Loading...

Credit Risk Measurement of Loan Portfolio Based on the Classification of Debtors Using Machine Learning

Ahmadnejad Saein, Mohammad Reza | 2020

506 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 53052 (44)
  4. University: Sharif University of Technology
  5. Department: Management and Economics
  6. Advisor(s): Zamani, Shiva; Haghpanah, Farshad
  7. Abstract:
  8. Credit risk is the possibility of a loss resulting from a borrower's failure to repay a loan or meet contractual obligations. All banks and financial institutions need to manage credit risk of their lending portfolios to maximize risk-adjusted rate of return and obey regulatory rules. The most commonly used method for determining credit risk is to calculate the maximum loss within the “Value at Risk” framework.Previous studies proposed different models for Credit VaR calculation like Vasicek Model and Credit Risk Plus Model. Furthermore, due to the high growth of computing power and easy access to information, the application of data-driven models such as Machine Learning has been increasing recently.In this thesis, we focused on using Machine Learning methods to satisfy the pre-assumptions of credit risk models and improve the assessment of Credit VaR for the loan portfolio of banks and financial institutions. We used Machine Learning models for pre-processing data to feed it into commonly used credit risk models.We developed Machine Learning models to divide a credit asset portfolio into risk homogeneous sub-portfolios then we calculated Credit VaR models’ parameters (e.g. default probabilities, recovery rates, and exposure at defaults) in each cluster. In the end, by calculation of aggregated Credit VaR for the credit asset portfolio, we found that using Machine Learning for making homogeneous sub-portfolios improves the accuracy of the mentioned Credit VaR Models
  9. Keywords:
  10. Value at Risk ; Machine Learning ; Artificial Neural Network ; Random Forest Algorithm ; Credit Risk ; Data Clustering

 Digital Object List

 Bookmark

No TOC