Private government,property rights and uncertain neighbourhood externalities: Evidence from gated communities

Abstract

Chinese

Economists traditionally view public and private land use regulation as alternatives to each other. An alternative view argues that public and private regulation are not equally suited to accomplish the same outcomes. In particular, government regulation is easily changed while private regulation is not, making the latter better suited to control future externality risk. One implication of the alternative view is that more risk-averse households are drawn to gated neighbourhoods while their less risk-averse counterparts are not. This paper exploits exogenous differences in neighbourhood amenity uncertainty created by public school attendance zone changes to test this prediction of the alternative view. The results show that greater exogenous amenity uncertainty yields stronger house price capitalisation in gated subdivisions than in open neighbourhoods, a pattern consistent with the risk-aversion sorting hypothesis. The results are robust and are consistent with the key implication of the alternative view of private regulation.

Keywords

house price capitalisation externality risk gated communities HOA private government

Introduction

The traditional view of land use regulation in the USA considers government regulation and private regulation in the form of deed restrictions as alternative approaches to accomplishing the same goals; either can be used to ameliorate neighbourhood externalities (Fischel, 1990). An alternative view argues that government and private regulation serve different purposes (Hughes and Turnbull, 1996a, 1996b); in this view private regulation can accomplish things that government regulation cannot. In particular, government regulation is easily changed over time. In contrast, the common law foundation of private regulation in the form of deed restrictions makes it difficult to modify once put into place. Hughes and Turnbull (1996b) argue that the difficulty of changing private regulation, considered a disadvantage in the traditional view (Fischel, 1990), actually makes it a better tool for controlling future externality risk. One implication of the alternative view is that households who are more risk averse are drawn to privately regulated gated neighbourhoods while their less risk-averse counterparts choose less restrictive environs. This paper examines whether or not this pattern holds, providing the first empirical test of the alternative view of regulation.

Gated communities exist in many countries, providing residents with amenities, privacy and a sense of security (Helsley and Strange, 1999; Lacour-Little and Malpezzi, 2009; Landman and Schontech, 2002; Le Goix, 2005; Le Goix and Vesselinov, 2013; Low, 2001; Wilson-Doenges, 2000).¹ Systems of property rights and their enforcement differ across countries, so it is not surprising that the role of gated communities in housing markets and their interaction with governmental powers vary across countries as well. Property law in this regard is well established in much of the USA; deed restrictions are private agreements that lie outside the sphere of local government power. Nonetheless, gated neighbourhoods can affect the social fabric of the broader community (Manzi and Smith-Bowers, 2005) and influence the demand for government-supplied services. Gated neighbourhoods can present a wider range of challenges for local government administration and urban planning in countries in which the legal distinction between private regulation and public regulation is not as clear-cut as in the USA (Liao et al., 2019). Since this paper focuses on testing whether the alternative view of private regulation is relevant or not, the distinction between private and public regulation and the results of this analysis are only relevant to regions with property law similar to that in the USA.

The gated developments in our sample are governed by homeowners’ associations (HOAs) with powers to enforce neighbourhood covenants regulating property development, maintenance and use in addition to their duties of managing common areas, recreational facilities, and roads, drainage and other physical infrastructure.² Applying the private regulation theory in Turnbull (1994) and Hughes and Turnbull (1996a, 1996b) to gated neighbourhoods, their alternative view predicts that households sort systematically across gated and open neighbourhoods according to their degree of neighbourhood amenity risk aversion. The next section explains how we test the model prediction using school quality uncertainty as an exogenous neighbourhood amenity risk similarly affecting both gated and open subdivisions in the same neighbourhood.

The paper makes two contributions to the literature. First, it applies the alternative view of private regulation to gated subdivisions, providing a general framework linking all of the traditional rationales for gated communities to the externality risk reduction rationale that is unique to the alternative view and shows how this opens the door for testing the validity of the alternative view. The second contribution of this paper is empirical; it presents the first empirical test of a key relationship predicted by the alternative theory of private regulation. School quality and property transactions data from Orange County, Florida, over 2001–2012 clearly show that the same level of neighbourhood amenity uncertainty in the form of education quality risk from school attendance zone instability reduces housing prices more in gated subdivisions than in open neighbourhoods, a result that is robust across neighbourhood income levels and market phases. The observed price effect is consistent with the household sorting predicted by the theory.

Household consumption risk aversion and housing price

This section outlines how the Hughes-Turnbull argument that private regulation reduces neighbourhood externality risk can be applied to the gated community environment to derive empirically testable implications regarding household sorting across neighbourhoods according to their attitude towards risk.³ Gated communities offer residents three different types of benefits: recreational or common area facilities (community goods or services); restrictions on neighbourhood properties and their use (reduction in neighbourhood externalities and externality risk); and greater privacy or safety (reduced accessibility to non-residents). This discussion focuses on the second factor, residents’ desires to control neighbourhood externalities and the associated consumption risk, as this is the essence of the alternative view.

An implication of bid price equilibrium is that houses in gated developments sell at a premium only if the additional benefits provided by the development are worth at least as much to households as the additional cost they must incur directly in the form of homeowner association and management fees. Previous studies of gated communities do find such premiums (Lacour-Little and Malpezzi, 2009; Le Goix and Vesselinov, 2013). This price pattern by itself, however, does not provide evidence of household sorting into these communities according to their degree of risk aversion.

According to the Hughes-Turnbull theory, stricter private regulation in the form of deed restrictions or covenants lowers the riskiness or uncertainty associated with externalities or neighbourhood amenities. Since households with greater risk aversion are willing to pay more than less risk-averse households to avoid or reduce consumption risk, more risk-averse households outbid less risk-averse households for houses located inside gated subdivisions. On the other hand, if households do not care about the risk (i.e. the alternative view of private regulation is not valid) then gated communities will not attract the more risk-averse households in equilibrium.

What this suggests is a way to test the predictions that gated neighbourhoods are populated by households with greater risk aversion. Consider an uncertain or risky neighbourhood amenity that is outside the control of the gated community HOA. Under the alternative view of private regulation, this exogenous amenity uncertainty or risk reduces housing prices more in gated than in open neighbourhoods in the same proximity, the differential effect reflecting the differences in consumption risk aversion of the marginal gated community household.

We take advantage of the setting created by public school attendance zone changes in Orange County, Florida, to empirically test the hypothesis that households living in gated communities exhibit greater risk aversion than households in open neighbourhoods. Recent studies provide strong evidence that households recognise uncertain school quality, whether from individual school performance uncertainty or attendance zone instability (Mothorpe, 2018; Turnbull et al., 2018). Given this starting point, the test offered here is conceptually straightforward. Because it is costly to move frequently, school attendance zone instability creates greater school quality uncertainty for households living in neighbourhoods likely to be reassigned to different schools than for households in areas unlikely to be reassigned. Attendance zone changes are made by the school system bureaucracy, making the risk of attendance zone changes (and the associated change in school quality) exogenous to established gated and open neighbourhoods. While gated developments may be able to reduce neighbourhood amenity risk arising from surrounding structures or neighbours’ lifestyles, uncertainty over location-specific education quality arising from reassignment to a different school cannot be reduced by locating inside a gated subdivision. The empirical test examines the price effects of school attendance zone instability; if households in gated subdivisions exhibit greater risk aversion, then such instability will have stronger capitalisation effects for houses in gated developments than for houses outside gated neighbourhoods.⁴

Data and empirical models

In order to test the hypothesis we draw data from a comprehensive database of gated communities and Homeowners Associations in Orange County, Florida, and map them at the parcel level. The parcel-level data are then linked to the Orange County Property Appraiser (OCPA) records of properties and sales occurring from August 2001, through August 2012. With this data set we are able to determine if a parcel is subject to an HOA, lies within a gated community, as well as the age and size of the HOA or gated community.

We only consider arm’s length transactions for single family detached houses. Each house transaction is geocoded to a specific elementary school attendance zone and census tract. We clean the OCPA data to delete sales with missing values, selling prices of US$1000 or less (which indicate administrative transactions), quit claims, as well as consecutive transactions occurring within 6 months that exhibit unusual prices. Finally, we trim the sample by deleting the observations that lie outside 3 standard deviations in a robust regression on the base model without school-related variables. Census tracts and year combinations reporting fewer than 20 sales are also removed. The resultant complete pooled sample covers 129,504 separate transactions.

Figure 1 provides a plot of the sample in Orange County. The central business district is roughly in the centre of the county. The tourist area with the theme parks and other attractions occupies the south-western part of the county. Green indicates transactions in gated subdivisions with stable school attendance zones, blue indicates gated subdivisions with multiple attendance zone changes, and light blue indicates non-gated areas with multiple attendance zone changes. It is clear that gated subdivisions experiencing multiple attendance zone changes are clustered in the eastern, middle western and south central parts of the county, while other locations experiencing multiple zone changes are spread fairly evenly throughout the heavily established part of the county, as expected. The clustering of gated locales with unstable attendance zones motivates the matching sample analysis we pursue later. One such locale is Lake Butler area (detailed inset map). It is a census-designated place (CDP) and unincorporated area, completely surrounding the town of Windermere, in Orange County. According to the census information the population was 15,400 at the 2010 census, up from 7062 at the 2000 census. The inset clearly shows the intermixing of all four types of observations; gated and non-gated neighbourhoods are often literally across the street from each other and from their counterparts experiencing multiple attendance zone changes.

Figure 1.

Orange county housing data and geographic distribution of gated subdivisions.

The hedonic price for detached single family houses is a function of the vectors of physical characteristics of the house: H; school characteristics including instability of attendance assignments: S; fixed effects for geographic location and year of sale: F; and community characteristics: N. The community characteristics vector includes our variables of interest described below.

lnSalePrice = δ + α H + γ S + φ F + β N + v

(1)

where δ is the regression constant and v the stochastic error. The dependent variable is the natural log of house selling price (SalePrice).

The variable Instability is a dummy variable indicating houses that have multiple elementary school attendance zone changes over the entire sample period. These houses experience greater school quality uncertainty, given their experience of being shifted from one school zone to another multiple times during the sample period. This variable proxies for the presence of greater location-specific amenity risk. This proxy assumes rational households evaluate neighbourhoods in terms of the likelihood they will be subject to attendance zone changes sometime during their children’s schooling years. Turnbull et al. (2018) find that this particular source of risk reduces house values in the broad market; the question here is whether it yields different price effects inside and outside gated developments.

Another key variable is Gated, a dummy variable indicating houses located inside a gated community. We expect the coefficient dummy variable for houses inside a gated community to be positive as long as the perceived benefits to residents exceed their out-of-pocket fees.

The interactive variable Instability*Gated is the main variable of interest for testing the alternative view of private regulation. When included in the price equation with Instability and Gated, the coefficient on this variable picks up differences in the school quality risk effect on prices inside gated neighbourhoods when compared with surrounding non-gated neighbourhoods. A negative coefficient is consistent with more risk-averse households sorting into gated neighbourhoods, as predicted by the alternative view of private regulation; a non-negative coefficient rejects this sorting pattern and the underlying theory of private regulation.

The other variables are standard to hedonic price functions. House characteristics include number of bedroom (Bedrooms), number of bathrooms (Bathrooms), house age (Age), total area in 1000 ft² (Total Area), and a set of dummy variables that control for building quality (Quality Index) provided by the OCPA. We also include a dummy variable for ownership status, Investment, to indicate houses that are not owner-occupied. Owner-occupied houses are eligible for a valuable homestead exemption in Florida so we can rely on owner-occupiers to self-identify and we use this information to identify all other houses as investment properties, which include traditional rental houses, periodic or vacation rental houses and second homes.

Previous research finds houses with unusual attributes sell for less (Haurin, 1988; Jud et al., 1996). To capture the atypicality effect we use a model based on Turnbull et al. (2006). This approach measures the extent to which a given house is either larger or smaller than the average living area in the surrounding neighbourhood. Denote the set of all houses within a 0.25-mile radius of the subject house i by J, the standardised measure of relative house size is:

Localsiz e_{i} = \frac{Livingare a_{i} - \sum_{j \in J} Livingare a_{j} / N_{j}}{\sum_{j \in J} Livingare a_{j} / N_{j}}

(2)

where N_j is the number of existing surrounding houses in the neighbourhood J. In order to allow for asymmetric relative house size effects on selling price, we define the variables Larger_i and Smaller_i as the absolute values of the positive and negative values of Localsize_i respectively:

Large r_{i} = {\begin{matrix} 0, if Localsiz e_{i} \leq 0 \\ | Localsize |, if Localsiz e_{i} > 0 \end{matrix}

(3)

Smalle r_{i} = {\begin{matrix} 0, if Localsiz e_{i} \geq 0 \\ | Localsize |, if Localsiz e_{i} < 0 \end{matrix}

(4)

An essential element of the alternative view is that households take into account both the mean and riskiness of possible outcomes over the consumption period. We already discussed the riskiness aspect. Looking at the mean, higher quality public schools represent greater expected neighbourhood amenity, which unambiguously increases the household’s bid price P. The theory does not predict that households in gated communities will necessarily be willing to pay more or less than otherwise identical households in open communities for location attributes (unrelated to risk) that cannot be affected by the presence of gates. Nonetheless, differences may exist and so we include public school performance as a location-specific attribute that varies across locations and is exogenous to all neighbourhoods.

We include measures of expected public school quality in the model. We obtain the elementary school test grades from the Florida Education Department and school attendance zone maps for each year from the Orange County Public school System (OCPS). We introduce several measures of school quality used in the extensive school quality capitalisation literature.⁵ The output-based measure is based on annual school performance in the state standardised math test, the percentage of students who perform at level 3 or higher in math (a State defined benchmark). Florida did not fundamentally change the method for rating school performance over the sample period, but it did change the test in the last two years of the sample period. Most schools experienced large declines in the math test scores under the new test. Therefore, in order to control these effects as well as other subtle changes in the tests or testing procedures over time, we normalise each school’s math test performance measure for each year using the school math grade divided by the average grade for all the schools in the Orange County Public School District in that year. The school score (Test Score) used in the regressions is the average of the school’s normalised scores across the entire sample period; this construct is intended to control for the expected test score, following Turnbull et al. (2018).

We use two other school quality variables in addition to test scores: student peer enrolment and an input-based quality measure based on the student/teacher ratio. These variables are obtained from the National Center for Education Statistics. We include the percentage of students on subsidised lunch (Free Lunch) as school peer effect variable and student/teacher ratio (S/T Ratio) as an input-based quality measure.⁶ The variable Alt School Options measures the number of available magnet, charter and private schools in each census tract as alternative school options open to students. The Free Lunch and S/T Ratio variables are school quality measures that have often been empirically important variables in the literature. The measure of available alternative schools is included as an additional attribute of neighbourhood characteristics, although previous research has not found that the availability of alternatives to public schools always has a consistent effect on house prices.

We also include a dummy variable for each of the 318 census block groups in order to pick up the effects of demographic factors, location and amenities that differ between communities that are not included in the model.

Subject houses are mapped into census block groups to construct a neighbourhood household income variable needed for further analysis of neighbourhood subsamples. We use the census neighbourhood variable median household income to partition the full sample into income-level subsamples in some of the analysis. Observations through 2005 use 2000 Census data; observations after 2005 use 2010 Census data. In that analysis, higher income neighbourhoods are defined as census block groups with median household income greater than the average of block groups in the sample; lower income neighbourhoods are defined as block groups with median household income less than the block group average.

Before turning to the empirical results, it is useful at this point to consider several empirical issues that sometimes arise in empirical hedonic analyses: omitted variables bias and spatial dependence and correlation. For example, if being in a gated community is correlated with an unobserved characteristic such as attractive views that we would expect to impact home price positively, then the estimated coefficient on the effect of Gated will be biased upwards because we are capturing the effect of attractive views in that estimate. Similar to Greenstone and Gayer (2009), we choose a fixed effects approach to overcome omitted variables bias. Fixed effects analysis controls for omitted variables by including a large set of dummy variables for small groups of observations, in our case, observations that are within a small geographic area. These dummy variables then pick up the effects of any time-invariant unobserved variables on house prices, allowing for unbiased estimates of any remaining variables that vary either amongst houses within the chosen scope of the fixed effects or over time.

In addition, using spatial fixed effects and error-clustering at the census tract level, following Davis (2004) and Heintzelman (2010) among others, helps address spatial correlation issues. Fixed effects coupled with clustered errors is equivalent to a version of the spatial weighting matrix that allows for spatial dependence at the scale of the fixed effects; observations within the same block group are assumed to be spatially dependent, but independence is assumed for any observations across block groups. Similarly, we allow error terms to be correlated for observations within tracts but assume zero correlation of the error terms for observations in different tracts. Clustered errors also adjust standard error estimates for possible heteroscedasticity.

Some researchers address these issues using difference-in-difference, or repeat sales techniques. However, repeat sales suffers from small sample sizes, possible selection bias associated with properties that sell repeatedly, and unmeasured changes in quality for the same house when improvements are made between transactions. Therefore, this study includes a matching methodology which reduces issues with omitted variable bias, allows larger sample sizes than repeat sales and produces similar indices (McMillen, 2012). In addition, a common concern in the treatment effects literature is systematic selection of individuals into either the treatment or control group, and in this literature propensity score models provide a standard solution for bias caused by a complex non-linear process of selection on observables (Heckman et al., 1998). Propensity score matching is based on the idea that the bias is reduced when the comparison of mean impact is performed using treated and control units, which are similar on the observables that influence selection. Rosenbaum and Rubin (1983) define the propensity score as the conditional probability of receiving a treatment given a vector of pre-treatment characteristics. Thus, propensity score matching allows us to determine the mean impact of treatment on the treated within a group of ‘very similar’ units.

Traditionally, the propensity score is used to divide the sample into equally spaced intervals or bins, and within each bin a test is conducted for whether the average propensity score of treated and control units differ statistically. If it differs, the interval is split again until the condition is satisfied. We employ a Kernel Matching method where each treatment observation is matched with a weighted average of all control units where weights are calculated as inverse of the Euclidean distance between the propensity scores of the two units. The Kernal Matching method efficiently uses all information to form a control or benchmark for each treatment observation, which is important when some treatment observations have few control or comparison observations with similar propensity scores.

Empirical results

Table 1 reports the variables, their description and summary statistics for the entire sample and gated and open subsamples. Comparing the two subsamples, houses in gated communities are generally pricier, larger and less likely to be owner-occupied than other houses. Some of the other comparisons may be surprising. For example, while houses in gated communities tend to be in the attendance zones of schools with higher test scores, they also tend to be in areas with greater instability in attendance zone boundaries and have fewer alternative school options available.

Table 1.

Variable definitions and sample summary statistics.

Variable name	Definition	Full sample		Non-gated subsample		Gated subsample		Differencein means
		Mean	Std. dev.	Mean	Std. dev.	Mean	Std. dev.	T-test
SalePrice	Selling price of house	191,039.10	113,179.80	180,984.70	101,727.10	310,621.90	163,044.00	−78.41
Bedrooms	Number of bedrooms	3.29	0.73	3.24	0.70	3.91	0.68	−95.03
Bathrooms	Number of bathrooms	2.11	0.64	2.06	0.59	2.83	0.76	−99.52
Age	Age of the house	19.57	14.55	20.75	14.49	5.57	4.63	243.28
Total Area	Square feet of total area under the roof	2368.99	901.92	2279.13	827.26	3437.88	1053.75	−1.1e+02
Investor	Dummy variable for investor owned property	0.39	0.49	0.39	0.49	0.44	0.50	−8.67
HOA	Dummy variable for Home Owners Association	0.68	0.46	0.66	0.47	1.00	0.00	−2.5e+02
Smaller	Negative deviations from local mean living area	0.15	0.17	0.16	0.17	0.02	0.07	166.46
Larger	Positive deviations from local mean living area	0.17	0.31	0.14	0.28	0.52	0.47	−80.54
Test Score	Mean value of normalised elementary math scores	1.04	0.18	1.03	0.18	1.14	0.15	−72.77
Instability	Dummy variable for multiple elementary school attendance zone changes in a given period	0.04	0.20	0.04	0.20	0.09	0.28	−16.88
Free Lunch	Portion of students on free or subsidised lunch	0.54	0.21	0.55	0.21	0.43	0.22	52.28
S/T Ratio	(Student/teacher ratio)/10	15.30	1.71	15.28	1.71	15.60	1.67	−18.54
Alt School Options	Number of charter, magnet, and private schools in a given census tract	1.42	1.72	1.44	1.74	1.21	1.44	15.16
Observations		129, 504		119,460		10,044

All equations include the complete set of property characteristics in Table 1 along with 318 census block fixed effects and year fixed effects. Table 2 reports key parameter estimates for the full sample. Column (1) presents a base model without gated community effects. Columns (2) and (3) introduce the variable Gated and related interactive variables into the model. Adding the variables has no dramatic effects on the other base model variable coefficient estimates. Location within a gated community generates well over a 4% or US$12,400 premium on average in both (2) and (3). This additional value is over and above the almost 4% premium associated with being in an HOA community, since all gated communities are HOA communities as well. Clearly, gated communities give households something more than what non-gated HOA communities offer. Gated community age does not seem to affect value, which may indicate that these communities are doing a good job maintaining neighbourhood quality.

Table 2.

Hedonic price estimates for 2001–2012.

Variables	(1)	(2)	(3)
Dependent variable	Ln(SalePrice)	Ln(SalePrice)	Ln(SalePrice)
House characteristics
Bedrooms	0.0043	0.0043	0.0046
	(0.0026)	(0.0026)	(0.0026)
Bathrooms	0.0482***	0.0475***	0.0473***
	(0.0041)	(0.0040)	(0.0040)
Age	−0.0046***	−0.0045***	−0.0045***
	(0.0002)	(0.0003)	(0.0003)
Total Area	0.0003***	0.0003***	0.0003***
	(0.0000)	(0.0000)	(0.0000)
Investor	−0.0357***	−0.0364***	−0.0361***
	(0.0028)	(0.0027)	(0.0028)
HOA	0.0438***	0.0392***	0.0380***
	(0.0070)	(0.0082)	(0.0081)
Smaller	−0.2524***	−0.2538***	−0.2553***
	(0.0140)	(0.0141)	(0.0138)
Larger	−0.0153	−0.0188**	−0.0196**
	(0.0084)	(0.0083)	(0.0081)
Quality_Index1	−0.0646***	−0.0646***	−0.0667***
	(0.0197)	(0.0202)	(0.0199)
Quality_Index2	−0.0835***	−0.0831***	−0.0850***
	(0.0174)	(0.0182)	(0.0183)
Quality_Index3	−0.0475***	−0.0459***	−0.0468***
	(0.0149)	(0.0158)	(0.0160)
Quality_Index4	−0.0260	−0.0246	−0.0259
	(0.0152)	(0.0161)	(0.0162)
Quality_Index5	−0.0041	−0.0029	−0.0049
	(0.0152)	(0.0164)	(0.0164)
School quality measures
Test Score	0.0704	0.0693	0.0688
	(0.0522)	(0.0523)	(0.0518)
Instability	−0.0087	−0.0119	0.0088
	(0.0138)	(0.0162)	(0.0106)
Free Lunch	−0.1377***	−0.1399***	−0.1435***
	(0.0345)	(0.0334)	(0.0323)
S/T Ratio	−0.0066***	−0.0067***	−0.0058***
	(0.0020)	(0.0020)	(0.0019)
Alt School Options	−0.0053	−0.0050	−0.0061
	(0.0047)	(0.0043)	(0.0045)
Variables of interest
Gated		0.0466***	0.0432***
		(0.0158)	(0.0164)
Gated Community Age		0.0001	0.0001
		(0.0002)	(0.0002)
Test Score*Gated			0.1046
			(0.0550)
Instability*Gated			−0.0828**
			(0.0418)
Free Lunch*Gated			0.0913
			(0.0689)
S/T Ratio*Gated			−0.0099**
			(0.0042)
Alt School Options*Gated			0.0079
			(0.0085)
Constant	11.7326***	11.7363***	11.7325***
	(0.0898)	(0.0889)	(0.0870)
Observations	129,504	129,504	129,504
R-squared	0.7721	0.7724	0.7727

Notes: Clustered standard error estimates in parentheses. Dummy variables for 318 census block groups and year sold are not reported in this table. **Indicates significance at the 5% level; ***indicates significance at the 1% level.

The variable most relevant to testing the alternative view of private regulation, however, is the interactive instability variable Instability*Gated. The estimates show that attendance zone instability significantly reduces prices inside gated communities by about US$25,400 relative to otherwise identical houses in open neighbourhoods. The effect of school quality risk clearly exceeds that of the school quality whether measured as test scores or student/teacher ratios. In any event, households inside gated communities are affected more by this source of exogenous location-specific risk than households in the broader market, a pattern consistent with the alternative view.⁷

There is nothing surprising in the estimates for the other variables included in the model. Looking at (1), the Investment property coefficient indicates about a 3.5% discount for non-owner-occupied properties. Since we cannot ascertain how many of the investment properties are tenant occupied or seasonal rentals and how many are purely second homes, the discount is likely picking up the rental externality associated with the former. Belonging to an HOA tends to increase value whether in a gated or open community. This is consistent with earlier literature (Do and Grudnitski, 1997; Rogers, 2006). Looking at the school quality measures, Test Score is insignificant, a result consistent with earlier observations that test scores are not always viewed by parents as reliable information (Zahirovic-Herbert and Turnbull, 2009). In any case, the peer effects and input-based quality measures Free Lunch and S/T Ratio have strong stable negative effects on prices, as expected. The interactive school quality term in column (3) indicates that gated households are more sensitive to this school quality measure than households in open neighbourhoods. Nonetheless, a one standard deviation increase in the student/teacher ratio lowers the price by about US$5100 for the mean valued house.

Neighbourhood income effects

Table 3 reports key estimates for higher and lower income neighbourhood subsamples, where higher (lower) income neighbourhoods are census blocks with median household income above (below) the county median income. The table reports results for the same three models in Table 2. Columns (4)–(6) and (7)–(9) clearly illustrate differences across these market segments that hold for all of the models. Test scores matter for the lower income subsample but do not for the higher income subsample. Free Lunch is significantly negative in all cases, but the indicated value discount is almost three times greater for higher income neighbourhoods than lower income neighbourhoods. Once again, the availability of alternative schools does not affect property values in this market in either segment.

Table 3.

Key parameter estimates by neighbourhood income.

Variables	Below median income sample			Above median income sample
Dependent variable	Ln(SalePrice)	Ln(SalePrice)	Ln(SalePrice)	Ln(SalePrice)	Ln(SalePrice)	Ln(SalePrice)
School quality measures
Test Score	0.1320**	0.1264**	0.1269**	0.0051	0.0168	0.0059
	(0.0545)	(0.0550)	(0.0555)	(0.1090)	(0.1063)	(0.0986)
Instability	−0.0066	0.0008	0.0023	−0.0128	−0.0226	0.0255
	(0.0124)	(0.0133)	(0.0134)	(0.0259)	(0.0279)	(0.0218)
Free Lunch	−0.0648***	−0.0663***	−0.0682***	−0.1880***	−0.1858***	−0.1958***
	(0.0250)	(0.0247)	(0.0247)	(0.0659)	(0.0626)	(0.0568)
S/T Ratio	−0.0038	−0.0035	−0.0030	−0.0080***	−0.0084***	−0.0076**
	(0.0022)	(0.0022)	(0.0023)	(0.0030)	(0.0031)	(0.0030)
Alt School Options	−0.0039	−0.0041	−0.0048	−0.0055	−0.0047	−0.0055
	(0.0034)	(0.0034)	(0.0034)	(0.0085)	(0.0075)	(0.0078)
Variables of interest
Gated		0.0497**	0.0700		0.0449***	0.0372**
		(0.0252)	(0.0356)		(0.0159)	(0.0147)
Gated Community Age		−0.0001	−0.0001		0.0006	0.0007
		(0.0002)	(0.0002)		(0.0004)	(0.0004)
Test Score*Gated			0.0771			0.0737
			(0.1124)			(0.0548)
Instability*Gated			−0.1039**			−0.1125**
			(0.0473)			(0.0462)
Free Lunch*Gated			0.0627			0.0917
			(0.1214)			(0.0699)
S/T Ratio*Gated			−0.0095			−0.0063
			(0.0073)			(0.0041)
Alt School Options*Gated			0.0161			0.0008
			(0.0135)			(0.0110)
Constant	11.5176***	11.5218***	11.5186***	11.9862***	11.9821***	11.9869***
	(0.0813)	(0.0811)	(0.0818)	(0.1634)	(0.1583)	(0.1469)
Observations	75,461	75,461	75,461	54,043	54,043	54,043
R-squared	0.7322	0.7325	0.7328	0.7661	0.7666	0.7671

Notes: Key parameter estimates from models (1)–(3) in Table 2 estimated on the indicated subsamples. Clustered standard error estimates in parentheses. **Indicates significance at the 5% level; ***indicates significance at the 1% level.

Turning to the gated community effects, in all cases houses in gated developments sell at a premium, with a greater premium in the lower income complete model (6) than the upper income model (9). The attendance zone instability stands out as the only significant interactive effect. In this regard, the income-level subsamples resemble the full sample in that this source of uncertainty regarding neighbourhood amenities significantly lowers property values inside gated communities while exhibiting negligible effects outside.

Property ownership effects

Table 4 reports the results stratified by ownership status. Looking first at the school quality effects, the pattern across owner-occupied and investment properties is similar, except for test scores. While test scores are not significant in the owner-occupied subsample, they are significantly positive in the investment subsample. The owner-occupied subsample results are not surprising in light of the existing body of evidence (Nguyen-Hoang and Yinger, 2011), but when coupled with the investment property outcome, the latter pattern is somewhat puzzling.

Table 4.

Key parameter estimates by ownership status.

Variables	Homeowner sample			Investor sample
Dependent variable	Ln(SalePrice)	Ln(SalePrice)	Ln(SalePrice)	Ln(SalePrice)	Ln(SalePrice)	Ln(SalePrice)
School quality measures
Test Score	0.0576	0.0602	0.0586	0.1044	0.0982	0.0998
	(0.0519)	(0.0523)	(0.0515)	(0.0585)	(0.0586)	(0.0586)
Instability	−0.0155	−0.0186	0.0041	0.0040	0.0004	0.0201
	(0.0168)	(0.0197)	(0.0103)	(0.0137)	(0.0149)	(0.0137)
Free Lunch	−0.1383***	−0.1401***	−0.1413***	−0.1193***	−0.1220***	−0.1263***
	(0.0348)	(0.0336)	(0.0320)	(0.0356)	(0.0347)	(0.0339)
S/T Ratio	−0.0062***	−0.0063***	−0.0057***	−0.0059**	−0.0059**	−0.0049**
	(0.0018)	(0.0018)	(0.0017)	(0.0024)	(0.0024)	(0.0023)
Alt School Options	−0.0012	−0.0008	−0.0015	−0.0082	−0.0078	−0.0094
	(0.0045)	(0.0041)	(0.0041)	(0.0054)	(0.0050)	(0.0052)
Variables of interest
Gated		0.0444***	0.0450***		0.0524***	0.0467**
		(0.0155)	(0.0161)		(0.0184)	(0.0190)
Gated Community Age		0.0002	0.0002		−0.0001	−0.0001
		(0.0002)	(0.0002)		(0.0002)	(0.0002)
Test Score*Gated			0.0798*			0.1406**
			(0.0466)			(0.0700)
Instability*Gated			−0.0933**			−0.0795
			(0.0420)			(0.0445)
Free Lunch*Gated			0.0396			0.1290
			(0.0648)			(0.0884)
S/T Ratio*Gated			−0.0067			−0.0138**
			(0.0038)			(0.0054)
Alt School Options*Gated			0.0044			0.0119
			(0.0077)			(0.0107)
Constant	11.7454***	11.7458***	11.7433***	11.6668***	11.6739***	11.6679***
	(0.0892)	(0.0882)	(0.0861)	(0.0992)	(0.0987)	(0.0964)
Observations	69,388	69,388	69,388	60,116	60,116	60,116
R-squared	0.7665	0.7669	0.7672	0.7826	0.7830	0.7833

In any case, the gated premium exceeds 4% in all models and subsamples. The differential gated community effect of the test score measure of school quality is positive in both subsamples, although significantly stronger in the investment property subsample. As before, the interactive attendance zone instability variable has a significant negative effect on house prices inside gated communities while having no stable effect outside. All of the full sample conclusions extend to these subsamples.

Market cycle effects

Capitalisation questions are closely tied to price discovery in the market, so it is natural to re-examine these price effects over different market phases when price discovery may be functioning with different degrees of efficiency. The FHFA Purchase Only house price index indicates that the Orlando-Kissimmee-Sanford MSA housing market peaked in the second quarter of 2007; the market bottomed out within two years of the peak, but it had not yet begun to recover significantly by the end of the sample period in 2012. Therefore, we use June 2007 to partition the sample period into rising and declining market subsamples.

Table 5 reports the key estimates for the complete model over the different market phases and across income-level subsamples for each. Not surprisingly, the school quality capitalisation effects estimates differ across market phases. This is qualitatively consistent with what Zabel (2015) finds for the Boston area. Looking at the variables of central interest, although some differences exist across market phases, what is interesting is the degree to which the results remain stable across subsamples. The gated community premium is not significantly different across market phases for the full sample and the higher income neighbourhood sample. This is not the case for the lower income subsample. Location inside a gated community leads to a well over 6% price premium in the rising market and no premium in the declining market for the lower income subsample. The evidence for the pooled sample and higher and lower income subsamples clearly shows that being located in a gated community offers no protection against wider market price depreciation in declining markets when compared with the depreciation experienced by houses outside gated developments. Further, prices suffered greater declines than the wider market during the falling market phase in lower income gated communities. Finally, the instability interaction term remains significantly negative (and numerically similar) for all cases except the lower income neighbourhood subsample in the declining market.

Table 5.

Key parameter estimates by housing market phase and neighbourhood income.

Variables	Declining housing market			Rising housing market
		Below median income	Above median income		Below median income	Above median income
Dependent variable	Ln(SalePrice)	Ln(SalePrice)	Ln(SalePrice)	Ln(SalePrice)	Ln(SalePrice)	Ln(SalePrice)
School quality measures
Test Score	−0.0059	0.0534	−0.0171	0.1683***	0.1321***	0.2282**
	(0.1049)	(0.1297)	(0.1379)	(0.0463)	(0.0450)	(0.1083)
Instability	0.0326	0.0067	0.0641*	−0.0093	−0.0065	−0.0089
	(0.0167)	(0.0135)	(0.0354)	(0.0118)	(0.0139)	(0.0189)
Free Lunch	−0.2483***	−0.2499***	−0.2124**	−0.0191	−0.0232	−0.0037
	(0.0708)	(0.0889)	(0.0931)	(0.0220)	(0.0198)	(0.0594)
S/T Ratio	−0.0028	−0.0016	−0.0029	−0.0017	−0.0009	−0.0048
	(0.0031)	(0.0039)	(0.0045)	(0.0015)	(0.0020)	(0.0030)
Alt School Options	−0.0018	−0.0011	−0.0027	−0.0004	0.0038	−0.0105
	(0.0063)	(0.0057)	(0.0102)	(0.0043)	(0.0032)	(0.0063)
Variables of interest
Gated	0.0368**	0.0908	0.0330**	0.0454***	0.0657**	0.0410**
	(0.0176)	(0.0568)	(0.0163)	(0.0156)	(0.0270)	(0.0160)
Gated CommunityAge	0.0003	0.0001	0.0007	−0.0000	−0.0002	0.0005
	(0.0003)	(0.0003)	(0.0005)	(0.0002)	(0.0002)	(0.0004)
Test Score*Gated	0.0575	−0.0434	0.0932	0.0338	0.0036	0.0230
	(0.0708)	(0.1301)	(0.0779)	(0.0461)	(0.1017)	(0.0427)
Instability*Gated	−0.1177**	−0.0258	−0.1356***	−0.0956***	−0.0727	−0.1011**
	(0.0479)	(0.0567)	(0.0471)	(0.0328)	(0.0396)	(0.0444)
Free Lunch*Gated	0.1192	−0.1357	0.1609**	−0.0028	0.0547	0.0127
	(0.0835)	(0.1145)	(0.0633)	(0.0578)	(0.1022)	(0.0831)
S/T Ratio*Gate	−0.0072	0.0045	−0.0102	−0.0022	−0.0044	−0.0006
	(0.0060)	(0.0099)	(0.0066)	(0.0036)	(0.0062)	(0.0032)
Alt School Options*Gated	0.0064	0.0437	−0.0050	0.0016	0.0047	−0.0002
	(0.0104)	(0.0237)	(0.0113)	(0.0082)	(0.0108)	(0.0115)
Constant	12.2778***	12.0436***	12.4483***	11.4896***	11.4735***	10.3900***
	(0.1458)	(0.2028)	(0.2041)	(0.0675)	(0.0683)	(0.1553)
Observations	43,430	18,981	24,449	86,074	56,480	29,594
R-squared	0.7696	0.6992	0.7614	0.7767	0.7304	0.7416

Notes: Key parameter estimates from model (3) in Table 2 estimated on the indicated subsamples. Clustered standard error estimates in parentheses. **Indicates significance at the 5% level; ***indicates significance at the 1% level.

Matched sample results

Recall that propensity score matching offers an empirical method to control for selection effects in the models. As a final robustness test, we re-estimate all of the models using the matched sample approach explained in the data and empirical modelling section. Table 6 reports matched sample results. For brevity, the table reports key parameter estimates only for the full model versions for each of the sample partitions examined earlier. Comparing these estimates with the estimates reported in previous tables, it is clear that nothing changes dramatically in the matched sample analysis. And one result that continues to stand out is the significantly negative coefficient on the interactive Instability*Gated variable – indicating a stronger price response to this source of uncertainty inside gated communities than observed outside gated communities.⁸ This key finding appears robust. It appears that households who are more consumption risk averse find gated communities more attractive than do households who are not as risk averse.

Table 6.

Matched sample key parameter estimates.

Variables	Full sample	Bellow median income	Above median income	Owner	Investor	Declining market	Rising market
Dependent variable	Ln(SalePrice)	Ln(SalePrice)	Ln(SalePrice)	Ln(SalePrice)	Ln(SalePrice)	Ln(SalePrice)	Ln(SalePrice)
School qualitymeasures
Test Score	0.0053	0.0296	−0.0072	0.0692	0.0196	0.1093	0.2264***
	(0.0883)	(0.0781)	(0.1080)	(0.0789)	(0.1294)	(0.1448)	(0.0726)
Instability	−0.0003	−0.0067	0.0113	0.0116	0.0007	0.0154	−0.0094
	(0.0104)	(0.0117)	(0.0126)	(0.0159)	(0.0149)	(0.0163)	(0.0151)
Free Lunch	−0.1751***	−0.1517***	−0.1923***	−0.0731**	−0.1764**	−0.1297	−0.0053
	(0.0530)	(0.0488)	(0.0612)	(0.0341)	(0.0708)	(0.0919)	(0.0377)
S/T Ratio	−0.0090***	−0.0073***	−0.0099***	−0.0070**	−0.0084**	−0.0018	−0.0036
	(0.0025)	(0.0022)	(0.0031)	(0.0030)	(0.0034)	(0.0042)	(0.0019)
Alt School Options	−0.0090	−0.0039	−0.0145**	−0.0043	−0.0094	−0.0046	−0.0051
	(0.0056)	(0.0052)	(0.0067)	(0.0050)	(0.0078)	(0.0075)	(0.0054)
Variables of interest
Gated	0.0374***	0.0461***	0.0310	0.0211	0.0480***	0.0442***	0.0347**
	(0.0136)	(0.0136)	(0.0162)	(0.0189)	(0.0145)	(0.0136)	(0.0143)
Gated CommunityAge	0.0007	0.0008**	0.0005	0.0004	0.0016**	0.0009	0.0006
	(0.0004)	(0.0004)	(0.0004)	(0.0004)	(0.0007)	(0.0005)	(0.0003)
Test Score*Gated	0.1233**	0.0881**	0.1703***	0.1232	0.1205**	0.1237	0.0488
	(0.0508)	(0.0438)	(0.0641)	(0.1176)	(0.0491)	(0.0733)	(0.0514)
Instability*Gated	−0.0847**	−0.0891**	−0.0858**	−0.1218***	−0.0937**	−0.1018**	−0.1016***
	(0.0417)	(0.0426)	(0.0429)	(0.0421)	(0.0421)	(0.0441)	(0.0333)
Free Lunch*Gated	0.0516	0.0195	0.0782	0.0111	0.0779	0.0596	−0.0056
	(0.0677)	(0.0665)	(0.0840)	(0.1283)	(0.0563)	(0.0919)	(0.0531)
S/T Ratio*Gated	−0.0097**	−0.0067	−0.0134**	−0.0079	−0.0096**	−0.0105	−0.0025
	(0.0040)	(0.0036)	(0.0052)	(0.0075)	(0.0037)	(0.0066)	(0.0039)
Alt School Options*Gated	0.0086	0.0067	0.0120	0.0183	0.0042	0.0035	0.0055
	(0.0081)	(0.0076)	(0.0101)	(0.0129)	(0.0105)	(0.0093)	(0.0080)
Constant	11.5335***	11.9439***	11.4576***	11.2080***	12.0021***	11.4089***	11.5423***
	(0.1783)	(0.1279)	(0.1703)	(0.1308)	(0.1978)	(0.2092)	(0.1159)
Observations	81,830	46,733	35,097	40,622	41,208	29,962	51,868
R-squared	0.7826	0.7790	0.7962	0.7511	0.7787	0.7806	0.7797

Notes: Key parameter estimates from model (3) in Table 2 estimated on the matched subsamples. Clustered standard error estimates in parentheses. **Indicates significance at the 5% level; ***indicates significance at the 1% level.

Conclusion

Deed restrictions and neighbourhood covenants are private market agreements to restrict property rights. The traditional view considers such private regulation as a substitute for government regulation. The alternative view offered by Hughes and Turnbull (1996a, 1996b) argues that grounding private regulation in property law makes private regulation different from government regulation. In particular, private regulation can have value to property owners over public regulation to the extent that restricting development, property maintenance and lifestyles places constraints on the behaviour of unknown future owners, reducing uncertainty over how the neighbourhood character will change over time. As explained here, one consequence is that households with stronger consumption risk aversion find deed-restricted neighbourhoods more valuable than do households with weaker risk aversion. This paper is the first to empirically test this prediction.

We use data drawn from single family house transactions in Orange County, Florida, to examine the house price effects of school quality uncertainty introduced by unstable school attendance zones. This source of uncertainty is exogenous to households inside and outside gated neighbourhoods and therefore provides a setting in which we can test the risk-aversion sorting hypothesis as it applies to gated developments. The empirical results clearly show that attendance zone instability reduces housing prices more in gated subdivisions than in other areas. The result is robust across neighbourhood income levels, owner-occupied or investor-owned properties, and rising or falling markets. The observed price effect is consistent with households sorting such that those with stronger consumption risk aversion are drawn to gated developments and those with weaker risk aversion, to open developments, as predicted by the alternative view of private regulation.

Supplemental Material

USJ872391_supplemental_material – Supplemental material for Private government, propertyrights and uncertain neighbourhood externalities: Evidence from gated communities

Supplemental material, USJ872391_supplemental_material for Private government, propertyrights and uncertain neighbourhood externalities: Evidence from gated communities by Geoffrey K Turnbull and Velma Zahirovic-Herbert in Urban Studies

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Velma Zahirovic-Herbert

Notes

References

Davis

(2004) The effect of health risk on house values: Evidence from a cancer cluster. American Economic Review 94(5): 1693–1704.

Grudnitski

(1997) The impact on house value of restrictions on rights of ownership: The case of an occupant’s age. Real Estate Economics 25(4): 683–693.

Fischel

(1990) Do Growth Controls Matter? A Review of the Empirical Evidence on the Effectiveness and Efficiency of Local Government Land Use Regulation. Cambridge, MA: Lincoln Institute of Land Policy.

Greenstone

Gayer

(2009) Quasi-experiments and experimental approaches to environmental economics. Journal of Environmental Economics and Management 57(1): 21–44.

Haurin

(1988) The duration of marketing time of residential housing. AREUEA Journal 16(4): 396–410.

Heckman

Ichimura

Todd

(1998) Matching as an econometric evaluation estimator. Review of Economic Studies 64: 261–294.

Heintzelman

(2010) Measuring the property-value effects of local land use and preservation referenda. Land Economics 81: 22–47.

Helsley

Strange

(1999) Gated communities and the economic geography of crime. Journal of Urban Economics 46(1): 80–105.

Hughes

Turnbull

(1996a) Restrictive land covenants. Journal of Real Estate Finance and Economics 12(1): 9–22.

10.

Hughes

Turnbull

(1996b) Uncertain neighborhood effects and restrictive covenants. Journal of Urban Economics 39(2): 160–172.

11.

Jud

Seaks

Winkler

(1996) Time on the market: The impact of residential brokerage. Journal of Real Estate Research 12(3): 447–458.

12.

Lacour-Little

Malpezzi

(2009) Gated streets and house prices. Journal of Housing Research 18(1): 19–43.

13.

Landman

Schontech

(2002) Urban fortresses: Gated communities as a reaction to crime. African Security Review 11(4): 71–85.

14.

Le Goix

(2005) Gated communities: Sprawl and social segregation in Southern California. Housing Studies 20(2): 323–343.

15.

Le Goix

Vesselinov

(2013) Gated communities and house prices: Suburban changes in Southern California, 1980–2008. International Journal of Urban and Regional Research 37(6): 2129–2151.

16.

Liao

Wehrhahn

Breitung

(2019) Urban planners and the production of gated communities in China: A structure-agency approach. Urban Studies 56(13) 2635–2653.

17.

Low

(2001) The edge and the center: Gated communities and the discourse of urban fear. American Anthropologist 103(1): 45–58.

18.

McMillen

(2012) Repeat sales as a matching estimator. Real Estate Economics 40(4): 745–772.

19.

Manzi

Smith-Bowers

(2005) Gated communities as club goods: Segregation or social cohesion? Housing Studies 20(2): 345–359.

20.

Mothorpe

(2018) The impact of uncertainty on school quality capitalization using the border method. Regional Science and Urban Economics 70: 127–141.

21.

Nguyen-Hoang

Yinger

(2011) The capitalization of school quality into house values: A review. Journal of Housing Economics 20(1): 30–48.

22.

Rogers

(2006) A market for institutions: Assessing the impact of restrictive covenants in housing. Land Economics 82(4): 500–512.

23.

Rosenbaum

Rubin

(1983) The central role of the propensity score in observational studies for causal effects. Biometrika 70(1): 41–55.

24.

Turnbull

(1994) Location and housing demand with endogenous consumption risk. Regional Science and Urban Economics 24(5): 543–563.

25.

Turnbull

Dombrow

Sirmans

(2006) Big house, little house: Relative size and value. Real Estate Economics 34(3): 439–456.

26.

Turnbull

Zahirovic-Herbert

Zheng

(2018) Uncertain school quality and house price: Theory and empirical evidence. Journal of Real Estate Finance and Economics 57(2): 167–191.

27.

Wilson-Doenges

(2000) An exploration of sense of community and fear of crime in gated communities. Environment and Behavior 32(5): 597–611.

28.

Zabel

(2015) The hedonic model and the housing cycle. Regional Science and Urban Economics 54: 74–86.

29.

Zahirovic-Herbert

Turnbull

(2009) Public school reform, expectations, and capitalization: What signals quality to homebuyers? Southern Economic Journal 75(4): 1094–1113.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.35 MB