Predicting Corporate Bankruptcy: An Evaluation of Alternative Statistical Frameworks

Abstract

Corporate bankruptcy prediction has attracted significant research attention from business academics, regulators and financial economists over the past five decades. However, much of this literature has relied on quite simplistic classifiers such as logistic regression and linear discriminant analysis (LDA). Based on a large sample of US corporate bankruptcies, we examine the predictive performance of 16 classifiers, ranging from the most restrictive classifiers (such as logit, probit and linear discriminant analysis) to more advanced techniques such as neural networks, support vector machines (SVMs) and "new age" statistical learning models including generalised boosting, AdaBoost and random forests. Consistent with the findings of Jones et al. (2015), we show that quite simple classifiers such as logit and LDA perform reasonably well in bankruptcy prediction. However, we recommend the use of "new age" classifiers in corporate bankruptcy modelling because: (1) they predict significantly better than all other classifiers on both the cross-sectional and longitudinal test samples; (2) the models may have considerable practical appeal because they are relatively easy to estimate and implement (for instance, they require minimal researcher intervention for data preparation, variable selection and model architecture specification); and (3) while the underlying model structures can be very complex, we demonstrate that "new age" classifiers have a reasonably good level of interpretability through such metrics as relative variable importances (RVIs).

Financial Risk Measurement and Prediction Modelling for Sustainable Development of Business Entities Using Regression Analysis

“…The reviews show that there are two groups of popular and promising tools within the bankruptcy prediction models research area, i.e., statistical tools (multiple discriminant analysis and logistic regression) and artificial intelligence tools (decision trees, neural networks, etc.). In this study, we test the use of a quite simple classifier, linear regression approach (similar to Guo et al [27]), for modelling the relationship between a scalar dependent variable and more explanatory variables (financial indicators) as it performs reasonably well in bankruptcy prediction, as proved by Jones et al [28]. Regression analysis if often use for bankruptcy prediction, the realized analysis is supported by the study of Calabrese et al [29] or latest researches in Romania [30] and Lithuania [31], which recommend regression models for bankruptcy prediction.…”

A machine learning analysis of citation impact among selected Pacific Basin journals

“…Gradient boosting machines are used extensively in the literature and are among the most powerful methods available in terms of predictive performance (see Hastie et al, 2009). Jones et al (2015Jones et al ( , 2017 provide evidence that the gradient boosting model and related machine learnings methods, including AdaBoost and random forests, strongly outperform conventional classifiers such as logit and probit (Hastie et al, 2009). TreeNet Ò and related machine learning methods have numerous advantages over classical parametric models such as OLS regression and logit.…”

Financial Bankruptcy across European Countries

“…As mentioned above, recent artificially intelligence expert system models would lightly outperform discriminant and logistic analysis (see, among others, Behr & Weinblat, 2016;Jones et al, 2017) but they are based on very complex underlying model structures; moreover, because of the strong imbalance of the data sets with regard to the bankruptcy status, standard implementations have to be modified to allow the estimation of realistic default propensities. Therefore, more advanced techniques would lightly outperform at the cost of artificial balancing of the data set with respect to the insolvency status, which would suggest some caution with the interpretability of the results.…”

“…Several types of AIES models have been implemented such as recursively partitioned decision trees, known as "random forest" approach (Behr & Weinblat, 2016), case-based reasoning models (Kolodner, 1993), neural networks (Odom & Sharda, 1990;Kim & Kang, 2010), genetic algorithms (Varetto, 1998;Shin & Lee, 2002) or rough sets model (Dimitras et al, 1999). As argued by Jones et al (2017), these more advanced techniques would lightly outperform discriminant and logistic analysis but at the cost of artificial balancing of the data set with respect to the insolvency status, which would suggest some caution with the interpretability of the results.…”