An Analysis of the Effectiveness of Bankruptcy Prediction Models – an Industry Approach

Abstract

Research background: Bankruptcy prediction models are frequently used in research. However, an industry approach is not often carried out. Due to this, this study included trends observable between the number of bankruptcies and its prediction by models. Purpose: The aim of the paper is to verify if changes in the number of actual bankruptcy in individual industries are properly predicted by the models. Also, if analyzed models are providing consistent information according to the risk of bankruptcy between industries. Research methodology: The data were collected from the Orbis database and the Coface reports. The period included in the study is 2014–2019. 5 Polish bankruptcy prediction models were used: these by D. Hadasik, E. Mączyńska and M. Zawadzki, M. Pogodzińska and S. Sojak, D. Wierzba and the Poznan one. Results: The analyzed models do not properly predict changes in the number of bankruptcy in individual industries, however, 3 out of 5 correctly predicted the trend for the entire sample. Analyzed models often provide inconsistent information. Hence, it seems sensible to use more than a few models in any further analyzes. Novelty: In the literature of the subject, there are often carried out analyses focused on the effectiveness of bankruptcy prediction models regarding individual companies. This research is focused on the prediction of changes in the number of companies to be considered as at bankruptcy risk between industries, and also on comparing these models.

A Financial Distress Prediction using a Non-stationary Dataset

“…The majority of FDP studies rely on private datasets, such as Compustat 2 : database containing fundamental financial and price data for active and inactive : database which has information on close to 450 million companies and entities across the globe [Succurro 2017, Pilch 2021]. These datasets often contain extensive information that could be organized chronologically, such as by year, semester or quarter.…”

“…Despite the interest of academics and practitioners in the topic [Kumbure et al 2022], the availability of datasets remains a barrier. The majority of the datasets are not publicly available [Barboza et al 2017, Barboza et al 2022, Bragoli et al 2022, Zou et al 2022, Chen et al 2022, Pilch 2021, while just few are public [Lombardo et al 2022, Liang et al 2016, Zieba et al 2016, Tang et al 2019. Additionally, most of these datasets have considered the data as stationary and have overlooked the time dependence aspect, which has been addressed in recent studies [Sun et al 2019, Shen et al 2020, Kim et al 2022.…”

A Binary Logistic Regression Model for Support Decision Making in Criminal Justice

“…In a developed smart society in all spheres of human activity, any systemic, balanced, gradual and planned changes are impossible without prior risk assessment and forecasting the future state of the system. The most effective means for this are the tools of mathematical and statistical methods and information technology (Dmytrów, Gdakowicz, Putek-Szeląg, 2019;Stanimir, 2020a;Stanimir, 2020b;Szkutnik, 2020;Bąk, Wawrzyniak, Sobolewski, 2020;Pilch, 2021;Berezka, Berezsky, Masliy, 2013;Kovalchuk, Shynkaryk, Masonkova, 2021). In many developed countries with high crime rates, such as the United States, statistical methods and information technologies such as artificial intelligence and big data are already used to support optimal decision-making in criminal justice.…”