Local government debt risk assessment: A deep learning-based perspective

Abstract

The research assesses local government debt risks in China with deep learning methods. We perform natural language processing and sentiment classification on all publicly available prefecture-level governments’ annual work reports from the previous three years. Then, for each of these cities, we calculated sentiment scores related to debt risks and examined the regional distribution of risks. Our empirical findings indicate that special attention should be paid to China's inland areas, where local government debt risks are highly concentrated. This paper extends the existing literature on discourse analysis with quantitative methods to the research of political economy.