What Is a Tree Worth? Green-City Strategies, Signaling and Housing Prices

Transcription

1 2008 V36 2: pp REAL ESTATE ECONOMICS What Is a Tree Worth? Green-City Strategies, Signaling and Housing Prices Susan M. Wachter and Grace Wong We investigate the correlation between curbside tree plantings and housing price movements in Philadelphia from 1998 to 2003, comparing two programs, one by the Philadelphia Horticultural Society (PHS) that requires block-group effort that focuses on low-income neighborhoods and the other by the Fairmount Park Commission that is individual-based without specific target areas. A 7 to 11% price differential is identified within 4,000 feet of the Fairmount tree plantings. We argue that this is largely driven by either social capital creation or a signaling mechanism, on top of an intrinsic tree value (around 2%). Findings using the PHS tree program suggest that development of social capital or environmentally conscious behavior might be a less important channel. Any positive changes brought by the PHS tree plantings were not detected with sufficient statistical power. Greening projects are generally viewed as a positive development in the neighborhood. Of all 32 elected mayors in the largest 100 metropolitan cities in 2005, 15 of them mentioned a greening program in their electoral platform. Environmental initiatives such as tree-planting programs that make environmentally sound behaviors easier to engage in and their economic returns more apparent may be more effective than public education efforts that focus on changing attitudes (Summit and Sommer 1997). While there has been some quantitative analysis on the value of trees in the urban planning literature (Anderson and Cordell 1985, 1988), complemented by survey evidence that inner city residents factor in the availability of trees in their residence location decisions (Getz, Karow and Kielbaso 1982), little has been said about the direct causal impact of greening programs on neighborhood desirability or housing values. In addition to their intrinsic value, greening programs might have a more interesting relationship with housing value by being a proxy for social and physical capital of housing in the neighborhood. For example, whether trees are planted on the curbside depends on whether the local residents desire trees and are able to follow through with the process of obtaining them. This endogenous process therefore reflects certain characteristics of the residents, such as low residential University of Pennsylvania, Philadelphia, PA or wachter@wharton.upenn.edu. University of Pennsylvania, Philadelphia, PA or wongg@wharton.upenn.edu. C 2008 American Real Estate and Urban Economics Association

2 214 Wachter and Wong mobility and civil participation, which might correlate with other characteristics such as vigilance on crime. DiPasquale and Glaeser (1999) provide evidence that, for example, homeownership correlates with a propensity to invest in local amenities and social capital. If these characteristics were not known to homebuyers (or the econometrician) before the plantings, then the tree plantings can be viewed as a signaling event. Examples include the degree of coercion among neighbors or details of the house condition. While this article primarily focuses on the empirical identification of signaling of resident characteristics in the housing market, it alludes to the productive signaling equilibria in Ben-Shahar (2004). In contrast to Spence (1973), Ben-Shahar (2004) outlines the conditions under which signaling can be productive. If the characteristics proxied by tree plantings were capitalized in housing prices but unobserved by the econometrician before the plantings, such as the general interior conditions of the house, controlling for the tree plantings in our price analysis alleviates an omitted variable bias, which can be in either direction. 1 Therefore the housing price differentials coincidental both in time and space with the tree plantings which we consider to be directly attributable to the plantings can be interpreted as compensating differentials in the housing market comprising the following: the intrinsic value of the trees, an omitted variable bias and a signaling of previously unknown characteristics of the residents. By using a geographic information system, we are able to map out positions of all new tree plantings and housing sales in Philadelphia between 1998 and This allows for a comparative study by both distances to tree plantings and by when the transaction took place relative to the tree planting. This means any confounding factor that affects housing prices at either a different time or location is removed from our estimate of the tree-related price differential. We concede that there still exists potential omitted variable bias in our tree-planting compensating differential and discuss it in detail in our empirical analysis. Our empirical strategy separates this article from much of the literature, such as Fu (2005) and Middleton, Murie and Groves (2005), which estimate the contemporaneous correlation of housing prices and characteristics such as racial diversity and neighborhood association membership. Our approach allows us to rule out factors that confound the social capital impact unless they coincide with the changes in social capital in both time and space. On a related note, 1 We consider factors observed by homebuyers prior to tree planting but not captured in our econometric framework as omitted variables. For example, a tree planting might be a result of the residents willingness to take care of the environment (positive, reverse causality), a need for cosmetic improvements to a rundown neighborhood (negative, reverse causality) or heterogeneous tastes (either direction, statistical correlation).

3 What Is a Tree Worth? Green-City Strategies, Signaling and Housing Prices 215 we perform a within-neighborhood (census tract) analysis to control for time effects and other city-wide unobservables in housing price movements. This means any city-wide housing price impact due to tree plantings will not be captured in our estimation. Our article further focuses on one interesting dimension of the resident characteristics the ability of residents to coordinate with each other to achieve an aim or an increase in social awareness that benefits other residents. Throughout this article we refer to this as social capital. Other components of social capital such as peer effects are less likely to correlate with tree plantings. Because the correlation of the tree plantings with social capital depends on the process of the particular program, we study two nonprofit tree-planting programs in Philadelphia, organized by the Fairmount Park Commission and the Philadelphia Horticultural Society (Fairmount and PHS programs henceforth), with contrasting aims and approaches (see the next section). We attempt to disentangle these explanations by carefully constructing the treatment and control groups. Primarily we investigate the tree plantings by the Fairmount Park Commission, which responds to requests of tree by individual Philadelphian city residents. A comparison between parcels sold within 1,000 feet of a between-sales Fairmount planting with other sales reveals a statistically significant planting-related premium between 7% and 11%. By comparing parcels along the propensity of ever getting a tree, we argue that the omitted variable bias is around 2%. Therefore 2 percentage points of the 7% to 11% premium can be attributed to the value of the trees. We find evidence for a significant and stable price differential within 4,000 feet of the planting, which is relatively widespread compared to the 1,000-feet impact of public community gardens identified by Been and Voicu (2006). To put this in context, the estimated price impact of a garage in our analysis is around 10 percentage points. This suggests significant economic importance of the characteristics of residents signaled by the tree plantings. 2 In order to shed some light on the importance of social capital among resident characteristics revealed by tree, we also explore data of another tree program in Philadelphia. The green city program under the PHS is neighborhood based and requires collective action by residents on the same block. It involves neighbors filling in paperwork and attending meetings together. If revelation or creation of social capital were the main driver behind the positive price change related to the Fairmount Park Commission tree plantings and if other previously unobserved 2 We thank the anonymous referee for pointing this out.

4 216 Wachter and Wong characteristics play a relatively small role, then we would expect a significant and positive price change related to a PHS planting, of a magnitude larger than what is found using Fairmount trees. When compared to all other housing sales in the city, houses within 1,000 feet of a PHS planting were sold at a discount. No strong evidence is found for positive changes related to PHS plantings, pointing to small effects, if any, brought by those plantings. One explanation is that the Fairmount tree-related price differential is mainly due to signaling of resident characteristics other than social capital or that other previously unobserved resident characteristics signaled through the PHS plantings are of a negative nature, given the aim of the PHS program to create a better living environment for struggling neighborhoods. In the latter case any positive impact of social capital is overcome by the negative attributes revealed to homebuyers at the same time. Another possibility is a lagged response to signaling in less affluent neighborhoods, which is not studied in this article due to data limitations. Similarly, the differential impact of the tree planting programs by various initial conditions of the neighborhoods for example, income, education and neighborhood coercion is not covered by this study. Tree-Planting Programs in Philadelphia This article studies the two different programs through which Philadelphia residents can apply for street (curbside) tree plantings. For various reasons, including potential noise production and damage of underground utility lines, residents are not allowed to plant street trees on their own. All tree plantings analyzed in this article are street plantings of young saplings. The Fairmount Park Commission tree-planting program allows approximately 900 individual homeowners throughout the city of Philadelphia to apply for street trees annually. Individual local residents initiate the application process by filling in a one-page form. There is generally a year-long wait between the receipt of the application and the tree planting, not including a waiting list for initial processing of the application. With the support of the city s Capital Program, trees are provided free of charge after approval and are distributed without preference throughout Philadelphia. The PHS is also the proprietor of a street tree-planting program. To initiate a tree planting under the PHS program, a local resident has to volunteer as the coordinator and collect signatures from other residents living on the same block. Typically 20 to 25 signatures are collected for each application. First, the PHS evaluates if there are at least 10 feasible tree sites on that block to create a critical mass. The feasibility analysis is to make sure that the tree planting will not negatively impact preexisting infrastructure, including utility lines or fire hydrants. Afterward the PHS arranges a meeting that applicants need to attend

5 What Is a Tree Worth? Green-City Strategies, Signaling and Housing Prices 217 and show reasonable support. The level of support is subjectively determined by the PHS agent; a sparse turnout (2 3 people) will cause the application to be put on hold. There is no fee to utilize the tree-planting program. Upon approval of their application, applicants undergo training and commit as neighborhood tree caretakers (called Tree Tenders ). The street tree programs facilitated by the Fairmont Park Commission and the PHS cater equally to the residents of Philadelphia but promote different levels of community involvement. Although the Fairmount tree program entails a delayed and competitive application process, all aspects of tree planting, including expenses, are managed at the organizational level. The PHS program involves individuals participating as a group and assuming much of the treeplanting and tree-caring responsibility. Data The housing price data set at hand has been filtered in order to include only arm s-length transactions. All transactions with the price of $500 or below were dropped, as were transactions where the seller and buyer was the same person. 3 Moreover, residential properties outside Philadelphia (property ZIP code not starting with 191 ) or those sold by the government, HUD or the sheriff were discarded. About 1,151 Fairmount tree plantings and 1,668 PHS tree plantings were merged by location to the housing sales data set. Table 1 shows the summary statistics for the full sample and those housing sales with a Fairmount or PHS tree planting within 100 to 500 feet between sales. The percentage of parcel sales that took place in proximity of a tree planting is shown in Table 2 by the relative timing of the sales and the plantings. Relative to all houses being sold, parcels situated within 100 to 500 feet within either type of tree plantings between sales are smaller in size. While houses near PHS plantings are more likely to be adjacent to vacant lots, the opposite is true for those near Fairmount plantings. The same difference in housing characteristics by the type of tree planting is also reflected by the sales price from the previous sale transaction. From a comparison of the differences between current and previous sales prices, it is evident that Fairmount plantings took place in fast-growing neighborhoods while PHS plantings typically occurred in areas with below-average price growth. This suggests systematic pretree planting differences that need to be controlled for in the price analysis. 3 The $500 cutoff was applied to eliminate extreme outliers. Alternative cutoff points ($1,000 and $3,000) were experimented with, but they do not affect the core results.

6 218 Wachter and Wong Table 1 Summary statistics Variable Obs Mean Std. Dev. Transaction Details Price 207,565 63, , Year of sale 207,565 1, Month of sale 207, Time gap between this and the previous 77, sale ( 000 days) Number of transactions per year 207, Parcel Structural Characteristics Square footage of lot 207,565 2, , Square footage of building 207,565 1, Number of floors in the property 207, Exterior Condition (0 = worst, 207, = best) = 1 if property is detached (single 207, house) = 1 if property is a row house 207, = 1 if property is semi-detached (e.g., 207, twin, duplex) = 1 if property has a miscellaneous 207, classification = 1 if property has an irregular 207, (non-rectangular) shape = 1 if property is above street level 207, = 1 if property has a nice view 207, = 1 if property originally had more 207, than one dwelling = 1 if property has central air 207, conditioning = 1 if property has qualitative amenities 207, = 1 if property has new improvements 207, = 1 if property is on a corner 207, = 1 if property has a garage 207, = 1 if parcel is adjacent to a vacant lot 207, = 1 if parcel borders a park 207, = 1 if a parcel is within 1 / 4 mile of a park 207, Number of fireplaces in the property 207, Floor area ratio of the property 207, Empirical Evidence Fairmount Plantings To investigate price movements correlated with tree plantings, the following regression is performed: P it = α + µp it 1 + βtree it + γ H i + C i + ξgap it + σ (P it 1 Gap it ) + Time t + ε it (1)

7 What Is a Tree Worth? Green-City Strategies, Signaling and Housing Prices 219 Table 2 Summary statistics tree plantings Obs Mean Std. Dev. Fairmount Tree Plantings % of parcels with a Fairmount tree planting within 100 ft between sales % of parcels with a Fairmount tree planting within 100 ft prior to the sale % of parcels that have ever had a Fairmount tree planting within 100 ft % of parcels with a Fairmount tree planting within 500 ft between sales % of parcels with a Fairmount tree planting within 500 ft prior to the sale % of parcels that have ever had a Fairmount tree planting within 500 ft % of parcels with a Fairmount tree planting within 1000 ft between sales % of parcels with a Fairmount tree planting within 1000 ft prior to the sale % of parcels that have ever had a Fairmount tree planting within 1000 ft % of parcels with a Fairmount tree planting within 2000 ft between sales % of parcels with a Fairmount tree planting within 2000 ft prior to the sale % of parcels that have ever had a Fairmount tree planting within 2000 ft % of parcels with a Fairmount tree planting within 4000 ft between sales % of parcels with a Fairmount tree planting within 4000 ft prior to the sale % of parcels that have ever had a Fairmount tree planting within 4000 ft PHS Tree Variables % of parcels with a PHS tree planting within 100 ft between sales % of parcels with a PHS tree planting within 100 ft prior to the sale % of parcels that have ever had a PHS tree planting within 100 ft % of parcels with a PHS tree planting within 500 ft between sales % of parcels with a PHS tree planting within 500 ft prior to the sale % of parcels that have ever had a PHS tree planting within 500 ft 207, , , , , , , , , , , , , , , , , , , , ,

8 220 Wachter and Wong Table 2 continued Obs Mean Std. Dev. % of parcels with a PHS tree planting within 1000 ft between sales % of parcels with a PHS tree planting within 1000 ft prior to the sale % of parcels that have ever had a PHS tree planting within 1000 ft % of parcels with a PHS tree planting within 2000 ft between sales % of parcels with a PHS tree planting within 2000 ft prior to the sale % of parcels that have ever had a PHS tree planting within 2000 ft % of parcels with a PHS tree planting within 4000 ft between sales % of parcels with a PHS tree planting within 4000 ft prior to the sale % of parcels that have ever had a PHS tree planting within 4000 ft 207, , , , , , , , , where P it and P it 1 are the real sales price and the previous real sales price in logs, α a constant term, H i a group of housing characteristics, C i the censustract fixed effects, Gap it a measure of time lag between current and previous sales and ε it an error term clustered at the census-tract level. 4 Time t stands for a group of time controls: year fixed effects, quarter fixed effects and a set of census tract-specific linear time trends. By controlling for census-tract fixed effects, all comparisons are within tracts. Any time-invariant factors are absorbed by lagged price P it 1. By regressing on sales prices while controlling for lagged price and the gap between sales, instead of regressing on a rate of price change, we allow for more flexibility in functional form. Table 3 makes a simple comparison between sales of parcels with a Fairmount tree planting within 1,000 feet either before or after sales with all other parcels. Columns 1 and 2 suggest that, relative to all other sales in Philadelphia, controlling for housing characteristics and seasonal effects as described above, sales price for parcels that were sold within 1,000 feet of previous or future Fairmount tree plantings are 2.30% higher. Remarkably, this differential does not diminish significantly when those sales are compared to sales within 4,000 feet of 4 Housing characteristics included in regressions presented in this article are those that demonstrate a stable and robust relationship with the dependent variable. Adding other controls decreases precision of estimates but does not change the main conclusions.

9 What Is a Tree Worth? Green-City Strategies, Signaling and Housing Prices 221 Table 3 Time-space comparison of real housing prices: Fairmount tree plantings within 1000 feet before or after sale =1 if there had ever been a tree planting within 1000 ft of the parcel (before or after sales) Log real price of parcel Dependent Variable: Log Real Price of Parcel =1 If There Had Ever =1 If There Had Ever Been a Tree Planting Been a Tree Planting within 4000 ft of the within 2000 ft of the Parcel (before or Parcel (before or Whole Sample after Sales) after Sales) (1) (2) (3) (4) (5) (6) (0.006) (0.006) (0.006) (0.006) (0.006) (0.006) (0.003) (0.004) (0.003) (0.004) (0.004) (0.005) at previous sale Log lot area (0.006) (0.006) (0.006) (0.006) (0.007) (0.007) Log floor area (0.010) (0.010) (0.010) (0.010) (0.011) (0.011) Number of stories in the parcel (0.006) (0.006) (0.006) (0.006) (0.007) (0.006) =1 if parcel has a frame exterior (0.015) (0.015) (0.016) (0.016) (0.019) (0.018) =1 if parcel has a stone exterior (0.016) (0.016) (0.017) (0.017) (0.018) (0.018) =1 if parcel has non-rectangular shape (0.011) (0.011) (0.011) (0.011) (0.012) (0.012) =1 if parcel has central air (0.053) (0.053) (0.054) (0.053) (0.061) (0.060) =1 if parcel has garage (0.007) (0.007) (0.007) (0.007) (0.007) (0.007) =1 if property is detached (0.011) (0.011) (0.012) (0.012) (0.013) (0.013) Time gap between this and previous sales (0.005) (0.005) (0.005) Interaction term between log real previous sale price and time gap (1.257) (1.279) (1.402) Year Dummies Yes Yes Yes Yes Yes Yes Quarter Dummies Yes Yes Yes Yes Yes Yes Trend Dummies No Yes No Yes No Yes Observations R-squared Standard errors in parentheses. Significant at 10%; significant at 5%; significant at 1%. previous or future Fairmount tree plantings and is reduced slightly when compared to sales within 2,000 feet of previous or future Fairmount tree plantings (columns 3 to 6). This suggests a baseline, within-tract price differential around 2% on average due to unobserved factors correlated to the propensity of ever

10 222 Wachter and Wong getting tree plantings within 1,000 feet. 5 This gives us an idea of the magnitude of the selection bias related to the Fairmount tree plantings. When the treatment group is defined as all parcel sales that occurred after a Fairmount tree planting within 1,000 feet, any price differential can be interpreted as a combination of the following: an omitted variable bias/proxy effect of the plantings and a signaling of physical and social capital, which coincide with the tree plantings in both space and time. The physical and social capital improvements can either be a real change in the quality of the parcel (e.g., the intrinsic value of the trees and promotion of environmentally conscious behavior) or a third factor effect/omitted variable bias (e.g., an exogenous increase in awareness that led to tree plantings and possibly other positive neighborhood changes). Table 4 presents the results. Relative to all other sales in the same census tract, parcel sales after a Fairmount tree planting within 1,000 feet were priced between 9 and 11% higher (columns 1 and 2). When compared to other parcel sales that also occurred after a Fairmount tree planting but were further away (2,000 feet and 4,000 feet, columns 3 to 6), this effect becomes close to zero and insignificant. This implies that the main factors that drive the 9 to 11% price differential work at a level above the immediate neighborhoods. A comparison of parcels with various distances from tree plantings instead of all parcels within the city boundaries by definition controls for all price determinants that are uniform to those parcels. This implies that, if the measured impact in columns 1 and 2 were wholly due to third factors other than tree plantings that are common to all parcels within 2,000 or 4,000 feet of the tree plantings, one would also expect the impact to diminish to zero in columns 3 to 6. However, the measured correlation using an after-planting indicator with prices in Table 4 is substantially higher than the selection bias estimated in Table 3. This supports the idea of a positive and relatively constant tree-related impact on prices within 4,000 feet. Compared to the differential found in Table 3, this differential is less likely to be driven by other factors unrelated to tree plantings because those factors would have to coincide with the tree plantings not only in space but also in time. Notably, it is also significantly larger in size. This lends support to the notion that either the tree plantings themselves or the drivers behind the tree plantings (e.g., a signaling motivation) led to price increases. One interesting aspect of the tree-related impact is the variation of the impact strength by proximity. In the case of most other microneighborhood amenities, 5 In results not shown, comparing before-planting sales of parcels that eventually had a planting within 1,000 feet and sales of parcels that never had a planting within 1,000 feet gives the same, statistically significant, baseline differential of 2%.

11 What Is a Tree Worth? Green-City Strategies, Signaling and Housing Prices 223 Table 4 Time-space comparison of real housing prices: Fairmount tree plantings within 1000 feet before sale =1 if there was a tree planting within 1000 ft of the parcel prior to this sale Dependent Variable: Log Real Price of Parcel If There Was a Tree If There Was a Tree Planting within 4000 ft Planting within 2000 ft of the Parcel of the Parcel Whole Sample Prior to This Sale Prior to This Sale (1) (2) (3) (4) (5) (6) (11.12) (9.24) (1.05) (0.07) (0.27) (0.34) Log real price of parcel at previous sale (72.06) (53.99) (23.27) (14.38) (22.65) (13.53) Log lot area (28.45) (28.64) (12.81) (12.05) (10.91) (10.12) Log floor area (34.34) (33.68) (15.54) (15.25) (13.37) (13.16) Number of stories in the parcel =1 if parcel has a frame exterior =1 if parcel has a stone exterior =1 if parcel has non-rectangular shape (5.71) (4.95) (2.29) (2.71) (0.65) (0.77) (1.05) (1.28) (0.24) (0.32) (0.19) (0.25) (3.60) (3.45) (1.83) (1.76) (1.18) (1.18) (3.28) (3.50) (0.19) (0.02) (0.40) (0.31) =1 if parcel has central air (3.35) (3.27) (0.37) (0.31) (0.84) (0.58) =1 if parcel has garage (15.65) (15.64) (5.64) (5.71) (4.14) (4.14) =1 if property is detached (1.14) (1.13) (1.05) (2.45) (0.63) (1.83) Time gap between this and previous sales (15.42) (7.97) (8.23) Interaction term between log real previous sale price and time gap (12.64) (6.96) (7.46) Census Tract Yes Yes Yes Yes Yes Yes Dummies Year Dummies Yes Yes Yes Yes Yes Yes Quarter Dummies Yes Yes Yes Yes Yes Yes Trend Census Tract No Yes No Yes No Yes Dummies Observations R-squared t statistics in parentheses. 2 significant at 10%; significant at 5%; significant at 1%.

12 224 Wachter and Wong such as grocery stores or subway stops, we expect both positive (e.g., proximity) and negative externalities (e.g., noise and congestion), which decline at different rates by distance (Li and Brown 1980). In the case of tree plantings, we do not expect any negative externalities so we model a simple linear decline of the proximity impact. Table A1 shows the results. There is evidence of a decreasing tree-related impact. Note that the coefficients of the other controls do not vary qualitatively or quantitatively by including the linear decline. Table 5 tells a similar story. Here the treatment group consists of the first parcel sales that occurred after a Fairmount tree planting, but not the second or third. In a comparison between the between-sales-planting treatment group and all other sales (again controlling for census tract fixed effect), the price differential is between 9% and 10%. This becomes insignificant when the comparison group is limited to sales that also occurred immediately after Fairmount tree plantings but were further away. Because this price differential is tied to the one-time shock of a tree planting, this is even stronger support for price increases that are driven by factors that either resulted from or resulted in tree plantings. The intrinsic value of a tree can be expected to affect house prices around a relatively small area. For example, Been and Voicu (2006) identify significant intrinsic value of public community gardens up to 1,000 feet. Tables 6 to 8 use a 100-foot distance cutoff to explore this. Notably, there is no significant price differential related to the likelihood of ever having a tree planting in close proximity (Table 6). 6 Compared to results in Table 3, this reinforces the idea that the process that generates tree plantings works at a more macrolevel. Table 7, where the treatment group consists of parcel sales after a Fairmount tree planting within 100 feet, presents results that are remarkably similar to those using a 1,000-foot cutoff in Table 4. The only difference is that the price differential as compared to the whole sample (columns 1 and 2) is about 2 percentage points higher. This suggests that the intrinsic value of a tree planting, which is more likely to vary in strength spatially than any physical and social capital related to the planting, is around 2%. Again, columns 3 to 6 indicate that the tree-price correlation is largely motivated at a more macro level. The lack of statistical power prevents precise comparisons between the parcels with plantings within 100 feet and those with plantings within 500 feet to 4,000 feet directly, but the point estimates in columns 3 to 6 are not far from 2%. A linear model of impact decline is modeled and presented in Table A2. While we find evidence of a decreasing impact, it is not as strong as the decline found in Table A1. 6 On average, a city block is one-tenth of a mile, which is 528 feet. A tree planting within 100 feet means a tree directly in front of the parcel or its immediate neighbors. In comparison, 4,000 feet is about eight city blocks.

13 What Is a Tree Worth? Green-City Strategies, Signaling and Housing Prices 225 Table 5 Time-space comparison of real housing prices: Fairmount tree plantings within 1000 feet between sales =1 if a tree planting occurred within 1000 ft of parcel between previous & current sales Log real price of parcel Dependent Variable: Log Real Price of Parcel If Tree Planting If Tree Planting Occurred within 4000 ft Occurred within 2000 ft of Parcel between of Parcel between Previous and Previous and Whole Sample Current Sales Current Sales (1) (2) (3) (4) (5) (6) (0.010) (0.010) (0.014) (0.014) (0.014) (0.014) (0.003) (0.004) (0.009) (0.012) (0.010) (0.014) at previous sale Log lot area (0.006) (0.006) (0.016) (0.016) (0.019) (0.019) Log floor area (0.010) (0.010) (0.023) (0.023) (0.026) (0.026) Number of stories in the parcel (0.006) (0.006) (0.011) (0.011) (0.013) (0.013) =1 if parcel has a frame exterior (0.015) (0.015) (0.040) (0.040) (0.047) (0.047) =1 if parcel has a stone exterior (0.016) (0.016) (0.041) (0.041) (0.046) (0.046) =1 if parcel has non-rectangular shape (0.011) (0.011) (0.028) (0.028) (0.033) (0.032) =1 if parcel has central air (0.053) (0.053) (0.151) (0.152) (0.168) (0.169) =1 if parcel has garage (0.007) (0.007) (0.017) (0.017) (0.020) (0.020) =1 if property is detached (0.011) (0.011) (0.025) (0.026) (0.029) (0.030) Time gap between this and previous sales Interaction term between log real previous sale price and time gap (0.005) (0.024) (0.028) (1.256) (6.509) (7.560) Year Dummies Yes Yes Yes Yes Yes Yes Quarter Dummies Yes Yes Yes Yes Yes Yes Trend Census Tract No Yes No Yes No Yes Dummies Observations R-squared Standard errors in parentheses. significant at 10%; significant at 5%; significant at 1%. Table 8 uses parcel sales with between-sales tree plantings as the treatment group. As argued before, the price differential between this treatment group and other sales is more closely tied to tree plantings than that found in Table 7. Evidence consistent with earlier results is found comparing Tables 7 and 8.

14 226 Wachter and Wong Table 6 Time-space comparison of real housing prices: Fairmount tree plantings within 100 feet before or after sale =1 if there had ever been a tree planting within 100 ft of the parcel (before or after sales) Dependent Variable: Log Real Price of Parcel =1 if There Had Ever =1 if There Had Ever =1 if There Had Ever =1ifThereHadEver Been a Tree Planting Been a Tree Planting Been a Tree Planting Been a Tree Planting within 4000 ft of the within 2000 ft of the within 1000 ft of the within 500 ft of the Parcel (before or Parcel (before or Parcel (before or Parcel (before or Whole Sample after Sales) after Sales) after Sales) after Sales) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (0.013) (0.013) (0.013) (0.013) (0.013) (0.013) (0.012) (0.012) (0.012) (0.012) Log real price of parcel at previous sale (0.003) (0.004) (0.003) (0.004) (0.004) (0.005) (0.005) (0.006) (0.007) (0.008) Log lot area (0.006) (0.006) (0.006) (0.006) (0.007) (0.007) (0.009) (0.009) (0.013) (0.013) Log floor area (0.010) (0.010) (0.010) (0.010) (0.011) (0.011) (0.013) (0.013) (0.018) (0.018) Number of stories in the parcel =1 if parcel has a frame exterior =1 if parcel has a stone exterior =1 if parcel has nonrectangular shape (0.006) (0.006) (0.006) (0.006) (0.007) (0.006) (0.008) (0.007) (0.010) (0.010) (0.015) (0.015) (0.016) (0.016) (0.019) (0.018) (0.024) (0.024) (0.038) (0.038) (0.016) (0.016) (0.017) (0.017) (0.018) (0.018) (0.022) (0.022) (0.033) (0.032) (0.011) (0.011) (0.011) (0.011) (0.012) (0.012) (0.015) (0.014) (0.022) (0.021)

15 What Is a Tree Worth? Green-City Strategies, Signaling and Housing Prices 227 Table 6 continued =1 if parcel has central air Dependent Variable: Log Real Price of Parcel =1 if There Had Ever =1 if There Had Ever =1 if There Had Ever =1 if There Had Ever Been a Tree Planting Been a Tree Planting Been a Tree Planting Been a Tree Planting within 4000 ft of the within 2000 ft of the within 1000 ft of the within 500 ft of the Parcel (before or Parcel (before or Parcel (before or Parcel (before or Whole Sample after Sales) after Sales) after Sales) after Sales) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (0.053) (0.053) (0.054) (0.053) (0.061) (0.060) (0.068) (0.067) (0.090) (0.088) =1 if parcel has garage (0.007) (0.007) (0.007) (0.007) (0.007) (0.007) (0.010) (0.009) (0.014) (0.013) =1 if property is detached (0.011) (0.011) (0.012) (0.012) (0.013) (0.013) (0.017) (0.017) (0.026) (0.026) Time gap between this and previous sales (0.005) (0.005) (0.005) (0.007) (0.010) Interaction term between (1.257) (1.279) (1.402) (1.763) (2.573) log real previous sale price and time gap Year Dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Quarter Dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Trend Dummies No Yes No Yes No Yes No Yes No Yes Observations R-squared Standard errors in parentheses. Significant at 10%; Significant at 5%; Significant at 1%.

16 228 Wachter and Wong Table 7 Time-space comparison of real housing prices: Fairmount tree plantings within 100 feet before sale Dependent Variable: Log Real Price of Parcel If There Was a Tree If There Was a Tree If There Was a Tree If There Was a Tree Planting within 4000 ft Planting within 2000 ft Planting within 1000 ft Planting within 500 ft of the Parcel of the Parcel of the Parcel of the Parcel Prior Whole Sample Prior to This Sale Prior to This Sale Prior to This Sale to This Sale (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) =1 if there was a tree planting within 100 ft of the parcel prior to this sale Log real price of parcel at previous sale (4.13) (3.57) (1.37) (1.09) (1.35) (1.18) (1.32) (1.48) (1.13) (1.24) (72.15) (54.04) (23.30) (14.37) (22.64) (13.51) (20.36) (12.25) (14.58) (8.00) Log lot area (28.42) (28.61) (12.79) (12.04) (10.89) (10.11) (7.12) (6.52) (5.48) (5.35) Log floor area Number of stories in the parcel =1 if parcel has a frame exterior =1 if parcel has a stone exterior =1 if parcel has non-rectangular shape (34.23) (33.60) (15.53) (15.24) (13.35) (13.14) (10.67) (10.46) (5.81) (5.90) (5.66) (4.90) (2.31) (2.72) (0.66) (0.78) (0.64) (0.57) (0.71) (0.53) (1.04) (1.28) (0.26) (0.33) (0.18) (0.24) (0.70) (1.00) (0.48) (0.22) (3.63) (3.47) (1.83) (1.76) (1.17) (1.16) (0.23) (0.14) (0.29) (0.04) (3.35) (3.54) (0.18) (0.03) (0.41) (0.33) (0.53) (0.86) (0.17) (0.40)

17 What Is a Tree Worth? Green-City Strategies, Signaling and Housing Prices 229 Table 7 continued Dependent Variable: Log Real Price of Parcel If There Was a Tree If There Was a Tree If There Was a Tree If There Was a Tree Planting within 4000 ft Planting within 2000 ft Planting within 1000 ft Planting within 500 ft of the Parcel of the Parcel of the Parcel of the Parcel Prior Whole Sample Prior to This Sale Prior to This Sale Prior to This Sale to This Sale (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) =1 if parcel has central air =1 if parcel has garage =1 if property is detached Time gap between this and previous sales Interaction term between log real previous sale price and time gap Census Tract Dummies (3.34) (3.27) (0.37) (0.33) (0.86) (0.60) (0.52) (0.56) (0.71) (0.81) (15.75) (15.74) (5.69) (5.72) (4.17) (4.14) (3.08) (3.06) (3.01) (2.53) (1.16) (1.18) (1.06) (2.44) (0.62) (1.81) (1.05) (1.56) (0.40) (0.52) (15.21) (7.97) (8.24) (6.13) (4.49) (12.48) (6.96) (7.48) (5.77) (4.34) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Year Dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Quarter Dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Trend Census Tract No Yes No Yes No Yes No Yes No Yes Dummies Observations R-squared t statistics in parentheses. 2 significant at 10%; significant at 5%; significant at 1%.

18 230 Wachter and Wong Table 8 Time-space comparison of real housing prices: Fairmount tree plantings within 100 feet between sales Dependent Variable: Log Real Price of Parcel If Tree Planting If Tree Planting If Tree Planting If Tree Planting Occurred within Occurred within Occurred within Occurred within 4000 ft of Parcel 2000 ft of Parcel 1000 ft of Parcel 500 ft of Parcel Between Previous Between Previous Between Previous Between Previous Whole Sample and Current Sales and Current Sales and Current Sales and Current Sales (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) =1 if a tree planting occurred within 100 ft of parcel between previous & current sales Log real price of parcel at previous sale (0.034) (0.034) (0.037) (0.037) (0.036) (0.035) (0.034) (0.034) (0.033) (0.033) (0.003) (0.004) (0.009) (0.012) (0.010) (0.014) (0.013) (0.018) (0.019) (0.027) Log lot area (0.006) (0.006) (0.016) (0.016) (0.019) (0.019) (0.025) (0.025) (0.040) (0.040) Log floor area (0.010) (0.010) (0.023) (0.023) (0.026) (0.026) (0.034) (0.034) (0.050) (0.050) Number of stories in the parcel =1 if parcel has a frame exterior =1 if parcel has a stone exterior =1 if parcel has non-rectangular shape (0.006) (0.006) (0.011) (0.011) (0.013) (0.013) (0.017) (0.017) (0.025) (0.025) (0.015) (0.015) (0.040) (0.040) (0.047) (0.047) (0.064) (0.064) (0.099) (0.105) (0.016) (0.016) (0.041) (0.041) (0.046) (0.045) (0.059) (0.060) (0.085) (0.088) (0.011) (0.011) (0.028) (0.028) (0.033) (0.032) (0.042) (0.041) (0.062) (0.063)

19 What Is a Tree Worth? Green-City Strategies, Signaling and Housing Prices 231 Table 8 continued Dependent Variable: log Real Price of Parcel If Tree Planting If Tree Planting If Tree Planting If Tree Planting Occurred within Occurred within Occurred within Occurred within 4000 ft of Parcel 2000 ft of Parcel 1000 ft of Parcel 500 ft of Parcel Between Previous Between Previous Between Previous Between Previous Whole Sample and Current Sales and Current Sales and Current Sales and Current Sales (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) =1 if parcel has central air =1 if parcel has garage =1 if property is detached Time gap between this and previous sales Interaction term between log real previous sale price and time gap (0.053) (0.053) (0.151) (0.151) (0.168) (0.169) (0.210) (0.208) (0.230) (0.231) (0.007) (0.007) (0.017) (0.017) (0.020) (0.020) (0.026) (0.026) (0.037) (0.038) (0.011) (0.011) (0.025) (0.026) (0.029) (0.030) (0.038) (0.041) (0.059) (0.064) (0.005) (0.024) (0.028) (0.037) (0.055) (1.257) (6.508) (7.559) (9.875) (14.446) Year Dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Quarter Dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Trend Census Tract No Yes No Yes No Yes No Yes No Yes Dummies Observations R-squared Standard errors in parentheses. Significant at 10%; significant at 5%; significant at 1%.

20 232 Wachter and Wong Slightly larger price differentials are identified when we focus on parcel sales with between-sales tree plantings, which diminish both in size and significance when we hold the timing of the planting constant and vary the distance to the plantings. In sum, we find that parcels located within 1,000 feet of between-sales Fairmount tree plantings sold at a 9 to 10% premium on average. The average premium for parcels within 100 feet of a between-sales planting is 11 to 13%. A comparison of these results points to an intrinsic value of a tree planting at around 2%. Using all parcel sales after a planting as the treatment group produces very similar results, with slightly smaller point estimates. Remarkably, these price differentials largely diminish when comparing the parcels at different distances but within 4,000 feet from the between-sales plantings. A baseline comparison of the propensity to have a tree planting in proximity, using all parcels that have ever had plantings within 100 feet (or 1,000 feet) as the treatment group, indicates that omitted variable bias is around 2%. 7 We argue that, after controlling for census tract fixed effects, parcel characteristics and time effects, it is unlikely that the substantial price differentials presented are due to some omitted factors unrelated to tree plantings but somehow coincide with them both in space and time. Therefore we conclude there is a significant price differential related to the tree plantings, between 7 to 11%, which is remarkably constant within 4,000 feet of the plantings. We posit that this is largely due to social capital (or other positive attitude) creation or a signaling effect, on the top of an intrinsic value of the plantings at around 2%. Empirical Evidence PHS Plantings The main purpose of investigating the PHS tree plantings is to exploit the differences in the application process as compared to the Fairmount plantings. As described in the introduction, PHS plantings involve a much more significant level of coordination and cooperation among neighbors in order to obtain a tree planting on the block they live on. If the revelation of social capital were a main driver behind the price differential identified in the previous section, we expect to find an even larger differential looking at PHS plantings. On the other hand, the intrinsic value of trees can be expected to be the same across the programs and both the signaling effect of other previously unknown resident characteristics and the omitted variable bias can point in either direction. We 7 We acknowledge that this is not a perfect measurement of the omitted variable bias, partly because parcels that have tree plantings in proximity in an out-of-sample year will be wrongly put in the control group instead of the treatment group.

21 What Is a Tree Worth? Green-City Strategies, Signaling and Housing Prices 233 treat the evidence from a comparison between the two greening programs as suggestive but not conclusive. Equation (1) is estimated as before using PHS plantings. One difficulty of exploring the relationship between PHS plantings and housing prices is the limited number of parcel sales close to the plantings and thus lower statistical power. Table 9 summarizes the results, each row showing the price differential comparing a given treatment group (row) with various control groups (column). In the last six columns the timing of the tree planting is held constant and the only factor that varies between the treatment and control groups is the distance to the planting: the control groups are further away from the plantings. These comparisons give us an idea how localized any tree-related price changes are. Row 1 confirms that the PHS program target the neighborhoods in need: parcels that have ever had a PHS planting within 1,000 feet sold at a discount of around 8%. This remains stable when parcels that have ever had a PHS planting within 2,000 feet or 4,000 feet are used as the control group, implying that the price discount is relatively localized at a 1,000-foot radius. Surprisingly, the parcels that have ever had a PHS planting within 100 feet, or those with a higher propensity to request a PHS planting, do not systematically sell at a significant discount (row 2). These together imply that most of the homeowners who took advantage of the PHS program in fact own houses of average quality in a below-average neighborhood of a 1,000-foot radius (about two city blocks). This reinforces the idea that a tree planting whether it is a Fairmount or a PHS planting relates to signaling of potentially positive resident characteristics. Row 3 shows a price discount between 6 to 11% for parcels that were sold after a PHS planting within 1,000 feet, which is not statistically different from that measured in row 1. Using parcel sales with between-sales tree plantings within 1,000 feet produces similar results (row 5). Therefore there is no strong evidence that PHS plantings decreased the negative price differential in a systematic way for parcels within 1,000 feet. Hence there is unlikely to be significant social capital creation during the process because this impact is expected to work at a multiblock level. This suggests that the significant price premium identified in the previous section is likely to be due to signaling of physical and other resident characteristics and that residents request trees from the Fairmount Park Commission to signal neighborhood qualities otherwise unobservable to homebuyers. Any positive impact of the PHS plantings within 100 feet is not precisely measured using this sample (rows 4 and 6). While the point estimates of the