affect the market. We present results of adaptive and non-adaptive simulations over a period of ten years of real data of a specific stock and show that the artificial agents, by displaying different and rich behaviours evolved throughout the simulations, are able to discover and refine novel and successful sets of market strategies that can outperform baseline strategies such as buy-and-hold or merely keeping money in the bank at a good rate of interest, even though the agents pay commission on every trade.
Mendeley saves you time finding and organizing research
Choose a citation style from the tabs below