In the 1930s, the federal government encouraged lending institutions to deny mortgages to people who lived in neighborhoods with large populations of immigrants and African Americans.
The practice became known as redlining because the government drew lines around certain neighborhoods, deeming them “hazardous,” and colored them red.
It wasn’t until 1968 that the Fair Housing Act outlawed this type of discrimination. But millions of mortgage records analyzed by Reveal from The Center for Investigative Reporting show that the legacy of redlining persists 50 years later: In dozens of cities across the country, African Americans, Latinos, Asians and Native Americans remain more likely to be denied a conventional mortgage than whites.
This disparity existed even after controlling for the applicants’ income, loan amount and certain neighborhood characteristics.
Reveal’s analysis exposed a pattern of denials in major metropolitan areas such as Atlanta, St. Louis and San Antonio and in smaller ones such as Chico, California; Iowa City, Iowa; and Mobile, Alabama.
Philadelphia became a focus of Reveal’s coverage because it consistently proved statistically significant, regardless of which variables were included – and because it has one of the widest lending disparities among the largest metro areas. Black applicants there were almost three times as likely to be denied a conventional home purchase loan as white applicants.
Reveal’s analysis was based on publicly available data released through the Home Mortgage Disclosure Act, or HDMA, and maintained by the Federal Financial Institutions Examination Council. The act, passed in 1975, requires mortgage lenders to report basic data about loan applications to ensure fair lending practices.
The HMDA dataset contains information about nearly every mortgage application in the country in a given year. It includes a wide range of information, from the lending institution that received the application to details about the applicant – such as race, ethnicity and income. It also includes the type of loan that was being sought, the loan amount and characteristics of the neighborhood where the property was located.
Reveal combed through 31 million of these records for 2015 and 2016.
The resulting analysis tested whether people of color were being locked out of homeownership by evaluating loans to purchase a home where the loan applicant intended to live. Only conventional mortgages were included for two reasons: to see how financial institutions acted when the government was not directly involved and because conventional loans typically offer borrowers a better deal.
How Reveal analyzed the data
To determine whether a disparity in lending existed, Reveal used a statistical technique called binary logistic regression. This type of regression assesses the relationship between multiple independent variables and a single binary output. In this case, the output was whether a mortgage was denied.
Lending institutions consider several factors when deciding whether to approve a loan. An applicant’s income is one example. Because income varies among people of different races and ethnicities, it’s important for any analysis of disparities to reduce the influence that race and ethnicity have on income by holding income constant among all applicants. A logistic regression controls for such factors.
Reveal separately analyzed data from 2015 and 2016, looking at nine independent variables against loans that were denied. Those factors included:
- Race/ethnicity of the applicant:
- American Indian or Alaskan Native
- Native Hawaiian or Pacific Islander
- Race and ethnicity not reported
- Gender of the applicant
- Whether there was a co-applicant
- Applicant’s income (includes a co-applicant’s income)
- Loan amount
- Ratio between the amount of the loan and the applicant’s income
- Racial and ethnic breakdown by percentage for each census tract
- Ratio of census tract median income to a metro area’s median income
- Regulating agency for the lending institution
What about credit scores?
Lending institutions say credit scores play an important role in their decision to approve or deny an application, providing crucial information about whether an applicant is likely to make his or her loan payments.
A credit score is assigned to a borrower by a credit rating agency. A higher score means the person makes consistent and on-time payments and is a good credit risk.
Despite its importance in lending decisions, credit score data is not included in the HMDA dataset. So Reveal couldn’t control for it as a variable. Banks consider the data proprietary, and the Freedom of Information Act specifically exempts certain financial information used by bank regulators from being released to the public.
Credit scores also carry their own set of problems. Studies have cited a relationship between lower credit scores and people of color. Credit scores don’t reflect on-time rent or utility bill payments, only those that are delinquent, which disproportionately affects people of color who may have less access to other types of credit, such as mortgages and credit cards.
How did Reveal choose which loans to focus on?
Reveal analyzed conventional loans for one- to four-unit properties where prospective borrowers said they would live, similar to the subset the Federal Reserve analyzes when it tracks lending trends. While a substantial number of applicants of color borrow through the Federal Housing Administration loan market instead, Reveal wanted to gauge relative access to a bank’s premium product – conventional mortgages – between applicants of color and white ones.
After conversations with former officials at the Department of Housing and Urban Development, Reveal opted to include only applications for which the lending institution either made the loan to the applicant or denied it, excluding withdrawn applications and other loan outcomes. This is the methodology that the Department of Justice uses to determine whether there is potential discrimination under the Fair Housing Act of 1968.
For more details about Reveal’s methodology, read the full white paper.