Incremental vs. Stock: Evaluating the Impact of VAT Credit Refunds on Corporate Productivity with Evidence from Difference-in-Differences and Machine Learning Heterogeneity Analysis

This paper re-examines the effects of incremental and stock value-added tax (VAT) credit refunds on enterprise total factor productivity (TFP). Based on updated panel data from Shanghai and Shenzhen listed companies covering both the incremental-refund window and the stock-refund window, a difference-in-differences (DID) model is used to identify average productivity effects. The results show that both refund policies significantly improve corporate TFP, and the updated estimates support the interpretation that tax refunds operate through R&D investment, financing constraints, productive investment, and firm-level absorptive capacity. The effects vary across regions and firms: fiscal pressure, marketization, profitability, growth, ownership, and innovation-related capacity help explain heterogeneous responses. To connect DID estimates with firm-level heterogeneity, Logistic Regression, Random Forest, and XGBoost models are further applied. Feature-importance and SHAP analyses show that ROA, ownership, growth, leverage, capital expenditure, financing constraints, and intangible assets are key predictors of policy response. The findings provide evidence for precision-oriented fiscal policy and are relevant to capital-intensive advanced manufacturing sectors, including optical communication and electromagnetic equipment industries.