BUILDING A BANKRUPTCY PREDICTION MODEL: COULD INFORMATION ABOUT PAST DEVELOPMENT INCREASE MODEL ACCURACY?

BUILDING A BANKRUPTCY PREDICTION MODEL: COULD INFORMATION ABOUT PAST DEVELOPMENT INCREASE MODEL ACCURACY?

Článek WoS

In most cases, bankruptcy models are based on financial indicators that describe the current condition or a certain area of financial health, such as profitability, indebtedness and so on, but they do not report on relevant past development. The main question of the research presented in this paper is whether information about past development could enhance the prediction accuracy of the bankruptcy prediction model. The aim of our research is to analyse the partial potential of financial indicators describing past development. Given that the threat of company bankruptcy is the result of a long-term process, the question arises as to whether it is possible to enhance the accuracy of a bankruptcy prediction model by using indicators monitoring the development of the company in the past. On a sample of 1,355 small and medium-sized Czech construction companies were taken into account during the period of 2011–2014. The study analysed two types of indicators – basic-form and change-form indicators. Basic-form indicators show the status of an indicator at a specific point in time; change-form indicators represent a modified base index of the basic-form ratio. The authors derived six different models for the purpose of comparing the two types of indicators. The authors used the method of stepwise discriminant analysis, both forward selection and backward elimination, to create the models. The accuracies of the resultant models were analysed using the methods of ROC curves and the Area Under Curve (AUC). The authors found that the model based solely on change-form indicators is not superior to the model based solely on basic-form indicators. However, the model using both types of indicators achieved a higher AUC in comparison with the models created with only one type of indicator.

In most cases, bankruptcy models are based on financial indicators that describe the current condition or a certain area of financial health, such as profitability, indebtedness and so on, but they do not report on relevant past development. The main question of the research presented in this paper is whether information about past development could enhance the prediction accuracy of the bankruptcy prediction model. The aim of our research is to analyse the partial potential of financial indicators describing past development. Given that the threat of company bankruptcy is the result of a long-term process, the question arises as to whether it is possible to enhance the accuracy of a bankruptcy prediction model by using indicators monitoring the development of the company in the past. On a sample of 1,355 small and medium-sized Czech construction companies were taken into account during the period of 2011–2014. The study analysed two types of indicators – basic-form and change-form indicators. Basic-form indicators show the status of an indicator at a specific point in time; change-form indicators represent a modified base index of the basic-form ratio. The authors derived six different models for the purpose of comparing the two types of indicators. The authors used the method of stepwise discriminant analysis, both forward selection and backward elimination, to create the models. The accuracies of the resultant models were analysed using the methods of ROC curves and the Area Under Curve (AUC). The authors found that the model based solely on change-form indicators is not superior to the model based solely on basic-form indicators. However, the model using both types of indicators achieved a higher AUC in comparison with the models created with only one type of indicator.

construction companies;bankruptcy prediction;dynamic indicators;model accuracy;multi-period transformation

construction companies;bankruptcy prediction;dynamic indicators;model accuracy;multi-period transformation

KARAS, M.; REŽŇÁKOVÁ, M.

2019

30.06.2018

2081-7452

Polish Journal of Management Studies

17

1

Polská republika

116

130

15

https://pjms.zim.pcz.pl/resources/html/article/details?id=174929