The impact of corporate governance ‘red flags’ on the probability of corporate failure.

No Thumbnail Available
Issue Date
Journal Title
Journal ISSN
Volume Title
The purpose of this study is two-fold. Firstly, this study aims to introduce a new perspective to the domain of bankruptcy and financial distress prediction. A data-set of European several European firms is constructed and processed using machine learning techniques. Subsequently, a logistic regression analysis is performed in order to construct a benchmark-model. The logistic regression provides a preliminary indication that corporate governance variables may be relevant non-financial predictors. What is more, the significance of the financial predictors deflated when macro-economic and governance variables were introduced. Hence, it is suggested that future research into bankruptcy and financial distress prediction should not merely focus on firm-specific accounting ratios, save an adequate proxy for firm size. The results indicate that corporate governance variables may form more consistent predictors as they describe more structural problems within a firm. These problems may relate to lack of monitoring capabilities, which may in turn induce opportunistic behavior by management. The main explanation provided within this study is that failing corporate governance may induce performance-harmful behavior by management. However, the nature of corporate failure is complex and may be characterized by non-linearities. The second purpose of this study to introduce machine learning models to the standard tool-kit within economic research. Therefore, the second (methodological) aim of this research is to show fellow researchers and students the value of machine learning in the area of accounting and finance research. In this respect, this study does balance both the advantages and limitations of machine learning. Although these models provide sophisticated methods to explain additional residual variance, they also are characterized by their lack of interpretability. To unfold the ‘black-box’ within machine learning, this study provides a method for analyzing the relative importance of each input feature. Relative feature importances allow this study to compare the pattern recognition behavior within machine learning models with the parameters estimated under logistic regression. This study observes that corporate governance become even more so relevant within the pattern recognition behavior of machine learning models. The emphasis on governance features seem also to have a ‘stabilizing’ effect on the discriminative power of the model to correctly classify the bankruptcy/ distressed class into the right category.  
Faculteit der Managementwetenschappen