Assessing the Role of Mortgage Fraud, Confluence, and Spillover in the Contemporary Foreclosure Crisis

Abstract

This study explores three features of the contemporary foreclosure crisis that have been highlighted in the literature but relatively neglected in existing empirical research. First, the study evaluates the capacity for levels of mortgage fraud to serve as a potentially important leading indicator of significant foreclosure activity. Second, the study examines the possibility that high foreclosure rates may exhibit spatial dependence, affecting the foreclosure realities of surrounding areas in ways similar to how they have been shown to influence housing prices. Finally, the research considers whether key factors emphasized in scholarly and policy discussions interacted to produce particularly high rates of foreclosure in some areas of the United States. The findings indicate that foreclosure rates in 2008 were significantly higher in United States. counties where “profit-motivated” mortgage fraud was more prevalent several years earlier (i.e., 2004–2006). This study also reveals that, net of a wide variety of other factors, high rates of foreclosure can adversely affect nearby counties by elevating their foreclosure rates. Overall, spatial variation in foreclosure rates appears to be due to additive effects of selected factors rather than interactions of those factors, although the study does show that affordable housing can lessen the tendency for high levels of subprime lending to translate into high foreclosure rates.

Keywords:

• foreclosure
• mortgage fraud
• spatial
• county

Acknowledgment

This project was supported by Award No. 2009-IJ-CX-0020 awarded by the National Institute of Justice, Office of Justice Programs, United States. Department of Justice. The opinions, findings, and conclusions or recommendations expressed in this publication/program/exhibition are those of the author(s) and do not necessarily reflect those of the Department of Justice. We owe a major debt of gratitude to Ann Fulmer, Vice President of Industry Relations for Interthinx®, for generously sharing data and invaluable insights on mortgage fraud. Thanks also to Shane De Zilwa for his help in generating the county-level mortgage fraud data used in the paper.

Notes

 1. For excellent general overviews of research on predictors of foreclosure in other historical eras, see Quercia and Stegman (1992) and Vandell (1995). Gruenstein and Herbert (2000), Immergluck and Smith (2004), and Zimmerman, Wyly, and Botein (2002) examine the spatial distribution of subprime mortgages and foreclosures in the late 1990s and early part of the 2000s.

 2. We exclude a small number of these counties (n = 16) because of missing data on other measures included in the study.

 3. An alternative would be to use decennial census data from 2000, which would be problematic because this predates the foreclosure crisis by several years and may not accurately reflect conditions present during the period leading up to the foreclosure crisis.

 4. There is no research of which we are aware that has formally compared foreclosure estimates from different sources, but some have questioned the reliability of RealtyTrac data for this purpose (Immergluck, 2010b). Given the empirical lacuna that exists on the matter, we used the county-level data assembled for our research and constructed MSA-level REO estimates comparable to the rates Immergluck computed from a private lender database (i.e., LPS). This analysis (not reported in tabular form) showed that the REO rates generated from the two sources are strongly correlated across large MSAs (r = .76).

 5. The numerator for the HAI equation is straightforward, but computing the denominator (i.e., qualifying income) requires several steps and some noteworthy assumptions. For full details of the computation, see http://www.realtor.org/topics/housing-affordability-index/methodology. To construct the county-level measure of HAI used in our analysis, we obtained median housing values from the ACS to gauge median home prices, and we assumed a 20% down payment, a principal and payment that cannot exceed 25% of median family income, and the average United States. interest rate for 2007 (6.337%).

 6. To avoid the loss of additional cases because of data suppression on the race and ethnicity items in the ACS three-year file, we draw the measures of racial composition and ethnicity from a five-year (2005–2009) pooled county-level file.

 7. This is also preferable to other possible options (e.g., an inverse distance or rook contiguity matrix) given the systematic omission of small counties in our data because of data suppression in the ACS, which results in spatial holes that can be problematic for applications of contiguous spatial weights matrices. This type of missing data also can skew spatial analyses based on a nearest-neighbor weight matrix, but this does not appear to be a major problem in our analysis. We constructed spatially lagged measures of logged foreclosure rates using the full sample of counties for which we have foreclosure data (n = 2,881) and the sample used in our analysis (n = 1,801). The resulting spatially weighted variables (one generated from a sample that contains almost all United States. counties and the other based on our sample) were very highly correlated (r = .91), suggesting that the spatial weights matrix used in our study does not introduce substantial bias because of the exclusion of counties with populations below 20,000 persons.

 8. We assessed the robustness of this finding in two ways. First, we reestimated the model using several alternative definitions of the nearest neighbor spatial weights matrix, ranging from the three to eight most proximate neighbors. These different specifications yielded virtually identical results both for the estimated spatial lag effect and for the estimated parameters of other explanatory variables. Second, we reestimated the model after excluding counties from Hawaii and Alaska. These states present unique problems for spatial analysis because of their separation from the remainder of the continent, but we retained them given that our analysis covers other issues as well for which these states can be informative (e.g., the more general assessment of the sources of foreclosure. The results were virtually identical to those reported in the table.

 9. A significant spatial lag effect can mimic other types of spatial relationships, perhaps most notably a spatial spillover effect of one or more explanatory variables. For example, counties in close proximity to areas with high and/or growing joblessness could experience higher rates of foreclosure (independent of their own employment conditions and other conditions) because joblessness in nearby areas depresses housing demand. We considered this possibility in our analysis by estimating a supplementary “spatial Durbin” model that parallels the specification reported in Table 2 except that it includes spatially weighted measures of unemployment rates and the change in unemployment rates. This model (not shown) revealed a statistically significant effect of a spatially weighted indicator of unemployment change on county foreclosure rates, but the magnitude of the spatial lag foreclosure effect and the patterns observed for other variables were virtually identical to those reported in Table 2.

10. There are a variety of other possible reasons for the divergence in our findings and those reported by Rugh and Massey (2010; e.g., the different units of analysis yield variations in available measures, they rely on 2000 census data to define segregation, while we use more recent data to do so), but it could simply reflect the different levels of aggregation used. We consider counties to be particularly useful for the task at hand because they often serve as the basic unit of government, they typically define features that are important considerations in moving decisions (e.g., school districts), they tend to be an important tax entity for homeowners, and they largely define local housing markets. Although MSAs also are meaningful units, they are much more heterogeneous than counties, and thus important relationships could be masked or misrepresented (see also Immergluck, 2010b).