Predicting Corporate Failure Using a Neural Network Approach

This paper investigates the performance of Artificial Neural Networks for the classification and subsequent prediction of business entities into failed and non-failed classes. Two techniques, back-propagation and Optimal Estimation Theory (OET), are used to train the neural networks to predict bankruptcy filings. The data are drawn from Compustat data tapes representing a cross-section of industries. The results obtained with the neural networks are compared with other well-known bankruptcy prediction techniques such as discriminant analysis, probit and logit, as well as against benchmarks provided by directly applying the bankruptcy prediction models developed by Altman (1968) and Ohlson (1980) to our data set. We control the degree of ‘disproportionate sampling’ by creating ‘training’ and ‘testing’ populations with proportions of bankrupt firms ranging from 1% to 50%. For each population, we apply each technique 50 times to determine stable accuracy rates in terms of Type I, Type II and Total Error. We show that the performance of various classification techniques, in terms of their classification errors, depends on the proportions of bankrupt firms in the training and testing data sets, the variables used in the models, and assumptions about the relative costs of Type I and Type II errors. The neural network solutions do not achieve the ‘magical’ results that literature in this field often promises, although there are notable 'pockets' of superior performance by the neural networks, depending on particular combinations of proportions of bankrupt firms in training and testing data sets and assumptions about the relative costs of Type I and Type II errors. However, since we tested only one architecture for the neural network, it will be necessary to investigate potential improvements in neural network performance through systematic changes in neural network architecture.     

DESIGNING AN IF–THEN RULES‐BASED ENSEMBLE OF HETEROGENEOUS BANKRUPTCY CLASSIFIERS: A GENETIC ALGORITHM APPROACH

This paper proposes a framework for an ensemble bankruptcy classifier that uses if–then rules to combine the outputs from a heterogeneous set of classifiers. A genetic algorithm (GA) induces the rules using an asymmetric, cost‐sensitive fitness function that includes accuracy and misclassification costs. The GA‐based ensemble classifier outperforms individual classifiers and ensemble classifiers generated by other methods. The results of the classifier are in the form of if–then rules. We apply the approach to a balanced dataset and an imbalanced dataset. Both are composed of firms subject to financial distress and cited in the US Securities and Exchange Commission's Accounting and Auditing Enforcement Releases. Copyright © 2014 John Wiley & Sons, Ltd.     

Model Predictions and Auditor Assessments of Going Concern Status

This paper compares the predictions of a bankruptcy prediction model and the assessments of auditors on the going concern status of a sample of 165 bankrupt companies and 165 matched non-bankrupt companies. Data from US companies for the period 1978 to 1985 were used. Probit analysis (with the weighted exogenous sampling maximum likelihood procedure) was applied to estimate the model parameters. The Lachenbruch U method hold-out accuracy rates of the model are 85.45% for bankrupt firms, 100.00% for non-bankrupt firms, and 99.91% overall. The corresponding accuracy rates of the auditors based on their audit reports are 54.37% for bankrupt firms, 100.00% for non-bankrupt firms, and 99.73% overall. The sensitivity of optimal cut-off points to misclassification costs of Type I and Type II errors was also considered. Results of the study suggest that bankruptcy prediction models can be useful to auditors in making going concern assessments. Further, such models can serve as analytical tools and defensive devices.     

An investigation of recent changes in going concern reporting decisions among Big N and non-Big N auditors

Corporate accounting failures and regulatory proceedings that led to the enactment of the Sarbanes–Oxley Act of 2002 increased the scrutiny of auditors. We investigate whether these events resulted in a change in auditor behavior with respect to going concern reporting. Generally speaking, we find that non-Big N auditors became more conservative while Big N auditors became more accurate. Specifically, non-Big N auditors issued more going concern opinions to both failing and non-failing clients post-2001, reducing their Type II misclassifications at the expense of increased Type I misclassifications. However, Big N auditors decreased their Type I misclassifications with no corresponding increase in Type II misclassifications. Thus, our findings suggest that increased auditor scrutiny resulted in performance improvements in the area of going concern reporting primarily for larger auditors. For smaller auditors, improved going concern accuracy for subsequently bankrupt clients came at the cost of more going concern opinions being issued to subsequently non-failing clients.     

Bankruptcy prediction for Korean firms after the 1997 financial crisis: using a multiple criteria linear programming data mining approach

The main purpose of this paper is to evaluate the data mining applications, such as classification, which have been used in previous bankruptcy prediction studies and credit rating studies. Our study proposes a multiple criteria linear programming (MCLP) method to predict bankruptcy using Korean bankruptcy data after the 1997 financial crisis. The results, of the MCLP approach in our Korean bankruptcy prediction study, show that our method performs as well as traditional multiple discriminant analysis or logit analysis using only financial data. In addition, our model??s overall prediction accuracy is comparable to those of decision tree or support vector machine approaches. However, our results are not generalizable because our data are from a special situation in Korea.     

Comparing the performance of market-based and accounting-based bankruptcy prediction models

Recently developed corporate bankruptcy prediction models adopt a contingent claims valuation approach. However, despite their theoretical appeal, tests of their performance compared with traditional simple accounting-ratio-based approaches are limited in the literature. We find the two approaches capture different aspects of bankruptcy risk, and while there is little difference in their predictive ability in the UK, the z-score approach leads to significantly greater bank profitability in conditions of differential decision error costs and competitive pricing regime.     

ZETATM analysis A new model to identify bankruptcy risk of corporations

The paper explores the development of a bankruptcy classification model which incorporates comprehensive inputs with respect to discriminant analysis and utilizes a sample of bankrupt firms essentially covering the period 1969–1975. Financial statement data and market related measures are transformed along guidelines suggested by traditional security analysis to promote comparability of companies and to reflect the most recent reporting standards so as to make the model relevant to future analysis. The results of the study are compared with alternative bankruptcy classification strategies via the explicit introduction of prior probabilities of group membership, observed accuracies, and estimates of costs of errors in misclassification. The latter is based on cost estimates derived from commercial bank lending errors. The results of the study indicate potential significant application to credit worthiness assessment, portfolio management, and to external and internal performance analysis.     

A simple correction of the WACC discount rate for default risk and bankruptcy costs

Standard discounted cash flow approaches suffer from a rudimental modeling of the possibility of a default, as the main characteristics such as the default probability and potential bankruptcy costs are commonly disregarded. This paper aims at providing a tractable extension of the well-known WACC approach for both default risk and bankruptcy costs. The corrected WACC discount rate reveals that default risk results in a systematically higher WACC because the tax component is scaled by the survivorship probability and an aditional component for bankruptcy costs must be added. This difference between the classical WACC discount rate and the simple modified WACC rate can be remarkable especially for firms from businesses with high bankruptcy costs and a relevant default probability.     

False (and Missed) Discoveries in Financial Economics

Multiple testing plagues many important questions in finance such as fund and factor selection. We propose a new way to calibrate both Type I and Type II errors. Next, using a double-bootstrap method, we establish a t-statistic hurdle that is associated with a specific false discovery rate (e.g., 5%). We also establish a hurdle that is associated with a certain acceptable ratio of misses to false discoveries (Type II error scaled by Type I error), which effectively allows for differential costs of the two types of mistakes. Evaluating current methods, we find that they lack power to detect outperforming managers.     

A NOTE ON THE OPTIMAL CUTOFF POINT IN BANKRUPTCY PREDICTION MODELS

A modified Bayesian decision model is derived in the study to systematically estimate an optimal cutoff point for bankruptcy prediction models. In addition, a loss function is implemented in the model so that the total error costs instead of the total error probability is minimized. Any dichotomous classification problem with unequal error costs would find this decision model useful.     

Off-site monitoring systems for predicting bank underperformance: a comparison of neural networks,discriminant analysis,and professional human judgment

This study compares the ability of discriminant analysis, neural networks, and professional human judgment methodologies in predicting commercial bank underperformance. Experience from the banking crisis of the 1980s and early 1990s suggest that improved prediction models are needed for helping prevent bank failures and promoting economic stability. Our research seeks to address this issue by exploring new prediction model techniques and comparing them to existing approaches. When comparing the predictive ability of all three models, the neural network model shows slightly better predictive ability than that of the regulators. Both the neural network model and regulators significantly outperform the benchmark discriminant analysis model's accuracy. These findings suggest that neural networks show promise as an off-site surveillance methodology. Factoring in the relative costs of the different types of misclassifications from each model also indicates that neural network models are better predictors, particularly when weighting Type I errors more heavily. Further research with neural networks in this field should yield workable models that greatly enhance the ability of regulators and bankers to identify and address weaknesses in banks before they approach failure. Copyright © 2001 John Wiley & Sons, Ltd.     

How Costly Is External Financing? Evidence from a Structural Estimation

We apply simulated method of moments to a dynamic model to infer the magnitude of financing costs. The model features endogenous investment, distributions, leverage, and default. The corporation faces taxation, costly bankruptcy, and linear‐quadratic equity flotation costs. For large (small) firms, estimated marginal equity flotation costs start at 5.0% (10.7%) and bankruptcy costs equal to 8.4% (15.1%) of capital. Estimated financing frictions are higher for low‐dividend firms and those identified as constrained by the Cleary and Whited‐Wu indexes. In simulated data, many common proxies for financing constraints actually decrease when we increase financing cost parameters.     

PREDICTING UNCERTAINTY AUDIT QUALIFICATIONS IN AUSTRALIA USING PUBLICLY AVAILABLE INFORMATION

This research project constructed a logit model to predict “subject to” qualified audit opinions using financial statement and market variables for 1,848 audit reports for Australian companies issued from 1984 to 1988. The model provided a better goodness of fit and was more efficient than two naive strategies for predicting “subject to” audit qualifications. The model explicitly incorporated the relative costs of Type II to Type I errors to account for the auditor's asymmetrical loss function. The model was reasonably accurate when a sensitivity analysis for the relative costs of Type II and Type I errors was considered. The accuracy rates for the estimation sample ranged from 70% to 95%. An inter-temporal holdout sample of 293 audit opinions for Australian firms issued during 1989 indicated that the model was useful for predicting “subject to” audit opinions. The accuracy rates for the holdout sample ranged from 72% to 90% over a range of relative Type II and Type I costs.     

Bankruptcy laws and debt renegotiation

This paper analyzes the effect of the toughness of bankruptcy law on the number of liquidations in a simple model of borrowing and lending with asymmetric information, where the creditor cannot credibly commit to liquidate the firm if the default occurs. In our setting we consider a bankruptcy law to be a one-dimensional variable that influences creditor's expectation value of collateral. We find that there is an interval of the bankruptcy law, where the number of liquidations decreases in the toughness of the bankruptcy law. We also find that if the liquidation costs are high, softer bankruptcy law is preferred.     

The identification of acquisition targets in the EU banking industry: An application of multicriteria approaches

In this paper we develop classification models for the identification of acquisition targets in the EU banking industry, incorporating financial variables that are mostly unique to the banking industry and originate from the CAMEL approach. A sample of 168 non-acquired banks matched with 168 acquired banks is used over the period 1998-2002, covering 15 EU countries. We compare and evaluate the relative efficiency of three multicriteria approaches, namely MHDIS, PAIRCLAS, and UTADIS, with all models developed and tested using a 10-fold cross validation approach. We find that the importance of the variables differs across the models. However, on the basis of univariate test and the results of the models we could state that in general after adjusting for the country where banks operate, acquired banks are less well capitalized and less cost and profit efficient. The results show that the developed models can achieve higher classification accuracies than a naïve model based on random assignments. Nevertheless, there is fair amount of misclassification that is hard to avoid given the nature of the problem, showing that as in previous studies for non-financial firms, the identification of acquisitions targets in banking is a difficult task.     

A quarterly time-series classifier based on a reduced-dimension generated rules method for identifying financial distress

Predicting financial distress has been and will remain an important and challenging issue. Many methods have been proposed to predict bankruptcies and detect financial crises, including conventional approaches and techniques involving artificial intelligence (AI). Financial distress information influences investor decisions, and investors depend on analysts’ opinions and subjective judgements in assessing such information, which sometimes results in investors making mistakes. In the light of the foregoing, this paper proposes a novel quarterly time series classifier, which reduces the sheer volume of high-dimensional data to be analysed and provides decision-makers with rules that can be used as a reference in assessing the financial situation of a company. This study employs the following six attribute selection methods to reduce the high-dimensional data: (1) the chi-square test, (2) information gain, (3) discriminant analysis, (4) logistic regression (LR) analysis, (5) support vector machine (SVM) and (6) the proposed Join method. After selecting attributes, this study utilises the rough set classifier to generate the rules of financial distress. To verify the proposed method, an empirically collected financial distress data-set is employed as the experimental sample and is compared with the decision tree, multilayer perceptron and SVM under Type I error, Type II error and accuracy criteria. Because financial distress data are quarterly time series data, this study conducts non-time series and time series (moving windows) experiments. The experimental results indicate that the LR and chi-square attribute selection combined with the rough set classifier outperform the listing methods under Type I, Type II error and accuracy criteria.     

An Empirical Estimation of Default Risk of the UK Real Estate Companies

Based on the Black and Scholes (Black, F., and M. Scholes. (1973). The Pricing of Options and Corporate Liabilities, Journal of Political Economy 81, 637–659) and Merton (Merton, R. C. (1974). On the Pricing of Corporate Debt: The Risk Structure of Interest Rates, Journal of Finance 29, 449–470) (BSM) contingent claims model, and KMV Corporation framework, we estimate the distance to default and the “risk neutral” default probabilities for a sample of 112 real estate companies over the period 1980 to 2001. Our empirical results classifies failed and non-failed companies into Type I error, cases that the BSM-type model fails to predict default when it did occur, and Type II error where BSM-type model predicts default when it did not occur. We find that none of the companies belong to the category of Type I error. Type II error is observed in 12 out of 112 companies. These results support the theoretical underpinnings of the BSM-type structural model in that the two driving forces of default are high leverage and high asset volatility.     

Presidential Address: Liquidity and Price Discovery

This paper examines the implications of market microstructure for asset pricing. I argue that asset pricing ignores the central fact that asset prices evolve in markets. Markets provide liquidity and price discovery, and I argue that asset pricing models need to be recast in broader terms to incorporate the transactions costs of liquidity and the risks of price discovery. I argue that symmetric information‐based asset pricing models do not work because they assume that the underlying problems of liquidity and price discovery have been solved. I develop an asymmetric information asset pricing model that incorporates these effects.     

Trade Credit Use as Firms Approach Default

Using a sample of distressed firms with information about suppliers, we document an average fall in the use of trade credit as firms approach bankruptcy compared to a control sample of nonbankrupt firms. However, we uncover a large degree of heterogeneity across suppliers. Suppliers facing high switching costs maintain their business ties with the distressed firms as they approach bankruptcy, and provide them more trade credit. Suppliers in concentrated markets provide temporary support to their clients. Overall, the findings of this paper suggest that switching costs are fundamental to explain whether suppliers provide liquidity to their distressed clients or not.     

Auditors’ going concern reporting accuracy during and after the global financial crisis

Motivated by concerns that one of the reasons for the Global Financial Crisis (GFC) is poor quality auditing, this study examines the accuracy of going concern modifications for a sample of United States (U.S.) companies in the pre-GFC (2005–2006), GFC (2007–2008), and post-GFC (2009–2010) periods. The results show that the type I misclassification is lower during the GFC but not different in the post-GFC period compared with the pre-GFC period. The type II misclassification is not significantly different in the GFC and post-GFC periods compared with the pre-GFC period. Additionally, the results suggest that non-Big 4 auditors, compared with Big 4 auditors, have become more conservative on clients’ going concern problems in the post-GFC period, which reduces their type II misclassification.