Underwriting, Automated Underwriting, and Discrimination

Scott Susin

Economist

FHEO Office of Systemic Investigations

HUD FHEO Policy Conference

July 22, 2010

Overview

• Underwriting

• 3 Cs: Credit, Capacity, Collateral

• Automated Underwriting -- what’s automated and what’s not

• Not: product choice, verification, appraisal, pricing, marginal/borderline applicants

• Facts and figures

• AUS reduced denial disparities? Who’s left out?

Underwriting Factors: Credit

• Foreclosures, bankruptcies, liens and/or judgments

• Mortgage delinquencies; Credit delinquencies, repossessions, collections, or charge-offs

• Credit accounts: type, age, limits, usage and status of revolving accounts

• Recent request for new credit

Combine into a score that predicts default (FICO, Fannie/Freddie proprietary sytems)

Underwriting Factors: Capacity

• Debt ratios:

• monthly housing expense-to-income ratio

• monthly debt payment-to-income ratio

• Salaried versus self-employed borrower

• Cash reserves

• Number of borrowers

Underwriting Factors: More Capacity

• Loan Characteristics:

• Product: a 15-, 20-, and 30-year fixed rate, a balloon/reset mortgage, an adjustable rate mortgage, etc.

• Purpose of Loan: purchase or refinance (cash-out or no cash-out)

Underwriting Factors: Collateral

• Borrower's total equity or down payment

• Appraisal

• Property type: a 1-unit or 2- to 4- unit detached property, Condominium Unit or Manufactured Home

• Property use: Primary Residence, Second Home or Investment Property

Automated Underwriting Systems

• Began to be adopted in mid-1990s, today used for almost every loan

• Computer balances different factors rather than human judgment

• Underwriting factors enter into a formula that predicts default

• Requires data on 100,000s or millions of loans and default outcomes to develop

• Fannie Mae: Desktop Underwriter

• Freddie Mac: Loan Prospector

Automated Underwriting Systems

• Feed in credit report, other underwriting factors, AUS provides decision

• Decision is Yes/No, Approve/Refer, not Score

• Decision has conditions (documentation)

“Computers Don’t Discriminate”What’s Not Automated

• Before AUS is run

• Choice of Product, Lender

• AUS says No (Refer)

• Manual Underwriting

• AUS says Yes (Accept)

• Income/Asset Verification

• Appraisal

• Independent of AUS

• Pricing

Choice of Lender & Product

• Choice of Product

• Often made by loan officer/broker

• Opportunity for steering

• e.g., Lenders where most borrowers don’t document income.

• Higher loan price but less work for lender

• Choice of Lender

• Steer to subprime division, lender

• E.g., Baltimore v. Wells Fargo charges Wells steered customers to subprime division

AUS returns “refer” – Manual Underwriting

• Explain circumstances

• Temporary illness, unemployment. Won’t recur.

• Borrower probably needs assistance making the case

• HDS testing study found that real estate brokers more likely to assist white homebuyers than minorities. Same for mortgage brokers, loan officers?

“Lenders Want to Make Loans”

But neither do they want to spend their time on loans that don’t close.

Brokers presumably make a judgment about how to allocate their time, and prejudices can easily enter into their decision

AUS Returns “Accept” – Verification Follows

• Two common reasons for a loan to be denied are: unable to verify income/assets

• Income can be complicated and time-consuming to verify

• Skilled trades

• Tips, commissions, bonuses

• Government programs such as disability

• Do LOs make as much effort to verify Minority borrower’s income as whites?

Income Verification

Potentially subjective judgments

• How much documentation is required?

• Letter from government verifying disability income, or from doctor too?

• Is income stable, likely to continue?

• Letter from employer required?

• NY Times: many lenders now assume that women on maternity leave won’t return to work

Pricing

• Pricing (interest rate, points, and fees) is not determined by AUS. It’s negotiable.

• Lenders would like a higher price

• Yield Spread Premiums or Overages

• Bonuses to broker/LO for selling a higher-rate loan

0.0

5.1

.15

.2.2

5

1990 1995 2000 2005 2010Year/Month

Black/White Hispanic/White

Purchase

-.1

0.1

.2

1990 1995 2000 2005 2010Year/Month

Black/White Hispanic/White

Refinance

Denial Rate Disparity, Seasonally Adjusted

16

The Unscored: Racial and Ethnic Patterns

Percent with No Credit Score

Excluding Unknown Race

Including Estimate for

Unknown Race

Non-Hispanic white 8.8% 18.9%

Black 17.6% 29.5%

Hispanic 12.8% 36.1%

Asian 8.8% 20.0%

American Indian 7.3% 47.3%

Unknown race 56.3%

Total 10.2% 22.9%

Source: Author's calculations from data in Federal Reserve Board Report to the Congress on Credit Scoring , Table 9.

Are the Unscored Creditworthy?

• Catch-22: It’s hard to know because there’s no data on them in credit files

• You’d expect:• Many have little experience paying bills

(young, thin files) • suggests less creditworthy

• Few have major derogatories (bankruptcy, foreclosure, collections) • If they defaulted, they’d have credit scores!

• suggests more creditworthy

Are the Unscored Creditworthy?

• Brookings examined consumers in a few states where utility bills are reported to credit bureaus• Those who have scores only because of utility bills

have about average delinquency rates (consistent with scores in the 680-740 range)

• So people in other states, without scores but with utility bills in their name, probably also have average scores.

• FTC examined use of credit scores to predict auto insurance claims.• Scores are very predictive of insurance claims.

• People without scores have about average claims risk.