1 Estmatng the Effect of Crme Rsk on Property Values and Tme on Market: Evdence from Megan s Law n Vrgna Raymond Brastow Federal Reserve Bank of Rchmond Raymond.Brastow@rch.frb.org Phone: Fax: E Byrd St Rchmond, VA Benne Waller* firstname.lastname@example.org Phone: Fax: Longwood Unversty 201 Hgh Street Farmvlle, VA Scott Wentland email@example.com Phone: Fax: Longwood Unversty 201 Hgh Street Farmvlle, VA * Contact author WORKING COPY: DO NOT CITE WITHOUT PERMISSION Keywords: Property values; Tme on market; Hedonc; Megan's Law
2 1 Abstract Ths paper explores the effect that lvng near a sex offender has on the marketablty of one s home. Specfcally, we estmate the mpact on a home s sales prce and the length of tme t takes for the home to sell. Snce the 1994 passage of Sexual Offender Act (known as Megan s Law), persons convcted of sex crmes have been requred to notfy local law enforcement about ther current domcle and any change of address. Snce then, sex offenders resdences have become publcly avalable nformaton allowng anyone to lookup whether a sex offender resdes nearby. Usng cross-sectonal data from a central Vrgna multple lstng servce we fnd that sexual offenders have robust and economcally large effects on nearby real estate. Our results ndcate that the presence of a nearby regstered sex offender reduces home values by approxmately 9%. Moreover, these same homes take as much as 10% longer to sell than homes not located near regstered sex offenders. These results prove robust over numerous specfcatons and modelng technques commonly found n the lterature.
3 2 Introducton The ntent of ths research s to examne the mpact of Megan s Law on the marketng duraton and sales prce of resdental real estate as observed n rural areas of central Vrgna. The emprcal real estate lterature has lnked a host of housng, property, and surroundng area characterstcs to sales prce and marketng duraton. However, relatvely few have studed the effect of Megan s Law on sales prce and none have studed the Law s effect on marketng duraton. Addtonally, prevous studes have examned the effect of sex offenders n prmarly urban settngs. For example, Lnden and Rockoff (2008) looked at data from Mecklenburg County, NC (a populous county contanng the cty of Charlotte), and Pope (2008) used data from Hllsborough County, FL (contanng Tampa). Ths paper flls ths gap n the lterature by estmatng the effect a nearby sex offender s resdence has on ts surroundng real estate market n relatvely rural areas of central Vrgna. We fnd a substantal dfference between our estmates of a sex offender s mpact n rural areas as compared to the estmates n studes of urban areas, suggestng a fundamental dfference between rural and urban areas n values placed on crme rsk mposed by nearby sex offenders. Lterature Revew Whle sex offender regstres are nothng new, the wdespread dssemnaton of such regstres on the nternet s a relatvely recent phenomenon. Most states now provde detaled data about the locatons, physcal descrptons, and pctures of nearby sex offenders, even ncludng detals about the charges for whch they have been convcted. Utlzng these new data sources, Pope (2008) and Lnden and Rockoff (2008) both fnd that propertes n close proxmty to a property that s lsted as the home address of a regstered sex offender wll suffer a loss n value. Both studes control for a varety of ndvdual home characterstcs as well as area fxed effects to control for heterogenety n ther respectve real estate markets. Indeed, the loss to property value that a regstered sex offender brngs dsspates as dstance from the sexual offenders address ncreases. However, each study uses sales and crme data from only a sngle county (Hllsborough, FL and Mecklenberg, NC, respectvely).
4 3 Whle the aforementoned studes analyze home values, they omt analyses of the home s marketng duraton. Yet, marketng duraton of a property has been examned and studed from numerous perspectves n pror real estate lterature. Belkn, Hempel and McLeavey (1976) were one of the frst to put forth the theory that lst prce, and changes to lst prce, drectly mpact the property s tme on market. They also ste market mperfectons, such as nadequate communcaton of prce changes, that may mpact tme on market of propertes. Mller (1978) fnds a postve relatonshp between tme on market and lst prce but also notes that a longer marketng duraton does not necessarly result n a hgher sales prce. Haurn (1988) has been well cted n the marketng duraton lterature for hs clam that the more atypcal a property the longer t wll reman on the market. Specfcally, he ctes unusual locaton as an example of an atypcal characterstc that may negatvely affect a property s marketng duraton. Turnbull and Dombrow (2007) present evdence that propertes located near other propertes lsted by the same agent are able to brng a hgher sales prce. The authors also fnd that the greater the dversty of lstngs by an agent, the longer ther lstngs stay on the market. Yang and Yavas (1995) suggest that hgher commsson rates for agents do not mpact the tme on market. However, they do suggest that a hgher commsson rate may sgnal that the property s more costly to sell because of ts locaton. They specfcally cte the example of a rural property beng more expensve to show than a property n the cty. Our study ncorporates methodologes found n these artcles regardng property values and marketng duraton and apples them to the analyss of the effect of sexual offenders on real estate marketablty. Data The data for ths research are from several sources. Informaton about sexual offenders s contaned n the Vrgna Sex Offender and Crmes Aganst Mnors Regstry and s avalable on a publc web ste mantaned by the Commonwealth of Vrgna. 1 Each observaton contans the regstered sex offender s current and pror addresses, along wth a number of other personal characterstcs (e.g. age, sex, race, and descrpton of the perpetrated crme). Data on real estate transactons consst of 1 See the followng webste:
5 4 observatons of resdental propertes on the market between July 1999 and June 2009 and comes from a multple lstng servce located n central Vrgna. The ntal housng data conssts of 21,453 observatons. After cullng for ncomplete, mssng or llogcal data that suggest data entry errors, the fnal data set conssts of 13,172 sold propertes. The data collected from the MLS nclude typcal property characterstcs (square footage, bedrooms and baths), and market and calendar nformaton (locaton, lst dates, length of lstng contract). Methodology We are nterested n the mpact regstered sex offenders have on real estate prces and the tme t takes to sell a gven home (.e. marketng duraton). After obtanng the longtude and lattude of the regstered sex offenders addresses, we use the great-crcle dstance formula to calculate the dstance from a regstered sex offender s home to a gven house on the market. Dstance = arccos [sn LAT π 1 sn LAT 2 + cos(lat 1) cos LAT 2 cos (LON 2 LON 1 ) (1) Followng Waller, Brastow, and Johnson (2009), the present study remans agnostc wth respect to the model specfcaton debate wthn the lterature. Indeed, we employ all three of the most common methodologes (descrbed below) n the related real estate lterature n order to determne the robustness of emprcal results. Lke Lnden and Rockoff (2008) and Pope (2008), we estmate the effect a sex offender has on surroundng real estate sale prces (wthn certan rad) for propertes located across the state of Vrgna. One key contrbuton of ths paper, however, s to answer an entrely new queston: do homes near regstered sex offenders also take longer to sell? To that end, the followng general modelng framework s presented: TOM X, L, Z, SO, SP) (2) (
6 5 SPˆ ( x, l, z, SO ) (3) Where: TOMand SP are vectors for property marketng tme and property sellng prce respectvely (expressed n natural log form), X and x are vectors of property characterstcs and other control varables, L and l are vectors for locaton control, Z and z are vectors that nclude varables such as the degree of over prcng and market condton, and SO, the varable of nterest, equals 1 f a regstered sex offender (contemporaneously) lves near home, 2 0 otherwse. We also estmate the sheer dstance between home and the nearest regstered sex offender as a contnuous varable. We report OLS coeffcent estmates of tme on market and sales prce [Equatons (2) and (3)] and Webull estmates for tme on market. A thrd model captures potental endogenety between sales prce and marketng tme. In ths model, SP s estmated n reduced form n Equaton (3) for a three stage least squares (3SLS) specfcaton, wth predcted values, SP ˆ, substtutng for SP n Equaton (2). In a 3SLS settng, (2) and (3) form the system of equatons between property prce and property marketng tme. In addton, 3SLS ncorporates an addtonal step wth seemngly unrelated regresson (SUR) 2 As we wll explan n the next secton, nearby can mean dfferent dstances to dfferent people. Below, we explore a number of dstances (or, more specfcally, rad) from the nearest sex offender (for example,.1 mle,.25 mle,.5 mle, and 1 mle).
7 6 estmaton. 3 Varables ncluded n the equatons above (vectors X, L, and Z ) follow exstng lterature and wll be dscussed n greater detal n the next secton. Results We fnd that f there s a sex offender regstered at a nearby resdence, then nearby homes sell for less and that those homes take longer to sell. However, the magntudes are strkng. Tables 2 (frst column), 3, and 4 show that a nearby sex offender (.e., one who resdes wthn one-tenth mle) reduces property values by approxmately 9% and ncreases the marketng duraton of a house by approxmately 10%. These are economcally meanngful effects and demonstrate that central Vrgna resdents assgn a large rsk to lvng near a convcted sex offender. To determne senstvty and robustness of the results, we estmated a number of model specfcatons. Pror studes have classfed nearby regstered sex offender n dfferent ways. Table 2 shows the effects of usng dfferent rad for measurng the effect a sex offender has on property values, holdng a number of property characterstcs constant. 4 Pope (2008) uses a dummy varable equal to 1 f a sex offender resdes wthn one-tenth mle (0.1 mles) of a property and another second dummy varable for two-tenths of a mle (0.2 mles). Lnden and Rockoff (2008) also use the 0.1 mle dummy, but dffer n ther addtonal use of a 0.3 mle dummy varable. The frst column of Table 2 shows perhaps the most strkng result of the paper: a regstered sex offender lvng wthn.1 mle of one s home wll reduce the value of surroundng propertes sold by about 9% (or, more precsely, 8.8%). Ths s more than twce the magntude of smlar estmates reported n Pope (2008) and Lnden and Rockoff (2008), suggestng a 3 Accordng to Belsley (1988), 3SLS has an edge on 2SLS n estmatng systems of equatons because t s more effcent, partcularly when there are strong nterrelatons among error terms. 4 Lke any multvarate analyss, we attempt to solate the effect a sex offender has by controllng for numerous property characterstcs that also affect our dependant varable (lke square footage, age, number of houses on the market, whether t s vacant, number of bedrooms, bathrooms, whether t has a pool, brck exteror, hardwood floors, walk-n closet, fnshed basement, gas freplace, paved drveway, fenced yard, and the acreage of the property). In addton, we control for the year and what tme of year t was sold n (.e. the season), the unemployment rate n Vrgna, whether t s a townhouse or condo, and area fxed effects (n most cases, cty/town). Most of these control varables are commonly used wthn the real estate lterature. Hence, we lmt our dscusson n ths paper prmarly to the varable of nterest: proxmty to a regstered sexual offender.
8 7 much hgher wllngness to pay to avod crme rsk n predomnantly rural central Vrgna. For the average homeowner, ths works out to be $14,826. To put ths n relatve terms, homeowners value avodng ths crme rsk more than they value a pool, a brck exteror, a walk n closet, a freplace, a paved drveway, or an addtonal bathroom. More precsely, homeowners value avodng ths rsk as much as 295 square feet n ther home (.e. homeowners mght be wllng to lve n a sgnfcantly smaller home f they could avod lvng so close to a sex offender). As the radus for the nearest sex offender wdens, Table 2 also shows that ths effect s stll present, albet dmnshed. A nearby sex offender lowers property values 8.5%, 6.6%, and 4.6% when you classfy nearby as resdng wthn 0.25, 0.5, and 1 mle respectvely. Ths result s also somewhat unque, gven that pror studes have not found ths effect on property values beyond 0.3 mles, suggestng that there s a sgnfcant dfference between the percepton of neghbor n rural and urban areas. Resdences tend to be less densely located n more rural areas lke central Vrgna than n urban areas lke Charlotte and Tampa. Hence, rural resdents may smply perceve homes wthn a larger radus as neghbors, resultng n a greater alertness to crme rsk over larger dstances. Some real estate studes (e.g. Rutherford, Sprnger, and Yavas (2005)) have chosen a Heckman selecton model (Heckman, 1979) to correct for sample selecton bas. There s good reason to suspect that our dependant varable, sales prce, s only observed for a restrcted, non-random sample. Qute smply, a house only has a sellng prce f t s actually sold. Hence, OLS estmates could be based because a number of houses lsted on the market are not sold (e.g. perhaps they are less lkely to be sold because a regstered sex offender lves nearby, whch would bas the estmates of the houses that actually sell). Table 3 shows the coeffcent estmates for the Heckman model, where the nverse Mlls rato corrects for selectvty bas by adjustng the condtonal error terms such that they have a zero mean (generated from a probt model, where the bnary dependent varable s whether the property has actually
9 8 sold). Thus, takng nto account potental sample selecton bas, nearby sex offenders stll mpose about an 8% dscount on surroundng propertes. 5 Ths general result remans robust even when we change the relevant area fxed effects. Lnden and Rockoff (2008), for example, use dummy varables for neghborhoods (probably the most relevant area varable n urban settngs), and other studes use smlar measures to hold constant one of the most mportant pllars of real estate value: locaton. Studes ncorporate area fxed effects n an attempt to control for unobserved heterogenety across these areas so that the explanatory varables effects are dentfed from varaton wthn a gven area (or even n a gven year, as s the case for year fxed effects). For most of our analyss up to ths pont, we have used ctes/towns n whch these propertes resde. In central Vrgna and possbly n other predomnantly rural areas, the relevant area metrc tends to be wder than n urban settngs. But, perhaps ths s too wde and there may stll substantal heterogenety wthn these areas. Table 3 shows that when we use zp codes or elementary school dstrcts as the relevant area metrc, we contnue to fnd a substantal dscount for homes located near a regstered sex offender. Whle sellers tend to sell ther propertes at substantally lower values when a regstered sex offender lves nearby, they may not be lowerng ther sales prce enough. Sellers and ther agents may have dffculty estmatng a property s expected value f a sex offender s near. That s, a reduced offer prce may not attract enough potental buyers, resultng n a longer marketng duraton of the home. Table 4 shows that homes located near (wthn.1 mle of) a regstered sex offender spend abut 10% more tme on the market. Ths works out to be about 13 days longer on the market than other smlar propertes, whch are also compettvely prced. In relatve terms, ths s roughly equvalent to sellng your home n the off season of fall or wnter (as compared to the summer or sprng). However, as s ndcated n the methodology secton, a number of studes n the real estate lterature employ other econometrc models to analyze tme on market. Gven the extensve debate wthn 5 The fact that lambda s statstcally sgnfcant ndcates selecton bas. Though, the dfferences between the OLS and Heckman estmates do not appear economcally sgnfcant, suggestng the selecton bas s present but not large.
10 9 the lterature, our paper ntends to reman neutral by employng three common technques for modelng tme on market: Heckman, Webull, and three-stage least squares. Moreover, ths neutralty comes wth an addtonal beneft of demonstratng that our general results are not senstve to the modelng technque, confrmng the robustness 6 of our fndngs. Table 4 shows that all technques are generally consstent wth the OLS fndng that regstered sex offenders ncrease marketng duraton of nearby propertes n central Vrgna (rangng from approxmately 6% to 10%). Concludng Remarks Ths study fnds that resdents n central Vrgna home sellers must absorb a relatvely large rsk premum when sellng property near a regstered sex offender. Alternatvely, home buyers are wllng to pay a premum to lve n safer areas. The qualtatve result s not surprsng and s entrely consstent wth prevous fndngs n smlar studes. However, the quanttatve result reveals an economcally large dfference between the rsk premum assocated wth sex offenders n a rural area lke central Vrgna versus urban areas lke Charlotte and Tampa. Moreover, our analyss of marketng duraton suggests that homes located near regstered sex offenders take longer to sell, sgnfyng a general reluctance to purchase propertes exposed to such rsks. Certanly, no one wants to lve near a regstered sex offender, but emprcal results ndcate that the more tghtly knt communtes of central Vrgna are wllng to pay more to avod such a rsk. Ths study s results are consstent wth the noton that resdents n rural areas consder larger areas when defnng who consttutes as a neghbor and assessng subsequent rsks. The results may also reflect 1) a greater averson to crme, 2) more households wth famles compared to more densely populated urban areas, or some combnaton of factors that contrbute to ths rather strkng magntude. Further research may shne lght on the dfferences between rural and urban areas and perhaps the sources of such dfferences. 6 The fnal verson of ths workng paper wll nclude addtonal qualtatve robustness checks that are not present n the present verson.
11 10 Exhbt 1: Varable legend Varable Tom Lprce lnlp Sprce Comm NoMkt Age Sqft Bedrooms Fullbath Halfbath Garage Fre Brck Vnyl Hardwood Ceramc Fullbase Vacant Area 1 Area 2 Area 3 Area 4 Area 5 Area 6 Lsttme Frmsd Wnter Sprng Summer Fall Defnton Tme on market measured from date of orgnal lstng to contract date Lstng prce ln(lstng prce) Sales prce Commsson rate Property sold n less than 4 days. Age of property Square footage of property Number of bedrooms Number of full bathrooms Number of half bathrooms Dummy varable, 1 f property has a garage, 0 otherwse Dummy varable, 1 f property has a freplace, 0 otherwse Dummy varable, 1 f property has brck exteror, 0 otherwse Dummy varable, 1 f property has vnyl floors, 0 otherwse Dummy varable, 1 f property has hardwood floors, 0 otherwse Dummy varable, 1 f property has ceramc floorng, 0 otherwse Dummy varable, 1 f property has full basement, 0 otherwse Dummy varable, 1 f property was vacant when lsted Dummy varable, 1 f property located n area 1, 0 otherwse Dummy varable, 1 f property located n area 2, 0 otherwse Dummy varable, 1 f property located n area 3, 0 otherwse Dummy varable, 1 f property located n area 4, 0 otherwse Dummy varable, 1 f property located n area 5, 0 otherwse Dummy varable, 1 f property located n area 6, 0 otherwse Chronologcal tme varable 30 year fxed rate mortgage at property contract date Dummy varable, 1 f property was lsted n wnter, 0 otherwse Dummy varable, 1 f property was lsted n sprng, 0 otherwse Dummy varable, 1 f property was lsted n summer, 0 otherwse Dummy varable, 1 f property was lsted n fall, 0 otherwse
12 11 Table 1a: Housng Descrptve Statstcs Varable Mean Std. Dev. Mn Max TOM Lprce Sprce Compensaton Nomkt Age Sqft Bedrooms Fullbath Halfbath Garage Fre Brck Vnylsdng Hardwood Ceramctle Fullbase Vacant Area Area Area Area Area Area Lsttme Frmsd Wnter Sprng Summer Fall Table 1b: Sex Offender Descrptve Statstcs Varable Obs Mean Std. Dev. Mn Max Male Black Whte Age Volent crme
13 12 Table 2 The effect of a Nearby Regstered Sex Offender on the Sellng Prce of a Home (1) - OLS (2) - OLS (3) - OLS (4) - OLS Independent Coeffcent t-stat Coeffcent t-stat Coeffcent t-stat Coeffcent t-stat Varables so_tenthmle so_quartermle so_halfmle so_onemle sqft lnage onmkt vacant bedrooms baths pool onestory brck hardwood walkncloset fnbase freplacegas paveddrve fencedyard condo townhouse fall wnter sprng acreage vaunemp constant Area Fxed Effects Year Fxed Effects Observatons R-squared *T-statstcs n all of the above regressons are robust
14 13 Table 3 The effect of a Nearby Regstered Sex Offender on the Sales Prce of a Home (1) - OLS (2) - Heckman (3) - OLS (4) - OLS Independent Varables Coeffcent t-stat Coeffcent* z-stat Coeffcent t-stat Coeffcent t-stat so_tenthmle sqft lnage onmkt vacant bedrooms baths pool onestory brck hardwood walkncloset fnbase freplacegas paveddrve fencedyard condo townhouse fall wnter sprng acreage vaunemp constant lambda (IMR) Cty/town Fxed Effects Zp Code Fxed Effects Elementary School Fxed Effects Year Fxed Effects Observatons R-squared n/a *Heckman Coeffcents (dy/dx) are calculated usng the mfx compute, pred(ycond) postestmaton command T-statstcs n all of the above regressons are robust
15 14 Table 4 The effect of a Nearby Regstered Sex Offender on the Marketng Duraton (or, Tme on Market) of a Home (1) - OLS (2) - Heckman (3) - Webull (4) 3SLS Independent Coeffcent t-stat Coeffcent* z-stat Coeffcent z-stat Coeffcent t-stat Varables + so_tenthmle lnlp sqft lnage onmkt vacant bedrooms baths pool onestory brck hardwood walkncloset fnbase freplacegas paveddrve fencedyard condo townhouse fall wnter sprng acreage vaunemp constant lambda Area Fxed Effects Year Fxed Effects Observatons Ch-squared n/a *Heckman Coeffcents (dy/dx) are calculated usng the mfx compute, pred(ycond) postestmaton command +Webull Coeffcents (dy/dx) are elastctes evaluated at ther respectve sample medans
17 16 References Belsley, Davd A., Two- or Three-stage Least Squares? Computatonal Economcs. 1988, 1, Belkn, J., D. J. Hempel and D. W. McLeavey, An Emprcal Study of Tme on Market Usng Multdmensonal Segmentaton of Housng Markets, Journal of the Amercan Real Estate and Urban Economcs Assocaton, 1976, 4, Haurn, D., The Duraton of Marketng Tme on Resdental Housng, AREUEA Journal 1988, 16(4): Heckman, J., Sample Selecton Bas as a Specfcaton Error, Econometrca, 1979, 47 (Feb.): Jud, Donald G., Terry G. Seaks, and Danel T. Wnkler, Tme on the Market: The Impact of Resdental Brokerage, The Journal of Real Estate Research 1996, 12(3): Knght, John R., Lstng Prce, Tme on Market and Ultmate Sellng Prce: Causes and Effects of Lstng Prce Changes, Real Estate Economcs 2002, 30(2): Lnden, L. and J.E. Rockoff, Estmates of the Impact of Crme Rsk on Property Values from Megan s Laws, Amercan Economc Revew, 2008, 98:3, Mller, N. G., Tme on the Market and Sellng Prce, AREUEA Journal, 1978, 6(2): Pope, J.C., Fear of crme and housng prces: Household reactons to sex offender regstres, Journal of Urban Economcs, 2008, 64, Rutherford, R.C., T.M. Sprnger, and A. Yavas. Conflcts Between Prncpals and Agents: Evdence from Resdental Brokerage. Journal of Fnancal Economcs, 2005, 76:3, Turnbull, G. K., and J. Dombrow. Indvdual Agents, Frms, and the Real Estate Brokerage Process. Journal of Real Estate Fnance and Economcs, 2007, 35:1, Yavas, A. and S. Yang. The Strategc Role of Lstng Prce n Marketng Real Estate: Theory and Evdence. Real Estate Economcs, 1995, 23:3,