Chinese companies distress prediction: an application of data envelopment analysisJ Oper Res Soc


Zhiyong Li, Jonathan Crook, Galina Andreeva
Strategy and Management / Management Science and Operations Research / Management Information Systems / Marketing


Chinese companies distress prediction: an application of data envelopment analysis

Zhiyong Li, Jonathan Crook and Galina Andreeva

Credit Research Centre, Business School, University of Edinburgh, UK

Bankruptcy prediction is a key part in corporate credit risk management. Traditional bankruptcy prediction models employ financial ratios or market prices to predict bankruptcy or financial distress prior to its occurrence. We investigate the predictive accuracy of corporate efficiency measures along with standard financial ratios in predicting corporate distress in Chinese companies. Data Envelopment

Analysis (DEA) is used to measure corporate efficiency. In contrast to previous applications of DEA in credit risk modelling where it was used to generate a single efficiency—Technical Efficiency (TE), we assume Variable Returns to Scale, and decompose TE into Pure Technical Efficiency and Scale

Efficiency. These measures are introduced into Logistic Regression to predict the probability of distress, along with the level of Returns to Scale. Effects of efficiency variables are allowed to vary across industries through the use of interaction terms, while the financial ratios are assumed to have the same effects across all sectors. The results show that the predictive power of the model is improved by this corporate efficiency information.

Journal of the Operational Research Society (2014) 65, 466–479. doi:10.1057/jors.2013.67

Published online 10 July 2013

Keywords: data envelopment analysis; efficiency; corporate credit risk modelling; financial distress


The recent financial crisis indicates the importance of credit risk management and the necessity of recognising early indicators of corporate financial distress in order to prevent potential losses. Credit scoring models are such tools to generate early signals of corporate bankruptcy, which have received academic attention since at least 1950s and are still widely used.

One of the main problems in failure prediction models is variable selection. Financial ratios that are the quotient of two items in financial statements are the most popular variables that have been considered in the literature.

Beaver (1966) was the first author to introduce financial ratios into bankruptcy prediction. In recent decades there have been a great number of bankruptcy prediction studies based on financial ratios using different statistical and machine-learning techniques. They are reviewed in Altman (1993), Balcaen and Ooghe (2006), Kumar and Ravi (2007), Bahrammirzaee (2010) and Verikas et al (2010).

Recent papers (eg Wang and Ma, 2011) also demonstrate that financial ratios are still dominating the variable selection. However, it is widely recognised that the main cause of the company’s financial failure is its poor management (Gestel et al, 2006). The quality of management can be measured by the company’s efficiency that compares outputs to inputs.

One way to assess the efficiency of an organisation relative to the most efficient one is to use Data Envelopment Analysis (DEA). A number of papers have used DEA efficiencies in corporate bankruptcy modelling (see next section). In this paper we use DEA to compute various measures of corporate efficiency that we then input as a variable in a standard classifier to see how well this enables one to predict financial distress. The paper makes a number of contributions. First, unlike previous papers on corporate failure modelling that simply use a single efficiency measure, we decompose this measure—Technical

Efficiency (TE) into Pure Technical Efficiency (PTE), which indicates the ability to improve efficiency by wisely allocating resources and applying new technology and

Scale Efficiency (SE), which measures the ability to achieve better efficiency by adjusting to its optimal scale, and examine how each of these separately contributes to predicting financial distress. Second, in contrast to most applications of DEA in financial distress prediction we assume variable rather than Constant Returns to Scale (CRS). Third, DEA can only meaningfully be carried out for a sample of firms that use the same or similar technology (Dyson et al, 2001) and our study is the first to meet this requirement in the context of mixed-industry bankruptcy prediction. While this reduces our sample size, by modifying the second stage logistic regression we are

Journal of the Operational Research Society (2014) 65, 466–479 © 2014 Operational Research Society Ltd. All rights reserved. 0160-5682/14 Correspondence: Zhiyong Li, Room 3.02, Business School, 29 Buccleuch

Place, Edinburgh EH8 9JS, UK.

E-mail: able to determine the effects of variables that are common across industries. Fourth, we add corroboratory evidence to the very few studies that, regardless of country, have explored the corporate efficiency as a predictive variable in a financial distress model.

The paper is organised as follows. The next section provides a comprehensive review of the application of

DEA in corporate distress prediction models. In the third section the methodology adopted in this research is presented. This is followed by the description of the data used in the empirical analysis and the subsequent section reports the results. The paper finishes with conclusions and recommendations.

Literature review

DEA is an optimising technique that measures the relative efficiencies of a group of companies or Decision Making

Units (DMUs) that use multiple inputs and produce multiple outputs. An efficient company uses less inputs to produce more outputs. Such efficiency is evaluated by the distance of a particular DMU to the efficient frontier (ideal position), which is based on its peers (other DMUs in the sample). The main idea and notation will be introduced in the next section, for more comprehensive explanation of

DEA see Cooper et al (2000).

DEA has been incorporated into the prediction of corporate distress (or bankruptcy) in two different ways.