Evaluating China's Central Environmental Inspection Policy Using Machine Learning and Augmented Synthetic Control Methods

To overcome key challenges in environmental policy evaluation we use machine learning based weather normalisation techniques to strip out the effect of weather on air pollution estimates. Combined with Augmented Synthetic Control Methods (ASCM) we provide a causal estimate of the impact of China’s decision to centralise environmental policy enforcement. Focusing on Hebei province we find that the recently introduced Central Environmental Inspection Policy led to a short term reduction in PM2.5 and SO2 immediately after the inspection. However, within 3 months of the inspection team leaving, pollution levels had returned to previous levels. Comparisons with Difference-in-Difference estimations show the importance of both weather normalising and using an ASCM approach, particularly in the absence of parallel pre-trends