Non-linear factor model for asset selection using multi objective genetic programming

Investors vary with respect to their expected return and aversion to associated risk, and hence also vary in their performance expectations of the stock market portfolios they hold. In this work we present an empirical study of the use of the Multiobjective genetic programming (MOGP) technique for a real world problem of portfolio optimisation on UK FTSE100 stocks. The MOGP evolves a nonlinear factor model of technical factors for asset ranking, and provides visual insight into the risk return trade-off involved in discovering an approximation to the risk-return efficient frontier of a portfolio optimization problem. We provide preliminary analysis of the set of factors chosen in the evolved solutions. We find evidence for the effect on risk-adjusted returns of firm size, return on equity and cash yield (and little or no evidence for the book to market ratio). Categories and Subject Descriptors I.2.M [Artificial Intelligence]: Miscellaneous General Terms Algorithms, Experimentati...

• Type: Conference
• Year: 2008
• Where: GECCO
• Authors: Ghada Hassan
• Keywords: Factor Models, GECCO 2008, Multiobjective Genetic Programming, Optimization, Portfolio Optimization