The Quality of Bankruptcy Data and its Impact on the Evaluation of Prediction Models: Creating and Testing a German Database

Abstract

This study assesses the impact of the quality of bankruptcy data on the estimation and evaluation of bankruptcy prediction models. To meet this objective, we develop a systematic methodology to obtain bankruptcy information from corporate news releases and public sources. Then, applying this methodology to the German market, we create a bankruptcy database that includes a higher number of and more accurate bankruptcy events as well as more accurate bankruptcy dates than those in the frequently used databases of Bureau van Dijk and Compustat Global. We use our bankruptcy data to conduct two empirical analyses. First, using our more accurate database, we compare the performance of several bankruptcy prediction models in the context of Germany. Second, we compare our database with Bureau van Dijk data and find that the quality of bankruptcy data significantly affects the parameter estimates and the out-of-sample evaluation of bankruptcy prediction models. Therefore, we suggest revising evidence presented by bankruptcy studies that use inaccurate bankruptcy information.