Can HMDA data herald neighborhood changes?
For years, researchers have used data collected under the Home Mortgage Disclosure Act (HMDA) to examine important fair lending issues. Could HMDA data also be used to collect neighborhood-level analyses?
- Assistant Vice President, Community Development
Published March 1, 2006 | March 2006 issue
For the community development field, there is great value in data resources that can be used to monitor neighborhood changes. This sort of data can measure the impact of targeted investments in urban neighborhoods and function as an early warning system if neighborhood conditions decline.
However, most data sets that are inexpensive and widely available to the public present two basic problems for those interested in conducting a neighborhood-level analysis: namely, the frequency and geographic detail of the data. Census data, for example, are widely available, easy to access, and offer a wide range of information at a detailed geographic level. But since they are only collected every ten years, the data do not lend themselves to research by those interested in the up-to-date monitoring of neighborhoods.1/ Other data sets that are updated more frequently, such as quarterly employment and unemployment information, present the opposite problem, as they are typically only available at higher levels of geographic aggregation, such as city or county level. (There are more rigorous ways to conduct neighborhood-level analyses, but they are often expensive and may be beyond the capacity of some community-based organizations to undertake. For more information, see the sidebar following this article.)
Some in the field suggest data from the Home Mortgage Disclosure Act (HMDA) could be used to fill this neighborhood-level information gap. HMDA, enacted in 1975, requires most mortgage lending institutions to collect and disclose information about their lending patterns. This information includes borrower demographics, such as income and race; and loan characteristics, such as property location and loan amount. Congress designed HMDA to 1) assist in determining whether financial institutions were meeting the housing needs of their service areas, 2) target investments where they are needed and 3) identify possible discriminatory lending patterns. (For information on recent changes to HMDA, see the sidebar below.)
For years, researchers have used HMDA data to examine important fair lending issues, particularly in the analysis of the spatial and racial distribution of lending and lending outcomes for a specific lender or within and across Metropolitan Statistical Areas (MSA). Could HMDA data also be used to conduct neighborhood-level analyses? Recent research suggests that they might.
Filling the gaps with HMDA
Under HMDA, lenders annually report data for individual loan applications, including information about the loan and the borrower, at the census tract level. Each fall, the Federal Financial Institutions Examination Council releases lender-reported application data for the previous calendar year for a small fee ($50 for the entire U.S. in 2005). The geographic level of detail provided in the HMDA data, combined with their low cost and annual release, are appealing characteristics for those interested in community and neighborhood development.
Recent regulatory changes expand HMDA data
Since the enactment of the Home Mortgage Disclosure Act (HMDA) in 1975, the Board of Governors of the Federal Reserve System (Board) has made periodic amendments to Regulation C, which implements the act. These changes have increased the variety and amount of information that mortgage lenders are required to collect and disclose under HMDA. With each round of revisions, HMDA data have become more comprehensive and, as a result, more useful to researchers who seek to analyze lending patterns and related issues. The most recent revisions to Regulation C, which took effect in 2004, expand the HMDA data set once again. Specifics:
New data fields. In a Board-driven effort to monitor a new and growing segment of the market that makes "high-priced" mortgage loans, the revised Regulation C adds three new fields under which lenders must disclose loan data: pricing information, including interest rates and fees, for loan applications above a specified rate threshold; the lien status of an application; and whether a loan is subject to the Home Ownership and Equity Protection Act. With regard to pricing information, lenders are required to report the spread between the annual percentage rate on an application and the rate on Treasury securities of comparable maturity for loans with spreads above designated thresholds. The thresholds for reporting differ by lien status: 3 percentage points for first liens and 5 percentage points for junior liens.
Improved data collection and consistency. The revised HMDA regulation requires lenders to designate a property code for each application, in order to distinguish between single-family, manufactured and multi-family housing types. It also requires lenders to use the race and ethnicity classification system common to other federal data collection systems, such as the U.S. Census. The system allows individuals to select more than one race under their specified ethnicity.
Additional information on HMDA changes is available through the Board at www.federalreserve.gov/publications_papers/bulletin/2005/
In a 2005 analysis of a long list of detailed neighborhood indicators, researchers George Galster, Chris Hayes and Jennifer Johnson2/ concluded that HMDA data are useful for monitoring several important dimensions of neighborhood conditions. They began their analysis by gathering dozens of public and proprietary variables that reflect social and economic conditions at the census tract level for five urban areas (Boston; Cleveland; Indianapolis; Oakland, Calif.; and Providence, R.I.) between 1993 and 1999. These cities are atypical in that they collected and organized extensive "neighborhood indicator" data in the mid-1990s. Galster, Hayes and Johnson augmented these 1993-1999 city-provided variables with additional census tract data from the 1990 Census. This rounded out their large set of neighborhood indicators.
Galster, Hayes and Johnson then looked to see if a few of their indicators might possess the following two key properties: 1) they are available cheaply, universally (i.e., widely enough to include most neighborhoods) and at least annually and 2) they reflect much of the information about neighborhood conditions contained in the large set of neighborhood indicators (most of which are not available cheaply, universally and annually). Variables with these two properties would be useful to city governments and neighborhood organizations because they would allow economical, annual monitoring of neighborhoods for possible changes in social and economic conditions—something that is generally difficult to do.
To find such variables, Galster, Hayes and Johnson proceeded in two steps. First, they used a statistical technique called factor analysis to identify six dimensions of neighborhood conditions that summarized most of the information in their full list of neighborhood indicators. They invented labels for each of these six dimensions—social Disadvantage, Housing Type and Tenure, Prestige, Business and Employment, Crime, and Housing Vacancy—based on the types of indicators from the full list that were most closely associated statistically with each dimension. For example, the Prestige dimension got its name from the fact that it mainly reflected indicator variables like the percentage of college-educated or professional persons and the median home value in the census tracts. Second, they used another statistical technique called regression analysis to test how well each of these six dimensions of neighborhood conditions correlated with each of the cheaply and universally available indicators, including indicators based on HMDA data.
For these researchers, HMDA data provided a useful proxy for three of the six dimensions of neighborhood conditions: Social Disadvantage, Prestige, and Housing Type and Tenure.3/
The Social Disadvantage dimension, which mainly reflects indicators such as welfare usage, unemployment, teen birthrates, and female-headed households, was proxied reasonably well by the two-year average rate of home purchase mortgage approvals, computed from HMDA data. The Prestige dimension, discussed above, was well proxied by the two-year median amount of loans for home purchase, computed from HMDA data for each census tract. Finally, the Housing Type and Tenure dimension of neighborhood conditions, which mainly reflects the proportion of single-family and owner-occupied homes in each census tract, was reasonably well proxied by the two-year average number of total mortgage loan applications, again computed from HMDA data.
It is important to remember that Galster, Hayes and Johnson's results are specific to the time periods and five urban areas for which they had data, so there is no guarantee that the same correlations will hold true for other time periods or urban areas. However, since the same HMDA variables worked consistently in their analysis, from 1993 to 1999 for all five urban areas, there is some reason to believe the same HMDA-based proxies may work elsewhere to highlight possible changes in neighborhood conditions that merit the attention of residents and the organizations and local governments that serve them.
So what could these HMDA-based proxies show for urban areas in the Ninth District? To explore that question, we turn to an analysis of conditions in the District's largest MSA. Our analysis uses data from 1992 to 2004 to calculate Galster, Hayes and Johnson's three HMDA-based proxies for the seven-county Twin Cities metropolitan region and selected subareas, with special focus on the two communities that constitute North Minneapolis. A caveat: It is important to keep in mind that the aim of this analysis is merely to provide indicators of neighborhood conditions. The HMDA-based proxies have not been specifically tested for validity in the Twin Cities and, at best, can only approximate movements in the full range of social indicators for which they stand.
Twin Cities analysis
The north side of Minneapolis, often called North Minneapolis for short, is the quarter of the city that lies north of downtown and west of the Mississippi River. It is made up of two smaller communities: Near North, which lies immediately north of downtown, and Camden, which lies between Near North and the city's northern border. It seems likely that annual monitoring of neighborhood conditions in North Minneapolis, of the type the HMDA data seem to make possible, would be of interest to its residents and those who serve them. On the one hand, many households in North Minneapolis appear to be financially vulnerable based on recent high concentrations of foreclosures in the area and, according to Census 2000, high concentrations of poverty, unemployment and low-income residents. On the other hand, this part of the city is potentially attractive for investment and the neighborhood changes it can bring, as it is close to downtown, has an affordable housing stock and has been targeted through recent private and public revitalization efforts (especially the Near North community).
An analysis of HMDA data for the six two-year time periods, from 1992-1993 to 2003-2004,4/ for the three HMDA-proxied neighborhood indicators—social Disadvantage, Prestige and Housing Type and Tenure—reveals important differences in level and trend for the census tracts that compose the north side of the city, as compared to the whole of Minneapolis, Hennepin County and the seven-county region. With regard to Social Disadvantage, the HMDA proxy (the proportion of originated home purchase applications) is consistently lower on the north side than in the three broader geographic areas (city, county and region) for all six two-year time periods. The rate for North Minneapolis reached its peak of 60 percent early in the 1994-1996 period, as it did in the broader geographic areas, and remained at least six percentage points below the rate for the city, county and region during the remaining five two-year periods. Within this area of the city, the origination rate for Camden was at least five percentage points higher than that for Near North. Thus, according to this HMDA proxy, the relative degree of social disadvantage on the north side changed very little over the period.
Despite no relative change in the Social Disadvantage proxy, the HMDA proxy for Prestige (median amount of loans for home purchase) reveals North Minneapolis started low but recently rose rapidly. The median loan amount for North Minneapolis dropped slightly between 1992-1993 and 1994-1996, as the median amount for the broader geographic areas increased. While the city's north side did not keep pace early on, the median loan amount for North Minneapolis has increased at a higher rate than for the three broader geographic areas since the 1997-1998 period. In addition, the total percent change over the six two-year periods (148 percent) outperformed the broader geographic areas. Both the Camden and Near North communities exhibited roughly the same pattern for median loan amounts during this period. However, Near North increased at a higher rate over the period (almost 200 percent) and had a higher median loan amount in 2003-2004 compared to Camden.
Lastly, the HMDA proxy for Housing Type and Tenure (the average number of total mortgage applications per census tract) for North Minneapolis reveals positive biennial percentage increases, even as other areas posted declines. The total change of almost 375 percent from 1992-1993 to 2003-2004 is well beyond the rate of increase for the three broader geographic areas. Within North Minneapolis, the per-tract application average in Camden outpaced Near North by a ratio of at least 2 to 1 during this period and accounted for much of the strength in the level of this indicator for North Minneapolis. Still, Near North did experience the highest rate of change, over 500 percent, between 1992-1993 and 2003-2004.
Conclusions and uncertainties
What do these findings imply about trends in these north side communities between 1992 and 2004? The tentative analysis of the three neighborhood indicators for North Minneapolis broadly indicates that the area may be transforming, primarily due to an increasingly vibrant housing market. If the findings of Galster, Hayes and Johnson can be applied to Minneapolis, the HMDA data proxies suggest that 1) the degree of Social Disadvantage historically present on the north side did not dramatically change between 1992 and 2004; 2) neighborhood Prestige grew slowly during the 1990s, but increased rapidly during the last six years; and 3) neighborhood housing activity has grown at a strong rate since the early 1990s, especially in Camden. While differences between Camden and Near North existed, both communities followed a trend line for the first two indicators that was common to all geographic levels.
While this analysis of HMDA data is useful, especially because it allows for a neighborhood-level analysis beyond the fixed time points of the 1990 Census and Census 2000, it raises a number of questions and concerns when applied. Most of these concerns focus on the interpretation of findings and, more importantly, the ability of HMDA data as structured by Galster, Hayes and Johnson to serve as proxies beyond the cities and period of time for which they were tested.
With regard to the first point, which is more important when using HMDA data as a proxy for an indicator, the level of change or the percentage change? In other words, using the North Minneapolis example, is it more important to focus on the fact that Camden accounted for the overall housing market increase for the area, or the fact that the percentage change over the time period was greatest in Near North? Whatever question is at hand will, most likely, determine which phenomenon is deemed more important.
Regarding the second point, the ability of these HMDA proxies to serve as predictors of neighborhood indicators, at least for the Twin Cities, is open to discussion. This uncertainty is mainly due to a basic assumption lying beneath any application of Galster, Hayes and Johnson's work to more recent years, that the mortgage lending market has remained constant since the mid-1990s. Two important changes since the mid- to late 1990s, namely the increase in the number of refinance applications and the rapid advent of subprime lending, have the potential to challenge the usefulness of one of the HMDA-based surrogates: the Housing Type and Tenure proxy.
As detailed above, the housing market in North Minneapolis, especially in the Camden community, generally outperformed the city, county and region in terms of per-tract applications. The arrival of subprime lending was an important mortgage market change in the late 1990s, especially for refinance applications, and appears to have had an effect on the Housing Type and Tenure proxy. Between 1992 and 2004, refinance applications reported by subprime lenders5/ increased by 600 percent, compared to 168 percent for prime lenders, to account for almost one-third of the refinance market in Minnesota.
Looking more closely at refinance and home purchase applications for this period reveals that beginning in 1999-2000, North Minneapolis experienced a strong period of refinance compared to the three broader geographic areas, which either experienced declines or stabilization. This pattern suggests that refinance activity in North Minneapolis census tracts may not follow a traditional pattern based on market interest rate fluctuations, but may be affected by the differential growth in subprime or perhaps predatory refinancing in North Minneapolis. This finding implies that changes in the way mortgage markets operate could impair the usefulness of this HMDA proxy and, at the very least, suggest that this measure may need further refinement.
Thus, it is important not to simply use the proxy findings without a deeper examination of underlying trends within the HMDA data. For example, an increase in the total number of loan applications reported under HMDA (i.e., an increase in the variable that forms the Housing Type and Tenure proxy) may indeed reveal that a neighborhood housing market has picked up steam. On the other hand, if this increase is largely due to a market change, such as the increase in subprime refinance seen in North Minneapolis, then the utility of this HMDA proxy as presently calculated could be problematic. In either case, it is important for anyone calculating these proxies to become familiar with what the detailed HMDA data reveal about mortgage lending at the neighborhood level. Not doing so could lead to the creation of an ineffective early warning system for neighborhoods.
Even with these reservations, the use of HMDA data as tools to monitor neighborhood conditions shows a great deal of promise. More work on when and how to apply and interpret these proxies in a neighborhood-level analysis is warranted.
1/ According to the U.S. Census Bureau, the American Community Survey may, if fully funded and implemented, eliminate the need for the long form in the 2010 Census and make data available more frequently.
2/ George Galster, Chris Hayes and Jennifer Johnson, "Identifying Robust, Parsimonious Neighborhood Indicators," Journal of Planning Education and Research (24) 2005, p. 265-280.
3/ The Business and Employment Activity dimension was proxied well by a count of businesses based on data collected by Dunn and Bradstreet. The count is available annually and fairly cheaply for most neighborhoods. For two dimensions—Crime and Housing Vacancy—Galster, Hayes and Johnson did not identify a cheaply and universally available annual proxy.
4/ HMDA data from 1995 for the Twin Cities are unavailable, so the six two-year time periods for this analysis are as follows: 1992-1993, 1994-1996, 1997-1998, 1999-2000, 2001-2002 and 2003-2004.
5/ Lender type is determined by a list of subprime and manufactured home lenders compiled annually by the U.S. Department of Housing and Urban Development for the years 1996-2002. For 2003 and 2004, the lenders listed for 2002 were used. For more information, visit www.huduser.org/datasets/manu.html.
Richmond study addresses reinvestment research challenges
Measuring the economic outcomes of investments in urban neighborhoods is difficult, if not impossible, to accomplish due to a number of challenges related to data collection and methodology. Data on neighborhood investment are typically collected—if at all—at a high level of aggregation, such as the census tract or neighborhood level, and rarely in a comprehensive manner. That makes it difficult to measure the outcome of investments precisely. In addition to these important data issues is a whole host of methodological concerns that can potentially limit the validity of an outcome analysis, such as the challenges of precisely attributing outcomes to specific investments or measuring the outcome of an investment at the right place and point in time (especially since its effects may appear after a number of years).
A recently released Federal Reserve Bank of Richmond study, The Ripple Effect: The Economic Impacts of Targeted Public and Nonprofit Investment on Neighborhood Development, aims to surmount these challenges in its analysis of targeted neighborhood investments in the city of Richmond, Va. At a Community Development Forum that the Federal Reserve Bank of Minneapolis and Twin Cities LISC cosponsored in December of 2005, Professor George Galster of Wayne State University, Greta Harris of Richmond LISC and Dan Tatar of the Federal Reserve Bank of Richmond discussed the study, along with Richmond's Neighborhoods in Bloom targeted investment program.
The Richmond study goes far beyond the usual anecdotes about the importance of community development work by quantifying the positive results of a well-defined, replicable investment strategy. In fact, key stakeholders in the effort believe this study can provide essential guidance to other communities about how best to accomplish and measure revitalization in America's urban centers and first-ring suburbs.
In brief, the study evaluates the outcomes of targeted investments made to specific neighborhoods in Richmond over a five-year period starting in 1999. The use of neighborhood-level investment data organized at the block level makes this research effort and study unique, as does the Adjusted Interrupted Time Series analysis model that researchers used to estimate the outcome of investments. Comparisons of housing price changes in Richmond's targeted and untargeted neighborhoods yield the following key findings: