Prediction of Corporate Bankruptcy: A Multi-class Approach

Research output: Contribution to conferencePaperpeer-review

Abstract

The purpose of this study is to predict corporate and to determine the attributes affecting the bankruptcy prediction in the U.S. publicly listed companies using gradient boosting (GBoosting) and extreme gradient boosting (XGBoosting) machine learning techniques. The research data consists of 118,514 firm-year observations for U.S. public companies from 1992 to 2019. The dataset comprised of various financial ratios, ownership concentration, executive compensation, market price variables, macroeconomic variables, and audit-related variables. The results of this study show that a multi-class and high dimensional setting using both GBoosting and XGBoosting provides a more accurate prediction of bankruptcy in terms of accuracy and Type I & Type II error rates compare to previous research. However, a comparison of the two algorithms shows that GBoosting has a slightly higher performance in terms of prediction accuracy that is potentially related to over-fitting. An analysis of relative variable importance indicates that industry variable, ownership concentration/structure, financial ratios, market price variables, macroeconomic variables (Moody’s seasoned AAA bond yield, and real GDP growth) have the highest rank among the predictors.

Conference

ConferenceAAA 2020 Virtual Annual Meeting and Conference on Teaching and Learning
Period8/7/208/13/20
Internet address

Keywords

  • Bankruptcy prediction
  • Gradient boosting
  • eXtreme gradient boosting
  • Multi-class classification

Fingerprint

Dive into the research topics of 'Prediction of Corporate Bankruptcy: A Multi-class Approach'. Together they form a unique fingerprint.

Cite this