Predicting Nature of Default using Machine Learning Techniques

Research output: Working paperOther research output

102 Downloads (Pure)


This paper presents machine learning techniques to help financial institutions model a loan’s nature of default and further incorporate nature of default in the prediction of loss given default models. Nature of default decribes the main reason why the lender puts the borrower in default. The comparison of different techniques show the decision tree approach as the best model, specifically the time it takes to default since a loan’s origination is the most important feature in distinguishing different default types. We find loans with longer time to default are more likely to emerge to bankruptcy; whereas loans defaulted shortly after origination are more likely to be sold at a discount, resulting in a material credit loss. We also find that trade finance loans are more likely to receive a specific provision or write-off from the lending bank when they default, possibly due to the significant decrease in collateral valuations when the company is in financial difficulty. The nature of default is also found to be a significant factor in predicting loss given default. The unique insight this paper provides, when compared to similar default and loss studies in the existing literature, lies with its specificity in the loan’s nature of default and its association with loss rates.
Original languageEnglish
Number of pages67
Publication statusIn preparation - 30 Apr 2021


  • loan default
  • nature of default
  • default reason
  • machine learning
  • Decision Tree
  • naive bayes
  • Random forests
  • multinomial regression
  • loss given default
  • recovery and resolution


Dive into the research topics of 'Predicting Nature of Default using Machine Learning Techniques'. Together they form a unique fingerprint.

Cite this