Portfolio optimization in dynamic environments using MemSPEAII

Abstract

This paper concerns the problem of portfolio optimization in dynamic environments using multi-objective evolutionary algorithms. Financial markets are characterized by volatility and uncertainty making portfolio optimization a challenging task. A novel multi-objective genetic programming algorithm is proposed, which is a memory enhanced version of the standard SPEAII algorithm. The proposed algorithm employs an explicit memory to store a number of non-dominated solutions. These solutions are reused in the later stages for adapting to the changing environments. A stock ranking based trading simulation is used for fitness evaluation and a probabilistic metric is employed to choose a solution from the Pareto Front. The experiments are performed on the constituent stocks of the S&P BSE FMCG Index. The results, evaluated using RMSE, MEA and cumulative returns, are very promising.