Abstract:

Bankruptcy prediction models based on the method of linear discriminant analysis (“indexes”) represent a widely used tool for analysing corporate failures or evaluating a Company’s financial health. However, such models are ineffective in alternative economic environments or industries. This is one of the reasons for the necessity of adjusting existing models or creating better new models. There are many papers on creating bankruptcy prediction models, but many of these papers deal with the problem of finding a better set of predictors or finding a better classification algorithm, while limited attention is paid to the problem of setting optimal cut-off scores or grey zone borders. As the grey zone problem can be viewed as a trade-off between a model’s accuracy and the number of companies that remain unevaluated, it represents a factor influencing model accuracy and effectiveness. In this paper we aim to suggest a criterion that can be used for deriving a grey zone that on one hand maximises model accuracy, while on the other hand minimises the number of unevaluated companies.