Robust portfolio selection with smart return prediction

Return prediction and portfolio decision are closely connected steps in forming optimal portfolios. However, existing studies often separate return prediction from portfolio decisions or overlook the high sensitivity of optimal portfolios to estimated returns, leading to poor out -of -sample performance. To address these two issues simultaneously, this paper proposes a robust portfolio framework that optimizes parameters in return prediction by maximizing robust mean-variance utility. The proposed framework effectively combines return prediction and portfolio decisions while mitigating the uncertainty in return prediction. Empirical analysis on S&P500 constituents demonstrates that, compared to the benchmarks, the proposed framework achieves lower risk and higher risk -adjusted returns, originating from more stable weight estimation and diversified portfolios. Further, we uncover that the market friction characteristics considerably contribute to the proposed strategies' performance. Our results suggest that integrating return prediction with the robust mean-variance models can significantly improve portfolio performance.