Text Mining and Reporting Quality in German Banks - A Cooccurrence and Sentiment Analysis

摘要

A bank's annual risk report intends to reduce the information asymmetry between the bank and its stakeholders. Using automated text mining measures, we assess the quality of the reports in terms of their fulfillment of regulatory requirement and identify its main drivers in a panel regression. On a set of 343 risk reports from 30 German banks between 2002 and 2013, we further perform a cooccurrence and sentiment analysis and determine several additional characteristics of the reports' text. Our methods detect discrepancies for the reports of distressed and non-distressed banks and also for different types of banks. Some of these discrepancies might indicate an intended concealment of certain risks of a bank. We find that our text mining measures explain the variance of the reporting quality to a large extent. The number of words is an important factor for the determination of risk reporting quality. The share of positive words in a report reduces its reporting quality on average.