Portfolio management using cyberagents

We present an Internet based multi-agent system for portfolio management. This system is designed to assist investment banking firms in serving small investors. In particular, we target mutual fund supermarkets which are essentially funds of funds. There are two types of agents: banker agents and investor agents. Banker agents assist mutual fund managers in devising a global efficient frontier from the individual risk-return characteristics of each of the funds from which a portfolio can be constructed. Each point on this efficient frontier constitutes a particular allocation of resources (i.e. a particular portfolio) among the funds. Investor agents offer personalized advice to each individual investor regarding the choice of a portfolio. Each investor agent gradually learns factors peculiar to the individual served; examples of such factors are risk tolerance, age, income, net worth and investment duration. The banker and investor agents (all implemented in Java) interact with each other over the Internet thereby allowing the investment firm to serve a large number of individual investors with minimal overhead. The various investor agents also learn from each other by swapping experiences in specially designed “forums” hosted by the concerned bank