An End-to-End Regime-Dependent Industry Rotation Strategy in China’s A-Share Market

This paper develops an integrated framework for industry rotation in China’s A-share market by combining industry-level factor construction, deep-factor extraction, market-regime prediction, directional classification, and regime-dependent risk parity. Industry signals are first formed from Alpha-style predictors aggregated from representative constituent stocks and then enriched by a feed-forward neural network that learns latent deep factors. The market is subsequently classified into four interpretable states defined by volatility and rotation speed, and next-period regime probabilities are forecast with XGBoost. These probabilities affect both the directional signal layer and the covariance structure used in the allocation layer. Out-of-sample evidence for 2023-2024 shows that the full framework delivers stronger return stability, a higher Sharpe ratio, and better drawdown control than simpler equal-weight or non-regime alternatives. The empirical results indicate that regime information is valuable not only for return prediction, but also for dynamic portfolio risk allocation.