Predicting CSI 800 Sector Index Returns in China ——A Multifactor Statistical and Regularized Learning Approach

Abstract

Industry allocation is critical to asset allocation and a major source of excess returns in active investing. This study explores sector return predictability in China with a multifactor framework. We employ monthly data spanning January 2019 to June 2026 for 11 CSI 800 primary sector indices and build 11 predictors, including macroeconomic indicators, Shibor, technical signals and the market factor. We estimate separate industry-level OLS and Lasso regression models, using January 2025–June 2026 as the out-of-sample test window. Static one-shot estimation generates negative out-of-sample R² for both models, a typical outcome in monthly return forecasting. We then implement rolling-window re-estimation with 72-, 60- and 48-month lookback windows to boost forecasts. The 72-month rolling Lasso attains an average out-of-sample R² of −0.07. Its Top-3 industry rotation strategy achieves 65.45% cumulative return and 48.03% excess return relative to the equal-weight benchmark, while rolling OLS yields only 37.46% cumulative return and 20.04% excess return. Given negative predictiveR², we evaluate cross-sectional ranking performance via monthly Spearman Rank IC. The 72-month rolling Lasso delivers a statistically significant mean Rank IC of 0.215 (t=2.96), whereas OLS’s 0.057 (t=0.60) lacks significance. Rolling estimation enables models to adapt to dynamic markets. Lasso demonstrates superior out-of-sample ranking power, consistent with its stronger industry rotation excess returns.