비모수기법을 이용한 기업도산예측 - DEA와 Negative DEA 교차층화기법의 개발 -
Nonparametric Approach to Bankruptcy Prediction: Developing Cross Peeling Technique integrating DEA and Negative DEA
민재형(서강대학교); 정철우(서강대학교)
35권 4호, 1157~1180쪽
초록
본 연구의 목적은 기업도산예측을 위한 방법으로서 Data Envelopment Analysis(DEA)와 Negative DEA를 함께 사용하여 예측력과 오분류비용을 개선할 수 있는 층화기법을 새로이 개발하는데 있다. 이를 위해 본 연구에서는 도산예측을위한 기존의 세 가지 비모수 접근방법(DEA와 Negative DEA의 최적절사점을 이용하는 방법, Negative DEA에 층화기법을 적용하는 방법, Simak의 층화기법)과 본 연구에서 개발한 교차층화기법(cross peeling technique) 등 네 가지방법을 도산예측에 적용하여 그 성과를 비교하였다. 실증분석 결과, DEA와 Negative DEA의 최적절사점을 이용한 도산예측방법의 경우, 최적절사점이 분석 시점에 따라 변동성이 높게 나타남으로써 그 기법을 실무에 적용하기에는 한계가있었다. 그리고 Negative DEA에 층화기법을 적용하는 방법과 Simak의 층화기법은 분류층을 몇 개로 하느냐에 따라예측력이 달라지는 문제가 발생하기 때문에 이 두 가지 기법을 이용하여 도산예측을 하기 위해서는 예측력을 높일 수 있는 최적의 분류층을 찾는 것이 중요한 문제로 부각됨을 알 수 있었다. 이에 반해 본 연구에서 개발한 교차층화기법은 적절한 분류층을 찾을 필요가 없으면서도 예측력 및 오분류비용 측면에서 기존의 두 층화기법과 비교해 향상된 결과를 가져옴을 확인할 수 있었다
Abstract
Many statistical models have been developed to predict a firm’s bankruptcy; however, the previous models have some weaknesses such that they need ex ante information of whether sample companies are bankrupted or not as well as ex post information of which factors affect the bankruptcy. Besides, the statistical models require that the data satisfy the assumption of linearity and normality, which rarely holds in reality. To overcome these weaknesses, DEA, a nonparametric method, has been suggested as an alternative to existing statistical models. DEA has strong points such that it can measure the financial soundness of firms without ex ante information, and it does not need any assumption like linearity and normality of the data, which other statistical models do. It can also use raw data of various numerical scales. Nevertheless, DEA-based models have the difficulty of taking undesirable outputs into account; hence they have been little used in the literature as an alternative method for bankruptcy prediction. The purpose of this paper is twofold. First, it is to suggest Negative DEA as a complementary method to normal DEA. Negative DEA is an approach of identifying the worst performers by placing them on the efficient frontier with taking undesirable outputs into account. Second, it is to develop “cross peeling technique”, a new method for bankruptcy prediction integrating DEA and Negative DEA, which shows improvement in terms of prediction power as well as misclassification cost over the existing methods. Specifically, we empirically compare the prediction performance of three existing methods (“a method employing respective cut-off points of DEA and Negative DEA”, “a method employing peeling technique applied to Negative DEA”, and “Simak’s peeling technique”) with the one of “cross peeling method” that is developed in this paper. From the experiment, it is shown that “cross peeling technique” is a superior method over the existing peeling techniques in terms of prediction power and total misclassification cost. In addition, it turns out to be a useful method for rating firms’ credit risk for risk management purpose. In conclusion, this paper makes contribution in two points. First, it reports the usefulness of nonparametric approach as an alternative method to the existing statistical methods. Second, it develops “cross peeling technique”, an improved method over the existing peeling techniques for practical bankruptcy prediction purpose.
- 발행기관:
- 한국경영학회
- 분류:
- 경영학