The accuracy and information content of two traditional statistical methods in forecasting bankruptcy: evidence from European countries

This study encompasses a treatment of underlying theory and specifications of various bankruptcy prediction models. It assesses the accuracy and information content of some of these models for publicly listed firms in the European Union. The models used to forecast bankruptcy include Altman (1968)’s Z-score model and the logistic regression model by Ohlson (1980). It is investigated what predictors are best, whether the model performance declines over time and which model does the best job of forecasting bankruptcies. The performance is assessed in an estimation sample (2005-2007) and in a hold-out sample (2011-2013). Model performance is evaluated alongside several criteria. It is found that both models lose a significant amount of predictive accuracy when applied out-of-sample, but they still carry information content. Many tests show that a significant amount of information is left uncaptured by the models. This generalizes through various sets of predictors. The findings are subjected to several benchmarks and robustness tests.