Multi-agent model for threshold constrained portfolio selection

Abstract

This paper presents a multi-agent model for the portfolio selection problem where every selected stock would have at least a specified fraction of the total investment. A system of agents divides the initial wealth and follows individual portfolio adjustment strategies starting with pseudo-random portfolios. Periodically, the agents share information about their performances, and change the composition of the portfolios leveraging experiences reported by other agents. A final threshold constrained portfolio is constructed by consolidating individual portfolios arrived at by the agents based on the past performance of the stocks. The portfolio suggested by the agent based model frequently outperforms the portfolios suggested by mean-variance models when tried out in real market.