# Toxic neighbors: foreclosures and short-sales spillover effects from the current housing-market crash.

I. INTRODUCTIONAs the national mortgage crisis continues, governments at all levels are considering undertaking ways to design homeowner-rescue and foreclosure-prevention policies. Justifications for such policies are based on the fact that foreclosures' negative impact goes beyond hurting those who are losing their homes. In addition to foreclosed homes being sold at a significant discount, they impose negative spillover effects on the values of the nearby properties. First, they cause physical blight. Delinquent mortgage and foreclosure process reduce the incentive of homeowners to invest in the interior and/or exterior of the property, thus reducing the attractiveness of the neighborhood to potential buyers. In addition, properties undergoing foreclosure most probably experience neglect, abandonment, and vandalism, which further alters a property's exterior. Second, homeowners' valuation of their own home is influenced by the price of neighboring homes. Thus, a decline in neighborhood home values due to foreclosure reduces the reservation and transaction price of non-foreclosed homes. Finally, an increase in foreclosures increases the total supply of available homes in the affected segment of the market. Given stable demand, prices of all properties will be lower in that segment of the market.

Over the past several years, a number of studies estimated the impact of foreclosure status on house price. This literature shows that mortgage defaults and resulting property foreclosures generate a selling price discount of about 7%-24%, depending on location within the United States. (1) More recently, a handful of studies concentrated on the spillover effects of foreclosed residential properties on the sales price of nearby nonforeclosure properties during stable housing markets. (2)

This study advances the knowledge of the short-term spillover effects for foreclosures and short sales during a severely "thin" market period. The study includes data of all single-family detached home transactions from January 2008 to the end of June 2009 in Las Vegas, Nevada. We find estimates of foreclosure spillover effects that are twice the size of previous estimates of the same market, from 1.1% to 2.9% per foreclosed home. Controlling for the overall market trend in prices, the neighborhood average price, and unobserved neighborhood characteristics significantly reduce the size of the effects. We find no additional spillover effects from short sales.

II. LITERATURE REVIEW

As mentioned earlier, only a handful of studies have looked at spillover effects of foreclosures. Immergluck and Smith (2006) examined the effect of 1997-1998 foreclosed properties on a sample of 9,600 homes that were sold in Chicago in 1999. They concluded that, on average, for each foreclosed property within 600 feet there was a 0.9% negative spillover effect on the values of nondistressed properties. Lin, Rosenblatt, and Yao (2009) estimated the spillover effect of foreclosed properties on the neighboring properties for Chicago. Using data on 1990-2006 foreclosed properties and year 2006 nondistressed sales, they found the spillover effect as large as 8.7% for foreclosures within 300 feet and 2 years of liquidation. The effect declined sharply within a distance of 2,700 feet and 5 years of liquidation. Schuetz, Been, and Ellen (2008) used data from 2000 to 2005 from New York and found a spillover effect of about 0.2% to 0.4% within 250 feet caused by neighbors filing for foreclosure. Leonard and Murdoch (2009) used 2005-2007 home transaction data in Dallas to estimate the effect of properties in some stage of default process (not sales of foreclosed homes) on nearby sales. They found a 0.5% effect within 250 feet.

The above papers estimate hedonic models that include the number of, or indicators for, foreclosed properties as additional independent variables. They do not control for spatial and time interdependence and/or for overall market trend in prices. In their most recent work, Harding, Rosenblatt, and Yao (2009) criticized hedonic specification on the grounds that it may suffer from omitted variable bias. They argued that the effects of unobserved house and neighborhood characteristics, as well as the effect of the overall local trend in market prices, are most likely picked up by the number of nearby foreclosures in a cross-sectional hedonic model, thus producing biased estimates of spillover effects. By using repeat-sales data from 1989 through 2007, they estimated spillover effects on price appreciation rates in seven metropolitan statistical areas (MSAs) and for the average over the seven MSAs. For the average of all MSAs, they found up to about a 1% spillover effect caused by foreclosed homes within a distance of 300 feet and around the time of foreclosure. The effect declined sharply with distance.

The 19-year period allows Harding, Rosenblatt, and Yao (2009) to obtain enough repeat-sales observations and to estimate "long-term" spillover effects on price appreciation. They estimated their model for seven different MSAs during mostly stable and/or up market periods. Nonetheless, they reported that the spillover effects were different across different local economies and housing-market conditions. Unlike a hedonic approach, the repeat-sales approach eliminates the need to control for house and neighborhood characteristics by assuming that there is no change in house and location characteristics over time and that house and location attribute prices stay constant over time.

A. The Las Vegas Market in the 2000s

The constant house and location characteristics assumptions are not reasonable when dealing with a "dynamic" market such as Las Vegas, where during most of the 2000s new neighborhoods were developed and the landscape of existing neighborhoods changed every few months. Between 2000 and the end of 2007, the population of Clark County increased by about half a million. The number of households increased by about 290,000. Housing stock increased from 500,000 to 731,000 and the median house price increased from $155,000 to $310,000. In addition, the long-term spillover effects, which are based on a stable housing-market period, are most likely different than those of short-term effects caused by housing-market crises. (3) In the short run, reasonable samples of repeat-sales data are not available. A modified hedonic specification (described in the methodology section below) is deemed appropriate.

Between early 2001 and 2007, Las Vegas was a "booming" city with strong housing development and house-price appreciation rates. Since November 2007, Las Vegas has been also among the hardest-hit economies and housing markets. Per Case-Shiller Home Price Index, between November 2007 and June 2009, price per square footage of single-family homes dropped by more than 47%. About 75% of single-family homes were sold under some sort of distress. According to RealtyTrac, Nevada is ranked number one in foreclosure rates; 1 out of every 76 homes is in foreclosure, followed by California with 1 out of every 176 homes. The U.S. average is 1 out of every 466 homes.

The sale of a distressed property is usually performed according to one of the following options: (1) the lender allows a preforeclosure short sale by the borrower; (2) the lender institutes the foreclosure process under a notice of default and the property is sold during the process by the borrower; and (3) the lender forecloses on the property, takes title, and sells the property in the market as real estate owned (REO). While previous studies estimated foreclosure spillover effects, no study has estimated the effects of short sales on nearby houses. (4) As mentioned before, homeowners' valuation of their own home is influenced by the price of neighboring homes. Although short sales may not cause as much spillover effect from physical neighborhood blight as REO, they will reduce the reservation and transaction price of nonforeclosed homes in the neighborhood. Thus, one would expect some, although a smaller one than REO, spillover effect from short sales. Differential spillover effects between the two cases would have both private- and public-sector policy implications.

[FIGURE 1 OMITTED]

In fact, the raw data that will be analyzed in the empirical section of this paper show significant own-price discounts between short and REO sales. Figure 1 shows the sales price per square footage for no-default, short sale, and REO sales. Prices of no-default sales increased from February 2008 to the end of May 2008 and show a linear declining trend thereafter. From January 2008 to June 2009, prices of no-default sales declined by about 25%. Prices of short sales and REO sales steadily declined from January 2008 to June 2009, respectively, 40% and 42%. Figure 1 reveals two interesting facts. First, depending on the month, own-price discounts are from 7% to 25% lower for short sales than for REO sales. Second, there is clearly a declining trend in prices of all types of sale, implying that unbiased estimates of the spillover effects must first isolate the overall market trend in prices. Finally, starting with May 2008, the three trends are almost parallel, indicating that at least in the short run there is a relatively "constant" per square footage own-price discount for distressed properties.

To obtain unbiased estimates of spillover effects of distressed properties on the price of nondistressed properties, we control for several empirically observable and unobservable effects. We include a monthly time trend variable to pick up the effects of the overall market-price trend of the broader market (Clark County, NV). We define "neighborhood" and control for the "neighborhood" price effect. That is, in addition to controlling for observable house characteristics and broader neighborhood characteristics by zip code, we use a weighted average price of narrowly defined neighborhoods within a set of defined physical and time distances. Furthermore, we control for unobserved neighborhood characteristics using a spatially correlated disturbance term model. Numbers of distressed properties sold around a nondistressed property sale, by distressed type, are obtained within a given physical and time distance (i.e., past 3 months and 0.1 of a mile, past 3 months and between 0.1 and 0.25 of a mile, etc.).

III. METHODOLOGY

A. The Model

We start with a standard log-linear hedonic specification model of house price. This specification is based on the idea that the price of a bundled good, such as a specific house, can be expressed as a product of a vector of the house and neighborhood characteristics (i.e., number of bathrooms) and the implicit prices of those characteristics (Rosen 1974). A modified general formulation of the hedonic pricing model to estimate spillover effects of the nearby distressed properties can be expressed as follows:

(1) [P.sub.n] = [X.sub.n][alpha] + [Z.sub.n][beta] + [C.sub.n][delta] + [u.sub.n],

where [P.sub.n] is an n x 1 vector of selling price (in natural log), [X.sub.n] an n x k matrix of k house characteristics affecting price, and [Z.sub.n] an n x g matrix of g neighborhood characteristics affecting price. [C.sub.n] is an n x l matrix of the neighborhood counts of the two types of distressed properties (short sale and REO) and their squared values, measured at l time and distance boundaries; [alpha], [beta], and [delta] are the parameters to be estimated; and [u.sub.n] is an n x 1 vector of regression disturbances. The ordinary least squares (OLS) estimation of Equation (1) assumes that [u.sub.n] is lid.

Equation (1) can be further modified to include the overall market time trend. (5) Even after controlling for the overall market trend, there are several methodological concerns with the OLS estimation of Equation (1) using a cross section of the houses. If not corrected for, estimates of spillover effects from Equation (1) would be biased. First, the equation does not include the partial effect of a "unique" localized neighborhood price. A large number of neighborhood foreclosures could be the result of declining localized home prices in the neighborhood. In other words, the selling price of a house is affected by a weighted average of selling prices of neighboring units. This form of interdependence is referred to as a spatial autoregressive model. To account for it, a spatially lagged dependent variable, the weighted average of house prices sold within a given time and distance in the neighborhood, must be included among the explanatory variables. (6) Such a model can be expressed as follows: (2)

[P.sub.n] = [X.sub.n][alpha] + [Z.sub.n][beta] + [C.sub.n][delta] + [lambda][T.sub.n] + [gamma][W.sub.n] [[bar.P].sub.n]] + [u.sub.n]

where [T.sub.n] is a monthly time trend to proxy for price trend, [W.sub.n] an n x n time and spatial distance weighting matrix of known] constants. and [[bar.P].sub.n] the time and spatially lagged price of neighbors. Of course, for any single observation, this variable is endogenous and is correlated with the disturbance term. Thus, OLS estimates of the parameters will be inconsistent.

Second, neither Equation (1) nor Equation (2) accounts for potential neighborhood unobservable characteristics, which cause spatial interdependence of the disturbance terms among the cross-sectional units. Unless the price equation model is perfectly specified, the estimated parameters are inefficient and potentially biased. (7)

An appropriate way to estimate house prices within a hedonic model is to use an autoregressive time and spatial model with autoregressive error terms. Kelejian and Prucha (1998) developed a procedure to estimate endogenous spatially lagged models with a spatially correlated disturbance term. We modify their model to allow time and spatial lag in forming matrix [W.sub.n]. Such a model can be expressed as follows:

(3) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

where [M.sub.n] is an n x n spatial weighting matrixes of known constants; [[bar.u].sub.n] is generally referred to as the spatial lag of [u.sub.n]; [gamma] and [rho] are autoregressive parameters to be estimated along with [alpha], [beta], [delta], and [lambda]; and [[epsilon].sub.n] is the n x 1 vector of iid disturbances.

Elements of [W.sub.n], [w.sub.ij], and [M.sub.n], [m.sub.ij], represent the spatial relationship between house i and house j. The elements [wi.sub.j] and [m.sub.ij] are nonzero only if prices of house i and house j are spatially correlated. Applying a Cochrane-Orcutt type transformation to equation system (3) yields

(4) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

Estimation of Equations (3) or (4) involves the endogenous spatially lagged price variable and nonlinearity of the autoregressive parameter [rho], preventing the use of OLS or traditional instrumental variable (IV) estimations. Kelejian and Prucha's (1998, 1999, 2004) theoretical work suggests the following generalized spatial two-stage least squares (GS2SLS) procedure, consisting of three steps. In the first step, the parameters in Equation (2) are estimated using the instrument(s) for the endogenous-lagged price variable. This step produces consistent estimators of the parameters, but does not utilize information with respect to the spatially correlated error term, thus the estimates are inefficient. The estimated residuals from the first step and a generalized method of moments (GMM) can be used in the second step to consistently estimate the autoregressive parameter [rho]. Given this estimate, the final step is to account for the spatial autocorrelation error term by making a Cochrane-Orcutt type transformation of the data and estimating Equation (4). The transformed model is estimated via a traditional 2SLS/IV that produces efficient results.

B. Creation of Spatial and Time Boundaries

To estimate the spillover effects, first we need to obtain the count of distressed sales by types, vector [C.sub.n], that neighbor a nondistressed sale. This requires the creation of relevant time and space boundaries. After considering previous research and the population density of Las Vegas, we will use a sliding neighborhood definition by calculating the number of distressed properties that were sold within the past 3 months and within 0.1 of a mile (528 feet), 0.25 of a mile (1,320 feet), and 0.5 of a mile (2,640 feet) of each observation. (8) That is, we create three separate mutually exclusive rings from 0 to 0.1 of a mile, from 0.1 to 0.25 of a mile, and from 0.25 to 0.5 of a mile. In addition, we calculate the counts for within the past 6 months of each observation and perform separate analyses for a 3- and 6-month time frame. We add 12 variables (two types of distressed for three distance rings and for two time frames) and their squared value to each observation in our data set.

To perform the above estimation procedure, we also need to obtain average-weight matrices [W.sub.n] and [M.sub.n], [W.sub.n] is designed to account for the most current neighborhood prices. As such, elements of [W.sub.n] include only the most recent past transactions, last 3 or 6 months in the defined distance neighborhood. Future sales are excluded. On the other hand, [M.sub.n] is designed to pick up the effect of any remaining unobserved neighborhood characteristics, for example the remaining spatial error dependence. As such, elements of [M.sub.n] include homes that were sold in the neighborhood both before and after the current transaction. For the formation of these matrices, we use all market transactions that have neighboring home sales within a 0.5 mile distance, 2,540 feet. (9)

To construct elements of [W.sub.n], [w.sub.ij], we use the inverse distance up to and including 0.5 of a mile (the outer-most distance ring) and 0 thereafter. To avoid a house predicting its own price, the diagonal elements of [W.sub.n] are set equal to 0. Each row of [W.sub.n] is then normalized by [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], so that each row of [W.sub.n], sums to 1. This process is referred to as row standardized. For examples of this approach see Pace and Barry (1997) and Pace and Gilley (1997). We use the same process for constructing matrix [M.sub.n].

IV. EMPIRICAL ANALYSIS

A. Data

The data set for the empirical estimation includes all detached single-family houses sold between January 2008 and June 2009 out of the Multiple Listing Services (MLS) sponsored by the Greater Las Vegas Association of Realtors[R] (GLVAR). Beginning with January 2008, the GLVAR provides information that shows whether the property was sold as a normal sale (no-default sales), a short sale, an in-the-process-of-foreclosure sale, or an REO sale. Variables short sale and REO were not available before January 2008. Thus, our analysis pertains to a downturned housing market. (10) We restricted the data set to properties sold up to $2,000,000 in the zip codes representing the Las Vegas metropolitan area. There were 36,093 detached single-family properties sold. (11) These observations are used to calculate distressed home counts, vector [C.sub.n], to construct the neighborhood weight matrices [W.sub.n], and [M.sub.n], and to estimate the autoregressive parameter, [rho]. That is, we are using complete market information to obtain the (1) distressed home counts for the estimation of spillover effects, (2) neighborhood weighted average price, and (3) unobserved neighborhood characteristics.

We have 8,517 observations on no-default sales for 18 months. Observations from the first 3 months (2,190) are deleted because of less than 3-month distressed property counts. An additional 65 no-default observations had no neighboring sales within 0.5 of a mile and the previous 3 months. The time and spatially weighted lagged price of neighbors, [[bar.P].sub.n] in Equation (4) cannot be calculated for these observations. Deleting these observations, we end up with 6,262 no-default sales observations. (12)

B. Variables

We model the natural logarithm of selling price net of seller's contribution to closing cost as a function of house physical characteristics ([X.sub.1]), and neighborhood characteristics ([Z.sub.1]). In addition, we control for the current market trend by including monthly time trend (T). To estimate the potential spillover effects on neighbors, we include a vector of the two types of distressed property counts and their square values within the three distance rings.

In addition to variables typically found in previous hedonic models, our house physical characteristics vector includes property condition (excellent, good, fair, and poor) recorded by the listing agent and some unique characteristics of the Las Vegas Valley, such as various house views (Table 1). Furthermore, we include property-occupancy status indicators such as vacant, owner occupied, and tenant occupied. Often distressed properties are sold as cash transactions. Previous research shows a cash discount for such transactions (Clauretie and Daneshvary, 2009, 2011). Our model includes an indicator reflecting cash versus mortgage transactions.

The neighborhood characteristics vector includes the percent of population ages 25-35, percent of population age 55 or older, percent of population with a high school diploma and percent with a college degree, and percent of households with a child at home, all measured on January 2008. The vector includes location indicators for five upscale large-planned communities (Summerline, Anthem, Lake Las Vegas, Seven Hills, and the Lakes). (13)

Variable names and descriptive statistics for the sample of 3- and 6-month spillover estimations are provided in Table 1. Table 1 shows that the average numbers of short-sale and REO neighbors increase with distance, because of a larger area of the outer rings, and with an increase in time frame. The overall average number of distressed neighbor sales is 9.2 for the 3-month spillover sample and 19 for the 6-month spillover sample.

As expected, the average property sale prices show a declining trend, reduced from $301,834 to $291,362 and from $133 per square foot to $126 per square foot. The average property had about 7.500 square feet, about 2,200 square feet of living space, and was about 11 years old. As Table l indicates, there are similar distributions between the two samples with respect to property and neighborhood characteristics, occupancy status, and property condition.

C. Results

As discussed in the methodology section and shown in Equations (2) through (4), accurate estimates of the spillover effects require that we control for the localized neighborhood prices. [[bar.P].sub.n]. This time and spatially lagged price variable is correlated with the error term and is endogenous, thus it must be instrumented in order to use IV/2SLS for Equation (2) or GS2SLS for Equation (4). Potentially valid instruments must correlate with P, but not correlate with the dependent variable P,. As suggested by Kelejian and Prucha (2004), spatially lagged values of a subset of the independent variables in the price equation are most likely good candidates. (14) For example, a time and spatially weighted average house size (square footage) in the neighborhood correlates with the neighborhood average price but is not likely to correlate with the reservation and transaction price of the current observation.

We tested the neighborhood weighted average of various subsets of the independent variables. Although several combinations passed as valid instruments, we take a parsimonious approach in order to avoid overidentification of the models. As instruments, we used the neighborhood W-weighted average of house size for the first step, Equation (2), and M-weighted average and M times W-weighted average for the final step, Equation (4). Table 2 shows the first-stage heteroskedasticity-robust statistics with respect to these instruments. The results indicate that the coefficients of the instrument(s) are highly significant. The partial [R.sup.2] attributable to the instrument(s) are from 46% to about 57%. F tests support the significance of the instrument(s). The [chi square] tests, as C statistics in Hayashi (2000), indicate that we cannot reject the hypothesis that these variables are orthogohal to the error term of the price equations. In addition, the coefficients of these variables, not shown in Table 2, are statistically insignificant in the estimated price equations.

To provide a baseline for comparison that allows us to investigate the impact of correcting for the overall market trend, the neighborhood average price variable and its endogeneity, and for neighborhood unobserved characteristics on the estimates of spillover effects, we first estimate the log of house price, Equation (1), by OLS. Then, we add the time trend and weighted average of neighborhood prices and estimate Equation (2) by OLS and IV/2SLS methods. Finally, we correct for the spatially correlated error term by performing the GS2SLS estimation. Estimated coefficients of the time trend, the spatially lagged weighted average of neighborhood prices, and the spillover effects for all four specifications for 3- and 6-month time lag samples are reported in Tables 3 and 4, respectively. In addition to the reported coefficients, each specification includes 22 houses and 10 neighborhood characteristics, 2 vacancy status, and the cash transaction indicator variables. The estimated coefficients of these 35 variables are not reported in Tables 3 and 4. The estimates of these coefficients for specification (3) in Tables 3 and 4 are reported in Appendix A, Table A1. The results of specifications (1), (2), and (4) are available upon request. The estimated coefficients are robust across specifications and are consistent with findings of previous research (Clauretie and Daneshvary 2011).

Concentrating on Table 3, the OLS results show that an additional REO property that was sold within the last 3 months and within 0.1 of a mile from a nondistressed property has about a 2.9% (albeit, declining) negative spillover effect. The effect declines to 1.9% and 1.3% as distance is increased to between 0.1 and 0.25 of a mile and between 0.25 and 0.5 of a mile, respectively. (15) These effects are significant at the 0.01% level. The estimates of short-sale spillover effects are mixed, with no effect within 0.1 of a mile and 1.2% and 0.007% effects between 0.1 and 0.25 of a mile and between 0.25 and 0.5 of a mile, respectively. These effects, however, are almost insignificant.

Controlling for the overall market trend and neighborhood prices, specification (2), reduces the negative spillover effects of REO within the first two distance rings by almost one-half. The spillover effects are now about 1.4% and 1%, respectively. The most-outer ring does not have a spillover effect. Short-sale spillovers are now insignificant and economically meaningless. (16) The coefficient of time trend itself is highly significant and indicates an average overall marketprice decline of about 1.6% per month. The OLS estimate of spatially lagged neighborhood price elasticity is about 0.26 and is statistically significant. Adjusted [R.sup.2]s for the OLS models are about 86 to 90. The third specification in Table 3 reports the result of the instrumental variable estimation, where the spatially lagged neighborhood price is endogenized. Notice that the estimated size of the neighborhood price elasticity is much smaller than that of the OLS estimate, 0.058. The size of the estimated coefficients of REO spillover effects and time trend are slightly higher than those of OLS estimates. Estimates of short-sale spillover effects are still insignificant and economically meaningless.

Our final specification in Table 3 reports the results of the GS2SLS estimation that endogenizes the neighborhood price variable and accounts for the unobserved neighborhood characteristics. The estimated autoregressive parameter, b, is 0.766 and is significant at the 0.05% level. Correcting for unobserved neighborhood characteristics reduces the neighborhood price elasticity to 0.024. It also reduces the size of REO spillover effects significantly. Again, there are no additional spillover effects resulting from short sales.

Table 4 reports the results of identical specifications as those in Table 3, except it estimates the 6-month spillover effects. The REO spillover effect estimates are highly significant, with correct signs, and are robust across specifications. Short sales do not have spillover effects. The overall pattern of results is very similar to that in Table 3, including the pattern of differences between the OLS, the IV, and the GS2SLS estimates. Notice that although the estimated autoregressive parameter, [??], is slightly smaller than that of 3-month estimate, 0.726 versus 0.766, it is still statistically significant. Comparison of specifications 3 and 4, in both Tables 3 and 4, reveals that correcting for spatial error term interdependence reduces the size of the neighborhood price elasticities and the spillover effects of REO but not their statistical significance.

It is possible that the neighborhood foreclosure and short-sale counts and price of the subject home may be affected by unobserved common factors, causing endogeneity of the REO and short-sale variables. We tested the IV and GS2SLS specifications, which are reported in Tables 3 and 4, for endogeneity of REO and short-sale count variables and for every time and distance ring. The p values of [chi square] tests range from .12 to .85, indicating that the null hypothesis of orthogonality of these variables to the error term cannot be rejected (Hayashi 2000). These finding are not surprising for the following two reasons. First, any unobserved variable most likely would cause endogeneity of REO and short sale in the neighborhood-weighted average price (the first stage) equation and not in the subject price (the second stage) equation. The neighborhood-weighted average price is already endogenized in our IV specification. Second, our final GS2SLS model corrects for any unobserved neighborhood effect, thus eliminating potential endogeneity because of omitted variables. We conclude that our estimates based on endogenized spatial autoregressive models with spatial autoregressive disturbances are robust. (17)

Almost all the past research estimated somewhat "long-term" spillover effects of foreclosed properties for different localities during a "stable" housing market. Our analyses pertain to a short-term effect during a "crashed" market. Las Vegas is one of the seven MSAs that are analyzed by Harding, Rosenblatt, and Yao (2009). They used repeat-sales data from 1990 to 2007 and estimated up to a 36-month spillover effect of foreclosed properties on nondistressed sales within four different distance rings. Comparing our results from the first distance ring (up to 528 feet) with the combined results to their first two rings (up to 300 feet and 300 to 500 feet) is perhaps the most adequate. For the first two distance rings, they found spillover effects of -0.52% and -0.46% between 3 and 6 months after foreclosure. Our estimates are twice as much. Our smallest estimates from the GS2SLS specification produce spillover effects of -1.1% (Tables 3 and 4) within the closest distance ring and within 3 months and 6 months after an REO sale.

Using our estimated coefficients of REO variables, the sales price, and the REO counts, we calculate potential losses for each observation. Table 5 provides the results for observations with at least one REO neighbor in any distance ring. It reports the least and the most affected observations, as well as the average loss. Within the 3-month framework, the average loss is $11,199 (about 3.8%) and the maximum loss is $117,666 (about 37.7%). The corresponding figures for the 6-month framework are $29,291 (about 10%) and $248,271 (about 79%), respectively. These estimates are, perhaps, the lower bound of the spillover effects of REOs. One may reasonably argue that the overall negative market trend in prices is also attributable to REO houses in the market. Including the estimated trend coefficient, the average percent losses for the 3- and the 6-month analysis are about 27% and 40%, respectively.

V. CONCLUSIONS

Based on the results reported in Tables 3 and 4, we conclude that hedonic models that include the number of distressed neighbors, a property's physical condition, extended house and neighborhood characteristics, the time trend, the endogenized time and spatially lagged neighborhood price, and that correct for error term interdependence produce robust short-term estimates of spillover effects for the period under consideration. Our results indicate significant negative spillover effects from REO. Controlling for the overall market trend, however, reduces the size of the negative externality attributable to REO per se, as does accounting for neighborhood prices and neighborhood unobservable characteristics. On the other hand, but for its impact on the overall market trend, short sales do not cause additional negative spillover effects on neighbors when REO effects are controlled.

According to the Center for Responsible Lending (2009), it is predicted that over the course of the next 4 years, foreclosures in the United States will affect the values of more than 90 million nearby homes, each losing more than $20,000. Our findings have several implications for policy-makers who are attempting to deal with such a crisis. First, the existence of foreclosure externality may be a good reason to call for both private, that is, lenders and public policies to prevent and to mitigate foreclosures. For example, lenders can implement a speedy process to mitigate default loans in return for a share of future home-price appreciations. Similarly, local, state, and federal governments may take measures to allow speedy loan modifications for primary residences and to prevent risky lending practices in the future. In addition, the overbuilding during most of the 2000s, as is shown by the low average age of foreclosed homes in our data set, took place during the period preceding the current foreclosures. Perhaps, local governments can take measures preventing local excess supply in the future.

Furthermore, we found statistically significant coefficients of the neighborhood-weighted average of the autoregressive parameter, O, as a measure of unobserved neighborhood characteristics. We also found significant change in the size of spillover effects of foreclosure between the specification that controls for unobserved neighborhood characteristics and the specification that does not. These findings imply that, even within a city, the size of the negative spillover effect is not the same for all neighborhoods. Unobserved heterogeneity among neighborhoods produces differential effects, a finding that may be considered when making policy decisions.

The findings of no spillover effects from short sales and significant effects of REO sales imply that short sales do not cause deterioration of neighborhood quality. Short sales also have certain advantages for the lender, the most obvious of which is avoiding additional carrying and transaction costs and the legal expenses of REO and their sale. (18) There are, however, a few disadvantages that may discourage both lenders and borrowers from practicing short sales. First, some of the short-sale agreements would eventually fail and become a foreclosure/REO, increasing the lender's cost associated with the final resolution. Second, a short sale is subject to potential moral hazard of encouraging more defaults. It is also subject to agency cost. That is, in the absence of effective deficiency judgment, the borrower will have no incentive to maximize the sales price. However, absent of an explicit statement in a short-sale agreement between the borrower and the lender, Nevada laws allow for deficiency judgment. The lender has 6 years to pursue the difference between the loan balance and the proceeds of the short sale. In case of foreclosure/REO, the lender is allowed to file for a deficiency judgment up to 6 months after the foreclosure is finalized. Such a differential legal treatment between short sales and foreclosures may discourage borrowers from seeking a short-sale solution over a foreclosure option. Given the absence of negative neighborhood spillover effects and potentially lower carrying, transaction, and legal costs to lenders, equalization of the allowable deficiency judgment period may be a sound public policy. Of course, such policy should be designed subject to preventive moral hazard measures.

ABBREVIATIONS

2SLS: Two-Stage Least Squares

GLVAR: Greater Las Vegas Association of Realtors

GMM: Generalized Method of Moments

GS2SLS: Generalized Spatial Two-Stage Least Squares

IV: Instrumental Variable

MLS: Multiple Listing Services

MSA: Metropolitan Statistical Area

OLS: Ordinary Least Squares

REO: Real Estate Owned

APPENDIX A

TABLE A1 Estimated Coefficients of House and Neighborhood Characteristics, Vacancy Status, and Cash Transaction Sample Variables 3-Month 6-Month Spillover Spillover Property physical characteristics Property age -0.004 *** -0.004 (5.75) (4.85) Square of age -0.001 -0.001 ** (1.49) (2.20) Building square footage 0.001 *** 0.001 *** (in [l0.sup.-3]) (38.14) (33.26) Square of building square -0.027 *** -0.028 *** footage (in 10-6) (16.88) (15.26) Lot square footage (in 0.001 *** 0.001 [l0.sup.-3]) (19.97) (16.45) Square of lot square -0.001 *** -0.001 footage (in [l0.sup.-6]) (7.37) (6.60) Number of bathrooms 0.044 *** 0.042 (9.37) (7.46) Number of bedrooms -0.033 *** -0.034 (9.26) (8.08) Property has a fireplace 0.047 *** 0.046 (12.13) (10.22) Property has a pool 0.090 *** 0.091 (13.66) (11.73) Property has a spa 0.043 *** 0.035 (6.19) (4.18) Number of garages 0.030 *** 0.031 *** (7.96) (7.19) Two-story building -0.066 *** -0.063 (10.79) (8.74) Has golf course view 0.246 *** 0.246 (22.41) (18.87) Has mountain view -0.015 *** -0.012 * (2.76) (1.90) Has strip view 0.055 *** 0.060 (4.67) (4.26) Has park view 0.001 -0.001 (0.09) (0.03) Has city view 0.031 *** 0.009 (3.00) (0.76) Has lake view 0.227 *** 0.240 (9.16) (7.47) Property condition Condition poor -0.371 *** -0.420 *** (15.50) (14.99) Condition fair -0.144 *** -0.160 *** (13.95) (13.05) Condition good -0.058 *** -0.064 *** (12.93) (12.10) Property neighborhood characteristics Percent age 25-35 -0.001 * -0.001 (1.85) (0.88) Percent age 55 or older -0.001 *** -0.001 * (2.86) (1.86) Percent with a high 0.003 *** 0.005 *** school diploma (3.34) (4.43) Percent with a college 0.011 *** 0.012 *** degree (13.51) (12.68) Percent with a child at -0.003 *** -0.003 home (5.60) (4.27) Summerline 0.041 *** 0.045 (4.72) (4.30) Anthem 0.037 *** 0.034 ** (3.33) (2.48) Lake Las Vegas 0.095 ** -0.001 (2.41) (0.03) Seven Hills -0.002 0.002 (0.13) (0.08) The Lakes 0.139 *** 0.136 *** (7.36) (5.91) Property occupancy status Vacant -0.050 *** -0.049 *** (10.60) (8.76) Occupied by a tenant -0.031 * -0.022 (1.90) (1.19) Sold cash -0.017 *** -0.019 (3.33) (3.30) Constant 10.596 *** 10.350 *** (64.95) (50.14) Notes: Taken from specification (3, IV) in Tables 3 and 4. t statistics are in parentheses. ***, **, *Significance at 1%, 5%, and 10% levels, respectively.

REFERENCES

Carroll, T. M., T. M. Clauretie, and H. R. Neill. "Effect of Foreclosure Status on Residential Selling Price: Comment." Journal of Real Estate Research, 13(1), 1997, 95-102.

Center for Responsible Lending, Soaring Spillover. "Accelerating Foreclosures to Cost Neighbors $502 Billion in 2009, Alone; 69.5 Million Homes Lose $7,200 on Average." 2009. Accessed May 7, 2009. http://www. responsiblelending.org/

Clauretie, T. M., and N. Daneshvary. "Estimating the House Foreclosure Discount Corrected for Spatial Price Interdependence and Endogeneity of Marketing Time." Real Estate Economics, 37(1), 2009, 43-67.

--. "The Optimal Choice for Lenders Facing Defaults: Short Sale, Foreclosure, or REO." The Journal of Real Estate Finance and Economics, 42(4), 2011. http:// www.springerlink.com/content/161724k2580u52w0/.

Forgey, F. A., R.C. Rutherford, and M. L. VanBuskirk. "Effect of Foreclosure Status on Residential Selling Price." Journal of Real Estate Research, 9(3), 1994, 313-18.

Hardin, W. G., and M. L. Wolverton. "The Relationship between Foreclosure Status and Apartment Price." Journal of Real Estate Research, 12(1), 1996, 101-09.

Harding, J. P., E. Rosenblatt, and V. W. Yao. "The Contagious Effect of Foreclosed Properties." Journal of Urban Economics, 66, 2009, 164-78.

Hayashi, F. Econometrics. Princeton, NJ: Princeton University Press, 2000.

Immergluck, D., and G. Smith. "The External Costs of Foreclosure: The Impact of Single-Family Mortgage Foreclosures on Property Values. Fannie Mae Foundation." Housing Policy Debate, 17(1), 2006, 57-79.

Kelejian, H. H., and R. I. Prucha. "A Generalized Spatial Two-Stage Least Squares Procedure for Estimating a Spatial Autoregressive Model with Autoregressive Disturbances." Journal of Real Estate Finance and Economics, 17, 1998, 99-121.

--. "A Generalized Moments Estimator for the Autoregressive Parameter in a Spatial Model." International Economic Review, 40, 1999, 509-33.

--. "Estimation of Simultaneous Systems of Spatially Interrelated Gross Sectional Equations." Journal of Econometrics, 118, 2004, 27-50.

Leonard, T., and J. C. Murdock. "The Neighborhood Effects of Foreclosure." Journal of Geographical Systems, 11(4), 2009, 317-22.

Lin, Z., E. Rosenblatt, and V. W. Yao. "Spillover Effects of Foreclosures on Neighborhood Property Values." Journal of Real Estate Finance and Economics, 38(4), 2009, 387-407.

Pace, R. K., and R. Barry. "Quick Computation of Spatial Autoregressive Estimates." Geographical Analysis, 29, 1997, 232-47.

Pace, R. K., and O. W. Gilley. "Using the Spatial Configuration of the Data to Improve Estimation." Journal of Real Estate Finance and Economics, 14, 1997, 333-40.

Pennington-Cross, A. "The Value of Foreclosed Property." Journal of Real Estate Research, 28(2), 2006, 193-214.

Rosen, S. "Hedonic Prices and Implicit Markets: Product Differentiation in Pure Competition." Journal of Political Economy, 82, 1974, 34-55.

Schuetz, J., V. Been, and I. G. Ellen. "Neighboring Effects of Concentrated Mortgage Foreclosures." Journal of Housing Economics, 17(4), 2008, 306-19.

Shilling, J. D., J. D. Benjamin, and C. F. Sirmans. "Estimating Net Realizable Value for Distressed Real Estate." Journal of Real Estate Research, 5(1), 1990, 129-39.

Springer, T.M. "Single Family Housing Transactions: Seller Motivations, Price, and Marketing Time." Journal of Real Estate Finance and Economics, 13(3), 1996, 237-54.

(1.) For examples see Shilling, Benjamin, and Sirmans (199l)). Forgey, Rutherford. and VanBuskirk (1994). Hardin and Wolverton (1996), Springer (1996), Carroll. Clauretie, and Neill (1997), Pennington-Cross (2006), and Clauretie and Daneshvary (2009, 2011).

(2.) For examples of this work see Immergluck and Smith (2006), Schuetz, Been, and Ellen (2008), Lin. Rosenblatt, and Yao (2009), Leonard and Murdock (2009), and Harding et al. (2009).

(3.) Previous studies utilized "affected" property data from 1989 to 2007 and either do not include a time boundary for foreclosed properties counts (affecting) or a boundary ranging from the prior 1 to 10 years.

(4.) In previous studies, it is often not clear whether the "foreclosure" status referred to a sale in the process of foreclosure or a sale as REO, or both. Our initial analyses, based on various tests of individual and sets of coefficients, revealed no statistically significant differences in spillover effects of REO and in the process of foreclosure. Thus, we combine these two categories of distressed. For simplicity, we will refer to this combined category as "REO."

(5.) As discussed by Harding, Rosenblatt, and Yao (2009), when the overall trend in prices is downward the likelihood of foreclosure increases. Thus, to obtain an unbiased estimate of spillover effects it is important to isolate the effect of the overall market trend in prices. Figure 1 shows a linear declining price trend in our sample. Thus, we include a linear time-trend variable in our regression analysis.

(6.) To see the reason and size of potential bias caused by omitting the lagged neighborhood price, assume that true population price is Equation (2). Instead, we estimate Equation (1). Then the estimated coefficient of distressed count is [delta] + [gamma][Cov([C.sub.n] x [[bar.P].sub.n])/Var([C.sub.n])]. The second term is the potential bias from omitting the lagged neighborhood price variable. The covariance is expected to be negative. It is also expected that [delta] < 0 and [gamma] > 0. Then the estimated negative spillover effect from Equation (1) overstates the true spillover effect. The same conclusion applies when the time trend variable is omitted.

(7.) The source of potential bias is the endogenous spatially lagged price variable. That is. if error terms associated with houses i and j are correlated, the price of house j, which is included in the calculation of the lagged explanatory variable for the price of house i, will be correlated with the error term in the price equation of the house i. Thus, the estimated coefficient of the lagged price variable will be biased.

(8.) We also calculate the count numbers for within 200 feet. Given a relatively low population density. 4,154 per square miles, in Las Vegas, the 200-feet calculation does not produce a sufficient number of neighbors. For comparison, Schuetz, Been, and Ellen (2008) used rings of 250-1.000 feet for New York. which has a population density of 27,440/per square mile. Actual transaction date and house latitude and longitude were used to form the rings.

(9.) [W.sub.n] is based on the past transactions within 0.5 of a mile and [M.sub.n] is based on till transactions within 0.5 of a mile.

(10.) Listings by MLS do not cover all public transactions. This may be more frequent for foreclosed and short-sale properties than no-defaults sales. Thus, samples from MLS data may have a larger undercount problem of distressed transactions than of no-default transactions. To the extent that this is true, the coefficient of spillover effects of distressed properties may suffer underestimation bias.

(11.) The sample consists of 8,517 no-default sales, 1,042 sales in the process of foreclosure, 3,535 short sales, and 22,999 REO sales, a total of 36,093. For the empirical analysis, in-the-process-of foreclosure and REO categories are combined. See Footnote 4.

(12.) For the 6-month time flame analyses, we deleted observations pertaining to the first 6 months and based our estimations on the last 12 months, a total of 4,639 observations.

(13.) The first five neighborhood variables for more than 60 zip code areas come from the data that are collected and analyzed by the Center for Business and Economic Research at the University of Nevada. Las Vegas. These variables would be more reflective of neighborhood characteristics if census block-group level data were available. The most recent block-group data, however, are available for 2000. As discussed earlier, Clark County changed significantly between 2000 and 2008, and the 2000 data would not be reflective of current neighborhood characteristics.

(14.) Kelejian and Prucha (2004) show that when the model involves a large number of the independent variables, the ideal instruments for [W.sub.n.],E([bar][P.sub.n]) and [M.sub.n] [W.sub.n] E([bar][P.sub.n]) are a subset of linearly independent columns of [X.sub.n], [W.sub.n][X.sub.n], [X.sub.n],[M.sub.n][X.sub.n], and [M.sub.n][W.sub.n][X.sub.n], respectively.

(15.) Given that the dependent variable, price, is measured in natural logarithm, the precise percentage effect on price within 0.1 of a mile is calculated as [Exp.sup.([beta])] - 1 = [Exp.sup.(-0.29)] - 1 = -2.86%. F tests indicate that the estimated spillover-effect differences between/among the distance rings are statistically significant.

(16.) These findings regarding the short-sale spillover effect may be a possible outcome of correlation between neighborhood REO and short-sale count variables. Of course, deletion of either the REO or short-sale count variables from the models would cause potential omitted variable bias. We investigated the issue in two ways. First, the partial correlation coefficients between the two count variables in various rings are between 3% and 35% and the Variable Inflationary Factors after the estimations are less than 10. Second, we restricted the sample to observations that had zero REO neighbors in all three rings and deleted REO variables from the model. The resulting sample size was 375 observations, hardly enough for meaningful testing of the issue. The coefficients of the short-sale count variables in this sample were still statistically and economically insignificant. Nonetheless, we caution the readers that our overall finding should be interpreted as: after controlling for the REO spillover effects, there are no additional spillover effects (partial effects) from short sales.

(17.) As pointed out by an anonymous referee. "in a market with 1,000 normal [no-default] sales on the market, 10 distressed sales may have less spillover than market with 50 normal sales." We performed a sensitivity analysis by calculating and incorporating the number of no-default home sales in a neighborhood for every time and distance ring. That is, six variables for each specification in Tables 3 and 4, three count variables, and their squared values. While the coefficients of the number of neighborhood no-default sales variables were positive and significant, the inclusion of the variables did not change, neither quantitatively nor qualitatively, the estimates of spillover effects of REO or short sale.

(18.) The State of Nevada allows both judicial and nonjudicial foreclosures. Judicial foreclosure can take place only in the absence of a sale clause in the loan document and when the lender sues the borrower in order to obtain a decree of foreclosure and order of sale. Judicial process is rarely used in Nevada. The most common practice is nonjudicial foreclosure under which the "power of sale" clause allows the lender to sell the property to recover the mortgage balance. There is no statutory right of redemption in Nevada and the debtor cannot regain possession.

NASSER DANESHVARY and TERRENCE M. CLAURETIE *

* Nasser Daneshvary would like to acknowledge the University of Nevada, Las Vegas, College of Business summer research financial support. We express our thanks to the participants of the Economics Department's Seminar Series at UNLV for their comments. We wish to thank anonymous referees of the journal for their comments and suggestions.

Daneshvary: Professor of Economics, Department of Economics, University of Nevada, Las Vegas. 4505 Maryland Parkway, Las Vegas, NV 89154. Phone 1-702-895-1660, Fax 1-702-895-1354. E-mail nasser.daneshvary@unlv.edu

Clauretie: Professor of Economics, Department of Economics, University of Nevada, Las Vegas, 4505 Maryland Parkway, Las Vegas, NV 89154. Phone 1-702-895-3223, Fax 1-702-895-4650, E-mail mike.clauretie@unlv.edu

doi: 10.1111/j.1465-7295.2011.00380.x

TABLE 1 Definition and Descriptive Statistics of Variables: Samples of No-Default Sales Samples 3-Month Spillover Variables M SD Number of neighbors with selling status No. of short-sale neighbors: 0-0.1 of 0.091 0.326 a mile distance No. of short-sale neighbors: 0.1-0.25 0.352 0.692 of a mile distance No. of short-sale neighbors: 0.25-0.5 1.033 1.359 of a mile distance No. of neighbors sold as REO: 0-0.1 of 0.557 0.999 a mile distance No. of neighbors sold as REO: 0.1-0.25 2.192 2.618 of a mile distance No. of neighbors sold as REO: 0.25-0.5 6.426 6.027 of a mile distance Selling price net of seller's $301,834 $208,302 contribution to closing costs Price per square footage $133.13 $43.57 Property condition indicators (assessed by the listing agent) Condition poor 0.008 Condition fair 0.045 -- Condition good 0.412 -- Condition excellent 0.536 -- Property occupancy status indicators Vacant 0.664 -- Occupied by owner 0.320 -- Occupied by a tenant 0.017 -- Sold cash 0.275 -- Property physical characteristics: Property age 11.297 10.417 Building square footage 2,200 883 Lot square footage 7,462 5,415 Number of bedrooms 3.395 0.836 Number of bathrooms 2.804 0.848 Number of garages 2.267 0.750 Property has a fireplace 0.756 -- Property has a pool 0.252 -- Property has a spa 0.194 -- Two-story building 0.478 -- Has golf course view 0.044 -- Has mountain view 0.244 -- Has strip view 0.040 -- Has park view 0.018 -- Has city view 0.056 -- Has lake view 0.008 -- Property neighborhood characteristics: Percent age 25-35 23.998 8.697 Percent age 55 or older 40.197 13.029 Percent with a high school diploma 52.399 7.123 Percent with a college degree 42.569 9.460 Percent with a child at home 29.311 7.632 Summerline 0.112 -- Anthem 0.059 -- Lake Las Vegas 0.003 -- Seven Hills 0.014 -- The Lakes 0.012 -- Number of observations 6,262 -- Samples 6-Month Spillover Variables M SD Number of neighbors with selling status No. of short-sale neighbors: 0-0.1 of 0.194 0.505 a mile distance No. of short-sale neighbors: 0.1-0.25 0.782 1.173 of a mile distance No. of short-sale neighbors: 0.25-0.5 2.223 2.345 of a mile distance No. of neighbors sold as REO: 0-0.1 of 1.161 1.695 a mile distance No. of neighbors sold as REO: 0.1-0.25 4.525 4.704 of a mile distance No. of neighbors sold as REO: 0.25-0.5 13.322 10.990 of a mile distance Selling price net of seller's $291,362 $209,854 contribution to closing costs Price per square footage $126.45 $41.87 Property condition indicators (assessed by the listing agent) Condition poor 0.008 -- Condition fair 0.044 -- Condition good 0.406 -- Condition excellent 0.543 -- Property occupancy status indicators Vacant 0.679 -- Occupied by owner 0.304 -- Occupied by a tenant 0.017 -- Sold cash 0.322 -- Property physical characteristics: Property age 11.229 10.491 Building square footage 2,222 904 Lot square footage 7,493 5,545 Number of bedrooms 3.417 0.842 Number of bathrooms 2.821 0.853 Number of garages 2.268 0.767 Property has a fireplace 0.760 -- Property has a pool 0.249 -- Property has a spa 0.189 -- Two-story building 0.491 -- Has golf course view 0.042 -- Has mountain view 0.238 -- Has strip view 0.038 -- Has park view 0.016 -- Has city view 0.055 -- Has lake view 0.006 -- Property neighborhood characteristics: Percent age 25-35 24.142 8.751 Percent age 55 or older 40.023 12.995 Percent with a high school diploma 52.523 7.064 Percent with a college degree 42.365 9.444 Percent with a child at home 29.411 7.689 Summerline 0.109 -- Anthem 0.055 -- Lake Las Vegas 0.003 -- Seven Hills 0.013 -- The Lakes 0.011 -- Number of observations 4,639 -- Source: Multiple Listing Services (MLS) of Greater Las Vegas Association of Realtors (GLVAR). TABLE 2 Statistics with Respect to the Excluded Instrument Variables: The First-Stage Heteroskedasticity-Robust Estimates Samples 3-Month 6-Month Spillover Spillover Statistics Estimation of Equation (2) Coefficient of neighborhood W-weighted 0.410 *** 0.407 *** average of house size (89.55) (77.37) (excluded instrument), in 103 Partial [R.sup.2] attributable to the 0.564 0.566 excluded instrument F value for the significance of the 8.019 *** -- instrument. [F.sub.(1,6212)] F value for the significance of the -- 5.985 *** instrument. [F.sub.(1,4589] p value of the [[chi square].sub.(1)] 0.359 0.439 test for orthogonality of the instruments Estimation of Equation (4) Coefficients of neighborhood M-weighted 1.731 *** 1.729 *** average of house size (excluded instrument), in [10.sup.3] (76.25) (73.12) Coefficient of neighborhood MW-weighted -1.915 *** -1.888 *** average of house size (excluded instrument), in [10.sup.3] (71.69) (69.32) Partial [R.sup.2]- attributable to the 0.487 0.543 two excluded instruments F value for the significance of the two 2.949 -- instruments. [F.sub.(1,6212)] F value for the joint significance of the -- 2.720 two instruments. [F.sub.(2,4588)] p value of the [[chi square].sub.(2)] 0.278 0.181 test for orthogonality of the two instruments Notes: In addition to the instrumental variables. each specification includes 22 houses and 10 neighborhood characteristics. 12 distressed neighborhood counts. 2 vacancy status, time trend, and the cash transaction indicator variables. Estimated coefficients of these variables are available upon request. t statistics are in parentheses. *** Significance at the 1% level. TABLE 3 Ordinary Least Squares (OLS), Instrumental Variable (IV), and Generalized Spatial Two-Stage Least Squares (GS2SLS) Estimates of the Log of Selling Price: 3-Month Spillover Effects of Distressed Sales (6,262 Observations on No-Default Sales from April 2008 to June 2009) Specifications (1, OLS) (2, OLS) Time trend -- -0.016 *** -- (31.73) Time and spatially weighted -- 0.258 *** lag of log selling price in -- (29.27) neighborhood Distressed property counts: No. of short-sale neighbors: 0.001 -0.002 0-0.1 of a mile distance (0.03) (0.18) No. of short-sale neighbors: -0.012 * -0.005 0.1-0.25 of a mile distance (1.70) (0.82) No. of short-sale neighbors: -0.007 * 0.003 0.25-0.5 of a mile distance (1.68) (1.00) No. of short-sale neighbors: 0.000 0.000 0-0.1 of a mile distance (0.00) (0.01) squared No. of short-sale neighbors: 0.004 0.001 0.1-0.25 of a mile distance (1.40) (0.52) squared No. of short-sale neighbors: 0.002 ** 0.000 0.25-0.5 of a mile distance (2.59) (0.12) squared No. of neighbors sold as REO: -0.029 *** -0.014 *** 0-0.1 of a mile distance (6.02) (3.42) No. of neighbors sold as REO: -0.019 *** -0.009 *** 0.1-0.25 of a mile distance (8.11) (4.80) No. of neighbors sold as REO: -0.013 *** -0.001 0.25-0.5 of a mile distance (11.31) (1.20) No. of neighbors sold as REO: 0.002 ** 0.001 0-0.1 of a mile distance (2.26) (0.72) squared No. of neighbors sold as REO: 0.001- 0.001 0.1-0.25 of a mile distance (5.37) (3.75) squared No. of neighbors sold as REO: 0.000 *** 0.000 0.25-0.5 of a mile distance squared Estimated autoregressive (5.43) (0.16) parameters, p Adjusted [R.sup.2] 0.863 0.908 (3, IV) (4, GS2SLS) Time trend -0.021 *** -0.022 *** (37.00) (46.41) Time and spatially weighted 0.058 *** 0.024 lag of log selling price in (4.84) (5.15) neighborhood Distressed property counts: No. of short-sale neighbors: -0.002 -0.004 0-0.1 of a mile distance (0.16) (0.31) No. of short-sale neighbors: -0.004 -0.002 0.1-0.25 of a mile distance (0.63) (0.43) No. of short-sale neighbors: 0.002 0.000 0.25-0.5 of a mile distance (0.61) (0.03) No. of short-sale neighbors: -0.001 -0.001 0-0.1 of a mile distance (0.10) (0.15) squared No. of short-sale neighbors: 0.002 0.000 0.1-0.25 of a mile distance (0.77) (0.01) squared No. of short-sale neighbors: 0.001 0.000 0.25-0.5 of a mile distance (1.05) (0.25) squared No. of neighbors sold as REO: -0.018 *** -0.011 *** 0-0.1 of a mile distance (4.27) (2.82) No. of neighbors sold as REO: -0.013 *** -0.007 *** 0.1-0.25 of a mile distance (6.45) (3.86) No. of neighbors sold as REO: -0.005 *** -0.002 0.25-0.5 of a mile distance (4.99) (1.62) No. of neighbors sold as REO: 0.001 0.000 0-0.1 of a mile distance (1.00) (0.01) squared No. of neighbors sold as REO: 0.001- 0.000 0.1-0.25 of a mile distance (4.47) (2.68) squared No. of neighbors sold as REO: 0.000 * 0.000 0.25-0.5 of a mile distance squared Estimated autoregressive (1.91) (0.27) parameters, p 0.766 ** (10.07) Adjusted [R.sup.2] -- -- Notes: In addition to these variables, each specification includes 22 house and 10 neighborhood characteristics, 2 vacancy status, and the cash transaction indicator variables. Estimated coefficients of these variables are available upon request. t statistics are in parentheses. *** Significance at the 10%, 5%, and 1% levels, respectively. TABLE 4 Ordinary Least Squares (OLS), Instrumental Variable (IV), and Generalized Spatial Two-Stage Least Squares (GS2SLS) Estimates of the Log of Selling Price: 6-Month Spillover Effects of Distressed Sales (4,639 Observations on No-Default Sales from July 2008 to June 2009) Specifications (1, OLS) (2, OLS) Time trend -- -0.017 *** -- (23.99) Time and spatially weighted lag of -- 0.298 *** log selling price in neighborhood Distressed property counts: -- (25.52) No. of short-sale neighbors: -0.009 -0.015 0-0.1 of a mile distance (0.76) (1.63) No. of short-sale neighbors: -0.012 ** -0.005 0.1-0.25 of a mile distance (2.27) (1.25) No. of short-sale neighbors: -0.006 * 0.000 0.25-0.5 of a mile distance (1.86) (0.20) No. of short-sale neighbors: 0.003 0.004 0-0.1 of a mile distance squared (0.61) (1.01) No. of short-sale neighbors: 0.003 *** 0.001 0.1-0.25 of a mile distance squared (2.87) (1.23) No. of short-sale neighbors: 0.001 *** 0.000 0.25-0.5 of a mile distance squared (2.87) (1.01) No. of neighbors sold as REO: -0.018 *** -0.012 *** 0-0.1 of a mile distance (4.90) (3.83) No. of neighbors sold as REO: -0.013 *** -0.008 *** 0.1-0.25 of a mile distance (7.27) (5.20) No. of neighbors sold as REO: -0.007 *** -0.002 ** 0.25-0.5 of a mile distance (7.90) (2.16) No. of neighbors sold as REO: 0.001 0.001 0-0.1 of a mile distance squared (1.55) (1.39) No. of neighbors sold as REO: 0.000 *** 0.000 *** 0.1-0.25 of a mile distance squared (5.57) (4.66) No. of neighbors sold as REO: 0.000 *** 0.000 0.25-0.5 of a mile distance squared (4.62) (1.52) Estimated autoregressive parameters, [rho] Adjusted [R.sup.2] 0.870 0.910 (3, IV) (4, GS2SLS) Time trend -0.023 *** -0.024 *** (29.09) (3.60) Time and spatially weighted lag of 0.069 *** 0.021 log selling price in neighborhood Distressed property counts: (4.28) (37.91) No. of short-sale neighbors: -0.012 -0.007 0-0.1 of a mile distance (1.24) (0.83) No. of short-sale neighbors: -0.004 -0.006 0.1-0.25 of a mile distance (0.94) (1.33) No. of short-sale neighbors: 0.000 -0.004 0.25-0.5 of a mile distance (0.11) (1.40) No. of short-sale neighbors: 0.003 -0.001 0-0.1 of a mile distance squared (0.83) (0.22) No. of short-sale neighbors: 0.002 0.001 0.1-0.25 of a mile distance squared (1.63) (0.62) No. of short-sale neighbors: 0.000 * 0.000 0.25-0.5 of a mile distance squared (1.71) (0.94) No. of neighbors sold as REO: -0.014 *** -0.010 0-0.1 of a mile distance (4.57) (3.29) No. of neighbors sold as REO: -0.010 *** -0.008 0.1-0.25 of a mile distance (6.80) (5.26) No. of neighbors sold as REO: -0.005 *** -0.003 0.25-0.5 of a mile distance (6.02) (3.46) No. of neighbors sold as REO: 0.001 0.000 0-0.1 of a mile distance squared (1.45) (0.66) No. of neighbors sold as REO: 0.000 *** 0.000 0.1-0.25 of a mile distance squared (5.47) (4.42) No. of neighbors sold as REO: 0.000 *** 0.000 0.25-0.5 of a mile distance squared (3.90) (2.60 Estimated autoregressive 0.726 ** parameters, [rho] (7.73) Adjusted [R.sup.2] -- -- Notes: In addition to these variables. each specification includes 22 houses and 10 neighborhood characteristics. 2 vacancy status, and the cash transaction indicator variables. Estimated coefficients of these variables are available upon request. t statistics are in parentheses. **, *** Significance at the 10%, 5%, and 1% levels, respectively. TABLE 5 Calculated Losses Because of REO neighbors Sample 3 Months Min Ave Max No. of neighbors sold as REO: 0 0.6 2 0-0.1 miles distance No. of neighbors sold as REO: 0 2.3 20 0.1-0.25 miles distance No. of neighbors sold as REO: 1 6.8 18 0.25-0.5 miles distance Net sale price $36,250 $296,259 $312,550 Total % loss because of REO -0.17 -3.78 -37.65 neighbors Total dollar amount loss -$61 -$11,199 -$117,666 because of REO neighbors Total % loss including trend -- -26.99% -- Total dollar loss including trend -- -$79,946 -- Number of observations with at -- 5,889 -- least one REO neighbor Sample 6 Months Ave Min Max No. of neighbors sold as REO: 0 1.2 4 0-0.1 miles distance No. of neighbors sold as REO: 0 4.6 34 0.1-0.25 miles distance No. of neighbors sold as REO: I 13.6 35 0.25-0.5 miles distance Net sale price $112,000 $290,280 $312,550 Total % loss because of REO -.29 -10.09 -79.43 neighbors Total dollar amount loss -$327 -$29,291 -$248,271 because of REO neighbors Total % loss including trend - -39.56% -- Total dollar loss including trend -- -$114,835 -- Number of observations with at - 4,541 -- least one REO neighbor Note: The calculations are based on the estimated coefficients of the GS2SLS specifications in Tables 3 and 4.

Printer friendly Cite/link Email Feedback | |

Author: | Daneshvary, Nasser; Clauretie, Terrence M. |
---|---|

Publication: | Economic Inquiry |

Article Type: | Statistical data |

Geographic Code: | 1USA |

Date: | Jan 1, 2012 |

Words: | 10774 |

Previous Article: | Talent and/or popularity: what does it take to be a superstar? |

Next Article: | On the search for talent in academic labor markets: easily observable later-graduate study outcomes as predictors of early-career publishing,... |

Topics: |