Cooperative Search Using Agents for Cardinality Constrained Portfolio Selection Problem

Authored by Ritesh Kumar, Subir Bhattacharya

Date Published: 2012-11

DOI: 10.1109/tsmcc.2012.2197388

Sponsors: No sponsors listed

Platforms: No platforms listed

Model Documentation: Other Narrative

Model Code URLs: Model code not found

Abstract

This paper presents an agent-based model to select an investment portfolio with a restriction on the number of stocks in it. Daily movements of all the stocks in the market for the past few years are assumed to be available. The scheme deploys a federally structured consortium of agents in the stock market at the start of the historical period. Each agent starts with a pseudorandom portfolio and follows individual investment strategies as it walks through the past data. The agents are designed to emulate some of the characteristics of human investors-adjusting the weights of the stocks based on its own attitude toward risk, occasionally dropping and adding stocks to the portfolio, etc. Periodically, the agents share information about their performances and can switch portfolios. A final cardinality constrained portfolio is constructed by consolidating individual portfolios arrived at by the agents working on the historical data of the stocks. When tested in real markets of the U.K. and Japan, the model suggested portfolios that were quite competitive to, and frequently better than, the portfolios suggested by the mean-variance models.

Tags

Social learning Agent-based systems cooperative search portfolio selection