I am using logistic regression to create a credit scorecard from past loan data. We will not approve loans in the future if the applicant has an insufficient credit score (no credit or insufficient data for a score from credit bureaus).

These borrowers were however granted loan approvals in the past; therefore, there is extensive data on these loans with insufficient credit scores. Should I include the past loan data of all these borrowers in the logistic regression model to train the model and compute the best model? Or omit them from the model since we will not approve them in the future?

  • 2
    $\begingroup$ Were the past loans given by the same banks that are currently making the evaluation? Do you have the borrowers credit scores from the time they received their previous loan, or only afterwards? $\endgroup$ – Michael Bishop Feb 4 '13 at 22:19

You should include them. If the intent of the model is to (in effect) predict who will be able to pay back loans, you need as full a dataset as possible and certainly including those cases that have failed before for the purpose of identifying how the explanatory variables lead to successful predictions of this.

I am assuming that the response variable in your logistic regression is success in paying back the loan, or something similar.

| cite | improve this answer | |
  • $\begingroup$ This should have one caveat though - there should be some measure of how representative their previous record is. For example a change in income could easily make previous defaults or successful repayments irrelevant. $\endgroup$ – probabilityislogic Feb 16 '13 at 22:29

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.