Predicting the credit risk of companies
: can structural models add value to the determination of probabilities of default in logit models?

  • Maximilian von Waldthausen (Student)

Student thesis: Master's Thesis

Abstract

This dissertation aims to investigate whether structural models have the ability to improve the predictive power of logit models. For this purpose, the distance-to-default (DD) metric, obtained from a simplified version of the structural model by Eisdorfer, Goyal and Zhdanov (2019), is integrated as another explanatory variable into the logit model by Campbell, Hilscher & Szilagyi (2008). A data set of 7.257 non-financial US-firms between 1983 and 2019 is used. The performance of the combined model is evaluated in both, in- and out-of-sample across time horizons of 6, 12, 24 and 36 months. In the in-sample evaluation, the distance-to-default reached significance for all considered time-horizons and led to an increase of the McFadden pseudo-R2. However, in the out-of-sample evaluation, the total number of right predictions decreased in the additive model compared to the benchmark model. These results suggest that the introduction of DD does not improve the predictive power of the logit model. As it was reasonable to assume that large parts of the DD variable were already captured by other explanatory variables, a follow up test was performed that did not verify this assumption. Based on these results, it is overall concluded that the distance-to-default obtained from the simplified EGZ model is not able to improve the performance of the CHS logit model.
Date of Award5 Feb 2021
Original languageEnglish
Awarding Institution
  • Universidade Católica Portuguesa
SupervisorDiana Bonfim (Supervisor)

Keywords

  • Credit risk
  • Default prediction
  • Distance-to-default
  • Logit models

Designation

  • Mestrado em Gestão e Administração de Empresas

Cite this

'