Bankruptcy Prediction Using Data Mining Tools

Bankruptcy Prediction Using Data Mining Tools

Wikil Kwak (University of Nebraska at Omaha, USA)
Copyright: © 2014 |Pages: 7
DOI: 10.4018/978-1-4666-5202-6.ch021

Chapter Preview



Altman (1968) used multiple discriminant analysis (MDA) by using financial ratios to predict a firm bankruptcy model. Altman et al. (1977) later proposed the ZETA model, but their assumptions of data being normally distributed can be a problem when applying this model.

Ohlson (1980) used a logit model without prior assumptions about the probability of bankruptcy or the distribution of predictor variables. However, hold-out sample tests, in general, are potentially upwardly biased (Grice and Dugan, 2001) and the differences in the macro economic factors are sensitive to specific time periods. For example, Grice and Ingram (2001) empirically tested Altman’s model using the 1988-1991 test period and reported that the overall correct classification rate dropped to 57.8%. Therefore, the test period may be sensitive to overall prediction rates.

Freed and Glover (1986) proposed linear programming (LP) to minimize misclassifications in linear discriminant analysis. Gupta et al. (1990) also proposed linear goal programming as an alternative to discriminant analysis. However, these approaches may not be manageable for the large-scale databases at that time. Pompe and Feelders (1997) compared machine learning, neural networks, and statistics using experiments to predict firm bankruptcy. However, their results are not conclusive to which methods outperform the other methods.

Shin and Lee (2002) proposed a genetic algorithm (GA) in bankruptcy prediction modeling using Korean manufacturing companies from 1995 to 1997. Their approach is capable of extracting rules that are easy to understand for users like expert systems, but by using only the manufacturing industry, there may be an upward bias for the prediction accuracy rate of 80.8%. Recent bankruptcy studies classified and predicted the final bankruptcy resolution using longitudinal study (for example, Barniv et al., 2002), but our focus should be prediction accuracy, flexibility of the model, and easy to apply using real world data without difficulty of computational complexity. The most prominent bankruptcy studies in accounting and finance are Altman’s and Ohlson’s and our study uses their financial variables. These variables are pulled from each bankrupt firm’s financial statements the year before they filed bankruptcy. If we want to improve the prediction accuracy, we could easily add a cash flows variable or non-financial variable such as stock market returns, missing dividend, and auditor changes.

Key Terms in this Chapter

Prediction: Try to predict future outcomes based on past financial and other data.

Auditor Change: Change of an auditor who examines the company’s financial statements and records.

Internal Control: A company’s maintenance of internal accounting system.

Bankruptcy: An extreme form of financial distress in which a firm seeks legal protection from its creditors and usually ceases to exist.

Data Mining: Finding out meaningful patterns from large amounts of data.

Multiple Criteria Linear Programming: Linear programming using multiple criteria and multiple constraints.

Logit Analysis: Classifies cases into category and primary tools for prediction studies for general purpose.

Complete Chapter List

Search this Book: