Tax Incentives, Financing Constraints, and Labor Structure Upgrading: Evidence from a Quasi-Natural Experiment and Random Forest Interpretation in China

Optimizing labor structure is critical for high-quality development, while tax incentives play a key role in guiding skilled employment. This paper investigates the impact of China's 2018 VAT credit refund policy on corporate labor structure using a quasi-natural experiment based on A-share listed firms from 2013 to 2023. We employ a difference-in-differences (DID) approach to estimate the policy effect and further introduce a machine learning module—Random Forest Regressor (RFR)—to explore non-linear interactions and variable importance related to skilled labor composition. Empirical results show that the VAT credit refund policy significantly increases the proportion of skilled labor, mainly by alleviating firms’ financing constraints. Robustness tests, including placebo, lagged variables, PSM-DID, and policy interference controls, confirm the findings. Heterogeneity analysis reveals that the effects are more pronounced in non-state-owned enterprises, eastern regions, and SMEs. The ML-based interpretability analysis supports these findings by identifying capital input, liquidity, and market valuation as key drivers of labor upgrading. This study offers new evidence on the policy's labor optimization mechanism and demonstrates the value of integrating interpretable machine learning into policy evaluation.