Dynamics of REIT Returns and Volatility: Analyzing Time-Varying Drivers Through an Explainable Machine Learning Approach

Abstract Real Estate Investment Trust (REIT) returns and volatility have been extensively studied, yet typically in isolation from each other. Given that returns and volatility are generally connected in the eyes of investors, we simultaneously analyze the drivers of REIT returns and volatility over the modern REIT era (1991–2022) using an eXtreme Gradient Boosting (XGBoost) machine learning algorithm. We enhance transparency and utility through the application of explainable artificial intelligence (XAI) techniques, particularly SHapley Additive exPlanations (SHAP) and Accumulated Local Effects (ALE), which unpack the decision-making process of the model. Our analysis reveals that while no single feature consistently dominates, the influence of various drivers fluctuates significantly over time. Notably, the importance of macroeconomic indicators generally diminishes, while REIT-specific characteristics become more influential during the sample period. Furthermore, market cycles (macroeconomic shocks) cause large deviations from otherwise long-run patterns. However, during these times of economic uncertainty, drivers of risk and return correlate more strongly in comparison to times of economic stability. Lastly, we find non-linearities in the way the drivers influence returns and volatility. These insights have significant implications for investors, policymakers, and researchers as they navigate the evolving landscape of real estate investments.