Foreclosure externalities: New evidence Kristopher Gerardi† FRB Atlanta

Eric Rosenblatt‡ Fannie Mae

∗

Paul S. Willen§ FRB Boston and NBER

Vincent Yao¶ Fannie Mae March 7, 2015 Abstract Policy makers have used externalities to justify government intervention in the foreclosure process. Using a new dataset that covers 15 of the largest Metropolitan Statistical Areas in the U.S. and a novel identification strategy, this paper provides new evidence on the size and source of these externalities. Our results show that a property in distress affects the value of neighboring properties from the time when the borrower becomes seriously delinquent on the mortgage until well after the bank sells the property to a new owner. Properties with seriously delinquent loans within 0.1 miles are found to decrease transaction prices of non-distressed properties by approximately one percent on average. The spillovers are found to dissipate rapidly with distance and completely disappear one year after the bank sells the property to a new homeowner. Importantly, we find that the size of the externality is sensitive to the condition of foreclosed properties, as bank-owned properties in poor condition lower nearby transaction prices by 2.6 percent on average while those in good condition marginally raise prices. We argue that the measured price spillovers are physical externalities caused by a lack of property maintenance and not pecuniary externalities that reflect local supply or demand shocks. ∗

The authors thank Chris Cunningham, Chris Foote, Scott Frame, Lauren Lambie-Hanson, Sam Kruger, Steve Ross, Joe Tracy, seminar participants at Hamilton College, the Furman Center at New York University, the Kansas City Fed, and the University of Connecticut for helpful comments and to Stefano Giglio and Susan Wachter for excellent discussions at the 2012 Spring HULM and 2012 Mid-year AREUEA meetings respectively. The views expressed in this paper are those of the authors and not the official position of Fannie Mae or any part of the Federal Reserve System. † Federal Reserve Bank of Atlanta, Research Department, 1000 Peachtree Street N.E., Atlanta, GA 30309, USA. [email protected] ‡ Fannie Mae. eric [email protected] § Federal Reserve Bank of Boston, Research Department, 600 Atlantic Avenue, Boston, MA 02210, USA. [email protected] ¶ Fannie Mae.vincent w [email protected]

1

Introduction

The existence of foreclosure externalities underpins either explicitly or implicitly many arguments for government intervention in the foreclosure process. For example, in a 2010 white paper, Federal Reserve Board economists wrote that: ...foreclosures can be a costly and inefficient way to resolve the inability of households to meet their mortgage payment obligations because they can result in “deadweight losses,” or costs that do not benefit anyone, including the neglect and deterioration of properties that often sit vacant for months (or even years) and the associated negative effects on neighborhoods. The financial crisis period was characterized by extremely high rates of mortgage delinquency and foreclosure: Figure 1 shows that between 2008 and 2012 millions of homes were in some stage of mortgage delinquency or foreclosure. While some areas were hit harder than others, foreclosures cut a wide swath and we show below that in 2010 almost 80 percent of residential sales occurred with at least one distressed property nearby. Given the huge number of foreclosures that characterized this time period, even small externalities could have significant economic impacts. Foreclosures can affect the well-being of residents of neighboring properties in many ways but in this paper we measure only those that affect the sale price of the property.1 A growing body of evidence has emerged showing that foreclosures decrease the market values of nearby, non-distressed properties. Estimates of the magnitude of the spillovers range widely. For example, Leonard and Murdoch (2009) finds that an additional foreclosure within 250 feet lowers transaction prices by approximately 0.5 percent in a sample of housing transactions in Dallas County, while Lin, Rosenblatt, and Yao (2009), using a sample of property transactions in Chicago in the early-to-mid 2000s, finds that an additional foreclosure within five blocks reduced prices by almost 9 percent. Other studies including Immergluck and Smith (2006), Rogers and Winter (2009), Campbell, Giglio, and Pathak (2011), and Harding, Rosenblatt, and Yao (2009) have found negative price spillovers of nearby foreclosures in the neighborhood of 1 percent.2 1

Other externalities that have been documented in the literature include increased neighborhood crime rates (Ellen, Lacoe, and Sharygin (2013)) as well as foreclosure contagion (Towe and Lawley (2013)). See U.S. Department of Housing and Urban Development (2010) for a detailed discussion of the various costs imposed on borrowers and lenders during the foreclosure process as well as a literature review of the empirical estimates of various types of foreclosure externalities. Frame (2010) provides a review of the empirical literature that tries to measure the negative price spillovers associated with the foreclosure process. 2 Immergluck and Smith (2006) considers property transactions in the city of Chicago in the late 1990s, Rogers and Winter (2009) focuses on transactions in St. Louis, Missouri from 1998–2007, Campbell, Giglio,

1

In addition to disagreement over quantitative magnitudes, there is also no consensus yet regarding the causal mechanisms that generate the estimated spillovers. Most papers estimating foreclosure price spillovers have used some variant of a hedonic pricing regression in which property transaction prices are regressed on a count of nearby foreclosures and a vector of controls for property and/or neighborhood characteristics (see equation (1) below), and have interpreted a negative correlation between prices and the number of nearby foreclosures as evidence of a highly localized, spillover externality. Yet, there has been little evidence on the cause of the foreclosure externality. There is some empirical evidence that the likely mechanism is a physical externality due to a lack of property maintenance.3 However, there is also evidence that suggests the mechanism is a supply effect in which a nearby foreclosure increases competition in a market resulting in lower transaction prices for non-distressed properties.4 Determining both the causal mechanism that leads to foreclosure price spillovers, and pinning down the quantitative magnitudes of spillovers is necessary for policy makers to be able to develop effective policy responses to future foreclosure crises. This paper uses a new, more nationally representative dataset that includes previously unavailable information on delinquent mortgages and on the physical condition of bankowned (REO) properties, and an identification strategy that addresses many of the severe econometric issues that have plagued the literature, to shed light on both the nature and magnitude of foreclosure price spillovers. Our identification strategy builds off of previous studies in the literature. Like Harding, Rosenblatt, and Yao (2009) we use the repeat-sales methodology to control for time invariant, unobserved, property-specific and neighborhood-specific factors that could cause omitted variable bias.5 However, the repeat-sales specification does not account for timevarying unobserved local shocks that might be correlated with house price dynamics and and Pathak (2011) uses the universe of transactions in the state of Massachusetts from 1989–2007, and Harding, Rosenblatt, and Yao (2009) uses a more nationally representative sample comprised of transactions from seven metropolitan statistical areas. 3 For example, Fisher, Lambie-Hanson, and Willen (Forthcoming), examines a sample of condominiums in Massachusetts. They focus on condominiums in an association and divide them up between those that are at the same address and those that are in the same association but at a different address. They find that only foreclosures at the same address exert an effect on the price and conclude that the large effects of condo foreclosures on the prices of nearby properties are due to a physical externality. 4 For example, Anenberg and Kung (2014) focus on sales of single-family properties in the San Francisco, Washington D.C., Chicago, and Phoenix metropolitan statistical areas over the period 2007–2009 and augments the typical hedonic pricing regression with property listings. The authors find that sellers decrease their listing prices in the same week that nearby bank-owned (REO) properties are listed, which they interpret as evidence of a supply effect. 5 The repeat-sales specification uses a sample of properties that transact at least twice during the sample period and involves taking differences in the dependent and independent variables across the multiple sale dates so that time-invariant property and neighborhood characteristics are eliminated from the estimation equation.

2

the number of nearby distressed properties. As we will see below, these unobserved, local supply and demand shocks are the primary threat to the identification of foreclosure price spillovers. To address this issue we include an innovative triple-interaction fixed effect in the repeat-sales specification that groups properties in the same small geographical area (a census block group) that transact in the same two quarters. This differences out all timevarying factors that are common to properties that transact in the same two quarters in the same small geography. We then follow the literature and measure nearby distressed properties using rings drawn around individual homes. The rings are smaller than the geography associated with the fixed effects to ensure that within a given fixed effect group, there is enough variation in the number of nearby distressed properties to estimate the model. Thus, our identification strategy compares house price growth for houses that are purchased in the same quarter and sold in the same quarter in the same small geography but with different numbers of nearby distressed properties. We argue that this identification strategy leaves us with a clean estimate of local price spillovers from distressed properties.6 We apply the methodology using a dataset that is unique in this literature. Previous researchers have linked sale prices of houses to nearby foreclosures but our data also include nearby properties where the borrower is delinquent on the mortgage but for which the lender has not yet pursued (and may never pursue) legal action. As we discuss in more detail below, delinquent borrowers are less likely to invest in their properties either because they anticipate losing the property eventually to foreclosure or because of the financial distress that is leading to the delinquency. The results point to a statistically significant, negative price spillover of nearby distressed properties. According to our estimates, an additional property within 0.1 miles of a nondistressed property transaction (approximately 528 feet) in which the homeowner has been behind by at least three mortgage payments for less than one year decreases the transaction price by an average of 0.6 percent. The negative spillover peaks when the borrower has been seriously delinquent for a significant length of time. An additional property in which the homeowner has been behind by at least three mortgage payments for more than one year decreases the sales price of non-distressed property transactions within 0.1 miles by an average of 1.2 percent. The negative price spillover dissipates slightly when the foreclosure 6

Previous studies in the literature have tried to control for endogeneity bias in a number of different ways. Harding, Rosenblatt, and Yao (2009) attempts to explicitly control for reverse causality bias by including a local market price index in their covariate set. Schuetz, Been, and Ellen (2008) and Campbell, Giglio, and Pathak (2011) include relatively disaggregated, geographic fixed effects in their hedonic specifications. Campbell, Giglio, and Pathak (2011) include census tract-by-year fixed effects in a hedonic specification, and is thus the closest analysis to ours in terms of econometric methodology. However, there are significant differences between the two studies, which we document in detail in the On-Line appendix (section A.2).

3

process is completed at which point the borrower is evicted and the property turned over to the lender, and disappears almost entirely one year after the lender sells the property to a new owner. Consistent with the previous literature, the estimates of price spillovers are very sensitive to distance, with the strongest spillovers occurring within a 0.1 mile radius, and completely disappearing at distances greater than 0.5 miles. Importantly, we find that the spillover estimates are sensitive to the physical condition of the distressed properties. REO properties in “poor” condition (according to the appraisal report commissioned by the bank at the time of foreclosure) lower the prices of nearby transactions by 2.6 percent on average. In contrast, REO properties in “average” condition lower nearby transaction prices by 1.5 percent, and REO properties in “good” condition have a small, positive impact on prices. Taken together, we argue that the results are most consistent with a highly localized, physical externality, caused by poor property maintenance, which we refer to hereafter as an investment externality. Without investment in maintenance, residential property depreciates at a relatively rapid rate. According to Harding, Rosenthal, and Sirmans (2007) the U.S. residential housing stock depreciates at approximately 2.5 percent per year gross of maintenance. Neither mortgage borrowers in delinquency nor lenders who take possession of a foreclosure have a strong incentive to properly maintain their properties. This can lead to physical deterioration, which, in turn, can reduce the value of nearby properties to potential buyers. The investment disincentive is arguably present both during the period in which the borrower is seriously delinquent and in foreclosure proceedings (we refer to this as the SDQ period) and during the REO period. There are at least two reasons why seriously delinquent borrowers are unlikely to maintain properties. The first is that many of them have suffered cash-flow-depleting life events and discount future consumption heavily relative to current consumption.7 Effectively, this raises the hurdle rate on any investment in the property. The second problem is that many seriously delinquent borrowers expect to lose their homes and, therefore, the long-term benefits of any investment would accrue to the ultimate owner of the property—the lender. After the foreclosure completion, there are also good reasons to expect underinvestment in maintenance. Narrowly, the lender does not obtain any consumption benefit from investment. More broadly, the scale of residential lending could lead to an information problem. The agent managing the property rarely has a large ownership stake in the property. This is obviously true when the agent is acting on behalf of a securitization trust, but it is even true when the agent works for a bank that owns the property. The issue is that the amount 7 See Gerardi et al. (2013) for direct evidence of the role of income and unemployment shocks on mortgage default rates.

4

of profitable investment in the property is a matter of discretion, and the owner of the property cannot be sure whether the manager has other incentives. As with all standard asymmetric information problems, the result would be a failure to exploit profitable gains from trade, in this case, investment in the property.8 Evidence from the existing literature supports the existence of an investment externality. First, there is ample evidence that foreclosed properties suffer from underinvestment. Recent studies include Pennington-Cross (2006), who finds that the price of a distressed property appreciates on average about 22 percent less than the prevailing metropolitan area price index in which the property is located. Campbell, Giglio, and Pathak (2011) find a 27 percent foreclosure discount for a sample of single-family properties in Massachusetts.9 Clauretie and Daneshvary (2009) also finds a foreclosure discount of slightly less than 10 percent in the Las Vegas metropolitan area between 2004 and 2007.10 Second, an older literature on subsidized investment showed that investment in property affects the value of nearby properties. Early studies in this literature used aggregated data and attempted to calculate whether subsidized housing raised or lowered nearby house prices.11 Galster, Tatian, and Smith (1999) developed a methodology to use transactions-level data to measure the impact of Section 8 housing on the sale prices of neighboring properties.12 Both Schwartz et al. (2006) and Rossi-Hansberg, Sarte, and Owens III (2010) used similar methodologies to examine programs that encouraged investment in individual properties and found important effects on prices. The policy implications of our findings are significant. In the aftermath of the recent foreclosure crisis many municipalities adopted vacant property registration ordinances (VPROs). VPROs are laws that typically require the formal registration of foreclosed properties, and require property owners (usually banks) to maintain and secure properties.13 The results of this analysis support policies like VPROs, and imply that policy makers should consider expanding their scale and scope. In addition, the finding that negative spillovers manifest well before completion of the foreclosure process when borrowers become seriously 8

Another way to put this is that the optimal mechanism for investment in single-family residential real estate is to sell the property to a small-scale investor who internalizes the costs and benefits. 9 The authors point to two additional datapoints in support of the theory that poor maintenance explains the discount. First, the foreclosure discount is larger for houses in poorer neighborhoods which they argue suggests vandalism. Second, they show that other “forced sales” resulting from death and bankruptcy generate much smaller discounts. One explanation for the second finding is that only foreclosure results in separation of ownership and control that removes much of the incentive to maintain a property. 10 In contrast, Harding, Rosenblatt, and Yao (2012) finds little evidence of a meaningful foreclosure discount, using both hedonic and repeat-sale methods. 11 See Matulef (1988) for a review of this early literature. 12 Its authors compared the sale prices of properties within 500 feet of a Section 8 site before and after the site transitioned to Section 8 and assumed that the difference in sale prices measures the treatment effect. 13 Immergluck, Lee, and Terranova (2012) provide a detailed summary of VPROs over the past decade.

5

delinquent implies that policies designed to delay foreclosure and prevent the ultimate eviction of a borrower may be misplaced. Gerardi, Lambie-Hanson, and Willen (2013) find no benefits from policies that prolong the foreclosure process and this paper finds nontrivial costs in the form of negative price spillovers that are present during the foreclosure process. The paper proceeds as follows. Section 2 presents the empirical methodology used in the analysis and contains a detailed discussion of the similarities and differences of the approach with previous studies in the literature. Section 3 describes the data used in the analysis, and Section 4 reports the empirical results. In Section 5 we discuss the potential interpretations of our empirical results and in Section 6 we discuss the economic importance of the spillover estimates. Finally, Section 7 provides a brief conclusion with a discussion of the policy implications of our analysis.

2

Empirical Methodology

To help illustrate the main differences and similarities between our empirical methodology and the methods used in previous studies of foreclosure price spillovers we derive our estimating equation starting from the traditional hedonic regression model that has become standard in the housing literature. Let us consider the following hedonic regression estimated on a sample of properties i ∈ I, located in geography j ∈ J, purchased in time period t:14 d,r ln(Pijt ) = δt + γj + αjt + βt · Xit + λd,r t · Nit + εijt

(1)

where the term δt represents unobserved aggregate factors that change over time that impact prices, γj represents unobserved time-invariant market-level factors that impact prices, and αjt corresponds to unobserved, time-varying market-level factors that impact prices (i.e. unobserved supply and/or demand shocks). Xit is a vector of property characteristics, while Nitd,r represents the number of properties in distressed state d within a radius r of the property transaction. If we observe the sample of properties transact twice, then we have a similar hedonic specification as equation (1) that applies to the second sale transactions occurring in period t + k: 14

The log-linear hedonic specification for modeling house prices is based on the idea that the logarithm of the price of a specific house can be expressed as a function of the inner product of a vector of characteristics and the market-determined shadow prices of those characteristics (for more detailed discussion and derivations see Rosen (1974)).

6

d,r ln(Pij,t+k ) = δt+k + γj + αj,t+k + βt+k · Xi,t+k + λd,r t+k · Ni,t+k + εij,t+k

(2)

Taking the difference in the two equations, and following the literature in assuming that property-level characteristics are time invariant (Xi,t+k = Xit = Xi ) and that the effects that they have on transaction prices are also time-invariant (βt+k = βt = β), yields the following repeat-sales specification:

ln



Pij,t+k Pijt



d,r d,r d,r = (δt+k − δt ) + (αj,t+k − αjt ) + λd,r t+k · Ni,t+k − λt · Nit + (εij,t+k − εijt ) (3)

where the time-invariant, unobserved local market factors, γj , as well as the property characteristics are differenced out.15 This is the main appeal of using a repeat-sales specification. To the extent that time-invariant property or neighborhood characteristics are correlated with both prices and nearby foreclosures, a repeat-sales specification will avoid the resulting omitted variable bias.16 Notice however, that we are still left with two terms that involve unobserved aggregate, time-varying factors (δt+k − δt ) and unobserved market-level, time-varying factors (αj,t+k − αjt ). If these are correlated with the number of nearby distressed properties, then estimation of equation (3) will still suffer from omitted variable bias. For example, unobserved negative demand shocks that lower market prices, which in turn cause foreclosures would create a bias due to reverse casuality.17 Similarly, unobserved positive supply shocks could also lead to lower prices and increased foreclosures producing spurious correlation. Finally, since foreclosure typically leads to the sale of the property, neighboring properties may sell at discounts because the addition of a property to the market gives potential buyers additional bargaining power. This has often been referred to as a “supply effect” of foreclosure in the literature and could also generate a negative correlation between prices and nearby foreclosures, confounding the measurement of price spillovers.18 This may lead 15

Like the previous literature, we are forced to assume that property characteristics do not change over time due to the fact that we do not have historical data on property attributes. However, we do relax the assumption that the shadow prices are constant over time, and estimate a version of equation (3) where we include the term (βt+k − βt ) · Xi,t . The estimation results are reported in the On-Line appendix, which demonstrate that the main results reported in the paper are not affected by this assumption. 16 In the empirical analysis below, we also estimate a hedonic specification and find significant differences between the two methodologies. This suggests that time-invariant property and neighborhood characteristics are important in this context. 17 The existing literature on mortgage default has found a large causal effect of house price movements on mortgage delinquency and foreclosure rates. See, for example, Gerardi, Shapiro, and Willen (2009). 18 Turnbull and Dombrow (2006) find that new listings cause both competitive and shopping externality effects on the market values of nearby properties. Vacant houses exert strong competitive effects in a declining market and exert shopping externality effects in a rising market.

7

to some confusion because the “supply effect” is often viewed as an example of a foreclosure externality. However, economic theory draws a sharp line between physical externalities and supply effects which, since Viner (1931), have been referred to as pecuniary externalities. In order to directly address these econometric issues caused by unobserved demand and supply shocks, we employ a fixed effects specification. Specifically, we include tripleinteraction fixed effects, where the three dimensions correspond to the market j, the first sale quarter t, and the second sale quarter t + k. To see how this triple-interaction fixed effect solves the issue of endogeneity bias, we apply the corresponding within transformation to equation (3), obtaining:

where

\  Pij,t+k \ d,r d,r d d,r ln = λd,r ijt,t+k t+k · Ni,t+k − λt · Nit + ε\ Pijt

(4)

    \  Cj,t,t+k X Pij,t+k 1 Pij,t+k Pij,t+k = ln − ln ln Pijt Pijt Cj,t,t+k i=1 Pijt

(5)

with Cjt,t+k defined as the total number of properties in geography j with first sale taking d \ d,r place in quarter t and second sale taking place in quarter t + k. The terms Nitd,r , N i,t+k , and ε\ ijt,t+k are defined similarly. The within transformation eliminates the unobserved time-varying aggregate and local market factors present in equation (3).19 In addition to including triple-interaction fixed effects, we make an additional assumption in taking equation (3) to the data. We assume that the effect of nearby distressed properties d,r 20 on prices is constant over time (λd,r t+k = λt = λ), so that our estimation equation becomes:

ln



Pij,t+k Pijt



d,r = ζj,t,t+k + λd,r · ∆Ni,t,t+k + ηij,t,t+k

(6)

d,r where ζj,t,t+k are the triple-interaction fixed effects, ∆Ni,t,t+k is the difference in the number

of properties in distressed state d within radius r between the two sale transactions, and ηij,t,t+k is a well-behaved residual. 19

To see this, note that Cjt 1 X 1 αj,t+k = ∗ Cjt ∗ αj,t+k = αj,t+k Cjt i=1 Cjt

since the α’s are constant for each property sold in market j in time period t. The same holds for the αjt , δt+k , and δt . 20 Harding et al., (2009), impose the same restriction in their analysis. In the On-Line appendix, we d,r estimate a version of equation (6) that includes both Nitd and Ni,t+k as separate terms, and thus does not d,r impose the constraint that λd,r = λ. We find that the estimated magnitudes (in absolute value) of t+k = λt d,r d,r λt+k and λt are quite close.

8

Since we are concerned about unobserved local market supply and demand shocks causing omitted variable bias or reverse causality, the fixed effects need to be specified at a geographic level that is at least as small as a local housing market. In this paper we include geographic fixed effects measured at the level of the census block group (CBG). A typical CBG is a very small geographic area, is composed of a relatively homogeneous housing stock, and has a relatively homogeneous population with respect to ethnic and economic characteristics.21 It is likely that the CBG is in most cases smaller in geographical terms than what we typically think of as a local housing market, and thus, the inclusion of fixed effects at the CBG level should control for any unobserved demand and supply shocks that impact local housing markets. Finally, in order to estimate equation (6), it is necessary to take a stand on exactly how Nitd,r is measured in the data. Most papers in the literature have focused on the number of foreclosure completions within a specific period of time around the non-distressed sale transaction. For example, Immergluck and Smith (2006) counts foreclosure deeds in the two years prior to the sale of non-distressed property; Harding, Rosenblatt, and Yao (2009) constructs a series of measures of foreclosure deeds in three month intervals before and after the sale; Rogers and Winter (2009) also uses counts of foreclosure deeds at different time intervals around the non-distressed property sale; and Campbell, Giglio, and Pathak (2011) counts all properties for which foreclosure proceedings have been completed. By focusing on properties that have completed the foreclosure process, these papers implicitly assume that the spillover externality does not occur until the foreclosure auction takes place22 , which is unlikely to be true if the underlying causal mechanism is an investment externality. To see this more clearly, consider a specification that includes the number of foreclosure completions that take place one year before the non-distressed sale of interest. Effectively, such a specification assumes that a property that was foreclosed on more than one year in the past plays absolutely no role whatsoever in the pricing of a nearby property. One might argue that exactly the opposite is true: the properties that produce the most blight, and that may be most likely to adversely impact surrounding values are the properties that lenders cannot sell. To make matters worse, the potential bias introduced by focusing exclusively on foreclosure completion flows instead of the stock of distressed properties is likely not constant over time or across locations. Foreclosure timelines differ widely across states and have slowed considerably through the recent boom/bust cycle, especially in states that 21

There are over 200,000 CBGs in the United States, with each group generally containing between 600 and 3,000 people. They are subsets of census tracts, which contain between 1,500 and 8,000 people. 22 An exception is Schuetz, Been, and Ellen (2008), which counts the number of foreclosure initiations, known as lis pendens filings in New York, in the 18 months prior to a non-distressed sale.

9

require judicial review.23 If, on the other hand, spillovers emanate from the completion of the foreclosure process and the subsequent increase in the inventory of homes on the market, the so-called “supply effect” of foreclosures, then we would expect nearby properties that have not yet fully completed the foreclosure process to have negligible price effects, and the assumption that only foreclosure completions impact nearby home values would be valid. The importance of these two channels also has implications for the distances over which we expect spillovers to operate. If a lack of property maintenance is the mechanism, then spillovers will likely operate over very short distances, while if the mechanism operates through a supply channel, then properties within the same local market will likely be affected, which would include properties at further distances away. While there is agreement in the literature in focusing on nearby foreclosure completions, there is significantly less agreement in the literature regarding the distances over which the spillovers are measured. For example, Immergluck and Smith (2006) focus on radii of 81 and 14 miles, Schuetz, Been, and Ellen (2008) consider mutually exclusive distances of within 250 feet, 250 - 500 feet, and 500 - 1000 feet from a property sale, Campbell, Giglio, and Pathak (2011) use radii of 1 and 41 miles, and Harding, Rosenblatt, and Yao (2009) consider distances of 0 - 300 feet, 10 300 - 500 feet, 500 - 1000 feet, and 1000 - 2000 feet. In this paper we integrate into the analysis information on the timing of mortgage delinquencies with information on the timing of foreclosure completions and REO property sales. This enables us to test whether nearby properties with homeowners in financial distress, but still living in their properties (i.e. either in the midst of the foreclosure process or not yet in foreclosure proceedings), exert negative spillovers on prices. We argue below that this feature of the analysis is important in helping to distinguish between alternative causal mechanisms. We consider numerous stages of distress that range from properties with mortgage borrowers who are only a few months behind on their payments all the way to properties that completed the foreclosure process and were sold by the lender in the distant past. Specifically, in our baseline specification, we include the number of properties with seriously delinquent mortgages (SDQs), which we define as properties owned by borrowers who have been delinquent 90 days or more on their mortgages for at least one year, the number of REO properties (“REO inventory”), and the number of properties sold by the lender to new arms-length homebuyers (“REO Sale”) in the 12 months before the nondistressed property transaction as well as the number between 12 and 24 months before the transaction. Furthermore, in variations of our baseline specification we also include the number of properties with mortgages that have been seriously delinquent for less than one 23 See Gerardi, Lambie-Hanson, and Willen (2013) for a discussion of foreclosure timelines across states and over time.

10

year and properties with mortgages that are fewer than 90 days delinquent, which we refer to as minor delinquencies. Thus, we allow for the possibility that the foreclosure spillover occurs well before the foreclosure is completed, when the borrower first becomes distressed. We also consider multiple distance measures to determine how the strength of price spill-overs changes with distance. For the majority of our analysis we divide the distressed properties into two different distance bins. We consider an inner ring of distressed properties within 0.10 miles and an outer ring between 0.10 and 0.25 miles. The reason for including two mutually exclusive geographic areas is that there is a strong positive correlation among counts of distressed properties at different distances from a given non-distressed sale transaction, as will become apparent in our discussion of the sample summary statistics below. As a result, if we are interested in the effect of distressed properties on the sale price of a property very close by, but omit the number of distressed properties farther away, our estimate will be biased upwards if the distressed properties farther away also exert a negative externality. We find that distressed properties as far away as 0.25 miles seem to exert a non-trivial negative effect on sale prices of non-distressed properties, while distressed properties at distances beyond 0.25 miles exert either no effect or an effect that is extremely small in magnitude and thus can be ignored. This finding (see our discussion of Table 7 below for more details) guides our decision to include counts of distressed properties in the second distance bin (between 0.10 and 0.25 mile) in our baseline specification. We include the various types of distressed property counts and the two distance intervals in the same estimation equation. Thus, our baseline estimation equation takes the following form: ln



Pij,t+K Pijt



= ζj,t,t+k +

XX d

r

 d,r + ηij,t,t+k λd,r · ∆Ni,t,t+k

(7)

where we consider the four measures of distress and two mutually exclusive distances discussed above, which yields eight different combinations of distress and distance as detailed in Table 1.

3

Data

Our sample of repeat sales includes all pairs of non-distressed transactions on single-family residential properties in 15 of the largest, metropolitan statistical areas (MSAs) in the U.S., pulled from public records purchased from Lender Processing Services (LPS). The sample is restricted to include transactions in which the first sale in the repeat-sale pair took place after 2001 and the second sale took place between 2006 and 2010, giving us a sample 11

that overlaps the pre-crisis and post-crisis periods.24 We exclude addresses that cannot be geocoded and transactions for which recorded prices or dates are missing, zero, or located in a thin market, which we define as a CBG in which there were fewer than five sales in a year. The final sample contains 950,234 repeat-sale pairs in 15 MSAs, and 16,932 CBGs, as reported in Panel A of Table 2.25 The bottom panel of Table 2 reports the distribution of observations in the repeat-sales sample by the year of first sale and the year of second sale, respectively. There are several notable patterns in the table. The sample of repeat sales gets smaller over time as national sales volumes fall. In 2006 and 2007, the modal sale occurred two years after purchase, increasing to three years in 2008, four years in 2009, and five years in 2010. The MSAs with the most observations are Phoenix, AZ, Washington, D.C., and Riverside, CA, which account for 16 percent, 13 percent, and 10 percent of the sample, respectively. Table 3 shows that the repeat-sales sample includes enormous variation in returns, which is not surprising, given that the dataset includes properties purchased in 2001 and sold in 2006 and also properties purchased in 2006 and sold in 2010. The public records data also contain information on basic property characteristics, including house size, lot size, property age, and number of bedrooms. These variables are summarized in Table 3. Using the public records, we can identify the date of the foreclosure deed,26 when the lender records transfer of ownership from the borrower, and the REO sale date, when an arms-length buyer takes ownership of the property. Using these flows, we can compute foreclosure inventory in a location at any point in time. The final sample contains 1.04 million foreclosure deeds and 1.15 million REO sales. To identify seriously delinquent properties, we use two methods. Our main approach augments the public records data with proprietary loan-level data from Fannie Mae, one of the two large government sponsored enterprises that provides credit risk insurance for a large fraction of the U.S. mortgage market. These data contain detailed payment histories for every loan insured by Fannie Mae in the 15 MSAs over the sample period. Both the 24

The previous literature has, for the most part, used pre-crisis sample periods. One exception is Campbell, Giglio, and Pathak (2011), which used data between 1987 and the first quarter of 2009, and thus captured a good portion of the crisis period. 25 Another important difference with much of the previous literature is the national representativeness of our data. Many papers have focused on a single state or even a single MSA. For example, Campbell, Giglio, and Pathak (2011) used data from the state of Massachusetts, Immergluck and Smith (2006) used data from Chicago, and Schuetz, Been, and Ellen (2008) used data from New York City. Harding, Rosenblatt, and Yao (2009), which used data from seven large MSAs, is probably the most nationally representative study. 26 The foreclosure deed corresponds either to the transfer of the property to the lender at auction, or if the auction is successful, to the transfer of the property directly to another arms-length buyer. The latter event is significantly less likely to occur than the former, but we are able to distinguish between the two events in the data.

12

Fannie Mae loan-level data and the LPS public records, property-level data have precise information on the physical location of each property, which allows us to precisely geocode the addresses in both datasets. The Fannie Mae dataset allows us to identify the first month in which a delinquent borrower enters serious delinquency (SDQ), which we define to be 90 days delinquent (typically three missed payments). An SDQ corresponds to the entire period in which the borrower is seriously delinquent before the foreclosure auction, and thus it covers both the time before the foreclosure process is initiated on a seriously delinquent borrower, and the time between the start of the foreclosure process and the end of the process (the auction). The data also allow us to identify the cumulative depth of delinquency at any point in time. Our Fannie Mae dataset contains approximately 1.12 million SDQs. However, because the Fannie Mae mortgage data do not cover the universe of all mortgages in our sample of 15 MSAs, we augment it with information from a more complete loan-level dataset.27 Unfortunately, this more complete loan-level data only contain limited information on the geographical location of the properties that collateralize each of the mortgages in the dataset.28 Thus, we are forced to use an indirect approach to identify additional SDQs. With the more complete loan-level dataset, we calculate, for each state, the distribution of the number of months that it takes for a mortgage to transition from serious delinquency to foreclosure completion. We then take the 25th percentile of these distributions and combine them with the information from the public records database on the date of the foreclosure auction to impute SDQ intervals. For example, the 25th percentile for California is four months. Thus, for each of the REO properties located in a California MSA in our sample, we assign an SDQ interval corresponding to the four months before the foreclosure auction dates. To be conservative, we use the 25th percentile as opposed to the median or average, as this means that 75 percent of foreclosures in California had a serious delinquency spell that lasted for more than four months. We call this variable “infilled” SDQs. This provides 726,547 additional SDQs. We then combine our infilled SDQs with the SDQs obtained from the Fannie Mae database to produce a more encompassing SDQ measure. For our analysis, we divide SDQs into “long SDQs” and “short SDQs,” depending on whether the borrower has been seriously delinquent for more than one year or not. For some regressions, we also look at “minor DQs,” which we define to be delinquencies of 60 days or fewer. As we mentioned above, we have some information on the condition 27

These data come from Lender Processing Services (LPS), which collects monthly information on loan characteristics and performance from the largest U.S. mortgage servicers. These data represent approximately 75-80% of the U.S. mortgage market. 28 The LPS servicer-based data provide information on the zip code in which the mortgage borrower resides, but do not provide any further geographic details.

13

of lender-owned properties. These data come from REO property appraisals ordered by lenders immediately after a property completes the foreclosure process. The full residential appraisal form provides a detailed description of the current condition of the property that typically includes a standard rating index with values of “great,” “good,” “average,” “fair,” and “poor.” For the appraisals that do not include the rating index, but do include detailed descriptions, we text-mined all of the keywords in the description and sorted them into the standard condition rating according to the Marshal and Swift Residential Cost Handbook used by appraisers. We characterize “great” and “good” as “above average,” and “fair” and “poor” as “below average.” In addition, we have collected property-level vacancy data from the U.S. Postal Service. Postmen classify each property as “occupied” or “vacant” each month based on whether or not mail is being picked up at each property. These data have been and are currently being used by mortgage servicers to identify loss mitigation opportunities. Panel A of Table 4 shows the percentage of property transactions in our sample that occurred with various numbers of distressed properties within 0.1 miles. More than half of the repeat-sale observations in our sample took place with at least one SDQ within 0.1 miles, and more than one-fifth took place with at least one long SDQ. Due mainly to lack of data, most papers that have tried to measure spillovers have been unable to identify and include in the estimation nearby distressed properties that have not fully completed the foreclosure process. These numbers suggest that this may have been an important omission.29 Panel B of Table 4 considers differences in the number of nearby distressed properties between the second and first sale in the repeat-sale pair, and shows that the vast majority of repeat sales were characterized by either no change in the number of nearby distressed properties or a greater number of nearby distressed properties at the time of the second sale compared to the first sale. This pattern is not surprising since the second sale year is constrained to be in the 2006–2010 period, which was characterized by extremely high delinquency and foreclosure rates. Panel C of Table 4 shows the fraction of sales that occurred with positive numbers of nearby distressed properties, stratified by second sale year. It shows that, not surprisingly, the incidence of sales with distressed properties nearby has increased significantly over time. Most dramatically, the proportion of properties with long SDQs nearby rose from less than 4 percent for repeat sales with second sale year in 2006 to almost 50 percent for repeat sales in 2010, reflecting both an increase in the likelihood that borrowers transitioned into serious delinquency over the course of the sample as well as an increase over time in the length of the foreclosure process. 29 These percentages are almost certainly lower bounds since our sample does not include performance information on all outstanding mortgages.

14

Finally, panel D of Table 4 displays correlations between our stock measures of distressed property and flow measures for both the close distance (within 0.10 miles) and the far distance (between 0.10 and 0.25 miles). Comparing the stock and flow measures of close distressed properties (first four rows and first four columns), it is apparent that they are all positively correlated. However, no two measures have a correlation higher than 0.50, a fact that emphasizes the importance of distinguishing between stocks and flows. We see similar patterns for the correlations among the different types of distressed properties at the far distance (last four rows and last four columns). However, the positive correlations between the number of the same types of distressed properties at different distances are very strong. For example, the correlation between REO inventory at the close distance and REO inventory at the far distance is 0.65, while the corresponding correlation for REO sales in the year before the repeat sale is 0.73. As we discussed above, these correlations emphasize the need to control for distressed properties at longer distances in the estimation, which we do in all of the regressions reported below.

4

Results

In this section we first report results from estimating the baseline specification (equation (7)) with the set of distressed variables listed in Table 1. We pool together all repeat sales to obtain an estimate of the average foreclosure price spillover over the 15 MSAs and various first sale and second sale years in our sample.30 All right-hand-side distress variables are measured as differences across the repeat sales. Standard errors are clustered at the MSA level in order to ensure that spatial and serial correlation in the residuals is not leading to artificially low standard errors and erroneous inference. In all of the regressions we adopt a parsimonious approach and use the unweighted number of each type of distressed property. This assumption is common in the literature.31 Finally we do not include any property or neighborhood controls in the repeat-sales regressions other than the fixed effects.32 Column (1) in Table 5 displays estimation results for the baseline specification with the 30

In the On-Line appendix we estimate separate regressions for each of the 15 MSAs in our sample and show how the spillover estimates change across MSAs. The estimation results are remarkably similar across MSAs, both qualitatively and quantitatively. 31 One notable exception is Campbell, Giglio, and Pathak (2011). See the discussion on page 2,125 for a detailed description of their weights. However, the authors show that their results are largely unchanged by using an unweighted approach. Furthermore, since theory does not provide much guidance on the appropriate weighting scheme to use and since our distance measures are approximate, we are concerned that any inference drawn from a complicated weighting scheme may be misleading, and therefore we choose to estimate unweighted regressions. 32 In the On-Line appendix we show results where we include controls for basic property characteristics. The results are not sensitive to this inclusion.

15

triple-interaction fixed effects at the CBG level. The right-hand-side variables of interest are nearby long SDQs from the Fannie Mae data and three measures from the public records: nearby REO inventory, the number of nearby REO properties sold one year prior to the nondistressed sale, and the number of nearby REO properties sold one to two years prior to the non-distressed sale. The distressed property variables are expressed as differences across the repeat sales within 0.10 miles of the non-distressed sale transactions, which we label “close” and between 0.10 and 0.25 miles, which we label “far.” The estimation results show a basic pattern that is present in all of the subsequent regression results: the estimates associated with nearby long SDQs, REO inventory, and recent REO sales are negative, statistically significant, and non-trivial in magnitude. The estimates associated with long SDQs are the largest (in absolute value), while the REO inventory coefficients are slightly smaller, and the estimates associated with nearby, recent REO property sales are the smallest. The estimates associated with nearby REO property sales that occurred more than one year in the past are also negative and statistically significant, but are close to zero. Another pattern that emerges is that the measures of close distressed properties have a slightly more negative effect on price growth than the measures of far distressed properties. This pattern is consistent with results in the previous literature on foreclosure spillovers and is a finding that we will explore in more detail below. According to the table, an additional long SDQ within 0.10 miles is estimated to decrease price growth on average by 1.2 percent while an additional long SDQ between 0.10 and 0.25 miles decreases price growth by 0.5 percent on average. The corresponding magnitudes for the REO inventory estimates are -1.0 percent and -0.6 percent respectively, while those for the recent REO property sales variable are -0.6 percent and -0.2 percent respectively. REO property sales that occurred between one and two years before the sale do not have an economically meaningful impact on prices for either distance metric. The inclusion of the set of triple-interaction fixed effects plays a significant role in the estimation results. The remaining columns in Table 5 show how sensitive the results are to the omission of the fixed effects as well as to changes in the geographic aggregation level of the fixed effects. Column (2) displays estimates of the baseline specification without any fixed effects whatsoever. The estimated magnitudes of price spillovers become extremely large (in absolute value). In the absence of fixed effects to control for unobserved local supply and demand shocks, an additional nearby long SDQ reduces price growth by approximately 5.5 percent on average, while an additional nearby REO property reduces price growth by 2.6 percent on average. Even nearby properties that the lender sold between one and two years earlier reduce price growth by a non-trivial amount (-0.5 percent). Column (3) in Table 5 includes a set of bilateral fixed effects that control for the quarter in 16

which the first sale occurred and the quarter in which the second sale occurred, but does not control for geography. Thus, this regression explicitly controls for time-varying aggregate shocks (the δs in equation (3)), but does not control for local time varying demand and supply shocks. The inclusion of the bilateral fixed effects substantially reduces the estimated price spillovers. For example, the magnitude of the coefficient estimate associated with long SDQs within 0.1 miles is reduced by almost two-thirds falling from -0.055 to -0.019. The difference in results between the specification with triple-interaction fixed effects measured at the CBG level and bilateral fixed effects is significant (column (1) versus column (3)), although it is not nearly as large as the difference between including bilateral fixed effects and no fixed effects. The spillover estimate associated with close long SDQs falls from -1.9 percent to -1.2 percent, and the estimate associated with close REO inventory falls from -1.7 percent to -1.0 percent. This confirms the intuition that we discussed above, where we argued that unobserved local demand and/or supply shocks could drive much of the observed negative correlation of nearby distressed properties and prices. The remaining columns in Table 5 display results for the baseline specification with triple-interaction fixed effects expressed at higher geographic aggregation levels than the CBG. Column (4) includes fixed effects measured at the MSA level, column (5) at the county level, and column (6) at the census tract level. The results are very similar to those from the specification that included fixed effects measured at the CBG level. These results suggest that the MSA is a sufficiently small geography to deal with reverse causality and omitted variable bias in this context. In Table 6, we exploit information about the condition of REO properties and vacancy status of distressed properties. As discussed in the previous section, for a subset of SDQs, we have information about the vacancy status of the properties, and for a subset of the REO properties we have information about condition. Since the vacancy and condition data are only well-populated beginning in 2009 and 2010, we focus on only repeat-sale observations in which the second sale year was either 2009 or 2010. According to Table 6, an additional close REO property in below average condition decreases price growth by approximately 2.6 percent, while an additional REO property in average condition reduces price growth by 1.5 percent on average. The estimated coefficient associated with above average REO properties is significantly positive (2 percent), albeit only marginally significant, which implies that nearby REO properties in good condition actually increase the sale price of non-distressed properties. The fact that the estimate associated with the missing category is significantly negative, but slightly lower than the estimate associated with average REO properties likely reflects the fact that many of the REO properties with missing condition data are in average to above average condition. One of the main takeaways from this table is that estimated 17

foreclosure spillovers are sensitive to the condition of nearby foreclosed properties. Below average REO properties exert an average negative spillover that is almost double that of a foreclosed property in average condition. In addition, unlike the estimates of the baseline specification in Table 5, the spillover effect for below average REO properties does not seem to diminish with distance, as evidenced by the estimate associated with below average REO properties between 0.1 and 0.25 miles away. Unlike the REO property condition results, the spillover estimates of SDQs do not appear to be sensitive to vacancy status. The results show that the coefficient estimate associated with vacant SDQ properties is approximately the same as that associated with occupied properties (-1.0 percent versus -0.9 percent).33 In Table 7 we look at the impact of distressed properties even farther away from the repeat-sale observations, by augmenting the baseline specification with two additional rings of distressed properties: 0.25–0.50 miles away and 0.50–1.0 miles away. Most papers in the literature have found negative price spillovers of nearby foreclosures that drop off rapidly in distance. The results in Table 7 are consistent with this finding as the coefficient estimates associated with distressed properties in the third and fourth rings are very close to zero. These results motivate our decision to include only the 0.10–0.25 mile ring in our baseline specification, as distressed properties farther than 0.25 miles do not exert an economically meaningful effect on non-distressed home sales. In Table 8 we test whether properties with moderately delinquent mortgages also exert price spillovers. The table reports estimation results in which we distinguish between short and long SDQs (SDQs that have lasted for less than one year and more than one year, respectively) as well as SDQs and minor DQs (loans that have missed less than 3 payments). The results in column (2) of the table show that the average estimated spillover effect of short SDQs is negative and statistically significant, but is about half the magnitude (in absolute value) as the estimated spillover effect from long SDQs (-0.6 percent versus -1.2 percent). In contrast, minor DQs are not found to negatively impact prices of nearby properties.34 Thus, it appears that negative price spillovers from nearby distressed properties first appear when borrowers are in serious delinquency, and not when borrowers have missed only one or two mortgage payments. In addition to the results on short SDQs and minor DQs, Table 8 reports estimation results from a series of alternatively specified regressions. At least two issues merit special 33

The reliability of the vacancy data could be an issue here however. The data on the occupancy status of a property in the foreclosure process come from observations of postal workers who deliver mail to the properties. The postal workers are simply reporting whether mail is picked up and whether the property looks abandoned or occupied, and thus there could be many instances of misclassification. 34 The point estimate associated with minor DQs is positive and statistically significant, but is quite small in magnitude.

18

attention. The first is a potential sample selection problem since our coverage of SDQs is only partial because the Fannie Mae proprietary mortgage data do not cover the entire mortgage market. Since unobserved SDQs are likely to be correlated with both the Fannie Mae SDQs and all other measures of distress, including the large number of REO properties from our public records database, sample selection bias here could potentially affect all of the estimates of interest. To address this problem, we use the infill method, which we discussed in detail above, to construct a more complete dataset of SDQs. In columns (3) and (4) of Table 8 we combine short and long SDQs into an overall SDQ measure and compare results for the sample of only SDQs identified in the Fannie Mae dataset and the sample of “infilled” SDQs. The results illustrate that using a broader measure of SDQs makes little difference to the spillover estimates. Overall, the coefficient estimate associated with SDQ decreases slightly in absolute value while the coefficient estimates associated with the other measures of distress are not significantly affected. The second specification issue is the choice of a repeat-sales versus a hedonic specification. With the exception of Harding, Rosenblatt, and Yao (2009), most papers in the literature have estimated hedonic specifications. As we showed in detail above in Section 2, the main difference between the specifications is the fact that the dependent and independent variables are expressed in differences instead of levels.35 In Table 8, we implement two alternative specifications to try to shed some light on the extent to which the two different approaches yield different estimates of price spillovers. In column (6), we show results from a hedonic model in which we specify both the dependent variable and the right-hand side variables in levels rather than differences. The results are fairly striking. The absolute magnitudes of all of the estimates increase dramatically. Long SDQs decrease prices by approximately 2.6 percent on average in the hedonic model compared to 1.2 percent in the repeat-sales model (shown in column (1)), while nearby REO properties decrease prices by 2.5 percent on average in the hedonic specification compared to only 1.0 percent in the repeat-sales model. In column (5) we consider a hybrid model that lies part way between the two specifications, in which we express the dependent variable in differences (price growth), but specify the nearby distressed property variables in levels (at the time of the second sale). This hybrid model generates slightly smaller spillover estimates (in absolute value) than the full repeat-sales specification, and significantly smaller estimates than the hedonic model. Thus, there do appear to be some important differences between the repeat-sales and 35

Another important difference is in sample selection. Estimating a repeat-sales specification requires a sample of properties that are observed to transact at least two times, compared to a hedonic specification which only requires a single transaction. Thus, repeat-sale samples are more restrictive and could result in potential sample selection issues. We do not explore this issue in this paper, choosing to estimate the hedonic specification on the same sample of repeat sales.

19

hedonic specifications, with the hedonic specification delivering significantly larger spillover estimates from all types of nearby distressed properties. Finally, there is evidence in the literature that foreclosure price spillovers are not linear in the number of nearby foreclosures. Schuetz, Been, and Ellen (2008) and Harding, Rosenblatt, and Yao (2009) both document spillover effects that are diminishing in the number of nearby foreclosures. In all of our specifications above we assumed that nearby distressed properties impact prices in a linear manner, so that the effect on prices from an additional nearby distressed property is the same in the case of zero versus one nearby distressed property and four versus five nearby distressed properties. In Figure 2 we display results for an alternative repeat-sales specification in which we allow for potential nonlinearities in spillovers.36 Specifically, we re-estimated our baseline specification including a series of indicator variables for each specific value of the difference in nearby distressed properties over the repeat-sale horizon. We group all repeat-sales observations with a difference of five or more nearby distressed properties (for both positive and negative values) due to the fact that we simply do not have many repeat-sales in the sample that have a difference of more than positive five or less than negative five distressed properties within 0.1 miles. The figure shows estimation results for each type of distressed property where the top panel shows results for the close distance and the bottom panel for the far distance. The first thing to note about the figure is that the spillover estimates are similar for long SDQs, inventory of REO properties, and REO property sales that took place up to one year before the non-distressed sale. However, there are important differences for the more extreme values of the distressed property variables for the close distance. For example, repeat-sales observations with five or more nearby recent REO property sales experienced approximately nine percent less price growth compared to observations with zero nearby recent REO property sales, while the corresponding estimate associated with nearby long SDQs is approximately four percent. The second notable observation from the figure is that the spillover estimates do appear to be fairly linear for values of the right-hand-side variables between -5 and 5, perhaps more so for the long distance than the short distance. The spillover estimates associated with long SDQs do appear to be nonlinear in nature however.37 In addition, consistent with the earlier reported results, the spillover estimates are economically small for REO sales that took place more than one year before the sale. 36

A more detailed table that shows the exact point estimates and standard errors is available in the On-Line appendix. 37 There appears to be diminishing marginal effects for positive values of nearby long SDQs, and increasing marginal effects for negative values. However, we do not have any repeat-sale observations in which there were three or more nearby long SDQs at the time of the first sale relative to the second sale, so we are reticent to make any strong claims.

20

5

Interpretation

In Section 4, we established some empirical facts. Houses that sell very close to all forms of distressed property appear to do so at slightly lower prices than otherwise similar properties in the same CBG that sell without the presence of nearby distressed properties. The effect appears when the borrower becomes seriously delinquent on his mortgage and disappears one year after the lender sells the foreclosed property to a new homeowner in an arms-length transaction. In this section we discuss the most plausible explanation for these patterns in the data. If our identification strategy is truly successful in dealing with reverse causality and omitted variable bias, then the most likely explanation for the empirical evidence in this paper is an investment externality effect. This refers to the tendency of financially distressed borrowers and lenders that do not derive consumption services from foreclosed properties to under-invest in property maintenance, leading to physical deterioration of the property and, in turn, causing a reduction in the values of nearby properties to potential buyers. There are two main results presented in Section 4 that support such an interpretation. First, there is the evidence on the condition of REO properties in Table 6. The fact that the coefficient estimate associated with nearby, below average, REO properties is significantly smaller than the coefficient estimate associated with nearby REO properties in average and above average condition is consistent with an investment externality interpretation. If instead we thought that nearby foreclosed properties were driving down prices by competing with non-distressed sales (the so-called “supply effect”), then we would expect, at the very least, that the distressed properties in above average condition would have the same effect as distressed properties in below average condition. Indeed, we might even expect the above-average distressed properties to generate even more competition, in which case we should expect above-average properties to generate negative effects on nearby prices rather than the positive effects we find in Table 6. The second piece of evidence that supports an investment externality interpretation is the pattern of the estimated coefficients for the different categories of property distress. Table 8 shows that nearby properties in which homeowners have only missed one or two mortgage payments (minor DQs) do not negatively impact prices, while nearby properties in which homeowners are seriously delinquent (more than three payments behind) do exert negative price effects. This is consistent with the idea that homeowners in serious delinquency do not have the financial resources and/or financial incentive to continue maintaining their homes. In addition, the results show that nearby properties sold out of REO more than one year before the non-distressed transactions of interest have negligible effects on prices. This is

21

consistent with new arms-length homebuyers having the financial resources and incentives to rehabilitate the properties. One important caveat about the analysis that must be pointed out is that our identification strategy, by construction, is meant to rule out price effects from unobserved supply and demand shocks. Therefore, we are unable to infer whether distressed properties also exert price effects through supply and demand channels. It is important to stress that we do not view this as a weakness of the analysis, but rather as a strength. In Section 2 we alluded to the fact that standard economic theory makes a crucial distinction between price effects caused by technological or physical externalities and price effects caused by supply and demand shocks, referred to in the literature as pecuniary externalities. The point of that discussion is that from a policy perspective, we should care more about non-pecuniary externalites compared to pecuniary externalities. In our context, while supply effects from increased foreclosure activity may certainly impact housing prices in a nontrivial way, the policy implications, if there are any, are not at all clear. In contrast, if there are nonpecuniary externalities associated with the foreclosure process, then there is likely a larger role for policy. Thus, the point of the identification strategy used in the analysis is to eliminate the pecuniary externalities that come in the form of supply and demand shocks and concentrate on the highly localized price effects that are likely to reflect non-pecuniary externalities.

6

Economic Importance

In this section we will briefly discuss our evidence on the economic importance of the negative price spillover estimates presented above. This is an extremely important issue from a policy perspective, since statistically significant, negative spillover estimates that are trivial in magnitude are not likely to be of interest to policy makers. We believe the evidence on economic importance is mixed. The results in Table 5, imply that an additional nearby property in the foreclosure process or property that has been foreclosed and is in the lender’s possession decreases prices by about one percent, on average. Figure 2 provides evidence that this spillover effect is roughly linear, at least for property sales with up to five nearby distressed properties. According to the figure having five or more nearby long SDQs decreases prices by about 4 percent while five or more nearby REO properties decreases prices by about 6 percent. These are certainly non-trivial magnitudes, however, one needs to remember that there are very few property sales that take place with that many distressed properties within 0.1 miles. Table 1 shows that in our repeat-sales sample, approximately 8 percent of transactions occurred with five or more nearby long 22

SDQs, and only 3 percent occurred with five or more nearby REO properties. On the other hand, 70 percent of transactions had no REO properties within 0.1 miles and about half of the transactions in the sample had no long SDQs nearby. Thus, these results suggest that while a few homeowners trying to sell their properties in hard hit neighborhoods may experience significant externalities from nearby distressed properties, most homeowners will not experience noticeable price spillovers. One important caveat to this conclusion comes from the results on the condition of REO properties. In Table 6 we saw that nearby REO properties in poor condition exerted significantly stronger negative price spillovers. The spillovers from REO properties in below average condition were almost twice as large (in absolute value) as the spillovers from REO properties in average condition. Thus, properties located in areas with many nearby distressed properties in poor states of maintenance, could experience large decreases in value.

7

Conclusion

In this paper, we use a new dataset that covers 15 of the largest MSAs in the U.S., spans the period before and after the 2008 financial crisis, and contains detailed information on mortgage delinquencies in addition to foreclosed properties to document some new facts about foreclosure price spillovers. We show that houses trade at slightly lower prices when there are homes nearby with delinquent homeowners, when there are homes nearby owned by lenders, and even when there are homes nearby recently sold by lenders in arm’s-length transactions. We show that nearby houses trade at substantially reduced prices when the lender-owned property is in below-average condition and at higher prices when it is in above average condition. We consider three possible explanations for the facts: supply effects, demand effects, and investment externalities, and argue that the third is the most plausible. Perhaps the most important conclusion that one should take from this analysis is that the effects of foreclosure and distressed property in general on the prices of neighboring homes are fairly small. We estimate the effect of a property in SDQ and a property in REO on the price of a home within 0.10 miles to be approximately -1.0 percent, an amount that would most likely go unnoticed by the typical seller who does not have many distressed homeowners living nearby. As Table 4 shows, the vast majority of properties sell without any distressed properties nearby, meaning that it is impossible to attribute more than a token amount of the collapse in prices in the 2006–2010 period to foreclosures. At the same time though, our results show that homeowners located in neighborhoods with significant numbers of distressed properties in poor states of maintenance may experience large negative 23

price externalities. Finally, the policy implications of even a small investment externality effect are important, especially in many areas of the U.S. that have been characterized by large numbers of distressed properties throughout the recent foreclosure crisis. Our results suggest two potentially effective policy avenues to minimize the spillover costs of foreclosure. First, policies aimed directly at preventing the rapid depreciation of properties owned by financially distressed borrowers and mortgage lenders are consistent with our results. A nice example of a current set of policies that address this issue are vacant property registration ordinances (VPROs), which are laws that usually require the formal registration of foreclosed properties, and often require banks to maintain and secure REO properties. A second type of policy that is also consistent with our results is one that minimizes the time that properties spend in serious delinquency and in the REO state. On one hand, this implies putting pressure on lenders to sell properties quickly. On the other hand, and perhaps much less palatably, it implies minimizing the time a borrower spends in serious delinquency, which means accelerating the foreclosure process.

24

References Anenberg, Eliot, and Edward Kung. 2014. “Estimates of the Size and Source of Price Declines due to Nearby Foreclosures.” American Economic Review 104(8): 2527–2551. Campbell, John Y, Stefano Giglio, and Parag Pathak. 2011. “Forced Sales and House Prices.” American Economic Review 101(5): 2108 – 2131. Clauretie, T.M., and N. Daneshvary. 2009. “Estimating the House Foreclosure Discount Corrected for Spatial Price Interdependence and Endogeneity of Marketing Time.” Real Estate Economics 37(1): 43–67. Ellen, Ingrid Gould, Johanna Lacoe, and Claudia Ayanna Sharygin. 2013. “Do Foreclosures Cause Crime?” Journal of Urban Economics 74(2): 59–70. Fisher, Lynn M, Lauren Lambie-Hanson, and Paul S Willen. Forthcoming. “The role of proximity in foreclosure externalities: evidence from condominiums.” American Economic Journal: Economic Policy. Frame, W.S. 2010. “Estimating the Effect of Mortgage Foreclosures on Nearby Property Values: A Critical Review of the Literature.” FRB Atlanta Economic Review. Galster, G.C., P. Tatian, and R. Smith. 1999. “The Impact of Neighbors Who Use Section 8 Certificates on Property Values.” Housing Policy Debate 10(4): 879–917. Gerardi, Kristopher, Kyle Herkenhoff, Lee O-Hanian, and Paul Willen. 2013. “Unemployment, Negative Equity, and Strategic Default.” Working Paper 2013-4. Federal Reserve Bank of Atlanta. Gerardi, Kristopher, Lauren Lambie-Hanson, and Paul S. Willen. 2013. “Do Borrower Rights Improve Borrower Outcomes? Evidence from the Foreclosure Process.” Journal of Urban Economics 73(1): 1–17. Gerardi, Kristopher., Adam H. Shapiro, and Paul S. Willen. 2009. “Decomposing the Foreclosure Crisis: House Price Depreciation Versus Bad Underwriting.” Working Paper 2009-25. Federal Reserve Bank of Atlanta. Harding, John P., Stuart S. Rosenthal, and C.F. Sirmans. 2007. “Depreciation of housing capital, maintenance, and house price inflation: Estimates from a repeat sales model.” Journal of Urban Economics 61(2): 193–217. Harding, J.P., E. Rosenblatt, and V. Yao. 2012. “The Foreclosure Discount: Myth or Reality?” Journal of Urban Economics 71(2): 204–218.

Harding, J.P., E. Rosenblatt, and V.W. Yao. 2009. “The Contagion Effect of Foreclosed Properties.” Journal of Urban Economics 66(3): 164–178. Immergluck, D., and G. Smith. 2006. “The External Costs of Foreclosure: The Impact of Single-family Mortgage Foreclosures on Property Values.” Housing Policy Debate 17(1): 57–79. Immergluck, Dan, Yun S. Lee, and Patrick Terranova. 2012. “Local Vacant Property Registration Ordinances in the U.S.: An Analysis of Growth, Regional Trends, and Some Key Characteristics.” Working Paper 2130775. SSRN. Leonard, Tammy, and James C. Murdoch. 2009. “The Neighborhood Effects of Foreclosure.” Journal of Geographical Systems 11(4): 317–332. Lin, Z., E. Rosenblatt, and V.W. Yao. 2009. “Spillover Effects of Foreclosures on Neighborhood Property Values.” The Journal of Real Estate Finance and Economics 38(4): 387–407. Matulef, Mark. 1988. “The Effects of Subsidized Housing on Property Values.” Journal of Housing 45: 286–287. Pennington-Cross, A. 2006. “The Value of Foreclosed Property.” Journal of Real Estate Research 28(2): 193–214. Rogers, William H., and William Winter. 2009. “The Impact of Foreclosures.” Journal of Real Estate Research 31(4): 455–479. Rosen, Sherwin. 1974. “Hedonic prices and implicit markets: product differentiation in pure competition.” The Journal of Political Economy 34–55. Rossi-Hansberg, E., P.D. Sarte, and R. Owens III. 2010. “Housing Externalities.” Journal of Political Economy 118(3). Schuetz, J., V. Been, and I.G. Ellen. 2008. “Neighborhood Effects of Concentrated Mortgage Foreclosures.” Journal of Housing Economics 17(4): 306–319. Schwartz, A.E., I.G. Ellen, I. Voicu, and M.H. Schill. 2006. “The External Effects of Placebased Subsidized Housing.” Regional Science and Urban Economics 36(6): 679–707. Towe, Charles, and Chad Lawley. 2013. “The Contagion Effect of Neighboring Foreclosures.” American Economic Journal: Economic Policy 5(2): 313–335.

Table 1: Measures of Distressed Property Measure of Distress Long REO REO REO

Distance < 0.1 miles 0.1-0.25 miles

SDQ λSDQ,close Inventory λREO,close SALE<1,close Sales < 1 yr λ Sales 1 − 2 yrs λSALE>1,close

λSDQ,far λREO,far λSALE<1,f ar λSALE>1,f ar

Notes: This table displays the regression coefficients of interest in our baseline specification given by:    XX Pij,t+k d,r λd,r · ∆Ni,t,t+k + ηij,t,t+k = ζj,t,t+k + ln Pijt r d

where the dependent variable is the price appreciation between the sale of property i between time t and t + k. Altogether we focus on eight parameters corresponding to four stages of property distress and two distance measures. “Long SDQ” refers to the number of properties for which a mortgage borrower has been at least 3 payments behind for one year, “REO Inventory” refers to the number of REO properties at the time of the non-distressed sale, “REO Sales < 1 yr” refers to the number of sales out of REO status that occurred in the twelve months before the non-distressed sale, and “REO Sales 1 − 2 yrs” refers to number of sales out of REO status that occurred between one and two years before the non-distressed sale. We measure each of these variables at two mutually exclusive distances from the non-distressed sale: < 0.1 miles and between 0.1 miles and 0.25 miles.

Turnbull, Geoffrey K., and Jonathan Dombrow. 2006. “Spatial Competition and Shopping Externalities: Evidence from the Housing Market.” Journal of Real Estate Finance and Economics 32: 391–408. U.S. Department of Housing and Urban Development. 2010. “Economic Impact Analysis of the FHA Refinance Program for Borrowers in Negative Equity Positions.” Tech. rep. Viner, Jacob. 1931. “Cost Curves And Supply Curves.” Zeitschrift f¨ ur National¨okonomie 101: 23–46.

Table 2: Distribution of Repeat-Sale Observations By MSA and Year Panel A: By Metropolitan Statistical Area # CBGs Atlanta Boston Chicago Detroit LV LA Miami New York Orlando Philadelphia Phoenix Riverside Seattle Tampa DC

1,020 1,029 1,139 1,060 438 2,228 825 2,023 383 1,485 1,069 885 836 745 1,767

Total

16,932

2006

Share of Repeat-Sale Transactions 2007 2008 2009 2010

26.8 19.5 24.2 18.4 26.3 24.4 33.8 21.5 29.6 24.8 26.2 27.9 28.5 30.1 26.3

23.0 21.0 19.8 20.2 15.5 16.3 20.0 22.3 17.8 23.5 18.3 15.0 24.6 18.8 19.2

17.7 18.1 16.9 20.9 17.7 16.2 13.0 19.7 13.3 18.0 17.2 16.3 16.1 14.6 16.8

14.9 21.0 20.3 20.7 20.5 22.1 16.2 20.5 17.8 18.2 18.9 19.7 15.8 18.1 19.4

17.6 20.4 18.9 19.9 19.9 21.0 17.0 16.1 21.5 15.5 19.4 21.2 15.1 18.5 18.4

Total 70,949 22,550 27,761 23,011 69,211 84,137 65,877 44,572 29,580 50,529 153,548 93,804 35,869 54,613 124,223

251,873 180,981 158,691 179,782 178,907 950,234 Panel B: By Year

Year of First Sale

2006

Year of Second Sale 2007 2008 2009 19,078 23,624 29,943 37,644 34,409 24,189 12,094 . . .

12,206 14,899 19,408 25,105 34,341 28,344 14,658 9,730 . .

11,814 14,634 19,175 25,126 36,163 32,530 14,944 13,041 12,355 .

2010 11,459 14,234 18,439 24,394 34,103 29,443 16,983 7,489 10,677 11,686

Total

%

85,199 104,914 136,283 170,295 195,459 134,427 58,679 30,260 23,032 11,686

9.0% 11.0% 14.3% 17.9% 20.6% 14.1% 6.2% 3.2% 2.4% 1.2%

2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

30,642 37,523 49,318 58,026 56,443 19,921 . . . .

Total %

251,873 180,981 158,691 179,782 178,907 950,234 100.0% 26.5% 19.0% 16.7% 18.9% 18.8% 100.0% .

Notes: This table reports the distribution of repeat-sale pairs of non-distressed transactions across the 15 metropolitan statistical areas (MSAs) in our sample. The data are drawn from public records purchased from Lender Processing Services (LPS). The repeat-sales sample includes all repeat-sale pairs in which the first sale occurred between 2001 and 2010 and the second sale between 2006 and 2010. The sample exclude addresses that cannot be geocoded and transactions for which recorded prices or dates are missing, zero, or located in a thin market, which we define as a census block group (CBG) in which there were fewer than five sales in a year.

Table 3: Basic Sample Summary Statistics Mean Price ($) Prior ($) Appreciation (%) Holding Period (qtrs)

Std

Min

p25

p50

p75

Max

331,217 252,565 4,500 180,900 270,400 400,000 9,775,000 349,559 279,454 4,500 187,000 285,000 425,000 9,500,000 0 40 -110 -20 10 30 90 13.9 8.4 1 8 13 20 39

Property Characteristics Size (square ft.) 1,973 Lot (square ft.) 12,424 Age (years) 23.8

849 23,014 21.9

500 500 0

1,372 5,200 7

1,770 7,405 16

2,371 10,890 36

32,222 435,600 150

Notes: This table displays summary statistics on the distributions of the transaction prices, price appreciation between repeat sales, quarters between repeat sales, and basic property characteristics for the sample of repeat sales used in the empirical analysis. The mean, standard deviation, minimum, 25th percentile (p25), 50th percentile (p50), 75th percentile (p75), and maximum of each variable are displayed. The repeat sales sample includes all repeat-sale pairs in which the first sale occurred between 2001 and 2010 and the second sale between 2006 and 2010. The sample exclude addresses that cannot be geo-coded and transactions for which recorded prices or dates are missing, zero, or located in a thin market, which we define as a census block group (CBG) in which there were fewer than five sales in a year. Price appreciation between repeat sales is winsorized at the first and ninety-ninth percentiles of the distribution.

Table 4: Summary Statistics of Nearby Distressed Properties for Repeat-Sales Observations Panel A: % of Sales with # distressed property within 0.1 miles 0 1 2 3 4 ≥5 SDQs Long SDQs Short SDQs Minor DQs REO Inventory REO Sales in last year REO Sales 1-2 years ago

47.4 79.7 52.2 13.2 70.0 71.3 78.7

21.6 11.4 23.2 9.8 15.2 12.3 11.1

11.9 4.4 11.8 8.9 6.3 5.2 3.8

6.8 2.2 6.1 8.1 3.3 3.0 2.0

4.2 1.1 3.2 7.3 1.9 2.0 1.2

8.2 1.4 3.6 52.8 3.3 6.2 3.2

Panel B: % of sales with difference in # distressed property within 0.1 miles <0 0 1 2 3 4

≥5

Long REO REO REO

1.2 2.1 5.2 2.7

SDQs Inventory Sales in last year Sales 1-2 years ago

0.3 2.4 8.3 9.2

82.6 75.5 66.9 73.0

9.4 11.5 10.8 9.4

3.7 4.7 4.5 3.2

1.8 2.4 2.6 1.6

1.0 1.4 1.7 1.0

Panel C: % of Sales with > 0... 2006 2007 2008 2009 2010

Total

SDQs Long SDQs Short SDQs Minor DQs REO Inventory REO Sales in last year REO Sales 1-2 years ago

52.6 20.4 47.8 86.8 30 28.7 21.3

40.2 3.7 39 84.4 14.2 7.8 12.2

40.2 4.5 38.7 84.4 18.9 8.3 7.7

49.6 13 45.1 86.5 37.3 29.6 8.9

65 38.9 54.5 89.1 41.9 52.7 29

72.9 47.7 65 90.8 45.1 54.1 51

Table 4 Continued

Panel D: Correlations SDQ inventory (< 0.1 miles) SDQ inventory (< 0.1 miles) REO inventory (< 0.1 miles) REO sales in last year (< 0.1 miles) REO sales 1-2 years ago (< 0.1 miles) SDQ inventory

REO inventory (< 0.1 miles)

REO sales in last year (< 0.1 miles)

SDQ REO sales 1-2 years ago inventory (< 0.1 miles) (0.1-0.25 miles)

REO inventory (0.1-0.25 miles)

REO sales in last year (0.1-0.25 miles)

REO sales 1-2 years ago (0.1-0.25 miles)

1 0.31

1

0.44

0.47

1

0.33

0.29

0.45

1

0.61

0.30

0.45

0.35

1

0.65

0.49

0.31

0.40

1

0.46

0.73

0.45

0.55

0.59

1

0.29

0.45

0.67

0.43

0.39

0.56

( 0.1 - 0.25 miles) REO inventory 0.32 ( 0.1 - 0.25 miles) REO sales in last year 0.44 ( 0.1 - 0.25 miles) REO sales 1-2 years ago 0.34 ( 0.1 - 0.25 miles)

Notes: This table displays detailed summary statistics on the number of nearby distressed properties to the non-distressed repeat-sales observations in the sample. A “short” SDQ is defined as a property in which the borrower has been in serious delinquency for less than one year, a “long” SDQ is defined as a serious delinquency that has lasted for more than one year, and a “minor delinquency” is defined as a property in which the borrower is only one or two payments behind. The repeat-sales sample includes all repeat-sale pairs in which the first sale occurred between 2001 and 2010 and the second sale between 2006 and 2010. The sample exclude addresses that cannot be geocoded and transactions for which recorded prices or dates are missing, zero, or located in a thin market, which we define as a census block group (CBG) in which there were fewer than five sales in a year.

1

Table 5: Baseline Specification. Dependent variable is price growth between repeat sales.

∆ Long SDQ (< 0.1 miles) ∆ in REO inventory (< 0.1 miles) ∆ REO sold in last year (< 0.1 miles) ∆ REO sold 1-2 years ago (< 0.1 miles) ∆ Long SDQ (0.1 − 0.25 miles) ∆ in REO inventory (0.1 − 0.25 miles) ∆ REO sold in last year (0.1 − 0.25 miles) ∆ REO sold 1-2 years ago (0.1 − 0.25 miles) Geographic Fixed Effect Time Period Fixed Effect # Fixed Effects # Observations Overall R2 Within R2

(1)

(2)

(3)

(4)

(5)

(6)

-0.012∗∗∗ (8.7) -0.010∗∗∗ (4.5) -0.006∗∗∗ (7.9) -0.001∗∗∗ (2.7) -0.005∗∗∗ (4.2) -0.006∗∗∗ (2.9) -0.002∗∗∗ (3.3) 0.001 (1.1)

-0.055∗∗∗ (8.8) -0.026∗∗∗ (5.5) -0.018∗∗∗ (9.5) -0.005∗∗ (3.3) -0.022∗∗∗ (5.6) -0.016∗∗∗ (4.3) -0.008∗∗∗ (4.8) 0.000 (0.0)

-0.019∗∗∗ (6.2) -0.017∗∗∗ (6.8) -0.009∗∗∗ (5.8) -0.002∗∗∗ (2.7) -0.008∗∗∗ (5.2) -0.012∗∗∗ (5.7) -0.004∗∗∗ (3.4) 0.000 (0.0)

-0.011∗∗∗ (6.6) -0.012∗∗∗ (4.4) -0.007∗∗∗ (5.4) -0.002∗∗∗ (2.7) -0.004∗∗∗ (3.1) -0.007∗∗∗ (2.7) -0.003∗∗∗ (2.9) 0.001 (1.1)

-0.011∗∗∗ (7.5) -0.010∗∗∗ (5.1) -0.005∗∗∗ (10.7) -0.001∗∗∗ (2.5) -0.005∗∗∗ (4.6) -0.006∗∗∗ (3.4) -0.002∗∗∗ (3.2) 0.001 (1.5)

-0.012∗∗∗ (8.7) -0.010∗∗∗ (4.5) -0.006∗∗∗ (7.9) -0.001∗∗∗ (2.7) -0.005∗∗∗ (4.2) -0.006∗∗∗ (2.9) -0.002∗∗∗ (3.3) 0.001 (1.1)

CBG Quarter-Year 745,940

None None 0

950,234 0.82 0.04

950,234 0.34 0.34

None MSA County Census Tract Quarter-Year Quarter-Year Quarter-Year Quarter-Year 590 8,850 57,125 250,364 950,234 0.73 0.15

950,234 0.82 0.05

950,234 0.83 0.06

950,234 0.82 0.04

Notes: This table displays results from the estimation of equation (7). In all regressions reported, the dependent variable is price appreciation over the repeat-sale interval. Column (1) is the baseline specification and includes triple-interaction fixed effects at the CBG level (i.e. CBG×quarter-ofsale1×quarter-of-sale2 fixed effects). Column (2) includes only bilateral fixed effects given by quarter-of-sale1×quarter-of-sale2, with no geographic component. The repeat-sales sample includes all repeat-sale pairs in which the first sale occurred between 2001 and 2010 and the second sale between 2006 and 2010. Columns (3), (4), and (5) include triple-interaction fixed effects at the MSA, County, and Census Tract levels, respectively. ∗ , ∗∗ , and ∗∗∗ denote statistical significance at the 10, 5, and 1 percent levels, respectively. The numbers in parentheses are t statistics. Standard errors are clustered at the MSA level.

Table 6: Effect of Property Condition and Vacancy Status based on 2009-2010 Sample. ∆ REO Inventory by Condition Below Average (< 0.1 miles) Average (< 0.1 miles) Above Average (< 0.1 miles) Missing Data (< 0.1 miles) Below Average (0.1 − 0.25 miles) Average (0.1 − 0.25 miles) Above Average (0.1 − 0.25 miles) Missing Data (0.1 − 0.25 miles) ∆ SDQs by Vacancy Status Occupied (< 0.1 miles) Vacant (< 0.1 miles) Missing Data (< 0.1 miles) Occupied (0.1 − 0.25 miles) Vacant (0.1 − 0.25 miles) Missing Data (0.1 − 0.25 miles) # Fixed Effects # Observations Overall R2 Within R2

-0.026∗∗∗ (4.3) -0.015∗∗∗ (7.8) 0.020∗ (2.9) -0.009∗∗ (4.0) -0.027∗∗∗ (8.4) -0.010∗∗∗ (4.3) 0.027∗∗∗ (5.2) -0.004* (2.4) -0.009∗∗ (4.0) -0.010∗∗ (3.9) -0.010∗∗∗ (4.7) -0.003 (1.5) -0.002 (0.7) -0.004∗ (2.5) 419,681 517,380 0.74 0.07

Note: This table displays results from the estimation of equation (7). The dependent variable is price appreciation over the repeat-sale interval, and the regression includes triple-interaction fixed effects at the CBG level (i.e. CBG×quarter-of-sale1×quarter-of-sale2 fixed effects). The sample consists of all repeatsales pairs for which the second sale took place in 2009 or 2010. The top panel in the table displays results for nearby REO inventory in different states of upkeep according to property appraisals performed at the time of foreclosure completion and acquisition of the property by the lender (see section 3 for more details). The bottom panel displays results for nearby SDQ inventory by vacancy status according to observations made by postal service workers (see section 3 for more details). ∗ , ∗∗ , and ∗∗∗ denote statistical significance at the 10, 5, and 1 percent levels, respectively. The numbers in parentheses are t statistics. Standard errors are clustered at the MSA level.

Table 7: Effect of REO/SDQs by Distance Difference in number of: ∆ Long SDQs ∆ REO inventory ∆ REO sold in: 0.0 - 0.10 miles 0.10 - 0.25 miles 0.25 - 0.50 miles 0.50 - 1.0 miles # Fixed Effects # Observations Overall R2 Within R2

-0.011∗∗∗ (8.5) -0.003∗∗∗ (4.5) -0.002∗∗∗ (5.2) 0.000 (0.8)

-0.008∗∗∗ (4.4) -0.003 (2.0) -0.002∗ (2.4) -0.001∗ (2.2)

Last year

1-2 years ago

-0.006∗∗∗ (9.0) -0.001∗ (2.4) -0.001∗∗ (3.3) 0.000 (0.9)

-0.002∗∗ (3.1) 0.001* (2.4) 0.000 (1.6) 0.000∗∗ (3.1)

745,940 950,234 0.83 0.04

Note: This table displays results from the estimation of equation (7). The dependent variable is price appreciation over the repeat-sale interval, and the regression includes triple interaction fixed effects at the CBG level (i.e. CBG×quarter-of-sale1×quarter-of-sale2 fixed effects). The repeat-sales sample includes all repeat-sale pairs in which the first sale occurred between 2001 and 2010 and the second sale between 2006 and 2010. ∗ , ∗∗ , and ∗∗∗ denote statistical significance at the 10, 5, and 1 percent levels, respectively. The numbers in parentheses are t statistics. Standard errors are clustered at the MSA level.

Table 8: Alternative Specifications.

(1) ∆ Minor DQs ∆ Long SDQs

-0.012∗∗∗ (8.7)

∆ Short SDQs

Baseline (2) (3)

(4)

0.003∗∗∗ (4.6) -0.012∗∗∗ (7.1) -0.006∗∗∗ (4.5)

∆ All SDQs

RHS in levels (5)

Hedonic (6)

-0.009∗∗∗ (7.0)

-0.026∗∗∗ (7.5)

-0.006∗ (2.8) -0.007∗∗∗ (4.9) 0 (0.6)

-0.025∗∗∗ (8.0) -0.011∗∗ (3.5) -0.004∗∗ (3.4)

745,940 950,234 0.82 0.03

745,940 950,234 0.74 0.19

-0.007∗∗∗ (6.1)

∆ All SDQs with Infill -0.010∗∗∗ (4.5) ∆ REO Prop. sold in past year -0.006∗∗∗ (7.9) ∆ REO Prop. sold 1-2 year ago -0.001∗ (2.7)

-0.010∗∗∗ (4.7) -0.006∗∗∗ (7.6) -0.001∗ (2.3)

-0.010∗∗∗ (4.6) -0.006∗∗∗ (7.1) -0.001∗ (2.4)

-0.006∗∗∗ (6.8) -0.009∗∗∗ (4.4) -0.006∗∗∗ (7.7) -0.001∗ (2.6)

# Fixed Effects # Observations Overall R2 Within R2

745,940 950,234 0.82 0.04

745,940 950,234 0.82 0.04

745,940 950,234 0.82 0.04

∆ REO Inventory

745,940 950,234 0.82 0.04

Note: This table displays results from the estimation of equation (7). The dependent variable is price appreciation over the repeat-sale interval, and all regressions include triple interaction fixed effects at the CBG level (i.e. CBG×quarter-of-sale1×quarter-of-sale2 fixed effects). The sample includes all repeat-sale pairs in which the first sale occurred between 2001 and 2010 and the second sale between 2006 and 2010. A “short” SDQ is defined as a property in which the borrower has been in serious delinquency for less than one year, a “long” SDQ is defined as a serious delinquency that has lasted for more than one year, and a “minor delinquency” is defined as a property in which the borrower is only one or two payments behind. See the discussion in section 3 for a description of the “infill” method used to calculate SDQs. ∗ , ∗∗ , and ∗∗∗ denote statistical significance at the 10, 5, and 1 percent levels, respectively. The numbers in parentheses are t statistics. Standard errors are clustered at the CBG level.

Distressed property counts (in millions)

Figure 1: The Stock of Distressed Properties by Stage of Distress

4 REO+FCL+>90DQց 3

REO+FCLց

2

1 ւREO

0.5 0

06

07

08

09

10

11

12

Notes: The data behind panel A come from Lender Processing Services, which is a servicer-based loanlevel dataset that covers approximately 75%-80% of the U.S. mortgage market. “REO” corresponds to the outstanding inventory of properties that are owned by the lender, “FCL” refers to the number of loans that are in the foreclosure process, and “90DQ” refers to the inventory of loans that are at least 3 mortgage payments behind.

Figure 2: Nonlinearities in Spillover Estimates Panel A: Close Distance 10.0% 8.0%

EstimatedPriceSpillover

6.0% 4.0% 2.0% 0.0% Ͳ2.0% Ͳ4.0% Ͳ6.0% Ͳ8.0% Ͳ10.0%

чͲ5

LongSDQs

REOInventory

REOSales<1Yr

REOSales1Ͳ2Yrs

Ͳ4

Ͳ3

Ͳ2

Ͳ1

0

1

2

3

4

ш5

3

4

ш5

ѐ#DistressedProperties(ч0.10Miles)

Panel B: Far Distance 10.0% 8.0%

EstimatedPriceSpillover

6.0% 4.0% 2.0% 0.0% Ͳ2.0% Ͳ4.0% Ͳ6.0% Ͳ8.0% Ͳ10.0%

чͲ5

LongSDQs

REOInventory

REOSales<1Yr

REOSales1Ͳ2Yrs

Ͳ4

Ͳ3

Ͳ2

Ͳ1

0

1

2

ѐ#DistressedProperties(0.10Ͳ 0.25Miles)

Note: This figure graphs the estimated coefficients from a regression where the dependent variable is price appreciation over the repeat-sale interval, and the independent variables include the four types of nearby distressed properties. Separate indicator variables are included for values of the right-hand-side variables between -5 and 5 with zero as the baseline (left out case). The regression includes triple interaction fixed effects at the CBG level (i.e. CBG×quarter-of-sale1×quarter-of-sale2 fixed effects). The sample includes all repeat-sale pairs in which the first sale occurred between 2001 and 2010 and the second sale between 2006 and 2010. The top panel displays estimates corresponding to the “close” distance (≤ 0.10 miles) while the bottom panel displays estimates corresponding to the “far” distance (0.10 - 0.25 miles). See the On-Line appendix for the corresponding table with standard errors.