Enhancing portfolio optimization in emerging markets: A cross-validation multi-target shrinkage approach

Abstract

This study formulates a novel portfolio optimization framework for emerging markets through the integration of cross-validation with a multi-target shrinkage estimator (CV-MTSE). The proposed method adaptively combines the sample covariance matrix with two structured targets, the Single Index Model and the Identity Matrix. Shrinkage intensities are optimized through a grid search-based cross-validation procedure. Using Vietnamese stock market data from 2013 to 2023, we compare CV-MTSE with traditional estimators such as SCM and equal-weighted. Empirical results demonstrate that CV-MTSE consistently achieves higher risk-adjusted returns and lower volatility particularly during stable market conditions. During periods of market stress, the equal-weighted MTSE model shows stronger robustness in term of volatility. These findings contributes to the literature on covariance matrix estimation and also has practical applications in portfolio management in emerging markets.