Portfolio Optimization Using ANNs and Mean-Semi Variance Markowitz Model: A Comparative Study of South Asian Economies

Abstract

This study uses many portfolio approaches to optimize mean semi-variance portfolios using artificial neural networks for South Asian investors. These methods include the mean-semi variance strategy, minimal variance approach, limited portfolios with transaction costs and turnover constraints, constrained portfolios with risk and return diversification limits, and the equally weighted approach. These portfolio strategies are analysed using the excess Sharpe ratio and the Information Ratio for financial efficiency and diversity, respectively, and validated using the 130/30 portfolio strategy. The Pakistan Stock Exchange (PSX), Bombay Stock Exchange (BSE), Dhaka Stock Exchange (DSE), Columbo Stock Exchange (CSE), and Nepal Stock Exchange (NEPSE) supply daily data for empirical research. We use a large data collection from 2017 to 2021. The study found that ANN-generated estimators using a mean semi-variance optimization strategy outperformed alternative portfolio optimization methods in South Asian markets. The research reveals that ANN-based returns outperform correlation statistics, descriptive analysis, and mean square prediction error (MPSE). The study suggests utilizing naive diversification as a benchmark for other portfolio optimization methodologies.