Heterogeneity in the Neighborhood Spillover Effects of. Foreclosed Properties

Transcription

1 Heterogeneity in the Neighborhood Spillover Effects of Foreclosed Properties Lei Zhang Edinboro University of Pennsylvania Tammy Leonard University of Texas at Dallas James C. Murdoch University of Texas at Dallas August 30, 2011 Abstract In this paper we examine heterogeneity in the simultaneous space-time impact of foreclosures on neighborhood property values. The heterogeneity arises from foreclosed properties that experienced different types of foreclosure outcomes. We find that for foreclosures that are not observed to have concluded with a market sale, the negative price impacts begin while the foreclosing household still has ownership of the property then start to diminish 6 months after the bank has taken possession of the property. However, for foreclosures which are observed to have a subsequent market sale, the price impacts do not occur until at least 6 months after the foreclosing household has lost ownership of the property. Further, houses that experience a market sale within 6 months of the household losing ownership of the property are associated with small neighborhood price effects. Corresponding author. Address: University of Texas at Dallas; School of Economic, Political and Policy Sciences; 800 W. Campbell Rd.; Richardson, TX Phone: (972) murdoch@utdallas.edu. 1

2 1 Introduction Several authors have found negative neighborhood price impacts associated with foreclosed properties (Harding et al., 2009; Leonard and Murdoch, 2009; Lin et al., 2009) but the geographic extent and magnitude of the price impacts vary. Lin et al. (2009) found that the most severe impact is an 8.7% discount on neighborhood property which is located within 100 meters of the foreclosed property and liquidated less than two years after the foreclosure. Leonard and Murdoch (2009) find similar distance effects but a relatively smaller neighborhood price discount. Only Harding et al. (2009) attempts to examine the joint variation in time and space of the neighborhood price impacts of foreclosure and while their results suggest a negative price impact that decays rapidly with distance from the foreclosed property, the magnitude of the price impact and the speed of decay differ by geographic region. There are many reasons to expect geographic variation in the price impacts including differences in housing market fundamentals, state laws and the density of foreclosures. Within a particular market, however, there may still be substantial variation in the time-space price effects associated with foreclosures. In other words, even controlling for housing market, legal and density differentials, the time-space price-foreclosure relationships may not all be the same. This heterogeneity may be expected because the foreclosure event is really just a part of a longer process and foreclosure processes can be substantially different even within small geographic regions. The role of the foreclosure process itself is important from a policy perspective. By knowing what aspects of the foreclosure process are related to the larger neighborhood price declines, we can imagine designing policy aimed at improving the efficiency of the foreclosure process and reducing the social costs associated with foreclosure. While there is a general consensus that foreclosed properties produce localized neighborhood price declines, the degree to which this effect varies across different types of foreclosed properties is unknown. Some foreclosures experience significant neglect and, almost certainly, generate negative price impacts. However, a negative price impact might also be expected for un-neglected 2

3 properties perhaps due to a negative stigma associated with foreclosure. Using data from Dallas County, Texas, we compare space-time econometric estimates of the neighborhood price impacts of foreclosure processes that terminate in a market sale of the property reported by the multiple listing service(mls) to those that do not clear through the MLS. This distinction facilitates an assessment of the potential benefits of maintaining foreclosed property and effectively returning it to the housing stock. 2 Background There are two theories for why neighborhood price declines might be observed when nearby properties experience foreclosure. The first, presented by Lin et al. (2009), is based on realestate pricing models that use the sale prices of comparable nearby properties to base list prices when properties are put on the market. To the extent that foreclosed properties usually sell at a discount, they depress the price of comparables generating reduced neighborhood home prices. Leonard and Murdoch (2009) propose a different mechanism one that is based on neighborhood condition. If properties going through the foreclosure process are maintained at lower standards, then the negative neighborhood price impacts reflect the real price declines associated with lower neighborhood quality. Harding et al. (2009) suggest this is the more likely mechanism since they find that the greatest neighborhood price impacts occur before the market sale of the foreclosed property; i.e., before the foreclosure sale price can be used as a comparable in real estate pricing models. Apparently, something is occurring during the foreclosure process that generates the price declines. However, not all properties experience the same foreclosure process; thus, we suspect that the effects of foreclosure might vary by different types of foreclosure processes. In Texas (the location of our study), the foreclosure process begins with a foreclosure notice (Notice of Default or Lis Pendens). Then, Texas law allows for two types of foreclosure: a nonjudicial foreclosure and a judicial foreclosure (Law, 2007). A nonjudicial foreclosure is 3

4 performed when the power of sale is present in the mortgage or deed of trust, allowing the lender to foreclose without obtaining a court order. A judicial foreclosure is primarily used when a government entity is seeking to collect property taxes. The government must file a lawsuit in order to have the home sold to collect the taxes (State Bar, 2011). In what follows, we are concerned with nonjudicial foreclosures. During the foreclosure process, a home can potentially be owned by three different parties. First, it is owned by the people that default on the mortgage. Second, the home will go to auction where either the bank owning the mortgage will end up with the property (REO) or another party will purchase it (Trustee Sale). Last, the property can be sold to someone that utilizes it for housing. When the home is transferred to a final user, we consider the foreclosure process to be over. 1 Even though in principle the foreclosure process appears quite uniform, the time between the initial default and re-purchase varies widely indicating heterogeneity in the foreclosure process. The potential for variability in foreclosure processes suggests that the neighborhood price impacts of foreclosure may well vary across both time and space within housing markets. Harding et al. (2009) examines the price impact of foreclosed properties in both dimensions with a repeat sales model and data from seven different MSAs. Besides the rich data set from multiple MSAs, one of the advantages of their method is that they have precise information about the timing of the changes in ownership of the foreclosed property. They are able to identify the time at which the property became an REO property and the exact time at which the property is resold to a third party (through a market REO-sale). They find that the strongest foreclosure price impacts occur around the time that the property changes hands from the foreclosing homeowner to the bank holding the failed mortgage. In the analysis that follows, we use Harding et al. (2009) s framework for controlling for 1 Sometimes, the foreclosure process terminates prior to a transfer of ownership either because the first owner becomes current on their mortgage or the lender modifies the loan. Modifications became increasingly utilized with the passage of new legislation in 2009 (e.g., Making Home Affordable Plan and the Home Affordable Modification Program) as a response to the financial crisis that began in 2007 (Stewart, 2010). However, the foreclosures we examine occur in 2008 or earlier; therefore they are not likely to be affected by this policy initiative. 4

5 the simultaneous space and time variations in the neighborhood price impacts. However, rather than repeat sales methodology we use spatial econometric modeling techniques as presented in Leonard and Murdoch (2009) to control for unobserved neighborhood characteristics and local house price trends that if not adjusted for likely bias the estimated price impact of foreclosure. 2 The spatial econometric approach for dealing with these unobserved neighborhood level attributes allows us to estimate the price impact of foreclosure using all housing sales transactions rather than just those for which we can observe a repeat sale. This is an especially important distinction when estimating the price impacts of foreclosure because the likelihood of observing a repeat sale may be related to the likelihood of having neighborhood foreclosures. 3 Data Our data comes from real-estate transactions and foreclosures recorded in Dallas County, Texas. Housing market sale price data for year 2004 to 2010 were obtained from the University of Texas at Dallas Real Estate Research Database. This database contains historical records on housing characteristics, appraised values and various types of sales. We use the sales prices for properties sold in 2008 as the base year because it allows us to examine the price impact of foreclosures that occurred in the wake of the financial crisis which began in The detailed variable descriptions are presented in Table 1. The average home sold for approximately $250,000, has 2,000 square feet of living area, two bathrooms, and is 35 years old. A list of foreclosures from RealtyTrac over the period from 2004 to 2010 was used to identify homes that foreclosed. The foreclosure data were geocoded and spatially merged to the real estate database to facilitate calculating neighborhood foreclosures and to identify completed foreclosures. Completed foreclosures are defined as foreclosures for which a market sale was found in the real estate data sometime after the foreclosure but before December 2010 (the last observation time in our sales data). Other, incomplete, foreclosures 2 See Kobie and Lee (2011) for another application using spatial econometric specifications. 5

6 are observed in the RealtyTrac listing but are not later found in the real estate database. At the time of our analysis, the resolutions of incomplete foreclosures is unknown. Incomplete foreclosures thus include all foreclosures which were not re-sold through a market transaction by December 2010 or a minimum of 2 years after the foreclosure began. In the 73,998 foreclosures observed from 2004 to 2010, 22,241, or 30%, are classified as complete with the balance, 70%, being incomplete. For each house that sold in 2008, the foreclosure data is summarized by counting the number of foreclosures in several time and distance threshold categories. We categorized the distance between each market sale property and foreclosed properties by using four concentric rings with different radii measured by Euclidean distance in feet: feet (ring 1), feet (ring 2), feet (ring 3), and feet (ring 4). The time effects were categorized according to the phase of the foreclosure process. Following Harding et al. (2009), we consider 13 time frames as shown in Figure 1. For complete foreclosures, we assume that the foreclosure process has ended when we observe a third-party market transaction. We break the time before foreclosure starts (F) into four quarterly periods, the time after F and before foreclosure sale (S) into at most five periods, and the time after S into four quarterly periods. Since not every loan passes through all post-foreclosure periods, once S occurs, we assign the date to one of the post-s categories. We define the last post-f and pre-s period to include all properties that are foreclosed more than 12 months before the sale. For incomplete foreclosures, there are two stages: pre-foreclosure and foreclosure process. Since for these properties we do not observe a market sale, we do not know when the foreclosure process ends. We count the number of foreclosures associated with each market sale by time and distance creating a total of 52 possible time-space buckets for complete foreclosures, and 36 buckets for incomplete foreclosures. Table 2 displays the summary information for the counts. We can see approximately 2% of the homes have complete foreclosure neighbors within 250 feet whose foreclosure sales are at the period of F-12 to F-9. The largest number 6

7 appears at the period of S to S+3, and distance 1000 to 1500, where about 54 out of 100 houses have complete foreclosure neighbors in that bucket. The last column of Table 2 is the mean ratio between incomplete and complete foreclosures. We can see that the magnitude of the incomplete foreclosures are all larger than complete foreclosures. If we look at the trend of mean ratio across time and space (Figure 2), we can see the mean ratios increase dramatically 6 months after default occurs. This is because complete foreclosures are moved to the post-foreclosure stage as soon as they are sold to a third-party via a market transaction. The large increase in mean ratios indicates that for many properties this occurs 6 to 9 months after default. This is expected since the foreclosure process in Texas averages 245 days from the initial default event to completed foreclosure (TDHCA, 2006). 4 Estimation and Results To identify the neighborhood time and space effects of foreclosures, we use hedonic price models (Rosen, 1974; Kim et al., 2003) controlling for property and neighborhood characteristics. The basic model is as follows: P 08 = βx 08 + γd 08 + ρ t W t P t + α dp C dp + δ dp I dp + ε (1) t=05 d=1 p=1 d=1 p=1 where P is the vector of natural log of sale prices and the subscript denotes the year (e.g., 08 for 2008). X is a matrix of property characteristics (e.g., living area in square feet, number of bathrooms, age of the house, etc.), and D is a matrix of dummy variables that control for fixed effects due to school districts and the month in which the sale took place. The monthly dummies control for any seasonal effects in home prices (Goodman, 1993). W t P t is the spatial weighted average of nearby sale prices in year t and is included in the model for t = 2005 through t = 2008 to control for the trend in neighborhood price levels. Each element of the spatial weight matrix W is calculated based on the inverse distance from sales in year t to sales in year For properties greater than 2000 feet away, the 7

8 corresponding element in the weight matrix is set to zero. Finally, all of the weights matrices are row standardized 3. Variable C dp is the number of complete foreclosures at the distance d and foreclosure period p, while I dp represents the incomplete foreclosure counts for each time-space bucket. The sets of coefficients, α dp and δ dp, measure the spillover effect of an extra complete or incomplete foreclosure, respectively, on the respective property value while holding other variables constant. Because P 08 appears on the left and the right hand side of the model, equation (1) is a classic spatial lag model that is not amenable to OLS estimation (Anselin, 1988). Instead maximum likelihood estimation is used to provide consistent parameter estimates (Lee, 2004). We estimate three different specifications. Model 1 only includes time effects. Thus, for Model 1, the vectors C and I in equation (1) include the counts of foreclosures at each time interval that are within 1,500 feet of the reference property. Next, Model 2 only considers distance effects and lumps all of the time effects together. For Model 2, C and I include the counts of foreclosures in each concentric ring around the reference property which are in any stage of the pre-foreclosure, foreclosure or post-foreclosure process. Finally, Model 3 is estimated with both time and distance effects simultaneously. All results except the simultaneous distance and time effects in Model 3 are reported in Table 3. The time-distance effects for Model 3 can be found in Tables 4 and 5. Note that the models explain more than 80% of the total variation in the dependent variable. To determine if there is a statistically significant difference in the way complete and incomplete foreclosures impact neighboring sales prices, we conduct a likelihood ratio (LR) test on model 3 4. The null hypothesis is that the coefficients of complete foreclosures are jointly not different from the coefficients of the incomplete foreclosures. The P-value of the χ 2 distribution with 9 degrees of freedom is less than 1% 5. Hence, we reject the null and 3 See Leonard and Murdoch (2009) for detailed explanation. 4 The LR tests results for model 1 and 2 are similar. 5 The restricted model is that 9 coefficients of incomplete foreclosure counts are the same as those of complete foreclosures. We calculate the log likelihoods for restricted L r and unrestricted model L u. Then, evaluate χ 2 = 2( L r L u ). 8

9 include both complete and incomplete foreclosure variables in our model. 4.1 Time Effects Turning now to the coefficient estimates, we see that in all models the estimated coefficients for house characteristics, school district fixed effects and time fixed effects are as expected. Additionally, the estimates on the local neighborhood price trends are all positive and statistically significant, highlighting the importance of including these variables. Figure 3 displays the estimated foreclosure effects resulting from foreclosures within 1,500 feet of a non-distressed sale as estimated in Model 1. All the coefficients are converted to percentage change of sales price using a logarithm transformation 6. The two series in the figure one for complete foreclosures, and the other for incomplete foreclosures clearly indicate that the effect varies with the period and type of the foreclosure. For complete foreclosures, the negative spillover effects do not significantly affect the neighborhood sales price until 6 months after the property becomes REO. The strongest effect appears 9 months after the foreclosure stage begins and diminishes until the foreclosed home is sold. Once the third party buyer takes possession of the property, the negative price effect stabilizes at a very small level and is insignificant 9 months into the post-foreclosure stage. On the other hand, the incomplete foreclosure effects begin much earlier. The first statistically significant negative price impact for incomplete foreclosures is seen as early as 6 months prior to the homeowner losing possession of the property. Additionally, the price impact of incomplete foreclosures is more prolonged. It continues through the remaining consecutive quarters and does not completely disappear over the period analyzed. The strongest effect appears right after the homeowner loses possession of the property, which is about 6 months earlier than the strongest effect of complete foreclosures. 6 Percentage change = e α 1. 9

10 4.2 Distance Effects Next, we examine the aggregate distance effects (Figure 4) as estimated in Model 2. For both complete and incomplete foreclosures, the spillover effect is strongest within 250 feet from the reference property. Within Ring 1, the price impact of incomplete foreclosures is larger (in absolute value) than the similar impact of complete foreclosures. However, the difference between complete and incomplete foreclosures diminishes in the subsequent rings. 4.3 Simultaneous Time and Distance Effects Finally, we examine the simultaneous time and space price impacts of neighborhood foreclosures in Model 3. Now there are a total of 52 different time-space buckets (13 time intervals X 4 distance intervals) for which effects are estimated and the coefficients for these foreclosure effects are presented in Tables 4 and 5 for complete and incomplete foreclosures, respectively. With so many different time-space buckets possible, it is more difficult to find statistically significant results. Therefore, we focus our discussion on the foreclosure spillover effects in ring 1. Recall that Figure 4 indicates that ring 1 is the location of the majority of the spillover effects. The effects for the outer rings are smaller, but generally support the same patterns and conclusions as the effects for ring 1. Figure 5 presents the percentage change in house prices as a result of additional complete or incomplete foreclosures in ring 1 based on the estimates from Model 3. The trend is similar to the aggregate time effects displayed in Figure 3, but the magnitude is much larger. This larger price impact is expected in Figure 5 because only the closest foreclosures (those occurring within 250 feet of the reference property) are included in the estimated effects. When other variables are held constant, each additional foreclosure within 250 feet of a house results in an expected property value decrease of up to 9.9% (complete) or 9.8% (incomplete). Given an average sales price of $250,826 in Dallas county, these imply decreases in value of approximately $24,831 and $24,580, respectively. While the magnitude of the price impacts are similar to Lin et al. (2009), there is the noted difference in the timing and duration of the price impact 10

11 between complete and incomplete foreclosures. In analyzing the results for Models 1 and 3, it becomes apparent that the price impacts for complete foreclosures begin 6-9 months into the foreclosure stage. This is also the time at which the mean ratio of incomplete to complete foreclosures begins to increase (see Figure 2). Thus, the timing of the foreclosure effects for complete foreclosures coincides with the time that many complete foreclosures are sold in the housing market so that they move out of the foreclosure stage and into the post-foreclosure stage. A logical next question then is if the price impact of complete foreclosures is observed only for those foreclosures which remain in the foreclosure process for an extended period. To explore this question, we estimate a fourth model which differs from Model 3 only in the treatment of complete foreclosures. In Model 4, the post-foreclosure phase is removed and complete foreclosures continue to be categorized in the foreclosure phase even after the market sale occurs grouping complete foreclosure that have sold with those that have not yet sold. The coefficient estimates for the foreclosure effects of Model 4 are presented in Tables 6 and 7 and the results for ring 1 are plotted in Figure 6 7. When the complete foreclosures that have sold on the market are grouped with those that have not yet sold, the price impact of foreclosure disappears. This suggests that complete foreclosures which sell within 6 months after the foreclosure stage begins have a very small and brief negative price externality while it is those foreclosures which take longer to sell that are responsible for most of the neighborhood price externality. 4.4 Robustness Check A potential concern with the models presented thus far is endogeneity between the occurrence of incomplete foreclosures and home sale prices. It is possible that there is some unobserved neighborhood characteristic which simultaneously increases the likelihood of observing an incomplete foreclosure and a lower home sale price. We included local neighborhood price trends from 2005 to 2008 in all of our models to correct for this possibility, but it is possible 7 The coefficient estimates for the other control variables do not differ substantially from those reported for Model 3. Results are available from the authors upon request 11

12 that some endogeneity remains. A Durbin-Wu-Hausman test indicates that the local neighborhood price trends do not fully explain the ratio of incomplete to complete foreclosures. 8 Since the potential endogeneity is related to an increased likelihood of observing one type of foreclosure over another, we estimated an additional model 9 that explicitly includes the percentage of foreclosures that are classified as incomplete within 2000 feet of each sale. The model also includes all of the other covariates found in Model 3. The estimated coefficent for the new variable measuring the percentage of incomplete foreclosures is and is significant at the 1% level. However, the presence of this previously omitted variable does not significantly change the neighborhood price impacts of foreclosure. All other coefficient estimates in this model are similar to those observed in Model 3. Thus, we conclude that a tendency for some neighborhoods to be associated with a larger number of incomplete foreclosures is not fully explained by the local neighborhood price trends which we include in all of our models, but it also does not account for the observed differences in the price impacts of complete and incomplete foreclosures. 5 Discussion and Conclusions The purpose of this study was to examine heterogeneity in the simultaneous space-time impact of foreclosures on neighborhood property values based on different types of foreclosure outcomes. We find considerable heterogeneity in price impacts, which may help explain variations in other published results of the price impact of foreclosed properties, underscoring the need to further examine the foreclosure process when developing policy recommendations for decreasing the costs associated with foreclosure. We find that for incomplete foreclosures, the price impacts begin while the foreclosing household still has ownership of the property and continue throughout the foreclosure process. However, for complete foreclosures, the 8 To test for this, we apply a Durbin-Wu-Hausman test to test if the ratio of the number of incomplete foreclosures to the number of both types of foreclosures within 2000 feet of a market sale is endogenous. Lagged spatial weighted average of neighborhood sale prices (our W t P t terms in equation (1)) were used as insturments. The test fails to reject the null hypothesis that the ratio is exogenous. 9 Results are available from the author upon request. 12

13 price impacts do not occur until at least 6 months after the foreclosing household has lost ownership of the property. Further, houses that sell within 6 months of the household losing ownership of the property are associated with very little neighborhood price externalities. These results should be viewed in light of the limitations of the study. First, the data is drawn from one county Dallas County, TX. Dallas County was not an area examined by Harding et al. (2009) and we confirm their general space-time results. Limiting our study to one area is advantageous for isolating the impact of heterogeneity in foreclosure outcomes, but certainly the external validity to other markets should be tested. Also, while we are able to identify third party market transactions for the complete foreclosures, we cannot identify the final resolution of incomplete foreclosures. More detailed data detailing the ownership of foreclosed properties at each stage of the foreclosure process would be extremely useful in further untangling the varied neighborhood price effects of foreclosure. Regardless of these limitations, the timing of the foreclosure effects estimated for the two types of foreclosure is both interesting and concerning. Leonard and Murdoch (2009) and Harding et al. (2009) both suggest inadequate property upkeep as a mechanism for the observed price impact of foreclosure. The results for incomplete foreclosure support this notion as the price impacts begin when the homeowner still has possession of the house and they start to decline (in absolute value) 6 months after another party (usually the bank) resumes ownership. However, for complete foreclosures which sell within 6 months of the begining of the foreclosure process, the timing of the small negative price impact is suggestive of a mechanism which works through real-estate pricing models based on comparables as suggested by Lin et al. (2009). Thus we find evidence in support of the two different mechanisms for the propagation of neighborhood price externalities depending upon the type of foreclosure process. The price impacts associated with complete foreclosures which do not sell quickly are more puzzling. For these complete foreclosures which take longer than 6 months to liquidate in the market, the price declines occur while the property is most likely REO status. For both incomplete foreclosures and complete foreclosures which take more than 6 13

14 months to liquidate, we find a deleterious price impact that cannot be attributed to realestate pricing models based on comparables because the impact occurs before the property enters the set of neighborhood comparables. This suggests a scope for policy action to reduce decay in the neighborhood or the property itself in order to reduce the propagation of the negative neighborhood price externality. However, the data limitations of this study do not allow us to identify whether policy would best be targeted at the neighborhood or individual property level. At best, we can definitively say that there is heterogeneity in the price impacts of foreclosures and one source of that heterogeneity is the foreclosure process itself even among foreclosures which proceed under the same set of laws, institutions and housing markets. Further study will be necessary to examine the underlying causes of long liquidation times for some properties in order to better identify the underlying causes for the heterogeneity which we elucidate. References Anselin, L. (1988): Spatial econometrics: methods and models, Dordrecht: Kluwer. Goodman, J. L. J. (1993): A Housing Market Matching Model of the Seasonality in Geographic Mobility, Journal of Real Estate Research, 8, 117. Harding, J. P., E. Rosenblatt, and V. W. Yao (2009): The contagion effect of foreclosed properties, Journal of Urban Economics, 66, Kim, C. W., T. T. Phipps, and L. Anselin (2003): Measuring the Benefits of Air Quality Improvement: A Spatial Hedonic Approach, Journal of Environmental Economics and Management, 45, Kobie, T. F. and S. Lee (2011): The Spatial-Temporal Impact of Residential Foreclosures on Single-Family Residential Property Values, Urban Affairs Review, 47, 3 30, /

15 Law, U. S. F. (2007): Texas Foreclosure Law Summary,. Lee, L.-F. (2004): Asymptotic Distributions of Quasi-Maximum Likelihood Estimators for Spatial Autoregressive Models, Econometrica, 72, Leonard, T. and J. Murdoch (2009): The neighborhood effects of foreclosure, Journal of Geographical Systems, 11, Lin, Z., E. Rosenblatt, and V. Yao (2009): Spillover Effects of Foreclosures on Neighborhood Property Values, The Journal of Real Estate Finance and Economics, 38, Rosen, S. (1974): Hedonic prices and implicit markets: product differentiation in perfect competition, Journal of Political Economy, 72, State Bar, o. T. (2011): Facing Foreclosure, Texas Bar Journal, 74, Stewart, L. (2010): 2009 State Residential Mortgage Foreclosure Laws, Tech. rep. TDHCA (2006): A Study of Residential Foreclosures in Texas, Tech. rep., Texas Department of Housing and Community Affairs. 15

16 Table 1: Variables, descriptions, and basic statistics 16 Variable Description Type Mean Std. Dev. saleprice Sales price Continuous 250, ,509 livarea Living area in sqft Continuous 2,147 1,087 baths Number of bathrooms Continuous effage Age of the house in yrs Continuous pool Pool Dummy story1 One story Dummy story stories Dummy slab Slab foundation Dummy centralheat Central heat Dummy onefire One fire place Dummy twofire Two and more fire places Dummy atgarg Attached garage Dummy atcp Attached carport Dummy cond1 - cond7 Condition from bad to good Dummies - - m1 - m11 Month of sale Dummies - - sd1 - sd13 School district Dummies - - Note: Statistics for conditions, months of sale, and school districts are omitted due to lack of space.

17 Table 2: Number of foreclosures by distance and time Phase of foreclosure Distance to obj. Complete Incomplete Ratio* Mean Std. Dev. Mean Std. Dev. F-12 to F-9 0 to F-12 to F to F-12 to F to F-12 to F to F-9 to F-6 0 to F-9 to F to F-9 to F to F-9 to F to F-6 to F-3 0 to F-6 to F to F-6 to F to F-6 to F to F-3 to F 0 to F-3 to F 250 to F-3 to F 500 to F-3 to F 1000 to F to F+3 0 to F to F to F to F to F to F to F+3 to F+6 0 to F+3 to F to F+3 to F to F+3 to F to F+6 to F+9 0 to F+6 to F to F+6 to F to F+6 to F to F+9 to F+12 0 to F+9 to F to F+9 to F to F+9 to F to >F+12 0 to >F to >F to >F to S to S+3 0 to S to S to S to S to S to S to

18 Table 2 continued Phase of foreclosure Distance to obj. Complete Incomplete Ratio* Mean Std. Dev. Mean Std. Dev. S+3 to S+6 0 to S+3 to S to S+3 to S to S+3 to S to S+6 to S+9 0 to S+6 to S to S+6 to S to S+6 to S to S+9 to S+12 0 to S+9 to S to S+9 to S to S+9 to S to * Ratio = Incomplete mean / Complete mean 18

19 Table 3: Regression results Model 1 Model 2 Model 3 ML time ML distance ML time and distance beta Std. Error beta Std. Error beta Std. Error livarea E-06*** E-06*** E-06*** baths *** *** *** effage 2.81E E E pool *** *** *** story *** *** *** story *** *** *** slab *** *** *** centralheat *** *** *** onefire *** *** *** twofire *** *** *** atgarg *** *** *** atcp *** *** *** dtcp * * * cond *** *** *** cond *** *** *** cond *** *** *** cond *** *** *** cond cond cond *** *** *** C C C C C C C *** C ** C ** C ** C *** C *** C I *** I I *** I *** I *** I *** 19

20 Table 3 continued Model 1 Model 2 Model 3 ML time ML distance ML time and distance beta Std. Error beta Std. Error beta Std. Error I *** I *** I *** C *** C * C C *** I *** I *** I *** I *** ρ *** *** *** ρ *** *** *** ρ *** *** *** Constant *** *** *** ρ *** *** *** R * Significant at 10%; ** Significant at 5%; *** Significant at 1% Notes: C.t and I.t represent time effects for complete and incomplete foreclosures at time t. C d. and I d. represent distance effects for complete and incomplete foreclosures in distance d. 20

21 Table 4: Regression results for time and distance effects (complete foreclosures) 21 0 to to to to 1500 Coeff. Std. Err. Coeff. Std. Err. Coeff. Std. Err. Coeff. Std. Err. F-12 to F F-9 to F *** F-6 to F * F-3 to F F to F F+3 to F * * * F+6 to F ** ** F+9 to F *** * >F ** S to S * *** S+3 to S ** ** S+6 to S *** S+9 to S * * Significant at 10%; ** Significant at 5%; *** Significant at 1%

22 Table 5: Regression results for time and distance effects (incomplete foreclosures) 0 to to to to 1500 Coeff. Std. Err. Coeff. Std. Err. Coeff. Std. Err. Coeff. Std. Err. F-12 to F *** *** ** * F-9 to F F-6 to F ** *** F-3 to F *** *** *** F to F *** ** *** *** F+3 to F *** * F+6 to F *** * F+9 to F *** * >F *** *** *** *** * Significant at 10%; ** Significant at 5%; *** Significant at 1% 22

23 Table 6: Regression results for model 4 (complete foreclosures) 0 to to to to 1500 Coeff. Std. Err. Coeff. Std. Err. Coeff. Std. Err. Coeff. Std. Err. F-12 to F F-9 to F *** F-6 to F * F-3 to F F to F F+3 to F *** ** ** F+6 to F ** F+9 to F *** >F *** *** * Significant at 10%; ** Significant at 5%; *** Significant at 1% 23

24 Table 7: Regression results for model 4 (incomplete foreclosures) 0 to to to to 1500 Coeff. Std. Err. Coeff. Std. Err. Coeff. Std. Err. Coeff. Std. Err. F-12 to F *** *** ** * F-9 to F F-6 to F ** *** F-3 to F *** *** *** F to F *** ** *** *** F+3 to F *** * F+6 to F *** * F+9 to F *** >F *** *** ** *** * Significant at 10%; ** Significant at 5%; *** Significant at 1% 24

25 Complete Foreclosure Pre-Foreclosure Foreclosure Post-Foreclosure F-12 F-9 F-6 F-3 F F+3 F+6 F+9 F+12 S S+3 S+6 S+9 S+12 Foreclosure Start Foreclosure Sale Incomplete Foreclosure F-12 F-9 F-6 F-3 F F+3 F+6 F+9 F+12 Figure 1: Foreclosure process. This figure shows different time periods for complete and incomplete foreclosure processes used in this paper. The two key reference dates for complete foreclosure are foreclosure start date (F) and foreclosure sale date (S). Incomplete foreclosure only has F date and does not have S date. 25

26 Ring 1 Ring 2 Ring 3 Ring Figure 2: Mean ratio trend over time. This figure shows the mean ratio between the number of incomplete foreclosures and the number of complete foreclosures for different rings over time. 26

27 0.5% 0.0% -0.5% -1.0% -1.5% -2.0% Complete Foreclosures Incomplete Foreclosures Figure 3: Aggregate time effect. This figure displays the estimated foreclosure effects aggregated across all four rings. Solid triangles and circles represent significance at 10% level. 27

28 0.0% -0.5% -1.0% -1.5% -2.0% -2.5% -3.0% -3.5% -4.0% -4.5% Ring 1 Ring 2 Ring 3 Ring 4-0.1% -0.3% -0.4% -0.4% -0.5% -0.5% -1.3% -4.0% Complete Foreclosure Incomplete Foreclosure Figure 4: Aggregate distance effect. This figure displays the estimated foreclosure effects aggregated across all time periods. 28

29 4% 2% 0% -2% -4% -6% -8% -10% -12% Complete Foreclosures Incomplete Foreclosures Figure 5: Time effect (ring 1). This figure displays the estimated foreclosure effects resulting from foreclosures within 250 feet (ring 1) of a non-distressed sale. Solid triangles and circles represent significance at 10% level. 29

30 4% 2% 0% -2% -4% -6% -8% -10% -12% Complete Foreclosures Incomplete Foreclosures Figure 6: Time effect for Model 4 (ring 1). This figure displays the estimated foreclosure effects from foreclosures within 250 feet (ring 1) of a non-distressed sale for Model 4. Solid triangles and circles represent significance at 10% level. 30