Use of Data Reduction Process to Bankruptcy Prediction: Evidence from an Emerging Market

Abstract

Predicting corporate bankruptcy has been an important challenging problem in research topic in accounting and finance. In bankruptcy prediction, researchers often confront a range of observations and variables which are often vast amount of financial ratios. By reducing variables and select relevant data from a given dataset, data reduction process can optimize bankruptcy prediction. This study addresses four well-known data reduction methods including t-test, correlation analysis, principal component analysis (PCA) and factor analysis (FA) and evaluated them in bankruptcy prediction in the Tehran Stock Exchange (TSE). To this end, considering 35 financial ratios, the results of data reduction methods were separately used to train Support Vector Machine (SVM) as the powerful prediction model. Regarding the empirical results, among the aforementioned methods, the t-test lead to the most prediction rate with 97.1% of predictability and PCA by 95.1% provides the next position.