CFP Risk Assessment & Underwriting 2 — Questions and Answers
Question 1: Which machine learning technique is most commonly used in fintech underwriting to detect non-linear relationships between borrower attributes and default probability?
- Linear regression
- Gradient boosting (Correct answer)
- Moving average
- Principal component analysis
Correct answer: Gradient boosting
Gradient boosting models (e.g., XGBoost, LightGBM) excel at capturing complex non-linear patterns in credit risk data.
Question 2: In open banking-based underwriting, what data source provides the most real-time view of a borrower's cash flow behavior?
- Credit bureau tradelines
- Bank transaction feeds via API (Correct answer)
- Tax return transcripts
- Employment verification letters
Correct answer: Bank transaction feeds via API
Bank transaction feeds accessed through open banking APIs offer up-to-date, granular cash flow data not available from static bureau reports.
Question 3: A fintech lender uses alternative data to underwrite thin-file applicants. Which regulatory concern is MOST relevant when using social media activity as a risk signal?
- Liquidity risk
- Fair lending and proxy discrimination (Correct answer)
- Concentration risk
- Operational risk
Correct answer: Fair lending and proxy discrimination
Social media data can inadvertently serve as proxies for protected classes, raising ECOA and Fair Housing Act compliance concerns.
Question 4: What does a Gini coefficient of 0.70 indicate about a credit scoring model's discriminatory power?
- Poor separation between good and bad borrowers
- Strong separation between good and bad borrowers (Correct answer)
- Perfect random classification
- Model is overfitted to training data
Correct answer: Strong separation between good and bad borrowers
A Gini coefficient of 0.70 (equivalent to AUC of 0.85) reflects strong ability to rank-order creditworthiness.
Question 5: Which underwriting approach best mitigates model risk when deploying an AI-based credit decision system?
- Using a single best-performing model
- Implementing a challenger model framework with ongoing monitoring (Correct answer)
- Relying solely on FICO scores
- Removing human review entirely
Correct answer: Implementing a challenger model framework with ongoing monitoring
A champion-challenger framework allows continuous comparison of model performance and catches degradation before it causes significant losses.
Question 6: In embedded finance underwriting, a merchant's point-of-sale sales data is used to extend working capital loans. This is an example of which underwriting innovation?
- Peer-to-peer lending
- Revenue-based underwriting (Correct answer)
- Collateral substitution
- Behavioral scoring
Correct answer: Revenue-based underwriting
Revenue-based underwriting uses real-time sales or revenue data to assess repayment capacity, especially for SMBs without strong credit histories.
Question 7: What is the primary purpose of a vintage analysis in fintech lending risk management?
- Determining the age of collateral assets
- Tracking loan cohort performance over time to identify underwriting quality trends (Correct answer)
- Calculating weighted average cost of capital
- Assessing liquidity ratios by loan age
Correct answer: Tracking loan cohort performance over time to identify underwriting quality trends
Vintage analysis groups loans by origination period and tracks their delinquency/loss curves to reveal whether underwriting standards are tightening or loosening.
CFP Risk Assessment & Underwriting 2
Which machine learning technique is most commonly used in fintech underwriting to detect non-linear relationships between borrower attributes and default probability?