Bootstrap variable selection for corporate distress prediction: evidence from the Chinese manufacturing industry

Abstract

ABSTRACT This study explores the corporate distress prediction of Chinese manufacturing firms over 2013–2018 by using a large number of financial predictors. We extend the bootstrap resampling procedure into corporate distress prediction, combining with the existing selection techniques such as best subset selection, forward stepwise selection, backward stepwise selection, and least absolute shrinkage and selection operator (LASSO). We provide empirical evidence of the advantage of LASSO in selecting key predictors. The models constructed by these four techniques all outperform the prominent two models in the literature. This study proves the effectiveness of combining the bootstrap resampling method with variable selection techniques to predict corporate distress.