Exploiting mixed-frequency characteristics in parametric Mean-Expected Shortfall portfolio selection

In this paper, we investigate the role of mixed-frequency characteristics in portfolio selection. We introduce a novel parametric Mean-Expected Shortfall model which establishes direct linkages between portfolio weights and mixed-frequency characteristics. The model solution is converted into a penalized MIDAS expectile regression problem, which not only reduces computational costs, but also identifies the crucial characteristics. An empirical analysis of the CSI 300 index reveals that incorporating mixed-frequency characteristics significantly enhances portfolio performance. The proposed method consistently outperforms other competing models by delivering substantially lower risk and markedly higher risk-adjusted returns. Further coefficient analysis highlights crucial characteristics exerting significant impacts on portfolio weights. Specifically, accumulation distribution and market value demonstrate positive influences, while moving averages, relative strength index, liquidity, and book-to-market ratio exhibit negative impacts. These findings provide investors with valuable tool for asset allocation, enhancing interpretability and reliability in complex data environments.