AI in Mortgage Lending Quiz

Answers and explanations appear as you take the quiz.

1. What does an automated underwriting system like Fannie Mae's Desktop Underwriter return?
   • A final, legally binding loan approval
   • A recommendation plus conditions for a human underwriter to clear
   • An appraisal report signed by a licensed appraiser
   • A credit score that replaces FICO
2. Why might a lender order a full appraisal even when an AVM estimate is available?
   • AVMs are illegal for refinance loans
   • The AVM's confidence measure for that property is low
   • Appraisals are always cheaper than AVMs
   • AVMs only work for commercial property
3. Which type of property is an AVM most likely to value poorly?
   • A typical house in a neighborhood with many recent sales
   • A condo in a large building with frequent resales
   • An unusual rural property with few comparable sales
   • A newer suburban home similar to its neighbors
4. What is the main job of intelligent document processing in a loan file?
   • Setting the loan's interest rate
   • Classifying pages, extracting fields and cross-checking figures
   • Negotiating the home's purchase price
   • Choosing which borrowers to market to
5. According to the CFPB's position, what happens when a lender's credit model is too complex to explain easily?
   • The lender is exempt from giving denial reasons
   • The lender must still give applicants specific reasons for a denial
   • The lender may cite 'algorithm decision' as the reason
   • The applicant must request reasons in writing within a year
6. Why can location data create fair-lending risk in mortgage models?
   • Location data is usually inaccurate
   • Because of redlining history, location can act as a proxy for race
   • Federal law bans using addresses in lending
   • Location has no link to property values
7. What lesson does the 2021 shutdown of Zillow Offers illustrate?
   • AVMs are always more accurate than appraisers
   • Valuation forecast errors can carry real financial cost
   • Home prices cannot be modeled at all
   • Document processing caused the losses
8. What is a 'less discriminatory alternative' search?
   • Looking for borrowers from underserved areas to market to
   • Finding a model that performs nearly as well with smaller disparities
   • Replacing all models with manual underwriting
   • Removing credit history from applications