Spot check on corporate disclosure quality: Insights from policy learning

Abstract

The China Securities Regulatory Commission (CSRC) carried out a Spot Check on corporate disclosure quality by random inspection of 5% of listed firms every year since 2016. This paper uses the policy learning method to evaluate and optimize the spot-check policy. Firstly, we employ double machine learning to evaluate the individual treatment effect of the policy. The results demonstrate that the treatment effects of the policy exhibit significant heterogeneity and do not significantly improve the overall corporate disclosure quality of listed firms. This may be because the proportion of firms that need to be randomly inspected is larger than 5% which is specified by the CSRC. Moreover, the limitations of random sampling hinder its ability to precisely identify the firms necessitating inspection, thereby reducing the efficiency of the inspection. Secondly, we apply the policy tree algorithm to optimize the spot-check policy. The improved policy can achieve a significant positive treatment effect, which is far greater than the treatment effect of the current random inspection. Finally, we evaluated the spillover effects of random inspections and determined that the existing random inspection policy does not exert a noteworthy deterrence influence on the peer firms of those inspected. Further optimization of the policy uncovered that the ineffectiveness of random inspections as a deterrent is primarily attributed to a low inspection rate. Specifically, an average inspection rate of at least 28% is required to achieve a substantial deterrent impact on the peer firms of the inspected entities.