关键词: 情感回避, 企业年报, 大语言模型, 违约风险

Abstract: There is an asymmetric emotional avoidance phenomenon in the Management Discussion and Analysis (MD&A) of corporate annual reports, which is regarded as abnormal in linguistics. Starting from linguistics, this study constructs an emotional avoidance index through large language models and explores the impact of emotional avoidance in the MD&A of Chinese listed companies on future default risks. Through methods such as regression analysis and instrumental variables, this study finds that emotional avoidance will increase the company's future default risks. Mechanism analysis shows that emotional avoidance affects corporate default risks by increasing financing constraints and causing analysts to underestimate. In addition, the study finds that emotional avoidance stems from the management's motivation for information manipulation, further indicating that emotional avoidance is the result of information manipulation. The findings of this paper enrich the regulatory methods for information disclosure of listed companies and provide new risk early warning tools for investors and regulators.

Key words: Emotional Avoidance, Annual Report, Large Language Model (LLM), Default risk