FRM Part II · FRM Exam Part II · Credit Scoring and Retail Credit Risk Management
A retail lender is building an application scorecard using logistic regression on a historical sample of approved loans, each labeled 'good' or 'bad' after 24 months. A risk analyst warns that the model may be biased when applied to the full applicant population. Which feature of the development process most directly causes this concern?
The main concern is sample selection bias: performance is observed only for previously accepted applicants, so the model may not represent rejected ones. Reject inference is used to address this. A binary outcome, WOE transformation and a fixed performance window are normal design choices, not the source of this bias.
- AReject inference is needed because outcomes are observed only for accepted applicantsCorrect
- BThe use of a binary dependent variable in logistic regression
- CThe use of weight of evidence to transform categorical variables
- DThe use of a 24-month performance window
Explanation
Outcomes are observed only for applicants who were previously approved, so the sample is not representative of all applicants (sample selection bias). Reject inference techniques attempt to correct for this. The binary outcome, WOE transformation and a performance window are standard design choices and do not create this selection problem.
Did you get it right without looking?
One question tells you little. A timed set on Credit Scoring and Retail Credit Risk Management shows your real accuracy, how long you take and where you lose marks.
More Credit Scoring and Retail Credit Risk Management questions
- A bank introduces a machine-learning model for retail credit decisions that is more accurate than its previous scorecard but is hard to inte…
- A bank wants to use a complex machine-learning model for retail credit decisions. Under supervisory model risk guidance, which step is most …
- A risk analyst compares the score distribution of loan applicants at model development with the current quarter's applicants. Population Sta…
- A bank wants to use a complex machine-learning model for retail credit decisions that outperforms its logistic scorecard but is hard to inte…
- A risk analyst argues that a retail mortgage portfolio of 50,000 loans needs far less single-name concentration monitoring than a corporate …
- A retail lender builds a scorecard using only approved applicants' repayment outcomes, then validates it on those same approved accounts. Wh…