Notes on HMDA Data
By Kathryn Pettit and Audrey Droesch, The Urban Institute

In general, the Federal Reserve Board requires depository lending institutions (banks, credit unions, and savings associations) to file under HMDA if they meet the following criteria:

1. A minimum level of assets ($33 million at the end of 2003 to report in 2004), ($36 million at the end of 2007 to report in 2008),
2. A home or branch office in metropolitan areas,
3. Origination of at least one home purchase or refinancing loan on a one-to-four-family dwelling, and
4. Any one of these conditions: the institution is federally insured or regulated; the mortgage loan is insured, guaranteed, or supplemented by a federal agency; or the loan is intended for sale to Fannie Mae or Freddie Mac.

For-profit non-depository institutions (e.g. mortgage companies) must file if they have:

1. Value of home purchase or refinancing loans equaling $25 million or more or equaling 10 percent or more of its loan originations
2. A home or branch office in metropolitan areas, or five or more home purchase or home loan applications, originations, or loan purchases for properties located in metropolitan areas.
3. Assets exceeding a minimum level ($10 million at the end of 2003 to report in 2004), ($10 million at the end of 2007 to report in 2008) or more than 100 home purchase or refinancing loan originations in the preceding calendar year.

Limitations of HMDA data

The HMDA data are less useful as indicators of demographic or economic change of neighborhoods with low homeownership rates.  Changes in home purchase loan amounts for 1-4 family structures may suggest changes in an area’s rent levels, but it is a very indirect measure.  In regards to larger structures, currently released HMDA data combine all multifamily housing loans, whether for purchase, refinancing or improvement.  This makes the loan amount very hard to interpret, since it mixes smaller home improvement loans with purchase loans that typically are larger.

Unfortunately, it was not possible before the 2004 data to determine from HMDA whether an individual loan is subprime. Instead the number of subprime loans were approximated by the number of loans originated by lenders that the U.S. Department of Housing and Urban Development (HUD) has identified as subprime specialists.  Though these "subprime lenders" focus on the subprime market, they may also offer traditional prime market loans.  The indicators based on loans from subprime lenders will therefore include some prime loans from subprime lenders and exclude subprime loans from institutions not identified on the subprime lenders list.

HUD compiled the subprime lenders list from industry trade publications, HMDA data analyses, and self-identification. In cases where lenders offered prime and subprime loans, HUD identified lenders as subprime lenders if they reported that these loans accounted for at least half of their conventional (i.e., not government-backed or insured) business.  HUD also uses feedback from lenders, policy analysts, and housing advocacy groups to update the list.  The subprime lender lists for 1993 to 2005 are available at http://www.huduser.org/datasets/manu.html.

The HMDA files contain three types of information describing the applicant or co-applicants.  Each record reports the applicant’s gross annual income, gender, and race.  While the first two fields are mostly complete, researchers and activists have been increasingly concerned about declining reporting of applicant race.  While lenders must request racial information for in-person loan applications, they have not been required to ask for race in mailed, phone, or Internet applications.  As these latter types of applications have become more commonplace, the proportion of applications of all purposes (i.e., purchase, refinance, home improvement, or multifamily) in the United States with missing race information gradually increased from 8 percent of the total in 1993 to 28 percent in 2002.  Over the same time period, home purchase applications exhibited a better response rate, but the records with missing race data still rose from 4 to 15 percent.  In response to this trend, lenders must request racial information for all applications after January 2003, regardless of the application method.   In the first year of the data reported under this new rule, the rate of missing race data declined ten percentage points to 17 percent for home loan applications of all purposes, and dropped three percentage points to 12 percent for home purchase applications.  Applicants may still refuse to provide their race.

"Mixed Race Pair" refers to two co-applicants who reported two different races.  Since 10 - 20 percent of loans do not report the borrower's race in a given year, we use the number of loans with a valid borrower's race as the denominator for this indicator.

Through 2002, lenders coded each application with the census tract as defined for the 1990 Census.  For 2003, they used the census tract as defined for the 2000 Census.   About half of all census tracts changed boundaries from 1990 to 2000, so loan data summarized to the census tract level before 2003 may not be directly comparable to the tract data in 2003 and after.  To address this issue, a population-based weighting system was used to make the data comparable from 1995 to 2003.  This system was jointly developed by The Urban Institute and GeoLytics, Inc. for the Neighborhood Change Database.  The weighting system re-allocates data reported using 1990 census tracts into 2000 census tract boundaries.  The results have been rounded to the nearest whole number.  As a result, the sum of the results for all individual neighborhoods may not add up to the total for the city.

Click here to access data defintions.

The above text is excerpted from: Pettit, Kathryn, and Audrey Droesch. 2005. A Guide to Home Mortgage Disclosure Act Data. Washington, DC: DataPlace.