Explore publications.

This study investigates portfolio optimization within the Mean–Variance framework using alternative estimators for expected returns and risk. Based on daily data of VN30 stocks from March 2021 to December 2025, nine portfolio strategies are constructed by combining three return estimators (historical mean, CAPM, and Exponential Moving Average) with three risk estimators (sample covariance, Ledoit–Wolf shrinkage covariance, and semicovariance). Portfolios are optimized by maximizing the Sharpe ratio under realistic weight constraints and a post-optimization sparsification procedure to improve implementability. Empirical results indicate that portfolios using EMA-based expected returns combined with Ledoit–Wolf or semicovariance estimators deliver more stable and superior out-of-sample performance over 1-, 3-, and 6-month horizons. In contrast, strategies relying on historical mean returns with sample covariance suffer from severe performance degradation, while CAPM-based approaches underperform consistently. These findings underscore the importance of adaptive return estimation and robust risk modeling in emerging equity markets and offer timely implications for portfolio construction in Vietnam as it prepares for a potential upgrade to emerging market status.