Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

The Resilience of the Premium for Homes in New Urbanist Neighborhoods

The Resilience of the Premium for Homes in New Urbanist Neighborhoods In this study, I analyze longitudinal differences in single-family home prices in two new urbanist neighborhoods versus surrounding conventional neighborhoods. Using data on 78,513 single-family home sales transactions in Montgomery County, Maryland, I adopt a novel multilevel methodology to assess the effects of neighborhood demographic, economic, and locational characteristics as well as the effects of year on home prices. Results support empirical evidence on the premium that buyers willingly pay for homes in new urbanist neighborhoods. Year effects indicate that the premium withstood the Great Recession quite well. New urbanism, also known as, traditional neighborhood development (TND), is a community development concept that advocates harmonizing demand for scarce development resources like land with a mix of uses—commercial and residential—associated with traditional neighborhoods (Fulton, 1996). By mixing, rather than separating, uses in pedestrian-oriented and transit-oriented spaces, TND adds value in the market according to its proponents (Congress for the New Urbanism, 2014). The Congress for the New Urbanism also suggests that homes in TND neighborhoods command a premium and retain value in the market because of walkable access to amenities (Congress for the New Urbanism, 2014). Indeed, the street networks in TND neighborhoods tend to be grid-like rather than curvilinear as in conventional neighborhoods, while the orientation of homes toward the street facilitates pedestrian trips to recreational or to shopping destinations (Plaut and Boarnet, 2003). Unfortunately, the literature on TNDs provides little empirical evidence on the willingness of buyers to pay such premiums over the long term. In this regard, the literature is limited in two fundamental ways: in temporal scale and in spatial scale. Taken together, these limitations make it difficult to ascertain the resiliency of these premiums. The temporal limitation relates to the fact that, at most, data for nine years of single-family home sales transactions are used to estimate TND premiums. While this is a reasonable time horizon to account for any randomness, more research with data from as long or longer periods of time is needed to make a definitive statement on the resiliency of TND premiums. The spatial limitation relates to the fact that the number of comparable neighborhoods used to estimate premiums is very small. Plaut and Boarnet (2003) used three neighborhoods (one TND and two conventional neighborhoods), Tu and Eppli (1999) used nine census J O SRE Vol . 8 No. 1 – 2016 146 Z o lni k tracts (one TND and eight conventional neighborhoods), and Tu and Eppli (2001) used three markets (three TNDs and surrounding conventional developments) to estimate TND premiums. A small number of comparable neighborhoods is limiting for the following reasons. While it is important that enough transactions are analyzed to account for randomness, it is the number of comparable neighborhoods that is the priority in the statistical estimation of neighborhood design effects, such as those exemplified in TNDs (Follain and Malpezzi, 1980). If there are not enough comparable neighborhoods, then it will be difficult to attribute any discernible differences in sales prices to the design features unique to TNDs. And, from a statistical perspective, a sample size of nine at the neighborhood level of analysis makes it a challenge to precisely estimate such differences. In this study, I attempt to address these temporal and spatial limitations in order to answer the following research questions. First, are TND premiums resilient over the long term? The long term in the study is ten years, which is enough to account for randomness and just slightly longer than the longest time horizon in the literature of nine years. The fact that the analysis covers a ten-year time period is important, but it is also important to acknowledge the importance of what occurred during those ten years. In fact, the time period from 2000 to 2009 coincides temporally with the Great Recession from December of 2007 to June of 2009; a time of marked volatility in regional real estate markets across the United States (Shiller, 2015). Second, what characteristics of single-family homes in TNDs have the largest impact on these premiums over this ten-year period of transactions? Third, and finally, how did these premiums change, if at all? The organization of the study is as follows. In the next sections, I briefly review the literature on the valuation of homes in TNDs as well as the opinions of developers. In the methodology section, I present the multilevel models in the study—a two-level model with a property level nested within a ZIP Code level and a three-level model with a property level nested within a year level nested within a ZIP Code level. In the data section, I name the sources of the property- level, the year-level, and the ZIP Code-level data and describe the dependent variables as well as the independent variables in the models. In the results section, I review the relevant results from the two- and the three-level models. In the discussion section, I relate the results to the research questions in the study and link the results to the literature on the valuation of homes in TNDs relative to homes in conventional neighborhoods. In the conclusion section, I highlight the contribution of the study and suggest one fruitful avenue for future research. Premium for Homes in New Urbanist Neighborhoods It is known that homes in TND neighborhoods can be more expensive that those in conventionally-developed neighborhoods (Song and Knapp, 2003). In addition, empirical evidence suggests that buyers willingly pay a premium for homes in such neighborhoods (Tu and Eppli, 1999, 2001; Plaut and Boarnet, 2003). T h e R esi lienc e o f t he P r e m i u m f o r H o m e s 147 However, it is also important for developers to know that those higher values are resilient; that is, the higher values relative to properties outside of TNDs persist over time. If so, then the development community can demonstrate to financers that not only do TND communities make more efficient use of development resources, they tend to retain their higher values (Gyourko and Rybczynski, 2000; Garde; 2006; Planning Design Group, 2007). If the focus of the research is on walkability, then the empirical evidence seems conclusive. Song and Querica (2008) found that greater pedestrian accessibility to commercial land uses garner a premium. Rauterkus and Miller (2011) found that the effect of walkability on land values is positive. Dong (2015) found that the effect of walkability, not other new urbanist neighborhood design features, on single-family home appreciation rates during the Great Recession is positive. However, if the focus of the research is on other neighborhood design features common to TNDs, such as development density, street connectivity, and land use mix, then the empirical evidence is inconclusive (Lang, 2005). Song and Querica (2008) also found that street connectivity garners a premium. However, they found that, consistent with older empirical evidence (Crecine, Davis, and Jackson, 1967; Reuter, 1973), land use mix garners no premium even though newer empirical evidence demonstrates both positive effects (Song and Knaap, 2003) and negative effects (Mahan, Polasky, and Adams, 2000). Further, low-development density, not high-development density, garners a premium. In low-income neighborhoods, where little research on the effects of different neighborhood design features is evident, assessments for infill-style developments are higher relative to TND-style and enclave-style developments (Ryan and Weber, 2007). Guttery (2002) found that homes with rear-entries facing alleyways, another design feature synonymous with TND-style development, sell for a $5,575 (5.3%) discount relative to homes with front-entry driveways. Designing and Developing TNDs Just as the valuation of TND homes is not conclusive, nor are the opinions of designers and developers unanimous. Surveys and interviews of developers, financiers, and investors from across the U.S. (n  23) who are familiar with TNDs or who have experience with TNDs (Gyourko and Rybczynski, 2000) suggest that financing TNDs is risky and costly, although neither financiers nor investors perceive costs to be prohibitive. The perception of risk is mostly attributable to the multiple land uses in TNDs. Surveys of professionals (n  169) from across the U.S. with experience in the design and the development of TNDs (Garde, 2006) suggest that designers, developers, and planners agree on the advantages of TNDs with regard to design, growth management, environmental preservation, and Not In My Back Yard (NIMBY) opposition. The same professionals also agree on the disadvantages of TNDs with regard to the restrictions of land use regulations, the resistance of developers, and the costs of construction. Indeed, 56% of developers agree that TNDs have higher construction costs and 63% of developers at least somewhat agree that TNDs are not sound investments. Interviews with the same professionals (n  11) suggest general J O SRE Vol . 8 No. 1 – 2016 148 Z o lni k skepticism with regard to TNDs (Garde, 2006). Developers are uncertain of the demand for TNDs relative to conventional suburban developments. Developers as well as lenders are also wary of the risks most associate with a relatively new product. The consensus of focus groups and surveys of professionals with experience in the development of TNDs in Florida is that ‘‘New Urbanism remains a complex, frequently misunderstood, and often challenging form of development that comprises only a fraction of the overall development landscape’’ (Planning Design Group, 2007, p. 2). And, ‘‘with the economy and housing markets slowing down and credit significantly tightening, it remains to be seen whether the market for these more expensive homes will hold or decline’’ (Planning Design Group, 2007, p. 9). The latter question coincides with the general trend towards greater methodological sophistication in the real estate valuation literature (Krause and Bitter, 2012) to more accurately estimate future changes given past volatility. To address the question of the resiliency of the premiums for homes in TND neighborhoods, I analyze differences in home prices between two contiguous TNDs—the Kentlands and the Lakelands—and surrounding conventional neighborhoods in Montgomery County, Maryland over a ten-year time period from 2000 to 2009. The Kentlands and the Lakelands were both designed by Duany Plater-Zyberk and Company in 1988 and in 1996, respectively. All told, the more than 3,000 residential units and the approximately 600,000 square feet of retail and commercial space in the Kentlands and in the Lakelands occupy approximately 695 acres in Gaithersburg, Maryland (Steuteville, 2010). Estimates of the mean premium buyers are willing to pay in the Kentlands are 13% ($24,603) (Eppli and Tu, 1999) and 12% (about $25,000) (Tu and Eppli, 1999). Estimates of the premium buyers are willing to pay in other TNDs are as follows. Eppli and Tu (1999) estimate that buyers pay a mean premium of: 25% ($30,690) in the TND of Harbor Town in Memphis, Tennessee; 4% ($5,157) in the TND of Laguna West in Elk Grove, California; and 9% ($16,334) in the TND of Southern Village in Memphis, Tennessee. Plaut and Boarnet (2003) estimate that buyers pay a mean premium of $8,229 in the TND of Central Carmel versus mean premiums of $6,690 and $6,008 in the conventional neighborhoods of Carmel and Denia, respectively, in Haifa, Israel. Methodology Jones and Bullen explain (1993) and illustrate (1994) the technical advantages of a multilevel approach to model home prices rather than assign dummy variables for different times or for different places in a multiple regression model of home prices. First, a multilevel approach explicitly accounts for autocorrelation, or nonindependence; that is, the sale prices of two homes from the same Zip Code are more alike than the sales prices of two homes drawn randomly. Second, a multilevel approach accurately estimates the effects of independent variables at higher levels of analysis such as median home value. Third, a multilevel approach is used to pool information from all of the Zip Codes in the study area to precisely estimate the mean home-price relation as well as the variation in the mean home- price relation both in ZIP Codes where home sales are many and in Zip Codes where home sales are few. T h e R esi lienc e o f t he P r e m i u m f o r H o m e s 149 The multilevel models in the study include a two-level model with a property ( p) level nested within a ZIP Code (z) level and a three-level model with a property ( p) level nested within a year ( y) level nested within a ZIP Code (z)level (Raudenbush and Bryk, 2002). Precedence for nesting time (year) within place (Zip Code) in the three-level model is from Jones and Bullen (1993, pp. 1412– 41): ‘‘[t]he purchase price of properties (level 1) are recorded for time periods (at level 2) for areas (at level 3). In such a three-level model, there is a separate level- 2 unit for each time period in each district. Such a formulation allows for complex space-time modelling, with potentially a separate trend in house-price inflation for each area.’’ The specifications of the two- and three-level models are in the Appendix. Data The characteristics that affect home prices fall into four groups: (1) accessibility characteristics, such as distance to popular destinations; (2) environmental characteristics; (3) physical characteristics, such as age; and (4) public sector characteristics, such as taxes and services (Brigham, 1965; Stull, 1975). Indeed, each home constitutes a bundle of the above characteristics. But, while the characteristics that affect home prices are known, home prices are not readily forecastable statistically even though one-year forecasts are more precise than ten- year forecasts (Shiller, 2015). For example, the most recent real estate boom— real home prices were up 85% from 1997 to 2006—was driven by regional real estate booms, like in Washington, DC, whose origins were difficult to explain. Overall, even with periodic regional volatility, the national real estate market is quite stable. Data for the property level are from Metropolitan Regional Information Systems, Inc. (MRIS) in Rockville, Maryland. MRIS is the largest Multiple Listing Service in the U.S. with 45,371 subscribers in the Middle Atlantic region, which includes the District of Columbia, Maryland, and Virginia, as well as portions of Delaware, Pennsylvania, and West Virginia (Metropolitan Regional Information Systems, 2013). The sample includes all transactions in Montgomery County, Maryland from January 1, 2000 to December 31, 2009. Exclusion of transactions with missing data left a subsample of 78,513 property sales. At the property level, the dependent variable is the sale price or the natural log of the sale price (Exhibit 1). The independent variables at the property level include the exterior, interior, and quality of the property as well as the subdivision of the property. The year level corresponds to the year of the property transaction. Demographic and economic data for the ZIP Code level are from the Environmental Systems Research Institute (2000; 2001; 2002; 2003; 2004; 2005; 2006; 2007; 2008; 2009). Demographic data includes the number of households, percent White, and median age in years. Economic data includes median home value in U.S. dollars. Location data are from a geographic information system (GIS) map document in ArcMap of ArcGIS 10.2.2 from esri, which include point, line, and polygon shapefiles for properties, METRO stations, streets, interstate highways, and ZIP Codes. The Beltway percentage is the percentage of J O SRE Vol . 8 No. 1 – 2016 150 Z o lni k Exhibit 1  Data Dictionary for Property, Year, and Zip Code Levels Level n Variables Description Property 78,513 Dependent Price Closing price in U.S. dollars. lnPrice Natural log of closing price in U.S. dollars. Independent Exterior Floors Number of floors. Parking If parking is included in sale price, then Parking  1, 0 otherwise. Type If property is detached, then Type  1; if property is a townhome, then Type  0. Interior Baths—Full Number of full baths. Baths—Half Number of half baths. Bedrooms Number of bedrooms. Fireplaces Number of fireplaces. Quality Age Age of property at closing in years. New If property is less than one year old at closing, then New  1, 0 otherwise. Subdivision TND If property is in the Kentlands or in the Lakelands, then TND 1, 0 otherwise. T h e R esi lienc e o f t he P r e m i u m f o r H o m e s 151 J O SRE Vol . 8 No. 1 – 2016 Exhibit 1  (continued) Data Dictionary for Property, Year, and Zip Code Levels Level n Variables Description Year 10 Independent Time Year Year property closed. ZIP Code 42 Independent Demographic Households Number of households. White Percent White. Median Age Median age in years. Economic Median Home Value Median home value in U.S. dollars. Location Beltway If property is within a ZIP Code inside the Capital Beltway, then Beltway  1, 0 otherwise. Distance to Central Business District Linear distance in miles from property to the CBD of the District of Columbia. Distance to Nearest METRO Station Linear distance in miles from property to nearest METRO station. 152 Z o lni k ZIP Codes in Montgomery County, Maryland that are inside the area surrounded by Interstate 495, also known as the Capital Beltway. The distance to the central business district is the linear distance in miles from the centroid of each ZIP Code polygon to the centroid of the District of Columbia polygon, which is contiguous to the Montgomery County, Maryland polygon. Distance to nearest METRO station is the linear distance in miles from the centroid of each ZIP Code polygon to the nearest METRO station point. The above independent variables account for the accessibility and for the physical characteristics that affect home prices, but not for the environmental or public sector characteristics that affect home prices. In the former case, data on the social and the physical characteristics of neighborhoods are not readily available even though the Smart Location Database from the U.S. Environmental Protection Agency is a potential data source. In the latter case, data on the public sector characteristics of homes such as taxes and services are readily available. However, Montgomery County, Maryland is the tax jurisdiction for all of the homes in the subsample, so taxes are invariant between ZIP Codes. Likewise, Montgomery County, Maryland is known for quality public schools, but the 133 elementary school, the 38 middle school, and the 25 high school districts overlap to such an extent that the differences between school districts are difficult to measure. The hypothesized effects of the independent variables at the property level of analysis are as follows. All else equal, sale prices for detached properties with parking and more floors are expected to be higher. Detached homes with more floors tend to be larger in terms of square footage of improvement and buyers, most of whom own private vehicles, value parking spaces. Sale prices for properties with more full baths and more half baths as well as more bedrooms and more fireplaces are also expected to be higher because such homes also tend to be larger in terms of square footage of improvement. The actual square footage of the improvement is the optimal interior, property-level independent variable to capture the effect of interior size on sale prices. However, MRIS is not a reliable source for square footage of improvement based on the large number of properties with missing data. The sale prices for newer properties are expected to be higher. Sale prices for TND properties in the Kentlands and in the Lakelands are expected to be higher than sale prices for conventional properties in the surrounding neighborhoods because of the TND premium. The hypothesized effects of the dummy variables at the year level of analysis are as follows. Coincident with the real estate bubble in the front half of the first decade and the real estate crash in the back half of the first decade, the coefficients for the year dummy variables are expected to change signs from negative to positive to negative before, during, and after the referent year of 2004, which represents the highest sales volume year between 2000 and 2009. The hypothesized effects of the independent variables at the ZIP Code level of analysis are as follows. The demographic independent variables control for between-ZIP Code differences in the number, race / ethnicity, and age of households. The economic independent variable controls for between-ZIP Code differences in home values. The location independent variables control for T h e R esi lienc e o f t he P r e m i u m f o r H o m e s 153 between-ZIP Code differences in accessibility to interstate highways, employment centers, and public transportation. All else equal, the sale prices for homes inside the Beltway, closer to the District of Columbia, and closer to METRO stations are expected to be higher. Results Two-Level Model of Properties and ZIP Codes Descriptive statistics for two levels of 2000 to 2009 data appear in Exhibit 2. Coefficient estimates for linear (Price) and natural log (lnPrice) dependent variables in the two-level models of properties nested within ZIP Codes appear in Exhibit 3. At the property level, all of the exterior, interior, quality, and subdivision independent variables are statistically significant at the 99.9% confidence level. Also, the signs and the magnitudes of all of the coefficients are consistent with expectations. One more floor increases sale prices by 4% ($7,063.87). If parking is included, then sale prices increase by 6% ($26,270.92). If the property is detached, then sale prices increase by 34% ($123,122.18). One more full bath, half bath, bedroom, and fireplace increases sale prices by 18% ($104,371.29), 7% ($44,021.46), 3% ($22,133.80), and 10% ($46,451.23), respectively. One more age-year decreases sale prices by 0.1% ($836.18). If the property is less than one year old, then sale prices increase by 7% ($94,139.39). If the property is in a TND, then sale prices increase by 28% ($111,593.77). At the ZIP Code level, one of the demographic, the economic, and one of the location independent variables are statistically significant at the 95%, the 99.9%, and the 99.9% confidence levels, respectively. However, the signs of these coefficients are consistent with expectations. If the median age in a ZIP Code increases by one standard deviation (3.91 years), then sale prices increase by $31,164.73. If median home value in a ZIP Code increases by one standard deviation ($173,515.11), then sale prices increase by $53,789.68. If the property is within a ZIP Code inside the Beltway, then sale prices increase by 19% ($115,845.98). Three-Level Model of Properties, Years, and ZIP Codes Descriptive statistics for three levels of 2000 to 2009 data appear in Exhibit 4. Coefficient estimates for linear (Price) and natural log (lnPrice) dependent variables in three-level models of properties nested within years nested within ZIP Codes appear in Exhibit 5. At the property level, as in the two-level model, all of the exterior (except parking), interior, quality, and subdivision independent variables are statistically significant at the 99.9% confidence level. Also, the signs and the magnitudes of all of the coefficients are consistent with expectations. One more floor increases sale prices by 4% ($9,114.33). If the property is detached, then sale prices increase by J O SRE Vol . 8 No. 1 – 2016 154 Z o lni k Exhibit 2  Descriptive Statistics for Two Levels of 2000 to 2009 Data Level Variables Mean Std. Dev. Min Max Property Dependent Price 458,801.41 273,031.70 89,000.00 3,200,000.00 lnPrice 12.89 0.62 1.00 5.00 Independent Exterior Floors 2.89 0.62 1.00 5.00 Parking (%) Yes 68.64 No 31.36 Type (%) Detached 64.20 Townhouse 35.80 Interior Baths—Full 2.34 0.78 1.00 5.00 Baths—Half 0.95 0.55 0.00 2.00 Bedrooms 3.68 0.85 1.00 6.00 Fireplaces 0.90 0.66 0.00 3.00 Quality Age (years) 28.49 18.83 0.00 83.00 New Yes 2.07 No 97.93 Subdivision TND Yes 1.49 No 98.51 T h e R esi lienc e o f t he P r e m i u m f o r H o m e s 155 J O SRE Vol . 8 No. 1 – 2016 Exhibit 2  (continued) Descriptive Statistics for Two Levels of 2000 to 2009 Data Level Variables Mean Std. Dev. Min Max ZIP Code Independent Demographic Households 8,214.32 6,814.26 86.50 24,220.70 White (%) 68.65 16.50 31.38 93.81 Median Age (years) 39.17 3.91 32.12 46.01 Economic Median Home Value ($) 491,266.86 173,515.11 249,624.86 847,488.57 Location Beltway (%) Inside 14.29 Outside 85.71 Distance to Central 16.37 7.16 5.40 30.59 Business District (miles) Distance to Nearest 5.25 4.10 0.58 15.32 METRO Station (miles) 156 Z o lni k Exhibit 3  Coefficient Estimates for Linear (Price) and Natural Log (lnPrice) Dependent Variables in Two-Level Model Price ($) lnPrice Level n Variables Coeff. t-Ratio Coeff. t-Ratio Property 78,513 Independent Exterior Floors 7,063.87 6.90*** 0.04 18.28*** Parking (Yes  1/No  0) 26,270.92 20.83*** 0.06 26.38*** Type (Detached  1 / Townhouse  0) 123,122.18 67.06*** 0.34 94.89*** Interior Baths—Full 104,371.29 97.78*** 0.18 85.29*** Baths—Half 44,021.46 34.17*** 0.07 26.62*** Bedrooms 22,133.80 22.68*** 0.03 17.37*** Fireplaces 46,451.23 44.07*** 0.10 47.49*** Quality Age 836.11 16.05*** 1E–03 12.45*** New (Yes  1/No  0) 94,177.15 21.63*** 0.07 7.86*** Subdivision TND (Yes  1/No  0) 111,593.77 21.02*** 0.28 27.54*** T h e R esi lienc e o f t he P r e m i u m f o r H o m e s 157 J O SRE Vol . 8 No. 1 – 2016 Exhibit 3  (continued) Coefficient Estimates for Linear (Price) and Natural Log (lnPrice) Dependent Variables in Two-Level Model Price ($) lnPrice Level n Variables Coeff. t-Ratio Coeff. t-Ratio Zip Code 42 Intercept 368,287.62 12.86*** 12.67 244.22*** Independent Demographic Households 1.50 0.93 3E–06 1.07 White 1,225.58 1.18 3E–03 1.35 Median Age 7,970.52 2.19* 0.02 2.63* Economic Median Home Value 0.31 3.95*** 0.00 3.34** Location Beltway (Inside  1 / Outside  0) 115,845.98 3.66*** 0.19 3.22** Distance to Central Business District 71.87 0.02 2E–03 0.28 Distance to Nearest METRO Station 736.50 0.14 0.01 0.57 Notes: * P  0.05. ** P  0.01. *** P  0.001. 158 Z o lni k Exhibit 4  Descriptive Statistics for Three Levels of 2000 to 2009 Data Level n Variables Mean Std. Dev. Min Max Property 78,513 Dependent Price 458,801.41 273,031.70 89,000.00 3,200,000.00 lnPrice 12.89 0.53 11.40 14.98 Independent Exterior Floors 2.89 0.62 1.00 5.00 Parking (%) Yes 68.64 No 31.36 Type (%) Detached 64.20 Townhouse 35.80 Interior Baths—Full 2.34 0.78 1.00 5.00 Baths—Half 0.95 0.55 0.00 2.00 Bedrooms 3.68 0.85 1.00 6.00 Fireplaces 0.90 0.66 0.00 3.00 Quality Age (years) 28.49 18.83 0.00 83.00 New Yes 2.07 No 97.93 Subdivision TND Yes 1.49 No 98.51 T h e R esi lienc e o f t he P r e m i u m f o r H o m e s 159 J O SRE Vol . 8 No. 1 – 2016 Exhibit 4  (continued) Descriptive Statistics for Three Levels of 2000 to 2009 Data Level n Variables Mean Std. Dev. Min Max Year 408 Independent Time Year (%) 2000 8.65 2001 9.90 2002 10.33 2003 10.95 2004 12.61 2005 12.32 2006 10.80 2007 8.44 2008 7.23 2009 8.77 ZIP Code 42 Independent Demographic Households 8,214.38 6,841.25 87.00 24,221.00 White (%) 68.66 16.50 31.40 93.80 Median Age (years) 39.17 3.92 32.10 46.00 Economic Median Home Value ($) 491,266.90 173,515.14 249,625.00 847,489.00 Location Beltway (%) Inside 19.05 Outside 80.95 Distance to Central 16.37 7.25 5.40 30.59 Business District (miles) Distance to Nearest 5.25 4.15 0.58 15.32 METRO Station (miles) 160 Z o lni k Exhibit 5  Coefficient Estimates for Linear (Price) and Natural Log (lnPrice) Dependent Variables in Three-Level Model Price ($) lnPrice Level n Variables Coeff. t-Ratio Coeff. t-Ratio Property 78,513 Independent Exterior Floors 9,114.33 11.68*** 0.04 37.51*** Parking (Yes  1/No  0) 1,735.14 1.79 4E–03 2.75** Type (Detached  1 / Townhouse  0) 150,651.10 107.28*** 0.40 198.21*** Interior Baths—Full 84,165.71 102.49*** 0.13 105.86*** Baths—Half 55,838.58 56.68*** 0.09 63.47*** Bedrooms 21,290.77 28.67*** 0.03 27.71*** Fireplaces 52,747.00 65.66*** 0.11 96.41*** Quality Age 2,564.84 62.57*** 6E–03 94.41*** New (Yes  1/No  0) 110,496.39 32.78*** 0.10 21.06*** Subdivision TND (Yes  1/No  0) 76,819.54 18.99*** 0.19 32.83*** Year 408 Independent Time Year 2000 228,898.70 28.33*** 0.59 48.10*** 2001 181,517.38 22.55*** 0.47 38.55*** 2002 132,712.71 16.56*** 0.32 26.21*** 2003 79,139.64 9.87*** 0.18 14.90*** 2004 Referent Referent 2005 91,330.52 11.50*** 0.17 14.23*** 2006 100,605.05 12.60*** 0.20 16.41*** 2007 85,009.21 10.58*** 0.17 13.78*** 2008 26,078.92 3.23 0.04 2.98** 2009 17,689.49 2.21* 0.07 5.61*** T h e R esi lienc e o f t he P r e m i u m f o r H o m e s 161 J O SRE Vol . 8 No. 1 – 2016 Exhibit 5  (continued) Coefficient Estimates for Linear (Price) and Natural Log (lnPrice) Dependent Variables in Three-Level Model Price ($) lnPrice Level n Variables Coeff. t-Ratio Coeff. t-Ratio Zip Code 42 Intercept 412,856.45 44.66*** 12.77 881.98*** Independent Demographic Households 2.17 1.48 0.00 1.99 White 1,270.72 1.34 3E–03 1.76 Median Age 8,043.23 2.44* 0.02 3.26 Economic Median Home Value 0.33 3.66*** 1E–06 4.57*** Location Beltway (Inside  1 / Outside  0) 118,006.81 4.08*** 0.19 4.21*** Distance to Central Business District 2,242.28 0.69 4E–03 0.72 Distance to Nearest METRO Station 28.76 0.01 4E–03 0.46 Notes: Referent is the year (2004) with the highest percentage of closings from 2000 to 2009 (12.61%). * P  0.05. ** P  0.01. *** P  0.001. 162 Z o lni k 40% ($150,651.10). One more full bath, half bath, bedroom, and fireplace increase sale prices by 13% ($84,165.71), 9% ($55,813.58), 3% ($21,290.77), and 11% ($52,747.00), respectively. One more age-year decreases sale prices by 0.6% ($2,564.84). If the property is less than one year old, then sale prices increase by 10% ($111,496.39). If the property is in a TND, then sale prices increase by 19% ($76,819.54). At the time level, all of the year dummy variables except for 2008 are statistically significant at the 99.9% confidence level. Also, the changes in the signs and the magnitudes of the coefficients are consistent with expectations. Sale prices increase from a trough in 2000 when sale prices are 59% ($228,898.70) lower than in the referent year of 2004 to a peak in 2006 when sale prices are 20% ($100,605.05) higher than in the referent year of 2004. After 2008, sale prices are again 7% ($17,689.49) lower than in the referent year of 2004. At the ZIP Code level, only one of the economic and only one of the location independent variables are statistically significant at the 99% and the 95% confidence levels, respectively. However, the signs of these coefficients are consistent with expectations. If median home value in a ZIP Code increases by one standard deviation ($24,354.94), then sale prices increase by $9,657.95. If the property is within a ZIP Code inside the Beltway, then sale prices increase by 17% ($126,726.80). Discussion Relating the results to the research questions in the study and linking the results to the literature on the valuation of homes in TNDs, the TND premium for homes in the Kentlands and in the Lakelands is resilient over the long term, notwithstanding the Great Recession. The mean TND premium for homes in the Kentlands and in the Lakelands ranges from 28% ($111,593.77) to 19% ($76,819.54) from the two-level to the three-level model specification, respectively. Regardless of the model specification, the results for the older Kentlands and for the newer Lakelands exceed the highest mean premiums for homes in the Kentlands from the literature; 13% from Eppli and Tu (1999) and 12% from Tu and Eppli (1999). Premiums are probably higher in the Kentlands and in the Lakelands versus the surrounding conventional neighborhoods because the former homes are larger in size in terms of number of full baths, number of half baths, number of fireplaces, and number of levels according to nonparametric and parametric test statistics. However, if the actual square footage of improvement is an independent variable to capture the effect of interior size on sale prices, premiums are not likely to be as great. Premiums are also probably higher in the Kentlands and in the Lakelands versus the surrounding conventional neighborhoods because a higher percentage of the former homes are new (5.23% vs. 2.03%). This follows from the finding that the characteristics of single-family homes that have the largest impact on the mean premium over the ten-year time period of transactions are, in order from largest effects to smallest effect, exterior effects (if the property is detached), T h e R esi lienc e o f t he P r e m i u m f o r H o m e s 163 interior effects (number of full baths), and quality effects (if the property is less than one year old at closing). Finally, premiums are probably higher in the Kentlands and in the Lakelands versus the surrounding conventional neighborhoods because the regional boom in home prices in the Washington, DC real estate market coincides with the first seven years—2000 to 2006—of the ten- year time period of transactions in the study. This follows from the finding that the mean closing price for homes in the Kentlands and in the Lakelands versus the surrounding conventional neighborhoods is higher in the former ($577,042.06 vs. $457,017.39). The TND home premium is probably lower in the three-level model specification than in the two-level model specification (19% vs. 28%, respectively) because in the three-level model specification, I explicitly control for the temporal effects on sale prices from 2000 to 2006 to 2009 when sale prices were down (59%), up (20%), and down again (7%) relative to the referent year of 2004. The two-level model specification distributes the year-level variance to the property and the ZIP Code levels. Nevertheless, a two-level model specification is statistically more valid than a cross-sectional approach where time is not a covariate or a level of analysis. The latter approach is not statistically valid because the subdivision independent variables in model specifications for 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, and 2009, respectively, are not comparable. Conclusion One contribution of the study to the TND literature is the resiliency of the premium buyers willingly pay for properties in TND neighborhoods. To that end, the study answers an important question in the TND literature on the long-term demand for what are known to be more expensive homes particularly when markets are slow and credit is tight. Another contribution of the study to the real estate valuation literature is the multilevel model specifications, which accurately contextualize property sales transactions in time and space. Indeed, the results indicate that the premiums for homes in TND neighborhoods are robust to the temporal and spatial differences in property sales transactions between TND neighborhoods and conventionally- developed neighborhoods. Overall, the specifications contribute to the trend towards methodological sophistication in the real estate valuation literature. To that end, the specifications show that time and space both enhance the accuracy of valuation models. One of the most fruitful avenues for future research on the resiliency of TND premiums is to use data on property sales transactions for more years and for more neighborhoods. One example, the ten-year time period of property sales transactions includes data from before, during, and after the Great Recession, but the after time period is short. In order to accurately estimate the effect of the Great Recession on property sales transactions, the time period after the Great Recession ought to include more post-recession years. Another example, the study area of J O SRE Vol . 8 No. 1 – 2016 164 Z o lni k Montgomery County, Maryland is on the fringe of the METRO system which, at least, partially explains why the effect of the locational independent variable METRO is not statistically different from zero in the two- and three-level model specifications, respectively. In order to accurately estimate the effect of METRO on property sales transactions, the study area ought to include property sales transactions for all of the county and county equivalent areas where there are METRO stations. The results clearly demonstrate to developers, investors, and lenders that buyers are willing to pay a premium for single-family homes in new urbanist neighborhoods even in the midst of a market downturn with tightening credit. Appendix The specification of the two-level model with a property ( p) level and a ZIP Code (z) level is as follows. Within each ZIP Code, closing prices are modeled as a function of property-level independent variables plus a property-level error term: Y     W  ...   W  r , (A1) pz 0z 1z 1pz Az Apz pz where Y is the closing price of property p in ZIP Code z;  is the y-intercept pz 0z term for ZIP Code z;  are a  1, ..., A property-level coefficients; W are a Az Apz 1, ..., A property-level independent variables; and r is the property-level pz random effect term. All of the regression coefficients at the property level are fixed; that is, invariant over ZIP Code z. A multilevel model in which the y-intercept is random and all of the regression coefficients at level-1 are fixed is known as a random-intercepts model. The model for variation between ZIP Codes is therefore as follows. For the ZIP Code effect  : 0z Z  ...   Z  u , (A2) 0z 00 01 1z 0BBz 0z where  is the y-intercept term for ZIP Code z;  are b  1, ..., B ZIP Code- 00 0B are b  1, ..., B ZIP Code-level independent variables; and level coefficients; Z Bz u is a ZIP Code-level random effect term. 0z The specification of the three-level model with a property ( p) level, a year ( y) level, and a ZIP Code (z) level is as follows. Within each ZIP Code, closing prices are modeled as a function of property-level independent variables plus a property- level error term: T h e R esi lienc e o f t he P r e m i u m f o r H o m e s 165 Y     W  ...   W  r , (A3) pyz 0yz 1yz 1pyz Ayz Apyz pyz where Y is the closing price of property p in year y and ZIP Code z;  is the pyz 0yz y-intercept term for year y in ZIP Code z;  are a  1, ..., A property-level Ayz coefficients; W are a  1, ..., A property-level independent variables; and r Apyz pyz is the property-level random effect term. Like the two-level model, the three-level model is a random-intercepts model; that is, all of the regression coefficients at the property level and at the year level are fixed. The model for variation among years within ZIP Codes is therefore as follows: X  ...   X  e , (A4) 0yz 00z 1z 1yz (C1)z (C1)yz 0yz where  is the y-intercept term for ZIP Code z;  are c  1, ..., C  1 00z (C1)z year-level coefficients; X are c  1, ..., C  1 dummy variables; and e is (C1)yz 0yz the year-level random effect term. The model for variation between ZIP Codes is as follows. For ZIP Code effect 00z Z  ...   Z  u , (A5) 00z 000 001 1z 00BBz 00z where  is the y-intercept term for ZIP Code z;  are b  1, ..., B ZIP Code- 000 00B level coefficients; Z are b  1, ..., B ZIP Code-level independent variables; and Bz u is a ZIP Code-level random effect term. 00z References Brigham, E. The Determinants of Residential Land Values. Land Economics, 1965, 41:4, 325–34. Congress for the New Urbanism. New Urbanism: It Just Performs Better. Chicago, IL: Congress for the New Urbanism, 2014. Crecine, J., O. Davis, and J. Jackson. Urban Property Markets: Some Empirical Results and Their Implications for Municipal Zoning. Journal of Law and Economics, 1967, 10, 79–99. Dong, H. Were Home Prices in New Urbanist Neighborhoods More Resilient in the Recent Housing Downturn? Journal of Planning Education and Research, 2015, 35:1, 5–18. Environmental Systems Research Institute. Community Sourcebook of Zip Code Demographics. Redlands, CA: Environmental Systems Research Institute, 2000. J O SRE Vol . 8 No. 1 – 2016 166 Z o lni k ——. Community Sourcebook of Zip Code Demographics. Redlands, CA: Environmental Systems Research Institute, 2001. ——. Sourcebook of Zip Code Demographics. Redlands, CA: Environmental Systems Research Institute, 2002. ——. The Sourcebook of Zip Code Demographics. Vienna, VA: Environmental Systems Research Institute, 2003. ——. Community Sourcebook of Zip Code Demographics. Vienna, VA: Environmental Systems Research Institute, 2004. ——. Community Sourcebook of Zip Code Demographics. Redlands, CA: Environmental Systems Research Institute, 2005. ——. Community Sourcebook of Zip Code Demographics. Redlands, CA: Environmental Systems Research Institute, 2006. ——. Community Sourcebook of Zip Code Demographics. Redlands, CA: Environmental Systems Research Institute, 2007. ——. Community Sourcebook of Zip Code Demographics. Redlands, CA: Environmental Systems Research Institute, 2008. ——. Sourcebook of Zip Code Demographics. Redlands, CA: Environmental Systems Research Institute, 2009. Eppli, M. and C. Tu. Valuing the New Urbanism: The Impact of the New Urbanism on Prices of Single-Family Homes. Washington, DC: Urban Land Institute, 1999. Follain, J. and S. Malpezzi. Dissecting Housing Value and Rent: Estimates of Hedonic Indexes for Thirty-Nine Large SMSAs. Washington, DC: Urban Institute, 1980. Fulton, W. The New Urbanism: Hope or Hype for American Communities? Cambridge, MD: The Lincoln Institute of Land Policy. 1996. Garde, A. Designing and Developing New Urbanism Projects in the United States: Insights and Implications. Journal of Urban Design, 2006, 11:1, 33–54. Guttery, R. The Effects of Subdivision Design on Housing Values: The Case of Alleyways. Journal of Real Estate Research, 2002, 23:3, 265–73. Gyourko, J. and W. Rybczynski. Financing New Urbanism Projects: Obstacles and Solutions. Housing Policy Debate, 2000, 11:3, 733–50. Jones, K. and N. Bullen. A Multi-level Analysis of the Variations in Domestic Property Prices: Southern England, 1980–87. Urban Studies, 1993, 30:8, 1409–26. ——. Contextual Models of Urban House Prices: A Comparison of Fixed- and Random- Coefficient Models Developed by Expansion. Economic Geography, 1994, 70:3, 252–72. Krause, A. and C. Bitter. Spatial Econometrics, Land Values and Sustainability: Trends in Real Estate Valuation Research. Cities, 2012, 29:S2, S19–S25. Lang, R. Valuing the Suburbs: Why Some ‘Improvements’ Lower Home Prices. Opolis, 2005, 1:1, 5–12. Mahan, B., S. Polasky, and R. Adams. Valuing Urban Wetlands: A Property Price Approach. Land Economics, 2000, 76:1, 100–13. Metropolitan Regional Information Systems. Vital Statistics. Available at: http: / / www.mris.com / about-mris / vital-statistics / , 2013. Planning Design Group. Economic Return on New Urbanism: A Summary of Focus Group Discussions of Developers and Practitioners of New Urbanism in Central Florida. Orlando, FL: Planning Design Group, 2007. Plaut, P. and M. Boarnet. New Urbanism and the Value of Neighborhood Design. Journal of Architectural and Planning Research, 2003, 20:3, 254–65. T h e R esi lienc e o f t he P r e m i u m f o r H o m e s 167 Raudenbush, S. and A. Bryk. Hierarchical Linear Models: Applications and Data Analysis Methods. Thousand Oaks, CA: Sage, 2002. Rauterkus, S. and N. Miller. Residential Land Values and Walkability. Journal of Sustainable Real Estate, 2011, 3:1, 23–43. Reuter, F. Externalities in Urban Property Markets: An Empirical Test of the Zoning Ordinance of Pittsburgh. Journal of Law and Economics, 1973, 16:2, 313–49. Ryan, B. and R. Weber. Valuing New Development in Distressed Urban Neighborhoods: Does Design Matter? Journal of the American Planning Association, 2007, 73:1, 100–11. Shiller, R. Irrational Exuberance. Princeton, NJ: Princeton University Press, 2015. Song, Y. and G. Knaap. New Urbanism and Housing Values: A Disaggregate Assessment. Journal of Urban Economics, 2003, 54:2, 213–38. Song, Y. and R. Quercia. How are Neighbourhood Design Features Valued across Different Neighbourhood Types? Journal of Housing and the Built Environment, 2008, 23:4, 297– Steuteville, R. Kentlands / Lakelands. Ithaca, NY: New Urban News Publications, 2010. Stull, W. Community Environment, Zoning, and the Market Value of Single-Family Homes. Journal of Law and Economics, 1975, 18:2, 535–57. Tu, C. and M. Eppli. Valuing New Urbanism: The Case of Kentlands. Real Estate Economics, 1999, 27:3, 425–51. ——. An Empirical Examination of Traditional Neighborhood Development. Real Estate Economics, 2001, 29:3, 485–501. The author wishes to thank the Center for Regional Analysis at George Mason University for assistance in data acquisition for this study. All articles published in JOSRE are distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Edmund Zolnik, George Mason University, Arlington, VA 22201 or ezolnik@ gmu.edu. J O SRE Vol . 8 No. 1 – 2016 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Journal of Sustainable Real Estate Taylor & Francis

The Resilience of the Premium for Homes in New Urbanist Neighborhoods

Zolnik, Edmund
Journal of Sustainable Real Estate , Volume 8 (1): 23 – Nov 1, 2016
Download PDF
23 pages

Loading next page...
 
/lp/taylor-francis/the-resilience-of-the-premium-for-homes-in-new-urbanist-neighborhoods-eoFju2s0iM

References

References for this paper are not available at this time. We will be adding them shortly, thank you for your patience.

Publisher
Taylor & Francis
Copyright
© 2016 American Real Estate Society
ISSN
1949-8284
DOI
10.1080/10835547.2016.12091882
Publisher site
See Article on Publisher Site

Abstract

In this study, I analyze longitudinal differences in single-family home prices in two new urbanist neighborhoods versus surrounding conventional neighborhoods. Using data on 78,513 single-family home sales transactions in Montgomery County, Maryland, I adopt a novel multilevel methodology to assess the effects of neighborhood demographic, economic, and locational characteristics as well as the effects of year on home prices. Results support empirical evidence on the premium that buyers willingly pay for homes in new urbanist neighborhoods. Year effects indicate that the premium withstood the Great Recession quite well. New urbanism, also known as, traditional neighborhood development (TND), is a community development concept that advocates harmonizing demand for scarce development resources like land with a mix of uses—commercial and residential—associated with traditional neighborhoods (Fulton, 1996). By mixing, rather than separating, uses in pedestrian-oriented and transit-oriented spaces, TND adds value in the market according to its proponents (Congress for the New Urbanism, 2014). The Congress for the New Urbanism also suggests that homes in TND neighborhoods command a premium and retain value in the market because of walkable access to amenities (Congress for the New Urbanism, 2014). Indeed, the street networks in TND neighborhoods tend to be grid-like rather than curvilinear as in conventional neighborhoods, while the orientation of homes toward the street facilitates pedestrian trips to recreational or to shopping destinations (Plaut and Boarnet, 2003). Unfortunately, the literature on TNDs provides little empirical evidence on the willingness of buyers to pay such premiums over the long term. In this regard, the literature is limited in two fundamental ways: in temporal scale and in spatial scale. Taken together, these limitations make it difficult to ascertain the resiliency of these premiums. The temporal limitation relates to the fact that, at most, data for nine years of single-family home sales transactions are used to estimate TND premiums. While this is a reasonable time horizon to account for any randomness, more research with data from as long or longer periods of time is needed to make a definitive statement on the resiliency of TND premiums. The spatial limitation relates to the fact that the number of comparable neighborhoods used to estimate premiums is very small. Plaut and Boarnet (2003) used three neighborhoods (one TND and two conventional neighborhoods), Tu and Eppli (1999) used nine census J O SRE Vol . 8 No. 1 – 2016 146 Z o lni k tracts (one TND and eight conventional neighborhoods), and Tu and Eppli (2001) used three markets (three TNDs and surrounding conventional developments) to estimate TND premiums. A small number of comparable neighborhoods is limiting for the following reasons. While it is important that enough transactions are analyzed to account for randomness, it is the number of comparable neighborhoods that is the priority in the statistical estimation of neighborhood design effects, such as those exemplified in TNDs (Follain and Malpezzi, 1980). If there are not enough comparable neighborhoods, then it will be difficult to attribute any discernible differences in sales prices to the design features unique to TNDs. And, from a statistical perspective, a sample size of nine at the neighborhood level of analysis makes it a challenge to precisely estimate such differences. In this study, I attempt to address these temporal and spatial limitations in order to answer the following research questions. First, are TND premiums resilient over the long term? The long term in the study is ten years, which is enough to account for randomness and just slightly longer than the longest time horizon in the literature of nine years. The fact that the analysis covers a ten-year time period is important, but it is also important to acknowledge the importance of what occurred during those ten years. In fact, the time period from 2000 to 2009 coincides temporally with the Great Recession from December of 2007 to June of 2009; a time of marked volatility in regional real estate markets across the United States (Shiller, 2015). Second, what characteristics of single-family homes in TNDs have the largest impact on these premiums over this ten-year period of transactions? Third, and finally, how did these premiums change, if at all? The organization of the study is as follows. In the next sections, I briefly review the literature on the valuation of homes in TNDs as well as the opinions of developers. In the methodology section, I present the multilevel models in the study—a two-level model with a property level nested within a ZIP Code level and a three-level model with a property level nested within a year level nested within a ZIP Code level. In the data section, I name the sources of the property- level, the year-level, and the ZIP Code-level data and describe the dependent variables as well as the independent variables in the models. In the results section, I review the relevant results from the two- and the three-level models. In the discussion section, I relate the results to the research questions in the study and link the results to the literature on the valuation of homes in TNDs relative to homes in conventional neighborhoods. In the conclusion section, I highlight the contribution of the study and suggest one fruitful avenue for future research. Premium for Homes in New Urbanist Neighborhoods It is known that homes in TND neighborhoods can be more expensive that those in conventionally-developed neighborhoods (Song and Knapp, 2003). In addition, empirical evidence suggests that buyers willingly pay a premium for homes in such neighborhoods (Tu and Eppli, 1999, 2001; Plaut and Boarnet, 2003). T h e R esi lienc e o f t he P r e m i u m f o r H o m e s 147 However, it is also important for developers to know that those higher values are resilient; that is, the higher values relative to properties outside of TNDs persist over time. If so, then the development community can demonstrate to financers that not only do TND communities make more efficient use of development resources, they tend to retain their higher values (Gyourko and Rybczynski, 2000; Garde; 2006; Planning Design Group, 2007). If the focus of the research is on walkability, then the empirical evidence seems conclusive. Song and Querica (2008) found that greater pedestrian accessibility to commercial land uses garner a premium. Rauterkus and Miller (2011) found that the effect of walkability on land values is positive. Dong (2015) found that the effect of walkability, not other new urbanist neighborhood design features, on single-family home appreciation rates during the Great Recession is positive. However, if the focus of the research is on other neighborhood design features common to TNDs, such as development density, street connectivity, and land use mix, then the empirical evidence is inconclusive (Lang, 2005). Song and Querica (2008) also found that street connectivity garners a premium. However, they found that, consistent with older empirical evidence (Crecine, Davis, and Jackson, 1967; Reuter, 1973), land use mix garners no premium even though newer empirical evidence demonstrates both positive effects (Song and Knaap, 2003) and negative effects (Mahan, Polasky, and Adams, 2000). Further, low-development density, not high-development density, garners a premium. In low-income neighborhoods, where little research on the effects of different neighborhood design features is evident, assessments for infill-style developments are higher relative to TND-style and enclave-style developments (Ryan and Weber, 2007). Guttery (2002) found that homes with rear-entries facing alleyways, another design feature synonymous with TND-style development, sell for a $5,575 (5.3%) discount relative to homes with front-entry driveways. Designing and Developing TNDs Just as the valuation of TND homes is not conclusive, nor are the opinions of designers and developers unanimous. Surveys and interviews of developers, financiers, and investors from across the U.S. (n  23) who are familiar with TNDs or who have experience with TNDs (Gyourko and Rybczynski, 2000) suggest that financing TNDs is risky and costly, although neither financiers nor investors perceive costs to be prohibitive. The perception of risk is mostly attributable to the multiple land uses in TNDs. Surveys of professionals (n  169) from across the U.S. with experience in the design and the development of TNDs (Garde, 2006) suggest that designers, developers, and planners agree on the advantages of TNDs with regard to design, growth management, environmental preservation, and Not In My Back Yard (NIMBY) opposition. The same professionals also agree on the disadvantages of TNDs with regard to the restrictions of land use regulations, the resistance of developers, and the costs of construction. Indeed, 56% of developers agree that TNDs have higher construction costs and 63% of developers at least somewhat agree that TNDs are not sound investments. Interviews with the same professionals (n  11) suggest general J O SRE Vol . 8 No. 1 – 2016 148 Z o lni k skepticism with regard to TNDs (Garde, 2006). Developers are uncertain of the demand for TNDs relative to conventional suburban developments. Developers as well as lenders are also wary of the risks most associate with a relatively new product. The consensus of focus groups and surveys of professionals with experience in the development of TNDs in Florida is that ‘‘New Urbanism remains a complex, frequently misunderstood, and often challenging form of development that comprises only a fraction of the overall development landscape’’ (Planning Design Group, 2007, p. 2). And, ‘‘with the economy and housing markets slowing down and credit significantly tightening, it remains to be seen whether the market for these more expensive homes will hold or decline’’ (Planning Design Group, 2007, p. 9). The latter question coincides with the general trend towards greater methodological sophistication in the real estate valuation literature (Krause and Bitter, 2012) to more accurately estimate future changes given past volatility. To address the question of the resiliency of the premiums for homes in TND neighborhoods, I analyze differences in home prices between two contiguous TNDs—the Kentlands and the Lakelands—and surrounding conventional neighborhoods in Montgomery County, Maryland over a ten-year time period from 2000 to 2009. The Kentlands and the Lakelands were both designed by Duany Plater-Zyberk and Company in 1988 and in 1996, respectively. All told, the more than 3,000 residential units and the approximately 600,000 square feet of retail and commercial space in the Kentlands and in the Lakelands occupy approximately 695 acres in Gaithersburg, Maryland (Steuteville, 2010). Estimates of the mean premium buyers are willing to pay in the Kentlands are 13% ($24,603) (Eppli and Tu, 1999) and 12% (about $25,000) (Tu and Eppli, 1999). Estimates of the premium buyers are willing to pay in other TNDs are as follows. Eppli and Tu (1999) estimate that buyers pay a mean premium of: 25% ($30,690) in the TND of Harbor Town in Memphis, Tennessee; 4% ($5,157) in the TND of Laguna West in Elk Grove, California; and 9% ($16,334) in the TND of Southern Village in Memphis, Tennessee. Plaut and Boarnet (2003) estimate that buyers pay a mean premium of $8,229 in the TND of Central Carmel versus mean premiums of $6,690 and $6,008 in the conventional neighborhoods of Carmel and Denia, respectively, in Haifa, Israel. Methodology Jones and Bullen explain (1993) and illustrate (1994) the technical advantages of a multilevel approach to model home prices rather than assign dummy variables for different times or for different places in a multiple regression model of home prices. First, a multilevel approach explicitly accounts for autocorrelation, or nonindependence; that is, the sale prices of two homes from the same Zip Code are more alike than the sales prices of two homes drawn randomly. Second, a multilevel approach accurately estimates the effects of independent variables at higher levels of analysis such as median home value. Third, a multilevel approach is used to pool information from all of the Zip Codes in the study area to precisely estimate the mean home-price relation as well as the variation in the mean home- price relation both in ZIP Codes where home sales are many and in Zip Codes where home sales are few. T h e R esi lienc e o f t he P r e m i u m f o r H o m e s 149 The multilevel models in the study include a two-level model with a property ( p) level nested within a ZIP Code (z) level and a three-level model with a property ( p) level nested within a year ( y) level nested within a ZIP Code (z)level (Raudenbush and Bryk, 2002). Precedence for nesting time (year) within place (Zip Code) in the three-level model is from Jones and Bullen (1993, pp. 1412– 41): ‘‘[t]he purchase price of properties (level 1) are recorded for time periods (at level 2) for areas (at level 3). In such a three-level model, there is a separate level- 2 unit for each time period in each district. Such a formulation allows for complex space-time modelling, with potentially a separate trend in house-price inflation for each area.’’ The specifications of the two- and three-level models are in the Appendix. Data The characteristics that affect home prices fall into four groups: (1) accessibility characteristics, such as distance to popular destinations; (2) environmental characteristics; (3) physical characteristics, such as age; and (4) public sector characteristics, such as taxes and services (Brigham, 1965; Stull, 1975). Indeed, each home constitutes a bundle of the above characteristics. But, while the characteristics that affect home prices are known, home prices are not readily forecastable statistically even though one-year forecasts are more precise than ten- year forecasts (Shiller, 2015). For example, the most recent real estate boom— real home prices were up 85% from 1997 to 2006—was driven by regional real estate booms, like in Washington, DC, whose origins were difficult to explain. Overall, even with periodic regional volatility, the national real estate market is quite stable. Data for the property level are from Metropolitan Regional Information Systems, Inc. (MRIS) in Rockville, Maryland. MRIS is the largest Multiple Listing Service in the U.S. with 45,371 subscribers in the Middle Atlantic region, which includes the District of Columbia, Maryland, and Virginia, as well as portions of Delaware, Pennsylvania, and West Virginia (Metropolitan Regional Information Systems, 2013). The sample includes all transactions in Montgomery County, Maryland from January 1, 2000 to December 31, 2009. Exclusion of transactions with missing data left a subsample of 78,513 property sales. At the property level, the dependent variable is the sale price or the natural log of the sale price (Exhibit 1). The independent variables at the property level include the exterior, interior, and quality of the property as well as the subdivision of the property. The year level corresponds to the year of the property transaction. Demographic and economic data for the ZIP Code level are from the Environmental Systems Research Institute (2000; 2001; 2002; 2003; 2004; 2005; 2006; 2007; 2008; 2009). Demographic data includes the number of households, percent White, and median age in years. Economic data includes median home value in U.S. dollars. Location data are from a geographic information system (GIS) map document in ArcMap of ArcGIS 10.2.2 from esri, which include point, line, and polygon shapefiles for properties, METRO stations, streets, interstate highways, and ZIP Codes. The Beltway percentage is the percentage of J O SRE Vol . 8 No. 1 – 2016 150 Z o lni k Exhibit 1  Data Dictionary for Property, Year, and Zip Code Levels Level n Variables Description Property 78,513 Dependent Price Closing price in U.S. dollars. lnPrice Natural log of closing price in U.S. dollars. Independent Exterior Floors Number of floors. Parking If parking is included in sale price, then Parking  1, 0 otherwise. Type If property is detached, then Type  1; if property is a townhome, then Type  0. Interior Baths—Full Number of full baths. Baths—Half Number of half baths. Bedrooms Number of bedrooms. Fireplaces Number of fireplaces. Quality Age Age of property at closing in years. New If property is less than one year old at closing, then New  1, 0 otherwise. Subdivision TND If property is in the Kentlands or in the Lakelands, then TND 1, 0 otherwise. T h e R esi lienc e o f t he P r e m i u m f o r H o m e s 151 J O SRE Vol . 8 No. 1 – 2016 Exhibit 1  (continued) Data Dictionary for Property, Year, and Zip Code Levels Level n Variables Description Year 10 Independent Time Year Year property closed. ZIP Code 42 Independent Demographic Households Number of households. White Percent White. Median Age Median age in years. Economic Median Home Value Median home value in U.S. dollars. Location Beltway If property is within a ZIP Code inside the Capital Beltway, then Beltway  1, 0 otherwise. Distance to Central Business District Linear distance in miles from property to the CBD of the District of Columbia. Distance to Nearest METRO Station Linear distance in miles from property to nearest METRO station. 152 Z o lni k ZIP Codes in Montgomery County, Maryland that are inside the area surrounded by Interstate 495, also known as the Capital Beltway. The distance to the central business district is the linear distance in miles from the centroid of each ZIP Code polygon to the centroid of the District of Columbia polygon, which is contiguous to the Montgomery County, Maryland polygon. Distance to nearest METRO station is the linear distance in miles from the centroid of each ZIP Code polygon to the nearest METRO station point. The above independent variables account for the accessibility and for the physical characteristics that affect home prices, but not for the environmental or public sector characteristics that affect home prices. In the former case, data on the social and the physical characteristics of neighborhoods are not readily available even though the Smart Location Database from the U.S. Environmental Protection Agency is a potential data source. In the latter case, data on the public sector characteristics of homes such as taxes and services are readily available. However, Montgomery County, Maryland is the tax jurisdiction for all of the homes in the subsample, so taxes are invariant between ZIP Codes. Likewise, Montgomery County, Maryland is known for quality public schools, but the 133 elementary school, the 38 middle school, and the 25 high school districts overlap to such an extent that the differences between school districts are difficult to measure. The hypothesized effects of the independent variables at the property level of analysis are as follows. All else equal, sale prices for detached properties with parking and more floors are expected to be higher. Detached homes with more floors tend to be larger in terms of square footage of improvement and buyers, most of whom own private vehicles, value parking spaces. Sale prices for properties with more full baths and more half baths as well as more bedrooms and more fireplaces are also expected to be higher because such homes also tend to be larger in terms of square footage of improvement. The actual square footage of the improvement is the optimal interior, property-level independent variable to capture the effect of interior size on sale prices. However, MRIS is not a reliable source for square footage of improvement based on the large number of properties with missing data. The sale prices for newer properties are expected to be higher. Sale prices for TND properties in the Kentlands and in the Lakelands are expected to be higher than sale prices for conventional properties in the surrounding neighborhoods because of the TND premium. The hypothesized effects of the dummy variables at the year level of analysis are as follows. Coincident with the real estate bubble in the front half of the first decade and the real estate crash in the back half of the first decade, the coefficients for the year dummy variables are expected to change signs from negative to positive to negative before, during, and after the referent year of 2004, which represents the highest sales volume year between 2000 and 2009. The hypothesized effects of the independent variables at the ZIP Code level of analysis are as follows. The demographic independent variables control for between-ZIP Code differences in the number, race / ethnicity, and age of households. The economic independent variable controls for between-ZIP Code differences in home values. The location independent variables control for T h e R esi lienc e o f t he P r e m i u m f o r H o m e s 153 between-ZIP Code differences in accessibility to interstate highways, employment centers, and public transportation. All else equal, the sale prices for homes inside the Beltway, closer to the District of Columbia, and closer to METRO stations are expected to be higher. Results Two-Level Model of Properties and ZIP Codes Descriptive statistics for two levels of 2000 to 2009 data appear in Exhibit 2. Coefficient estimates for linear (Price) and natural log (lnPrice) dependent variables in the two-level models of properties nested within ZIP Codes appear in Exhibit 3. At the property level, all of the exterior, interior, quality, and subdivision independent variables are statistically significant at the 99.9% confidence level. Also, the signs and the magnitudes of all of the coefficients are consistent with expectations. One more floor increases sale prices by 4% ($7,063.87). If parking is included, then sale prices increase by 6% ($26,270.92). If the property is detached, then sale prices increase by 34% ($123,122.18). One more full bath, half bath, bedroom, and fireplace increases sale prices by 18% ($104,371.29), 7% ($44,021.46), 3% ($22,133.80), and 10% ($46,451.23), respectively. One more age-year decreases sale prices by 0.1% ($836.18). If the property is less than one year old, then sale prices increase by 7% ($94,139.39). If the property is in a TND, then sale prices increase by 28% ($111,593.77). At the ZIP Code level, one of the demographic, the economic, and one of the location independent variables are statistically significant at the 95%, the 99.9%, and the 99.9% confidence levels, respectively. However, the signs of these coefficients are consistent with expectations. If the median age in a ZIP Code increases by one standard deviation (3.91 years), then sale prices increase by $31,164.73. If median home value in a ZIP Code increases by one standard deviation ($173,515.11), then sale prices increase by $53,789.68. If the property is within a ZIP Code inside the Beltway, then sale prices increase by 19% ($115,845.98). Three-Level Model of Properties, Years, and ZIP Codes Descriptive statistics for three levels of 2000 to 2009 data appear in Exhibit 4. Coefficient estimates for linear (Price) and natural log (lnPrice) dependent variables in three-level models of properties nested within years nested within ZIP Codes appear in Exhibit 5. At the property level, as in the two-level model, all of the exterior (except parking), interior, quality, and subdivision independent variables are statistically significant at the 99.9% confidence level. Also, the signs and the magnitudes of all of the coefficients are consistent with expectations. One more floor increases sale prices by 4% ($9,114.33). If the property is detached, then sale prices increase by J O SRE Vol . 8 No. 1 – 2016 154 Z o lni k Exhibit 2  Descriptive Statistics for Two Levels of 2000 to 2009 Data Level Variables Mean Std. Dev. Min Max Property Dependent Price 458,801.41 273,031.70 89,000.00 3,200,000.00 lnPrice 12.89 0.62 1.00 5.00 Independent Exterior Floors 2.89 0.62 1.00 5.00 Parking (%) Yes 68.64 No 31.36 Type (%) Detached 64.20 Townhouse 35.80 Interior Baths—Full 2.34 0.78 1.00 5.00 Baths—Half 0.95 0.55 0.00 2.00 Bedrooms 3.68 0.85 1.00 6.00 Fireplaces 0.90 0.66 0.00 3.00 Quality Age (years) 28.49 18.83 0.00 83.00 New Yes 2.07 No 97.93 Subdivision TND Yes 1.49 No 98.51 T h e R esi lienc e o f t he P r e m i u m f o r H o m e s 155 J O SRE Vol . 8 No. 1 – 2016 Exhibit 2  (continued) Descriptive Statistics for Two Levels of 2000 to 2009 Data Level Variables Mean Std. Dev. Min Max ZIP Code Independent Demographic Households 8,214.32 6,814.26 86.50 24,220.70 White (%) 68.65 16.50 31.38 93.81 Median Age (years) 39.17 3.91 32.12 46.01 Economic Median Home Value ($) 491,266.86 173,515.11 249,624.86 847,488.57 Location Beltway (%) Inside 14.29 Outside 85.71 Distance to Central 16.37 7.16 5.40 30.59 Business District (miles) Distance to Nearest 5.25 4.10 0.58 15.32 METRO Station (miles) 156 Z o lni k Exhibit 3  Coefficient Estimates for Linear (Price) and Natural Log (lnPrice) Dependent Variables in Two-Level Model Price ($) lnPrice Level n Variables Coeff. t-Ratio Coeff. t-Ratio Property 78,513 Independent Exterior Floors 7,063.87 6.90*** 0.04 18.28*** Parking (Yes  1/No  0) 26,270.92 20.83*** 0.06 26.38*** Type (Detached  1 / Townhouse  0) 123,122.18 67.06*** 0.34 94.89*** Interior Baths—Full 104,371.29 97.78*** 0.18 85.29*** Baths—Half 44,021.46 34.17*** 0.07 26.62*** Bedrooms 22,133.80 22.68*** 0.03 17.37*** Fireplaces 46,451.23 44.07*** 0.10 47.49*** Quality Age 836.11 16.05*** 1E–03 12.45*** New (Yes  1/No  0) 94,177.15 21.63*** 0.07 7.86*** Subdivision TND (Yes  1/No  0) 111,593.77 21.02*** 0.28 27.54*** T h e R esi lienc e o f t he P r e m i u m f o r H o m e s 157 J O SRE Vol . 8 No. 1 – 2016 Exhibit 3  (continued) Coefficient Estimates for Linear (Price) and Natural Log (lnPrice) Dependent Variables in Two-Level Model Price ($) lnPrice Level n Variables Coeff. t-Ratio Coeff. t-Ratio Zip Code 42 Intercept 368,287.62 12.86*** 12.67 244.22*** Independent Demographic Households 1.50 0.93 3E–06 1.07 White 1,225.58 1.18 3E–03 1.35 Median Age 7,970.52 2.19* 0.02 2.63* Economic Median Home Value 0.31 3.95*** 0.00 3.34** Location Beltway (Inside  1 / Outside  0) 115,845.98 3.66*** 0.19 3.22** Distance to Central Business District 71.87 0.02 2E–03 0.28 Distance to Nearest METRO Station 736.50 0.14 0.01 0.57 Notes: * P  0.05. ** P  0.01. *** P  0.001. 158 Z o lni k Exhibit 4  Descriptive Statistics for Three Levels of 2000 to 2009 Data Level n Variables Mean Std. Dev. Min Max Property 78,513 Dependent Price 458,801.41 273,031.70 89,000.00 3,200,000.00 lnPrice 12.89 0.53 11.40 14.98 Independent Exterior Floors 2.89 0.62 1.00 5.00 Parking (%) Yes 68.64 No 31.36 Type (%) Detached 64.20 Townhouse 35.80 Interior Baths—Full 2.34 0.78 1.00 5.00 Baths—Half 0.95 0.55 0.00 2.00 Bedrooms 3.68 0.85 1.00 6.00 Fireplaces 0.90 0.66 0.00 3.00 Quality Age (years) 28.49 18.83 0.00 83.00 New Yes 2.07 No 97.93 Subdivision TND Yes 1.49 No 98.51 T h e R esi lienc e o f t he P r e m i u m f o r H o m e s 159 J O SRE Vol . 8 No. 1 – 2016 Exhibit 4  (continued) Descriptive Statistics for Three Levels of 2000 to 2009 Data Level n Variables Mean Std. Dev. Min Max Year 408 Independent Time Year (%) 2000 8.65 2001 9.90 2002 10.33 2003 10.95 2004 12.61 2005 12.32 2006 10.80 2007 8.44 2008 7.23 2009 8.77 ZIP Code 42 Independent Demographic Households 8,214.38 6,841.25 87.00 24,221.00 White (%) 68.66 16.50 31.40 93.80 Median Age (years) 39.17 3.92 32.10 46.00 Economic Median Home Value ($) 491,266.90 173,515.14 249,625.00 847,489.00 Location Beltway (%) Inside 19.05 Outside 80.95 Distance to Central 16.37 7.25 5.40 30.59 Business District (miles) Distance to Nearest 5.25 4.15 0.58 15.32 METRO Station (miles) 160 Z o lni k Exhibit 5  Coefficient Estimates for Linear (Price) and Natural Log (lnPrice) Dependent Variables in Three-Level Model Price ($) lnPrice Level n Variables Coeff. t-Ratio Coeff. t-Ratio Property 78,513 Independent Exterior Floors 9,114.33 11.68*** 0.04 37.51*** Parking (Yes  1/No  0) 1,735.14 1.79 4E–03 2.75** Type (Detached  1 / Townhouse  0) 150,651.10 107.28*** 0.40 198.21*** Interior Baths—Full 84,165.71 102.49*** 0.13 105.86*** Baths—Half 55,838.58 56.68*** 0.09 63.47*** Bedrooms 21,290.77 28.67*** 0.03 27.71*** Fireplaces 52,747.00 65.66*** 0.11 96.41*** Quality Age 2,564.84 62.57*** 6E–03 94.41*** New (Yes  1/No  0) 110,496.39 32.78*** 0.10 21.06*** Subdivision TND (Yes  1/No  0) 76,819.54 18.99*** 0.19 32.83*** Year 408 Independent Time Year 2000 228,898.70 28.33*** 0.59 48.10*** 2001 181,517.38 22.55*** 0.47 38.55*** 2002 132,712.71 16.56*** 0.32 26.21*** 2003 79,139.64 9.87*** 0.18 14.90*** 2004 Referent Referent 2005 91,330.52 11.50*** 0.17 14.23*** 2006 100,605.05 12.60*** 0.20 16.41*** 2007 85,009.21 10.58*** 0.17 13.78*** 2008 26,078.92 3.23 0.04 2.98** 2009 17,689.49 2.21* 0.07 5.61*** T h e R esi lienc e o f t he P r e m i u m f o r H o m e s 161 J O SRE Vol . 8 No. 1 – 2016 Exhibit 5  (continued) Coefficient Estimates for Linear (Price) and Natural Log (lnPrice) Dependent Variables in Three-Level Model Price ($) lnPrice Level n Variables Coeff. t-Ratio Coeff. t-Ratio Zip Code 42 Intercept 412,856.45 44.66*** 12.77 881.98*** Independent Demographic Households 2.17 1.48 0.00 1.99 White 1,270.72 1.34 3E–03 1.76 Median Age 8,043.23 2.44* 0.02 3.26 Economic Median Home Value 0.33 3.66*** 1E–06 4.57*** Location Beltway (Inside  1 / Outside  0) 118,006.81 4.08*** 0.19 4.21*** Distance to Central Business District 2,242.28 0.69 4E–03 0.72 Distance to Nearest METRO Station 28.76 0.01 4E–03 0.46 Notes: Referent is the year (2004) with the highest percentage of closings from 2000 to 2009 (12.61%). * P  0.05. ** P  0.01. *** P  0.001. 162 Z o lni k 40% ($150,651.10). One more full bath, half bath, bedroom, and fireplace increase sale prices by 13% ($84,165.71), 9% ($55,813.58), 3% ($21,290.77), and 11% ($52,747.00), respectively. One more age-year decreases sale prices by 0.6% ($2,564.84). If the property is less than one year old, then sale prices increase by 10% ($111,496.39). If the property is in a TND, then sale prices increase by 19% ($76,819.54). At the time level, all of the year dummy variables except for 2008 are statistically significant at the 99.9% confidence level. Also, the changes in the signs and the magnitudes of the coefficients are consistent with expectations. Sale prices increase from a trough in 2000 when sale prices are 59% ($228,898.70) lower than in the referent year of 2004 to a peak in 2006 when sale prices are 20% ($100,605.05) higher than in the referent year of 2004. After 2008, sale prices are again 7% ($17,689.49) lower than in the referent year of 2004. At the ZIP Code level, only one of the economic and only one of the location independent variables are statistically significant at the 99% and the 95% confidence levels, respectively. However, the signs of these coefficients are consistent with expectations. If median home value in a ZIP Code increases by one standard deviation ($24,354.94), then sale prices increase by $9,657.95. If the property is within a ZIP Code inside the Beltway, then sale prices increase by 17% ($126,726.80). Discussion Relating the results to the research questions in the study and linking the results to the literature on the valuation of homes in TNDs, the TND premium for homes in the Kentlands and in the Lakelands is resilient over the long term, notwithstanding the Great Recession. The mean TND premium for homes in the Kentlands and in the Lakelands ranges from 28% ($111,593.77) to 19% ($76,819.54) from the two-level to the three-level model specification, respectively. Regardless of the model specification, the results for the older Kentlands and for the newer Lakelands exceed the highest mean premiums for homes in the Kentlands from the literature; 13% from Eppli and Tu (1999) and 12% from Tu and Eppli (1999). Premiums are probably higher in the Kentlands and in the Lakelands versus the surrounding conventional neighborhoods because the former homes are larger in size in terms of number of full baths, number of half baths, number of fireplaces, and number of levels according to nonparametric and parametric test statistics. However, if the actual square footage of improvement is an independent variable to capture the effect of interior size on sale prices, premiums are not likely to be as great. Premiums are also probably higher in the Kentlands and in the Lakelands versus the surrounding conventional neighborhoods because a higher percentage of the former homes are new (5.23% vs. 2.03%). This follows from the finding that the characteristics of single-family homes that have the largest impact on the mean premium over the ten-year time period of transactions are, in order from largest effects to smallest effect, exterior effects (if the property is detached), T h e R esi lienc e o f t he P r e m i u m f o r H o m e s 163 interior effects (number of full baths), and quality effects (if the property is less than one year old at closing). Finally, premiums are probably higher in the Kentlands and in the Lakelands versus the surrounding conventional neighborhoods because the regional boom in home prices in the Washington, DC real estate market coincides with the first seven years—2000 to 2006—of the ten- year time period of transactions in the study. This follows from the finding that the mean closing price for homes in the Kentlands and in the Lakelands versus the surrounding conventional neighborhoods is higher in the former ($577,042.06 vs. $457,017.39). The TND home premium is probably lower in the three-level model specification than in the two-level model specification (19% vs. 28%, respectively) because in the three-level model specification, I explicitly control for the temporal effects on sale prices from 2000 to 2006 to 2009 when sale prices were down (59%), up (20%), and down again (7%) relative to the referent year of 2004. The two-level model specification distributes the year-level variance to the property and the ZIP Code levels. Nevertheless, a two-level model specification is statistically more valid than a cross-sectional approach where time is not a covariate or a level of analysis. The latter approach is not statistically valid because the subdivision independent variables in model specifications for 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, and 2009, respectively, are not comparable. Conclusion One contribution of the study to the TND literature is the resiliency of the premium buyers willingly pay for properties in TND neighborhoods. To that end, the study answers an important question in the TND literature on the long-term demand for what are known to be more expensive homes particularly when markets are slow and credit is tight. Another contribution of the study to the real estate valuation literature is the multilevel model specifications, which accurately contextualize property sales transactions in time and space. Indeed, the results indicate that the premiums for homes in TND neighborhoods are robust to the temporal and spatial differences in property sales transactions between TND neighborhoods and conventionally- developed neighborhoods. Overall, the specifications contribute to the trend towards methodological sophistication in the real estate valuation literature. To that end, the specifications show that time and space both enhance the accuracy of valuation models. One of the most fruitful avenues for future research on the resiliency of TND premiums is to use data on property sales transactions for more years and for more neighborhoods. One example, the ten-year time period of property sales transactions includes data from before, during, and after the Great Recession, but the after time period is short. In order to accurately estimate the effect of the Great Recession on property sales transactions, the time period after the Great Recession ought to include more post-recession years. Another example, the study area of J O SRE Vol . 8 No. 1 – 2016 164 Z o lni k Montgomery County, Maryland is on the fringe of the METRO system which, at least, partially explains why the effect of the locational independent variable METRO is not statistically different from zero in the two- and three-level model specifications, respectively. In order to accurately estimate the effect of METRO on property sales transactions, the study area ought to include property sales transactions for all of the county and county equivalent areas where there are METRO stations. The results clearly demonstrate to developers, investors, and lenders that buyers are willing to pay a premium for single-family homes in new urbanist neighborhoods even in the midst of a market downturn with tightening credit. Appendix The specification of the two-level model with a property ( p) level and a ZIP Code (z) level is as follows. Within each ZIP Code, closing prices are modeled as a function of property-level independent variables plus a property-level error term: Y     W  ...   W  r , (A1) pz 0z 1z 1pz Az Apz pz where Y is the closing price of property p in ZIP Code z;  is the y-intercept pz 0z term for ZIP Code z;  are a  1, ..., A property-level coefficients; W are a Az Apz 1, ..., A property-level independent variables; and r is the property-level pz random effect term. All of the regression coefficients at the property level are fixed; that is, invariant over ZIP Code z. A multilevel model in which the y-intercept is random and all of the regression coefficients at level-1 are fixed is known as a random-intercepts model. The model for variation between ZIP Codes is therefore as follows. For the ZIP Code effect  : 0z Z  ...   Z  u , (A2) 0z 00 01 1z 0BBz 0z where  is the y-intercept term for ZIP Code z;  are b  1, ..., B ZIP Code- 00 0B are b  1, ..., B ZIP Code-level independent variables; and level coefficients; Z Bz u is a ZIP Code-level random effect term. 0z The specification of the three-level model with a property ( p) level, a year ( y) level, and a ZIP Code (z) level is as follows. Within each ZIP Code, closing prices are modeled as a function of property-level independent variables plus a property- level error term: T h e R esi lienc e o f t he P r e m i u m f o r H o m e s 165 Y     W  ...   W  r , (A3) pyz 0yz 1yz 1pyz Ayz Apyz pyz where Y is the closing price of property p in year y and ZIP Code z;  is the pyz 0yz y-intercept term for year y in ZIP Code z;  are a  1, ..., A property-level Ayz coefficients; W are a  1, ..., A property-level independent variables; and r Apyz pyz is the property-level random effect term. Like the two-level model, the three-level model is a random-intercepts model; that is, all of the regression coefficients at the property level and at the year level are fixed. The model for variation among years within ZIP Codes is therefore as follows: X  ...   X  e , (A4) 0yz 00z 1z 1yz (C1)z (C1)yz 0yz where  is the y-intercept term for ZIP Code z;  are c  1, ..., C  1 00z (C1)z year-level coefficients; X are c  1, ..., C  1 dummy variables; and e is (C1)yz 0yz the year-level random effect term. The model for variation between ZIP Codes is as follows. For ZIP Code effect 00z Z  ...   Z  u , (A5) 00z 000 001 1z 00BBz 00z where  is the y-intercept term for ZIP Code z;  are b  1, ..., B ZIP Code- 000 00B level coefficients; Z are b  1, ..., B ZIP Code-level independent variables; and Bz u is a ZIP Code-level random effect term. 00z References Brigham, E. The Determinants of Residential Land Values. Land Economics, 1965, 41:4, 325–34. Congress for the New Urbanism. New Urbanism: It Just Performs Better. Chicago, IL: Congress for the New Urbanism, 2014. Crecine, J., O. Davis, and J. Jackson. Urban Property Markets: Some Empirical Results and Their Implications for Municipal Zoning. Journal of Law and Economics, 1967, 10, 79–99. Dong, H. Were Home Prices in New Urbanist Neighborhoods More Resilient in the Recent Housing Downturn? Journal of Planning Education and Research, 2015, 35:1, 5–18. Environmental Systems Research Institute. Community Sourcebook of Zip Code Demographics. Redlands, CA: Environmental Systems Research Institute, 2000. J O SRE Vol . 8 No. 1 – 2016 166 Z o lni k ——. Community Sourcebook of Zip Code Demographics. Redlands, CA: Environmental Systems Research Institute, 2001. ——. Sourcebook of Zip Code Demographics. Redlands, CA: Environmental Systems Research Institute, 2002. ——. The Sourcebook of Zip Code Demographics. Vienna, VA: Environmental Systems Research Institute, 2003. ——. Community Sourcebook of Zip Code Demographics. Vienna, VA: Environmental Systems Research Institute, 2004. ——. Community Sourcebook of Zip Code Demographics. Redlands, CA: Environmental Systems Research Institute, 2005. ——. Community Sourcebook of Zip Code Demographics. Redlands, CA: Environmental Systems Research Institute, 2006. ——. Community Sourcebook of Zip Code Demographics. Redlands, CA: Environmental Systems Research Institute, 2007. ——. Community Sourcebook of Zip Code Demographics. Redlands, CA: Environmental Systems Research Institute, 2008. ——. Sourcebook of Zip Code Demographics. Redlands, CA: Environmental Systems Research Institute, 2009. Eppli, M. and C. Tu. Valuing the New Urbanism: The Impact of the New Urbanism on Prices of Single-Family Homes. Washington, DC: Urban Land Institute, 1999. Follain, J. and S. Malpezzi. Dissecting Housing Value and Rent: Estimates of Hedonic Indexes for Thirty-Nine Large SMSAs. Washington, DC: Urban Institute, 1980. Fulton, W. The New Urbanism: Hope or Hype for American Communities? Cambridge, MD: The Lincoln Institute of Land Policy. 1996. Garde, A. Designing and Developing New Urbanism Projects in the United States: Insights and Implications. Journal of Urban Design, 2006, 11:1, 33–54. Guttery, R. The Effects of Subdivision Design on Housing Values: The Case of Alleyways. Journal of Real Estate Research, 2002, 23:3, 265–73. Gyourko, J. and W. Rybczynski. Financing New Urbanism Projects: Obstacles and Solutions. Housing Policy Debate, 2000, 11:3, 733–50. Jones, K. and N. Bullen. A Multi-level Analysis of the Variations in Domestic Property Prices: Southern England, 1980–87. Urban Studies, 1993, 30:8, 1409–26. ——. Contextual Models of Urban House Prices: A Comparison of Fixed- and Random- Coefficient Models Developed by Expansion. Economic Geography, 1994, 70:3, 252–72. Krause, A. and C. Bitter. Spatial Econometrics, Land Values and Sustainability: Trends in Real Estate Valuation Research. Cities, 2012, 29:S2, S19–S25. Lang, R. Valuing the Suburbs: Why Some ‘Improvements’ Lower Home Prices. Opolis, 2005, 1:1, 5–12. Mahan, B., S. Polasky, and R. Adams. Valuing Urban Wetlands: A Property Price Approach. Land Economics, 2000, 76:1, 100–13. Metropolitan Regional Information Systems. Vital Statistics. Available at: http: / / www.mris.com / about-mris / vital-statistics / , 2013. Planning Design Group. Economic Return on New Urbanism: A Summary of Focus Group Discussions of Developers and Practitioners of New Urbanism in Central Florida. Orlando, FL: Planning Design Group, 2007. Plaut, P. and M. Boarnet. New Urbanism and the Value of Neighborhood Design. Journal of Architectural and Planning Research, 2003, 20:3, 254–65. T h e R esi lienc e o f t he P r e m i u m f o r H o m e s 167 Raudenbush, S. and A. Bryk. Hierarchical Linear Models: Applications and Data Analysis Methods. Thousand Oaks, CA: Sage, 2002. Rauterkus, S. and N. Miller. Residential Land Values and Walkability. Journal of Sustainable Real Estate, 2011, 3:1, 23–43. Reuter, F. Externalities in Urban Property Markets: An Empirical Test of the Zoning Ordinance of Pittsburgh. Journal of Law and Economics, 1973, 16:2, 313–49. Ryan, B. and R. Weber. Valuing New Development in Distressed Urban Neighborhoods: Does Design Matter? Journal of the American Planning Association, 2007, 73:1, 100–11. Shiller, R. Irrational Exuberance. Princeton, NJ: Princeton University Press, 2015. Song, Y. and G. Knaap. New Urbanism and Housing Values: A Disaggregate Assessment. Journal of Urban Economics, 2003, 54:2, 213–38. Song, Y. and R. Quercia. How are Neighbourhood Design Features Valued across Different Neighbourhood Types? Journal of Housing and the Built Environment, 2008, 23:4, 297– Steuteville, R. Kentlands / Lakelands. Ithaca, NY: New Urban News Publications, 2010. Stull, W. Community Environment, Zoning, and the Market Value of Single-Family Homes. Journal of Law and Economics, 1975, 18:2, 535–57. Tu, C. and M. Eppli. Valuing New Urbanism: The Case of Kentlands. Real Estate Economics, 1999, 27:3, 425–51. ——. An Empirical Examination of Traditional Neighborhood Development. Real Estate Economics, 2001, 29:3, 485–501. The author wishes to thank the Center for Regional Analysis at George Mason University for assistance in data acquisition for this study. All articles published in JOSRE are distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Edmund Zolnik, George Mason University, Arlington, VA 22201 or ezolnik@ gmu.edu. J O SRE Vol . 8 No. 1 – 2016

Journal

Journal of Sustainable Real Estate – Taylor & Francis

Published: Nov 1, 2016

There are no references for this article.