Berkeley Haas’s Morse: FinTech Also Practices Housing Discrimination

Poets&Quants Professor of the Week Adair Morse of Berkeley’s Haas School of Business

Can an algorithm discriminate against minorities?

Apparently, it can and does, according to research by Professor Adair Morse of the Haas
School of Business at the University of California at Berkeley and three other Berkeley
professors.

In their working paper, “Consumer Lending in the Era of FinTech,” Morse, and her co-authors
Richard Stanton and Nancy Wallace of Haas and Robert Bartlett of UC Berkeley School of Law,
discovered that even so-called fintech (short for “financial technology”) lenders, driven by computer
algorithms, charge as big a premium to African American and Hispanic applicants as face-to-face lenders,
who would presumably be more driven by their prejudices.

MORTGAGE LENDING IS RAPIDLY MOVING ONLINE

But the algo-driven fintech lenders are moving in the right direction, the researchers found, and show
no greater propensity than human mortgage bankers to reject minority borrowers’ loan applications.
Mortgage lending is rapidly moving online, led by firms like Rocket Mortgage, an entirely digital division
of Quicken Loans, the nation’s largest mortgage lender. Some 45% of the 2,098 large U.S. mortgage
lenders the researchers examined offered completely online or app-based mortgage contracting.

“Nearly all of the big banks and most small lenders now act as fintech,” Adair and her colleagues write.
The U.S. residential mortgage market is dominated by the government-sponsored entities (GSEs),
Federal National Mortgage Association (Fannie Mae) and Federal Home Loan Mortgage Corporation
(Freddie Mac). Fannie, Freddie, and another GSE, Government National Mortgage Association (Ginnie
Mae), invest in or insure more than 90% of all U.S. mortgages.

Fannie and Freddie won’t buy or guarantee mortgages that don’t meet their lending standards,
which usually raises the cost of such “nonconforming” loans to borrowers. Those uniform standards
remove much of the subjectivity from lending decisions, which is where discrimination tends to creep in.
Quantitative standards, of course, are especially amenable to fintech companies and their algorithm-
driven decision-making.

PROFS USED MACHINE-LEARNING TECHNIQUES TO STUDY 8.7 MILLION LOAN APPLICATIONS

Since the financial crisis, the researchers write, Fannie and Freddie have “fully [determined] the price of
credit risk by their role as guarantors. In particular, the GSEs produce a predetermined grid pricing that
prices credit risk” based on loan-to-value ratios and credit scores.

That “grid pricing” enabled Morse and her colleagues to quantify the extent of housing discrimination.
They used machine-learning techniques to create a dataset that for the first time linked income,
ethnicity, and loan-to-value ratio, as well as such specific factors as coupon and loan amount, for five
million accepted loan applications and 3.7 million rejected loan applications between 2007 and 2015.

They looked at only 30-year fixed-rate mortgages on single-family homes securitized by the GSEs, whose
median loan amount was just over $100,000, and they zeroed in on borrowers who had FICO scores
between 630 and 770.

MORTGAGE ‘DISCRIMINATION’ ON 11% TO 17% OF LENDERS’ AVERAGE PROFIT PER LOAN

After stripping out other factors like differences in risk among mortgages issued to different groups, the
researchers found “discrimination…of 5.6 to 8.8 bps for purchase mortgages, or approximately 11-17%
of lenders’ average profit per loan.” And there was virtually no difference between traditional and
fintech lenders. “Interest rate discrimination is almost identical for fintech lenders (5.3 basis points extra paid by minorities) as for the overall set of lenders (5.6 basis points),” they wrote.

Their conclusion: Fintechs do not discriminate less than face-to-face lenders. “It must be the case that any loan officer discrimination that is removed by not seeing faces is captured by algorithms that better predict which borrowers can be captured at higher rates,” they write, suggesting that algorithms identify loans to
minority borrowers as potentially more profitable and so charge them slightly higher rates.
But the aggregate cost is high. Since “each extra basis point in mortgage interest charged due to
discrimination costs minority mortgage holders approximately $100 million per year,” that means
minority borrowers pay $500 million a year more in interest than white borrowers at fintech and
traditional lenders.

The good news is that they discrimination seems to be declining, which, the authors argue, “could be
due to competition from the platforms and/or the ease of shopping-around made possible by online
applications….” And of course, they found fintech lenders don’t discriminate in their decisions to accept
or reject new mortgage applications.

Morse, associate professor of finance, began her career as an entrepreneur in Poland. She studies a
wide range of topics, from entrepreneurship to household finance to tax policy. She got her BA at
Colgate University, two masters’ degrees at Purdue and a Ph.D. in finance from the Ross School of
Business at the University of Michigan. She teaches classes in New Venture Funding and Impact
Investing, and before Haas taught at the University of Chicago Booth School of Business.

Howard R. Gold is a contributing writer for Poets&Quants and a columnist at MarketWatch. He also has written for Forbes, Barron’s, and USAToday. Follow him on Twitter @howardrgold.

Poets&Quants’ Professor of the Week Series

Columbia’S Amit K. Khandelwal

Columbia’s Khandelwal: Trump’s Trade War Hurts the Ones He Loves