A multi-industry bankruptcy prediction model using back-propagation neural network and multivariate discriminant analysis

Abstract

The accurate prediction of corporate bankruptcy for the firms in different industries is of a great concern to investors and creditors, as the reduction of creditors’ risk and a considerable amount of saving for an industry economy can be possible. This paper presents a multi-industry investigation of the bankruptcy of Korean companies using back-propagation neural network (BNN). The industries include construction, retail, and manufacturing. The study intends to suggest the industry specific model to predict bankruptcy by selecting appropriate independent variables. The prediction accuracy of BNN is compared to that of multivariate discriminant analysis.

The results indicate that prediction using industry sample outperforms the prediction using the entire sample which is not classified according to industry by 6–12%. The prediction accuracy of bankruptcy using BNN is greater than that of MDA. The study suggests insights for the practical industry model for bankruptcy prediction.

Introduction

Prediction of corporate bankruptcy is of a great concern to investors/creditors, borrowing firms, and governments. As a result of the collapse of Enron, many voices have called for a revolution of existing bankruptcy warning systems to detect or prevent bankruptcy problems in real time. Bankruptcy can happen to any organizations because the business environment is increasingly undergoing uncertainty and competition these days. The reduction of creditors’ risk and a considerable amount of saving for an economy can be possible from even a slight improvement with respect to assessing credit risk. An improvement in accuracy of even a fraction of a percent in scoring models to estimate the probability of default leads to enormous future savings for the credit industry (West, Dellana, & Qian, 2005). Assessment of bankruptcy offers invaluable information by which governments, investors, shareholders and the management can make their financial decisions in order to prevent possible losses. The study of bankruptcy provides an early warning signal and detects areas of weaknesses. Accurate bankruptcy prediction usually leads to many benefits such as cost reduction in credit analysis, better monitoring, and an increased debt collection rate.

A number of publications have pursued this subject and extending conventional models for prediction during the past 50 years. The number of bankruptcy prediction models has grown enormously due to the growing availability of data and the development of improved econometrical techniques during the 1980s and 1990s. Most of this work has been influentially led by a small number of early papers (e.g., Altman, 1968, Ohlson, 1980, Zavgren, 1985) on US quoted companies. The methods for bankruptcy prediction can be grouped in two categories: statistical and artificial intelligence models. The first group consists of Logit, multivariate discriminant analysis, etc. The tool first applied to bankruptcy prediction was the univariate data analysis proposed by Beaver (1966), which was followed by the multi-variate discriminant analysis and regressions (Ohlson, 1980). The second group includes neural networks (Chauhan et al., 2009, Cho et al., 2009, Pendharkar, 2005, Tseng and Hu, 2010), genetic algorithms (Etemadi et al., 2009, Lensberg et al., 2006), and support vector machine (Min and Lee, 2005, Yang et al., 2011), and case based reasoning (Cho Hong & Ha, 2010). While some of these models show high predictive accuracy levels, the absence of bankruptcy theory makes attempts to establish a generally accepted model for bankruptcy prediction unsuccessful.

Although the discriminant analysis and linear regression model have become the most commonly used in bankruptcy prediction, their inherent drawbacks of statistical assumptions such as linearity, normality and independence among variables have constrained both applications. Recent trends in the development of artificial intelligence have brought forth new alternatives in solving nonlinear problems. The expert system, fuzzy logic, and neural networks are a great help to a manager in predicting bankruptcy making decisions. Neural networks have many different topologies for problem dissimilarities. Among them, back-propagation is the most well known and commonly used, categorized as one of the supervised learning models. It draws the mapping function between the input and output from the provided data set. The back propagation neural network (BPN) usually contains one input layer, one or two hidden layer(s) and one output layer. Each layer of a neural network structure has several units and output units of a layer are input units of its next layer. The purpose of back-propagation training is to produce the weight of each edge, in order to minimize the squared error sum between the actual value and the predicted value.

Previous studies on neural network applications for bankruptcy prediction have been targeting single industry or not investigated the industry difference in bankruptcy prediction. For example, He and Kamath (2005) evaluated the effectiveness of two successful bankruptcy models by Ohlson (1980) and Shumway (2001) with the help of a mixed industry sample in discriminating between bankrupt and non bankrupt firms from an individual industry the equipment & machinery manufacturing (EMM) industry. Dewaelheyns and Van Hulle (2006) suggested that models involving both bankruptcy variables defined at subsidiary level and at group level provide a substantially better fit and classification performance. These studies did not examine the difference in industries in terms of independent variables, prediction accuracy, and practically usable models. It is still elusive whether generic prediction models are still successful in predicting individual industry. Further, although there exist a number of studies on using neural networks on bankruptcy prediction, the studies on using multi-industry data and developing models for multi-industry are almost rare. This paper intends to fill this void. This paper focuses on the different back propagation neural network (BNN) models for construction, retail, and manufacturing industries. This study intends to show the optimal tested for each industry by using the bankruptcy data of Korean companies. The study includes the comparison of predictive accuracy with multivariate discriminant analysis (MDA) and shows the implications of difference in prediction models and results.