Corruption Detection Through Textual Analysis: Evidence From Eurozone Banks

Abstract

This research investigates the disclosure of banking institutions by analyzing their annual reports to identify the determinants capable of signaling possible corruption scandals. A textual analysis was conducted on the financial reports of 42 Eurozone banks from the period 2013 to 2022. Drawing on impression management theory, we combine an advanced large language model (LLM) and dictionary approach to extract and analyze the governance-related textual content of the banks in the sample. Machine learning algorithms—including random forests, support vector machines, gradient boosting, and naive Bayes classifiers—and logistic regression have been employed to verify whether disclosure indicators allow for the identification of corruption scandals from a preventive perspective. Our findings show that specific textual measures can be used to analyze the association between disclosure and corruption events and as predictive tools to detect corruption scandals before they become public domain. Our study has several implications, particularly for supervisors and investors who can proactively leverage our findings to identify possible corruption scandals in banks by analyzing their financial disclosures.