«Moving to opportunity voluMe 14, nuMber 2 • 2012 U.S. Department of Housing and Urban Development | Office of Policy Development and Research ...»
The cities in Group 2 contain neighborhoods with extremely high delinquency rates, where as many as one out of every three or four borrowers is delinquent. Restoring stability to these neighborhoods will be a special challenge, requiring particularly intensive or imaginative strategies.
These metropolitan areas also are characterized by steep declines in house prices or by high unemployment. The extent of market-driven recovery will be tied to long-term population and employment prospects for the city or region.
Group 3 A third grouping is distinguished by a highly positively skewed, long, or fat-tailed distribution.
Most metropolitan areas in this group have mean delinquency rates in the moderate range and multiple high-delinquency neighborhoods, which may be clustered together. Delinquency rate dispersion is more one-sided than for Group 2, as reflected in the measures of skewness and kurtosis. This group includes Atlanta, Baltimore, Chicago, Miami, and New York.
Local economic trends, property age and condition, and long-run neighborhood conditions such as high vacancy rates before the mortgage crisis will influence the extent or pace of market-driven recovery, as emphasized by Newburger (2010).
Miami is an example of a metropolitan area with both widespread high delinquency and substantial positive skewness. It has by far the highest mean delinquency rate among cities in Group 3 and is closer to Group 2 in this respect. Chicago is more typical of Group 3. A delinquency map of Chicago shows many high-delinquency neighborhoods, mostly clustered on the city’s south side and into neighboring areas southeast of the city, including Gary, Indiana. Chicago has a kurtosis value of 12 and Miami’s measured kurtosis is 7.9, both well above the 5.7 sample average or the 3 associated with a normal distribution.
In the case of Miami, where high delinquency rates are widespread throughout the city and its environs, a regional perspective is required, as with the cities in Group 2. In a city more typical of Group 3, such as Chicago, the focus can be on the neighborhoods constituting the highdelinquency tail of the distribution.
Most metropolitan areas in this group have numerous high-delinquency-rate neighborhoods, requiring a planning perspective that encompasses sizable sections of the city or region. In these cases, strategies to address foreclosure and REO will have to be scalable, as discussed for Group 2.27 The cities in Group 3, like those in Group 2, contain neighborhoods with extremely high delinquency rates, presenting a special challenge.
Often, the higher delinquency neighborhoods will be those where subprime lending was concentrated. Thus, strategies to prevent foreclosure, such as loan modification to reduce the payment burden on households with high-cost subprime loans, could help stem neighborhood decline.
As with Group 1, assessing the potential for spillover effects that could cause the foreclosure problem to expand into adjacent neighborhoods, and taking countermeasures as needed, would be advisable. Again, targeted use of NSP funds to acquire and rehabilitate properties is a possible containment strategy.
Group 4 Group 4 consists of metropolitan areas with low-to-moderate mean delinquency rates, high positive skewness, and steep gradient around the peak-delinquency neighborhood. Low-to-moderate delinquency neighborhoods predominate in these MSAs. As reflected in the skewness measure, however, some neighborhoods will have distinctly higher delinquency, and at least one spatial outlier neighborhood is characterized by a high gradient value.
In general, metropolitan areas in Group 4 have fewer and less extreme high-delinquency neighborhoods than those in Group 3. They tend to have more high-delinquency pockets, or more spatial separation of high-delinquency neighborhoods, in comparison with Group 1. They also are distinguished by the outlier neighborhood having a much higher delinquency rate than neighboring ZIP A few cities in Group 3 (Hartford, Oklahoma City, Rochester, and Syracuse) have a relatively low mean delinquency rate.
Thus, although the neighborhood delinquency rate distribution is positively skewed, relatively few neighborhoods have high or very high delinquency. From a policy perspective, these cities more closely resemble those in Group 4, although they lack the gradient or spatial outlier aspect.
258 Refereed Papers Geographic Patterns of Serious Mortgage Delinquency: Cross-MSA Comparisons Codes, which suggests that the high-delinquency pockets are relatively self-contained (spillover is limited). In many, if not most, cases, the high-delinquency neighborhoods reflect concentrations of subprime lending.
Areas in this group include Albany, Barnstable, Minneapolis-St. Paul, and Virginia Beach. The delinquency map for Minneapolis, for example, shows two distinct high-delinquency pockets, one on the east side of St. Paul and another in northwest Minneapolis, extending north over the city boundary into the lower suburbs. They are relatively self-contained, largely surrounded by areas with much lower delinquency rates.
As with Group 1, neighborhood effects of delinquency and foreclosure would be limited to the higher delinquency pockets, which should then receive particular attention. The policy considerations noted for Group 1 apply to Group 4, with two nuances. First, the spatially separated highdelinquency pockets that are more characteristic of Group 4 may not be amenable to the same responses. Second, the high gradient measure suggests that containing the foreclosure problem may be of less concern.
Group 5 Group 5 is dominated by the gradient measure. Unlike the cities in Group 4, the cities in Group 5 have more or less symmetrical delinquency rate distributions, but, like those cities, they have a high gradient measure. All metropolitan areas slotted to this group have low-to-moderate delinquency means except for Riverside-San Bernardino, which may more appropriately belong to Group 2 (highmean-delinquency cities), and fell into this group only because of an extreme outlier ZIP Code.
The large gradient suggests that the ZIP Code with the highest delinquency rate is isolated from other problematic ZIP Codes. It is possible that this ZIP Code is one of several problematic neighborhoods that are not near each other or that the MSA does not have many neighborhoods with very high delinquency rates.
The density plots for Buffalo and Charlotte, selected for exhibit 4, suggest that policy implications for this group vary, depending on the nature of the outlier ZIP Code and on potential effects of foreclosure in other, higher delinquency neighborhoods. For example, the Buffalo MSA has a single outlier ZIP Code that is associated with the large gradient but, reflecting the distribution’s symmetry, also has substantial mass in neighborhoods with delinquency rates above 10 percent. A delinquency map of Buffalo indicates that the outlier ZIP Code is in the Niagara Falls area, where the delinquency rate is higher than in the other portions of the MSA beyond central Buffalo, whereas a large portion of urban Buffalo has moderately high delinquency rates. Thus, if the neighborhood foreclosure rates are considered problematic, Buffalo may require scalable strategies analogous to those discussed for Group 3. Charlotte, on the other hand, has no ZIP Code with a delinquency rate of 9 percent or more, which suggests that effects on neighborhood stability may not be a concern.
Group 6 The sixth group is the largest cluster. Group 6 consists of metropolitan areas that have low-tomoderate scores for all components; examples include Philadelphia, Pittsburgh, Sacramento, and Washington, D.C. Most cities in this group have moderate mean and skewness. A few, such as
Cityscape 259Brown, Chen, Narragon, and Calem
Sacramento and Tampa, have high means but are distinguished from cities in Groups 2 and 3 by lower standard deviation and skewness; that is, less heterogeneity of neighborhood default rates, without the extremes associated with Groups 2 and 3.28 The density plots for Sacramento and Washington, D.C., selected for exhibit 4, illustrate the relatively compact, mildly skewed delinquency distributions that characterize most cities belonging to this group. A delinquency map of Washington, D.C., shows scattered high-delinquency neighborhoods, mostly adjacent to and east of the city or in outlying suburbs to the southwest of the city. A delinquency map of Sacramento illustrates the different case of widespread high delinquency rates through much of the metropolitan area.
Policy implications for these cities vary with the share of delinquencies in high-delinquency neighborhoods. The more typical metropolitan areas in this group, such as Washington, D.C., have a moderate delinquency mean and some scattered high-delinquency neighborhoods, largely tied to subprime concentrations. As with Group 1, neighborhood effects of delinquency and foreclosure would be limited to the higher delinquency pockets, which should then receive particular attention. Cities in this group, such as Sacramento, with high delinquency means and widespread high delinquency rates, require a citywide or regional perspective, comparable with that of Group 2.
Regression Analysis of Spatial Characteristics Although this article’s primary objective is to classify cities according to spatial characteristics of mortgage delinquency, a secondary goal is to examine the housing market and economic conditions that influence these characteristics. As we emphasized in the preceding section, understanding these factors is important for developing appropriate policy responses. For example, a high foreclosure rate in a lower income neighborhood that is a consequence of concentrations of high-risk lending to vulnerable borrowers might require a different response than would a spike in foreclosures in a far suburb resulting from overbuilding during the housing boom.
In this section, we develop an exploratory, multivariate regression analysis relating the spatial characteristics to economic and housing market conditions across metropolitan areas. The analysis highlights the contribution of subprime lending patterns and identifies the aspects of a metropolitan area’s delinquency patterns that are most closely tied to the housing market cycle and to economic conditions. This analysis is a preliminary attempt to identify some basic relationships; it is not intended to be comprehensive.
First, we introduce a set of potential explanatory variables that we classify into three groups:
(1) subprime lending spatial distribution measures, (2) housing market factors, and (3) other economic factors. Next, we estimate regression equations for each of the four principal components characterizing the spatial distributions.29 The cluster classifications for these cities are robust to using predicted values for their components from the regression analysis in place of actual values.
To select efficiently among the large number of potential explanatory variables, we initially use a stepwise regression procedure for each of the four principal components. Because stepwise regression may generate some arbitrary selections, we subsequently evaluate and test the robustness of the resulting variable selections to the inclusion of omitted variables and adjust the specifications as appropriate. We dropped a few variables where the selection was questionable because of marginal statistical significance and colinearity or redundancy with other included variables.
260 Refereed Papers Geographic Patterns of Serious Mortgage Delinquency: Cross-MSA Comparisons Subprime Spatial Distribution Because subprime loans are disproportionately represented among delinquent mortgages, we expect that distributional moments and spatial patterns of previous subprime lending activity in a
metropolitan area influence mortgage delinquency patterns. We describe the characteristics of subprime lending across ZIP Codes using measures analogous to those used for mortgage delinquency:
mean, standard deviation, kurtosis, and skewness for percent of active loans that are subprime (weighting by active subprime count), as well as spatial autocorrelation and gradient measures.
Exhibit 5 provides summary statistics for the eight analysis variables.
Housing Market Variables We expect distributional moments and spatial patterns of mortgage delinquency to be tied to housing market activity. For example, delinquency rates will be higher in cities with more rapidly depreciating home values during 2007 and 2008.
Variables associated with the housing market boom and bust considered in the regression analysis include (1) annual home price appreciation rate from the third quarter of 2005 through the third
quarter of 2006 and from the third quarter of 2006 through the third quarter of 2008 in each MSA, (2) annual change in MSA housing starts over these periods, (3) the percentage of MSA home purchase loans in 2005 and 2006 that were for nonprimary residence (investment property or second home), and (4) the National Association of REALTORS® housing affordability index for the third quarter of 2005 and the third quarter of 2006.
We also construct measures of the spatial distribution of housing market activity for inclusion in our regression equations. Specifically, we calculate the distributional moments of percent change in home purchase loan originations from the third quarter of 2005 through the third quarter of 2006 and from the third quarter of 2006 through the third quarter of 2007 in each MSA: mean, standard deviation, kurtosis, and skewness (weighting by ex ante number of originations), along with the spatial autocorrelation measures. Spatial patterns of home purchase lending activity during the housing boom or at the beginning of the downturn may help differentiate neighborhoods where the market “overheated,” as reflected in subsequent mortgage delinquency patterns.