Nonlinear Pricing Mechanism of China's Public REITs: A Study on Predicting REITs Returns Based on Explainable Machine Learning

To address the research gap where the nonlinear pricing mechanism of China's public REITs market remains unelucidated, this study constructs an explainable machine learning (XAI) analytical framework, innovatively realizing a full-process research of "prediction-attribution-discovery". Based on the daily trading data (2021-2024) of 40 public REITs, the XGBoost algorithm is adopted to build a return prediction model. The pure long-only portfolio (ML_top20) constructed from this model achieved a cumulative return of 41.29% (Sharpe ratio = 0.12) during the out-of-sample testing period (2022-2024), significantly outperforming the market benchmark portfolio which recorded a cumulative return of -9.51% (Sharpe ratio = -0.03). Comprehensive analysis using machine learning interpretation tools reveals that stock market return, momentum effect, Treasury bond yield, and daily price range constitute the core driving factors, while feature interaction and nonlinear effects dominate the REITs price formation mechanism. The findings of this study suggest that a new generation of pricing models integrating interaction factors and nonlinear responses should be developed, providing support for the establishment of cross-market risk early warning mechanisms.