Longitudinal design as an extension to standard banking approaches to companies' financial performance assessment and prediction

Abstract

Financial distress modelling used for loan-approval decision making has received a lot of attention from both practitioners in commercial banks and academics over the last decades. Several modelling techniques accepted by both sides have emerged from the literature review. Surprisingly, such models are usually developed on cross-sectional data with lagged variables. This study extends traditional cross-sectional framework by employing longitudinal design in connection with generalised linear mixed models (logistic regression). Results are summarised by receiver operating characteristic curves. This approach allows direct comparison of techniques with different complexity and can incorporate different cost of misclassification (granted loan to defaulted company and loan rejection to healthy company). Our results obtained on sample of 278 companies suggest that longitudinal approach allows identification of company-specific behaviour. As analysis of receiver operation characteristic shows, this turns in high predictive power of bankruptcy classification on one year predictive horizon.