How to build Loan Default Forecasting model?

data_science

#1

I need some help in one of my office projects.
I am building a loan level default probability based model for 24 months (cumulative model) . So, I followed an approach of developing a cumulative default probability model for 6 months and then extrapolate the prediction from months 1 to 24.

I have used Bayes theorem for extrapolation of 6th month prediction to months 1- 24. But, it doesn’t seem to work fine with the actuals. Can you please a decision state algorithm like Bayes theorem, which I can use?


#2

Hi Vajravi,
Other 6 methods you can try for forecasting .

hope it will work!


#3

Best is to try to build a logistic regression. That is what I used and it worked.

Since you already have a model in place, extrapolating it might not be a good way.

  1. 6 months to 2 years: We use extrapolation beyond 2 years. Say I want to have a cumulative for 5 years or 10 years. The reason is, the new accounts take time to mature. If you already have seasoned accounts with you, go ahead, else please make a model for 2 years.
  2. Cumulative: It is but a summation of your defaults over the previous months. Try to decrease your monthly error. We used a panel data. For each account we took the performance across various months and then rolled them up on a monthly level.

Hope it helps and answers your question.