Profiles for predicting financial distress using company final accounts

Abstract

Lately, financial distress prediction has become an issue of concern for decision makers, especially because of the early warnings of possible bankruptcy yielded by such prediction models. Thus, this study aims to offer a practical solution to predict corporate distress by focusing on the development of an integrated decision support system and on analysing the effectiveness of several techniques, including Decision Trees, Naïve Bayes, as well as Artificial Neural Networks. Signs of business failure are in most cases evident long before official bankruptcy occurs. Through the use of 96 indicators, including financial ratios, industry-related variables and fraud red-flags, this study attempts to analyse financial records in order to predict the sustainability of a given firm. Key profitability ratios were also compared to market averages. Two pattern analysis techniques were used: comparison to the previous year, or comparison to the first year of financial records available; combined with our proposed model of checking both difference and magnitude of change across the comparisons, the study reached an F1 Score of up to 88.7%. The research findings over four real-life datasets confirmed the strength and ability of the proposed model in predicting eminent business failure. Data used spanned from three to five financial years. The dependent variable is the Active or Failed status. The study finds that a model based on previous-year analysis performs better than a model based on base-year analysis. It was also found that in most cases, the inclusion of different aspects of a company's upkeep (profitability, solvency, leverage, management efficiency, industry specifics) lead to more accurate results. Finally, the study can be extended by predicting the degree to which a company is failing. This would be quite advantageous as it would indicate to what extent firms have to go, to improve their financial status.