Credit Scoring & Probability of Default Flashcards
7 cards from real CCP practice questions. Tap to flip, then mark Knew It or Still Learning — missed cards come back until you master them.
Read the first 7 Credit Scoring & Probability of Default flashcards as text
A credit analyst is evaluating a borrower with a FICO score of 680 and a debt-to-income (DTI) ratio of 48%. Which factor most concerns regulators under Qualified Mortgage (QM) rules?
Answer: DTI exceeding 43% is a key threshold that may disqualify the loan as a QM
Under the original QM rule, a DTI ratio above 43% generally disqualifies a mortgage from safe harbor QM status, although this has been updated to an APR-based threshold.
What is the difference between a 'scoreable' and 'non-scoreable' population in credit risk management?
Answer: Scoreable means enough credit history exists to generate a reliable score; non-scoreable lacks sufficient data
A scoreable population has sufficient credit bureau data for the model to produce a reliable score, while non-scoreable individuals lack the minimum required tradeline history.
Under point-in-time (PIT) PD estimation, how does a recession affect estimated PDs compared to a TTC approach?
Answer: PIT PDs spike during recessions, while TTC PDs remain more stable
PIT PDs reflect current macroeconomic conditions and therefore rise sharply in recessions, while TTC PDs are smoothed averages that change more gradually.
A model developer uses a 70/30 train-test split to validate a scorecard. What is the PRIMARY purpose of the holdout (test) set?
Answer: To provide an unbiased estimate of model performance on unseen data
The holdout set is kept separate from training to give an unbiased assessment of how the model will generalize to new, unseen applicants.
Which of the following is an example of an 'application score' as opposed to a 'behavioral score'?
Answer: A score generated at the point of loan application using bureau and application data
Application scores are generated at the time of origination using bureau data and application-supplied information, before any account performance data exists.
A bank's credit model shows a KS of 45 on the development sample but only 30 on the holdout sample. What does this discrepancy suggest?
Answer: The model is overfitting the development sample
A large KS drop from development to holdout indicates overfitting — the model learned patterns specific to the training data that don't generalize well.
Which method is commonly used to handle missing values for a numeric credit variable (e.g., 'months since last delinquency') when building a scorecard?
Answer: Create a separate missing-value bin and assign WoE-derived points to it
Creating a dedicated missing-value bin captures the predictive signal in missingness itself — borrowers with no delinquency history often behave differently from those with a known record.