Predicting the outcome following bankruptcy filing: a three-state classification using neural networks

This paper uses artificial neural networks (ANNs), multi-state ordered logit and nonparametric multiple discriminant analysis (NPDA) for predicting the three-state outcome of bankruptcy filing. The study compares the classification accuracy of these procedures. It differs from previous studies on predicting financial distress by focusing on the firm after the filing of bankruptcy using accounting data, market data, and court-related information. Following the filing and through court approval the bankruptcy is resolved as firms are either acquired by other firms, emerging as independent operating entities, or liquidated. Distinguishing this three-state outcome is more complex than discriminating between healthy and financially distressed firms. Models suggested in previous studies for predicting the two-group financial distress perform poorly for our three-state scenario. Therefore, we develop models which focus on characteristics relevant for the bankruptcy resolution. We use a sample of 237 publicly traded firms which have complete data. For the entire sample and estimation samples, ANNs provide significantly better three-state classification than logit and NPDA. However, for some holdout samples the differences in classification accuracies are statistically insignificant. © 1997 John Wiley & Sons, Ltd.     

A Note on Selecting a Response Measure for Financial Distress

Since 1966, researchers have examined financial distress prediction models to determine the usefulness of accounting information to lenders. These researchers primarily used legal bankruptcy as the response variable for economic financial distress, or included legal bankruptcy with other events in dichotomous prediction models. However, theoretical models of financial distress normally define financial distress as an economic event, the inability to pay debts when due (insolvency). This study uses a loan default/accommodation response variable as a proxy for the inability to pay debts when due. The purpose of this note is to empirically test whether or not using the inability of a firm to pay debts when due, loan default/accommodation, as a response measure produces different results than using legal bankruptcy as the response measure. The study's empirical results show that legal bankruptcy and loan default/accommodation financial distress prediction models produce different statistical results, thus suggesting that the responses measure different constructs. A loan default/accommodation model also fits the data better than a bankrupt model. Our results suggest that a loan default/accommodation response may be a more appropriate measure to determine which accounting information is most useful to lenders in evaluating a firm's credit risk.     

Bankruptcy prediction using fuzzy convolutional neural networks

We propose a combined method for bankruptcy prediction based on fuzzy set qualitative comparative analysis (fsQCA) and convolutional neural networks (CNN). Currently, CNNs are being applied to various fields, and in some areas are providing higher performance than traditional models. In our proposed method, a CNN uses calibrated variables from fuzzy sets to improve performance accuracy. In addition, there are no published studies on the effect of feature selection at the input level of convolutional neural networks. Therefore, this study compares four well-known feature selection methods used in financial distress prediction, (t-test, stepdisc discriminant analysis, stepwise logistic regression and partial least square discriminant analysis) to investigate their effect on classification performance. The results show that fuzzy convolutional neural networks (FCNN) lead to better performance than when using traditional methods.     

Predicting Bankruptcy Resolution

This study examines classification and prediction of the bankruptcy resolution event. Filing of bankruptcy is resolved through one of three alternative resolutions: acquisition, emergence or liquidation. Predicting the final bankruptcy resolution has not been examined in the prior accounting and finance literature. This post-bankruptcy classification and prediction of the final resolution is harder than discriminating between healthy and bankrupt firms because all filing firms are already in financial distress. Motivation for predicting the final resolution is developed and enhanced. A sample of 237 firms filing for bankruptcy is used. Classification and prediction accuracies are determined using a logit model. A ten-variable, three-group resolution logit model, which includes five accounting and five non-accounting variables is developed. The model correctly classifies 62 percent of the firms, significantly better than a random classification. We conclude that non-accounting data add relevant information to financial accounting data for predicting post bankruptcy resolution. Further, public policy implications for investors, researchers, bankruptcy judges, claimants and other stakeholders are discussed.     

The Limitations of Bankruptcy Prediction Models: Some Cautions for the Researcher

The purpose of this study is to demonstrate potential problems associated with the use of bankruptcy prediction models in current research. The tests in this study demonstrate the problems that may arise when bankruptcy prediction models are inappropriately applied. This analysis evaluated the Zmijewski (1984) and Ohlson (1980) models using time periods, industries, and financial distress situations other than those used to originally develop the models. The findings indicated that both models were sensitive to time periods. That is, the accuracy of the models declined when applied to time periods different from those used to develop the models. The findings also suggest that the accuracy of each model continues to decline moving from the 1988–1991 to the 1992–1999 sample period. Additionally, Ohlson's (Zmijewski's) model was (was not) sensitive to industry classifications. The findings of this study also suggest that the Ohlson and Zmijewski models are not sensitive to financial distress situations other than those used to develop the models. Thus, the models appear to be more generally useful for predicting financial distress, not just bankruptcy.In sum, the results of this study suggest that researchers should use bankruptcy prediction models cautiously. Applying the models to time periods and industries other than those used to develop the models may result in a significant decline in the models' accuracies. Additionally, some bankruptcy prediction models may be more appropriate for evaluating various forms of financial distress as opposed to just bankruptcy. To avoid erroneous applications of bankruptcy prediction models in the future, it is necessary for researchers not only to understand the uses of prediction models, but also to understand the limitations of the models.     

A horse race of models and estimation methods for predicting bankruptcy

We use a comprehensive set of performance metrics to analyze the improvement in the classification power and prediction accuracy of various bankruptcy prediction models after adding governance variables and/or varying the estimation method used. In a sample covering bankruptcies of U.S. public firms in the period 2000 to 2015, we find that the addition of governance variables significantly improves the performance of all bankruptcy prediction models. We also find that the additional explanatory power provided by governance measures improves the further the firm is from bankruptcy, which suggests that governance variables may provide earlier and more accurate warning of the firm's bankruptcy potential. Our findings show that the performance of any bankruptcy prediction model is significantly affected by the estimation method used. We find that regardless of the bankruptcy model, hazard analysis provides the best classification and out-of-sample forecast accuracy among the parametric methods. Furthermore, non-parametric methods such as neural networks, data envelopment analysis or classification and regression trees appear to provide comparable and sometimes superior classification accuracy to hazard analysis. Lastly, we use the dynamic panel generalized methods of moments model to address concerns raised in prior studies about the susceptibility of similar studies to endogeneity issues and find that our findings continue to hold.     

Empirical comparison of hazard models in predicting SMEs failure

This study aims to shed light on the debate concerning the choice between discrete-time and continuous-time hazard models in making bankruptcy or any binary prediction using interval censored data. Building on the theoretical suggestions from various disciplines, we empirically compare widely used discrete-time hazard models (with logit and clog-log links) and the continuous-time Cox Proportional Hazards (CPH) model in predicting bankruptcy and financial distress of the United States Small and Medium-sized Enterprises (SMEs). Consistent with the theoretical arguments, we report that discrete-time hazard models are superior to the continuous-time CPH model in making binary predictions using interval censored data. Moreover, hazard models developed using a failure definition based jointly on bankruptcy laws and firms’ financial health exhibit superior goodness of fit and classification measures, in comparison to models that employ a failure definition based either on bankruptcy laws or firms’ financial health alone.     

THE SIGNIFICANCE OF BASE YEAR IN DEVELOPING FAILURE PREDICTION MODELS

Most failure prediction models developed over the past fifteen years have been based on firms' financial characteristics one year before they collapsed. This article reports the results of a study employing multiple discriminant analysis on a sample of fifty-three British companies which failed over the period 1960–1971. Models were developed on various financial ratios and an industry indicator for each of the five years, prior to failure, and the results in fact show that the model developed from data five years before failure performed at least as well as that derived one year prior to bankruptcy.     

Bankruptcy prediction of financially stressed firms: an examination of the predictive accuracy of artificial neural networks

This is an extension of prior studies that have used artificial neural networks to predict bankruptcy. The incremental contribution of this study is threefold. First, we use only financially stressed firms in our control sample. This enables the models to more closely approximate the actual decision processes of auditors and other interested parties. Second, we develop a more parsimonious model using qualitative ‘bad news’ variables that prior research indicates measure financial distress. Past research has focused on the ‘usefulness’ of accounting numbers and therefore often ignored non‐accounting variables that may contribute to the classification accuracy of the distress prediction models. In addition, rather than use multiple financial ratios, we include a single variable of financial distress using the Zmijewski distress score that incorporates ratios measuring profitability, liquidity, and solvency. Finally, we develop and test a genetic algorithm neural network model. We examine its predictive ability to that of a backpropagation neural network and a model using multiple discriminant analysis. The results indicate that the misclassification cost of the genetic algorithm‐based neural network was the lowest among the models. Copyright © 2001 John Wiley & Sons, Ltd.     

Corporate bankruptcy prediction in Japan

The purpose of this study is to highlight the financial characteristics of failed firms in Japan, and to construct corporate bankruptcy prediction models with greater prediction accuracy. Our principal component analysis indicated that failed firms in Japan could be classified into two groups: a group having negative financial structures and a group having a declining flow of funds. Additionally, they can be classified into two other different categories of groups: one whose financial position during three years before shows a ‘V’ shape and another group that shows a ‘XXX’ shape.Our discriminant analysis indicated that improved prediction accuracy could be obtained by using, as predictor variables, both ratios and absolute amounts based on cash base financial statement data three years before failure. This data was adjusted to properly reflect the exceptions, reservations, and qualifications appearing in the audit reports and those based on accrual base financial statement data.     

Features selection,data mining and finacial risk classification: a comparative study

The aim of this paper is to compare several predictive models that combine features selection techniques with data mining classifiers in the context of credit risk assessment in terms of accuracy, sensitivity and specificity statistics. The t‐statistic, Battacharrayia statistic, the area between the receiver operating characteristic, Wilcoxon statistic, relative entropy, and genetic algorithms were used for the features selection task. The selected features are used to train the support vector machine (SVM) classifier, backpropagation neural network, radial basis function neural network, linear discriminant analysis and naive Bayes classifier. Results from three datasets using a 10‐fold cross‐validation technique showed that the SVM provides the best accuracy under all features selections techniques adopted in the study for all three datasets. Therefore, the SVM is an attractive classifier to be used in real applications for bankruptcy prediction in corporate finance and financial risk management in financial institutions. In addition, we found that our best results are superior to earlier studies on the same datasets.     

系统性风险与企业财务危机预警——基于前沿机器学习的新视角

本文采用Logit回归模型以及随机森林模型、梯度提升模型等前沿机器学习方法,深入考察系统性风险指标对我国企业财务危机的预测能力。结果表明,系统性风险对中下游企业的财务危机具有显著的预测能力,而基于因子分析构建的系统性风险指标,结合随机森林模型可取得更好的预测效果。本文进一步区分财务危机的不同成因并发现,基于随机森林模型和Logit回归模型的预测框架能够对我国大多数财务危机事件进行有效预警。在此基础上,本文对我国上市企业监管提出相关建议,从而为完善金融风险处置机制提供一定参考。     

THE IMPLICATIONS OF DISTRESS PREDICTION MODELS FOR CORPORATE LENDING

Three literatures on financial distress prediction are reviewed and evaluated in terms of their usefulness in credit analysis. These are (1) bankruptcy prediction models initiated by Altman, (2) behavioural studies of the ability of loan officers and others to make accurate predictions of bankruptcy and (3) attempts to simulate loan officers' judgements. It is concluded that techniques based on judgement simulation are likely to be the most effective decision aids.     

Factors Affecting the Probability of Bankruptcy: A Managerial Decision Based Approach

The majority of classification models developed have used a pool of financial ratios combined with statistical variable selection techniques to maximize the accuracy of the classifier constructed. Rather than follow this approach, this article seeks to provide an explicit economic basis for the selection of variables for inclusion in bankruptcy models. This search to develop an economic theory of bankruptcy augments the existing bankruptcy prediction literature. Variables which occur in bankruptcy probability expressions derived from the solution of a stochastic optimizing model of firm behaviour are 'proxied' by variables constructed from financial statement data. The random nature of the lifetime of a single firm provides the rationale for the use of duration or hazard-based statistical methods in the validation of the derived bankruptcy probability expressions. Results of the validation exercise confirm that the majority of variables included in the empirical hazard formulation behave in a way that is consistent with the model of the firm. The results highlight the need for developments in the measurement of earnings dispersion.     

PROBABILISTIC APPROACHES FOR CREDIT SCREENING AND BANKRUPTCY PREDICTION

Three probabilistic neural network approaches are used for credit screening and bankruptcy prediction: a logistic regression neural network (LRNN), a probabilistic neural network (PNN) and a semi‐supervised expectation maximization‐based neural network. Using real‐world bankruptcy prediction and credit screening datasets, we compare the three probabilistic approaches using various performance criteria of sensitivity, specificity, accuracy, decile lift and area under receiver operating characteristics (ROC) curves. The results of our experiments indicate that the PNN outperforms the other two techniques for decile lift and specificity performance metric. Using the area under ROC curve, we find that for bankruptcy prediction data the PNN outperforms the other two approaches when false positive rates (FPRs) are less than 40 %. LRNN outperforms the other two techniques for FPRs higher than 40 % for bankruptcy data. We observe that the LRNN results are very sensitive to the ratio of examples belonging to two classes in training data and there is a tendency to overfit training data. Copyright © 2012 John Wiley & Sons, Ltd.     

A comparison of alternative bankruptcy prediction models

Early models of bankruptcy prediction employed financial ratios drawn from pre-bankruptcy financial statements and performed well both in-sample and out-of-sample. Since then there has been an ongoing effort in the literature to develop models with even greater predictive performance. A significant innovation in the literature was the introduction into bankruptcy prediction models of capital market data such as excess stock returns and stock return volatility, along with the application of the Black–Scholes–Merton option-pricing model. In this note, we test five key bankruptcy models from the literature using an up-to-date data set and find that they each contain unique information regarding the probability of bankruptcy but that their performance varies over time. We build a new model comprising key variables from each of the five models and add a new variable that proxies for the degree of diversification within the firm. The degree of diversification is shown to be negatively associated with the risk of bankruptcy. This more general model outperforms the existing models in a variety of in-sample and out-of-sample tests.     

Bankruptcy probability changes and the differential informativeness of bond upgrades and downgrades

Prior studies have found that stock returns around announcements of bond upgrades are insignificant, but that stock prices respond negatively to announcements of bond downgrades. This asymmetric stock market reaction suggests either that bond downgrades are timelier than upgrades, or that voluntary disclosures by managers preempt upgrades but not downgrades. This study investigates these conjectures by examining changes in firms’ probabilities of bankruptcy (assessed using bankruptcy prediction models) and voluntary disclosure activity around rating change announcements. The results indicate that the assessed probability of bankruptcy decreases before bond upgrades, but not after. By contrast, the assessed probability of bankruptcy increases both before and after bond downgrades. We also find that controlling for potential wealth-transfer related rating actions, which can impact stock returns differently, does not alter our results. Tests of press releases and earnings forecasts issued by firms suggest that the differential informativeness of upgrades and downgrades is not caused by differences in pre-rating change voluntary disclosures by upgraded and downgraded firms. The results support the hypothesis that downgrades are timelier than upgrades.     

RST–GCBR‐CLUSTERING‐BASED RGA–SVM MODEL FOR CORPORATE FAILURE PREDICTION

Corporate bankruptcy is perceived as a shocking event. Several researchers focused on the prediction of these phenomena using various methods aiming to avoid high generated costs. In this paper, a new hybrid approach is proposed to deal with corporate failure prediction. Based on financial ratios as input data and in order to predict if the business unit will fail or not, our approach integrates rough set theory, Gaussian case‐based reasoning‐clustering, real‐valued genetic algorithm with support vector machines. This combination is justified by a high accuracy rate, reaching 100% at 1 year before failure and 94.0925% at 3 years before failure. Copyright © 2011 John Wiley & Sons, Ltd.     

Bankruptcy prediction: Application of the Taylor's expansion in logistic regression

The purpose of the present study is to test whether Taylor's series expansion can be used to solve the problem associated with the functional form of bankruptcy prediction models. To avoid the problems associated with the normality of variables, the logistic model to describe the insolvency risk is applied. Taylor's expansion is then used to approximate the exponent of the logistic function, or the logit. The cash to total assets, cash flow to total assets, and shareholder's equity to total assets ratios operationalize the factors affecting the insolvency risk. The usefulness of Taylor's model in bankruptcy prediction is evaluated applying the logistic regression model to the data from the Compustat database. The classification accuracy in the test data for the first and second years before bankruptcy show that the classification accuracy of a simple financial ratio model can be increased using the second-order and interaction terms of these ratios. However, in the third year, for the test data, Taylor's expansion is not able to increase the classification accuracy when compared with the first-order model.