Incremental information value of alternative data and return forecasting in an emerging crude oil futures market: evidence from China

This study explores whether alternative data provide incremental predictive information for crude oil return forecasting in emerging markets, focusing on China’s Shanghai International Energy Exchange (INE) crude oil futures market. We construct a comprehensive forecasting framework that integrates traditional econometric benchmarks, mixed-frequency specifications, and deep learning models, and evaluate predictive performance through return forecasting, interval prediction for uncertainty quantification, a mean-variance framework for economic value assessment, and interpretable machine learning for information contribution analysis. The results show that deep learning models generally achieve better out-of-sample forecast accuracy and directional predictability than conventional linear benchmarks. Incorporating alternative data provides incremental predictive information beyond traditional predictors and improves the economic performance of forecasting-based portfolios. Furthermore, feature ablation and interpretability analyses reveal that different categories of alternative data provide heterogeneous incremental predictive signals, with their relative importance varying across market conditions. These findings enhance the understanding of how alternative information contributes to return predictability and information processing in the INE market, and offer practical implications for forecasting, risk assessment, and information infrastructure development in this market.