For its mortgage analysis, the Center used federal data collected under the Home Mortgage Disclosure Act (HMDA), enacted by Congress in 1975. The data included more than 350 million mortgage applications covering 1994 through 2007, the most recent year available. The loan-to-income analysis encompassed data from these years. The top 25 high-interest lender list is based on data for the years 2005 through 2007.
The act requires lenders to submit mortgage data to the Federal Financial Institutions Examination Council under rules devised by the Federal Reserve Board. While some small lenders are exempt, more than 8,500 lenders currently report details of the mortgage applications they receive, allowing researchers to track such items as the amount of money requested, the income of the borrower, whether the lender approved the loan, and what the loan was used for. Lenders report loans mostly originated in metropolitan areas, leaving out rural lending, and federal researchers estimate the data capture about 80 percent of all home mortgages. HMDA also requires collection of demographic information such as the race and sex of the borrower — data designed to track and expose discriminatory lending practices. The Center acquired its HMDA data from the National Institute for Computer-Assisted Reporting, a non-profit organization that supports the data needs of journalists.
In choosing its data and methodology, the Center relied on a study by Chris Mayer of the Columbia Business School and Karen Pence, a Federal Reserve economist. In their work Subprime Mortgages: What, Where, and to Whom?, they examined — among other research questions — which data sources likely cover most of the subprime market.
One of those sources dates from 2004, after the Federal Reserve Board changed the rules for reporting HMDA data. With the growth this decade in high-interest loans, including those labeled “subprime,” the Fed began requiring lenders to report a “spread” on their interest rate for approved loans if the rate was three or more percentage points above Treasury securities of comparable maturity at the time the loans were originated. Research showed that at least 98 percent of subprime loans would be captured by the three points and above requirement. The objective, according to the Fed, “was to cover substantially all of the subprime mortgage market while generally avoiding coverage of prime loans.” Thus, for the purpose of this report, the Center defined “subprime” as a spread of three points or higher in the HMDA data.
The Center also examined two other sources in its efforts to document the universe of subprime lending. One of those sources is a list of subprime lenders prepared annually by Randy Scheessele, director of the Mortgage Market Analysis Division at the U.S. Department of Housing and Urban Development. The HUD list, combined with the HMDA data, captured much of the subprime lending in the 1990s, when most of the subprime lenders were primarily engaged in that category of lending. But the subprime market changed this decade, which meant the HUD list might under report the subprime volume, as Scheessele indicated in an e-mail message. “The subprime lender list is a list that identifies subprime lender specialists,” Scheessele wrote. “There are prime lenders who do a significant number of subprime loans but it is not their specialty. Therefore, HMDA data could potentially underestimate the number of subprime loans.”
The Center also looked at data collected privately and sold for use in the real estate industry. Private data collectors rely on self-reporting by the lenders, whose definitions of what is a “subprime” or even a high-interest loan can vary significantly.
Mayer and Pence compared results using the HUD list, private vendor data on securitized mortgages, and the HMDA spread data. “The HMDA higher-priced measure may provide the most comprehensive coverage,” the authors concluded.
Relying on Mayer and Pence’s study, the Center analyzed high-interest, first-lien conventional loans with a spread of three points or higher from 2005 through 2007 for its top 25 list. Only loans collateralized by one-to-four-family properties were included, dropping loans collateralized by manufactured housing. Data from 2004 — the first year that the spreads are reported in the data — were not included because of a suspected high amount of noncompliance with the new HMDA reporting rule and an interest rate yield curve making that year’s spread data less reliable. With much better compliance and the yield problem diminishing, the 2005 through 2007 data have proved much more reliable. That time period captured the height of the subprime lending boom and its collapse, and the spread data allowed the Center to track and quantify the volume of high-interest loans for each lender and rank them.
To look at risks borrowers have taken on through the years — and the lenders who have stoked those risks — the Center looked at first-lien conventional mortgages from 1994 through 2007. Although lien status was not added to the data until 2004, the Center followed Mayer and Pence’s methods and dropped mortgages with loan amounts of less than $25,000, adjusted for inflation each year to 2006 dollars. The Center then calculated a median loan amount backed by median income for each year to reduce the effect of outliers.
The heat maps for each of the top 25 subprime lenders were generated using Palantir Government software from Palantir Technologies on the sample of 2005 through 2007 high-interest mortgages from HMDA. This software mapped the census tract of each subprime loan based on the latitude and longitude derived from the mapping files of the U.S. Census Bureau. It placed a grid over the map and aggregated all the loans in each box of the grid. Then it computed the maximum and minimum number of loans in any box. Each box on the grid is colored according to its position on this scale as shown in each map legend. Colors are keyed using a “heat” scale, with red showing the highest subprime lending and blue showing the lowest. Visual interpolation is used to smooth the transition between the colors of the boxes on the map.
To calculate Federal campaign contributions, the Center used the CQ Money Line database to compile all reported expenditures, from the 1994 through 2008 cycles, identified as being from the parent company or its major subsidiaries, employees of those companies, or political action committees affiliated with those companies. For the overall totals, we removed any contributions from employees to their own corporate PACs, but included contributions from those PACs to other candidates and committees.
To document federal lobbying by individual firms, the Center compiled all lobbying expenditures reported by the company (and major subsidiaries) and separately by its contract lobbyists, in the Lobbying Disclosure Act Database, hosted by the Senate Office of Public Records.
Center reporters attempted to obtain comment from every CEO and company named in the profiles. For firms still active, we reached out to their corporate communications departments. In all, the Center contacted 13 companies (or their successor companies) and more than two dozen former CEOs, their attorneys, or spokespeople. Four CEOs and three companies offered no comment; five companies sent statements.
In its reporting of the project’s main findings, the Center contacted banks that received federal bailout funds and financed subprime lenders. Staff members also contacted the Department of the Treasury and the Federal Reserve. Neither responded to a request for comment.
Help support this work
Public Integrity doesn’t have paywalls and doesn’t accept advertising so that our investigative reporting can have the widest possible impact on addressing inequality in the U.S. Our work is possible thanks to support from people like you.