Selection of balanced structure samples in corporate bankruptcy prediction

Abstract

Selecting samples is one of the methodological issues in the case of corporate failure prediction. In practice, when constructing a sample of balanced structure (50% of failed and 50% of non-failed firms), the most popular approach is based on subjective pairings of available bankrupt companies with non-bankrupt ones. This method is sometimes called pair-matched sampling. However, samples obtained in this way are not independent random ones. For this reason, other techniques should be taken into consideration. The simplest solution seems to be the employment of random sampling with replacement. The paper presents a comparative study of prognostic capabilities of four types of bankruptcy prediction models obtained as a result of applying two sampling techniques (pair-matched sampling and random sampling with replacement). The conducted analysis is based on financial data of Polish manufacturing companies.