Portfolio Selection and Optimization through Neural Networks and Markowitz Model: A Case of Pakistan Stock Exchange Listed Companies

Abstract

This paper used artificial neural networks (ANNs) time series predictor for approximating returns of Pakistan Stock Exchange (PSX) listed 100 companies. These projected returns are then substituted into expected returns in the Markowitz’s Mean Variance (MV) portfolio Model. For comparison empirical data used is closing prices of PSX listed stocks, Karachi Inter Bank Offer Rates (KIBOR) as risk free rate and KSE-all share index as benchmark. The Portfolio returns are compared for two datasets by employing various constraints like budget, transaction costs, and turnover constraints. The value of portfolios is measured through Sharpe ratio and Information ratio. Both Sharpe and Information ratios support use of ANNs as return predictor and optimisation tool over simple MV model implemented for empirical data as well as predicted data. ANNs framework performed better in both Long and Short positions and its portfolio returns are significantly higher as compared with MV.