Skip to content

FRM Part II · FRM Exam Part II · Credit Scoring and Rating

A lender built its scorecard on accepted applicants only, since default outcomes are observed only for loans that were granted. Applicants previously rejected under the old policy were never observed. A validation team warns the model may perform poorly on the through-the-door population. Which technique is a standard attempt to address this problem?

Reject inference is the standard remedy. It assigns inferred good or bad outcomes to previously rejected applicants so the development sample better represents the through-the-door population, reducing sample selection bias caused by training only on accepted loans.

  1. AIncreasing the number of bins used for each predictor
  2. BReject inference, assigning likely good or bad outcomes to rejected applicants to reduce sample selection biasCorrect
  3. CReplacing the logistic model with a linear probability model
  4. DRemoving all variables used in the old acceptance policy from the model

Explanation

Training only on accepted accounts creates sample selection bias, because rejected applicants differ systematically. Reject inference methods such as augmentation or parceling infer outcomes for rejects so the model reflects the full applicant population. More bins, a different functional form, or dropping variables do not address the missing population.

Did you get it right without looking?

One question tells you little. A timed set on Credit Scoring and Rating shows your real accuracy, how long you take and where you lose marks.

Practise Credit Scoring and RatingFree diagnostic mock

More Credit Scoring and Rating questions