Study on the impact mechanism of agricultural e-commerce subsidies on energy efficiency based on IV-2SLS and Random forest

In the context of digital-economy and agricultural-modernization integration, rural e-commerce subsidy policies are an important lever for energy-efficiency optimization. Currently, agricultural e-commerce may reduce unit output energy intensity via standardized logistics and clean technology, but cold-chain equipment popularization increases power consumption, with a non-linear action mechanism. Existing studies use linear regression, which struggles with endogenous variables and high-dimensional data. This study innovatively combines the IV-2SLS and random forest (RF) algorithm. Using 2020–2024 panel data of 28 provinces and terrain complexity and historical infrastructure as exogenous variables, it analyzes subsidy policies' causal effects on energy efficiency. Experimental results show the instrumental variable method overcomes policy endogeneity, passing the weak instrumental variable test. The RF model identifies key mediating variables. Heterogeneity analysis reveals that a 10% increase in cold-chain Subsidies boosts photovoltaic penetration by 12.6% but raises refrigerated storage energy consumption by 8.3%, with net energy-efficiency gains showing an east-high, west-low regional difference. The study also finds that the subsidy policy forms a technology diffusion channel through the agricultural machinery service outsourcing market, increasing small and medium-sized farmers' clean-energy equipment adoption by 23.7%. However, more logistics nodes increase diesel transportation equipment by 15.4%, showing an energy rebound in the green transformation.