An assessment and comparison of bankruptcy prediction models in forecasting the financial distress of JSE-listed companies over a twenty-year period (2000 to 2020)

Thesis / Dissertation


Permanent link to this Item
Journal Title
Link to Journal
Journal ISSN
Volume Title
This study aimed to test the reliability of various models in predicting the failure of JSE-listed companies. Spanning a period of twenty years between the years 2000 and 2020, a sample of 156 companies was considered, with variables extending across financial information, non- financial information, as well as macroeconomic indicators. The timespan considered for the assessment is particularly significant considering that it encompasses two periods of catastrophic negative market downturns. This includes the 2008 financial crisis and the impacts of the Covid-19 pandemic in 2020. Furthermore, the introduction of new international accounting standards in the latter part of this period, implemented to address issues of risk and transparency in financial reporting, had a marked impact on accounting ratios. Accounting ratios have traditionally been used as inputs in distress prediction models. Considering this context, a study of the comparative performance of various models was undertaken, with the model set including multiple discriminant analysis, logit, probit, recursive partitioning and non-financial models. What was particularly noteworthy in this research was the inclusion of models developed by South African researchers in the model set. In respect of the multiple discriminant analysis, logit and probit models, the results demonstrated a predictive accuracy rate below those surfaced in previous studies, with accuracy rates averaging between 55% and 70%. These models were cumbered predominantly by Type I errors. The application of a model which included both financial and non-financial variables demonstrated more favourable results at an accuracy rate of 73%. The recursive partitioning model, however, which comprehended a high ratio of cashflow- related variables, yielded the highest accuracy, at 83%. The model, with its cash focus and its unique approach of considering the cumulative impact of variables instead of basing the predictive outcomes on the performance of a single financial year, tended not to fall prey to the error-types prevalent in the other models in the model set. The output of this research affirmed the importance of the traditional and rudimentary marker between distressed and non-distressed firms. That is, the abundance of cash or the lack thereof is the key differentiating mark between failure and success. The research also highlighted the importance of considering the cumulative impact of variables when forecasting the failure or success of companies, instead of basing predictive outcomes on the performance of a single period. Furthermore, this research confirmed what had been established in previous studies. That is, the size of the firm is a significant predictor of bankruptcy. This study also attempted to reassess the outcome of distress prediction models, by adjusting for the impact of changes in accounting. The impact on the predictive accuracy of the models by normalising for accounting changes, however, was inconclusive. This was mainly due to the extent of Type I errors across the multiple discriminant analysis, logit and probit models, in conjunction with the low prevalence of failed companies towards the latter part of the period considered in this research when changes in accounting standards were introduced. Further research is required in this area to understand the impact of accounting changes on traditional distress prediction models and potentially to revise these distress models, in order to yield higher predictive accuracies.