Predicting Bankruptcy with Semi-Parametric Single-Index Model

Arjana Brezigar Masten,Igor Masten

DOI: https://doi.org/10.1080/1331677X.2012.11517497

2012-01-01

Abstract:Abstract Semi-parametric methods are virtually neglected in the bankruptcy prediction literature. This paper compares the logit model, as the standard parametric model for bankruptcy prediction, to the semi-parametric model developed by Klein and Spady (1993). Special care is devoted to the effect of choice-based sampling prediction accuracy. The choice of the sampling and estimation method lead to a similar trade offs. Using choice-based sampling and logit model leads to minimization of risk exposure. Samples unbalanced across groups and the semi-parametric method allow for better overall prediction accuracy and thus profit maximization.