Chinese stock return predictability: A time–frequency and shrinkage modeling approach

Accurate identification of dominant predictors remains pivotal for Chinese stock return forecasting amidst complex multi-timescale interactions. Utilizing 2010–2025 monthly data from 42 predictors across financial macroeconomic and technical, groups, we develop a multi-timescale framework combining wavelet decomposition with group-shrinkage models. Key findings are threefold. First, models with exogenous predictors show time-varying performance, and quantile-based regularization consistently improves accuracy. Second, predictive power is heterogeneous: MKT, ERP, HSCE, and SMB remain strong, whereas technical indicators are unstable. Third, across time scales, short-term components dominate, and group-shrinkage of long-term decomposed predictors further enhances portfolio outcomes. Methodologically, quantile-based regularization techniques enhance model robustness under extreme market conditions. Notably, the substantial economic gains further validate the superior forecasting performance of our comprehensive forecasting framework. These findings offer actionable guidance for policymakers and market participants by highlighting the value of integrating financial and macroeconomic predictors' short-to-medium-term dynamics into stock return forecasting and early-warning systems.