CCP Credit Scoring & Probability of Default 3 — Questions and Answers
Question 1: A bank is building a PD model and must choose between a vintage analysis and a roll-rate analysis. What is the PRIMARY purpose of vintage analysis?
- To measure the speed at which borrowers migrate between delinquency buckets
- To track cohorts of loans originated in the same period and observe cumulative default rates over time (Correct answer)
- To estimate loss given default for each loan vintage
- To compare interest rate spreads across origination years
Correct answer: To track cohorts of loans originated in the same period and observe cumulative default rates over time
Vintage analysis groups loans by origination period and tracks their cumulative default performance, revealing seasoning curves and underwriting quality trends.
Question 2: Under Basel III, what is the minimum PD floor that must be applied to retail exposures when using the Internal Ratings-Based (IRB) approach?
- 0.01%
- 0.03% (Correct answer)
- 0.10%
- 0.25%
Correct answer: 0.03%
Basel III sets a minimum PD floor of 0.03% for retail exposures under the IRB approach to avoid unrealistically low capital charges.
Question 3: What is the primary risk of using 'thin file' applicants (those with limited credit history) in scorecard development?
- They inflate the Gini coefficient artificially
- The model may be biased because their performance data is sparse or absent (Correct answer)
- They always have higher default rates, skewing the bad rate upward
- They are excluded by FCRA regulations from credit model datasets
Correct answer: The model may be biased because their performance data is sparse or absent
Thin-file borrowers have insufficient historical data, so models built without accounting for them may perform poorly when applied to this segment.
Question 4: A scorecard developer calculates an Information Value (IV) of 0.48 for a variable. How should this be interpreted?
- Weak predictor — should be excluded from the model
- Medium predictor — acceptable for inclusion
- Strong predictor — highly predictive and should be included
- Too strong — the variable is likely leaking target information and should be investigated (Correct answer)
Correct answer: Too strong — the variable is likely leaking target information and should be investigated
An IV above ~0.3–0.5 is often considered suspiciously high and may indicate target leakage or overfitting, requiring investigation before inclusion.
Question 5: Which of the following credit bureau attributes is typically the STRONGEST predictor of default in a generic scorecard?
- Number of credit inquiries in the last 6 months
- Total outstanding balance across all trades
- Worst delinquency in the past 24 months (Correct answer)
- Age of oldest open account
Correct answer: Worst delinquency in the past 24 months
Historical delinquency severity (worst status) is consistently found to be the single strongest predictor of future default in credit scoring research.
Question 6: What does 'population stability index' (PSI) measure in credit scoring?
- The stability of PD estimates across economic cycles
- The shift in score distribution between development and current deployment populations (Correct answer)
- The consistency of approval rates across demographic groups
- The volatility of default rates within a single vintage
Correct answer: The shift in score distribution between development and current deployment populations
PSI compares score distributions at development versus current deployment to detect whether the population has shifted significantly, triggering model redevelopment needs.
Question 7: In a credit scorecard, what is a 'characteristic' and what is an 'attribute'?
- Characteristic = the model output score; attribute = the raw input variable
- Characteristic = the predictor variable (e.g., delinquency history); attribute = a specific bin or category of that variable (Correct answer)
- Characteristic = a demographic feature; attribute = a financial feature
- Characteristic = the dependent variable; attribute = the independent variable
Correct answer: Characteristic = the predictor variable (e.g., delinquency history); attribute = a specific bin or category of that variable
In scorecard terminology, a characteristic is the input variable (e.g., 'months since last delinquency'), and each attribute is a distinct grouping or bin of that variable assigned scorecard points.
CCP Credit Scoring & Probability of Default 3
A bank is building a PD model and must choose between a vintage analysis and a roll-rate analysis.
What is the PRIMARY purpose of vintage analysis?