Economic Returns to Energy-Efficient Investments in the Housing Market: Evidence from Singapore

Transcription

1 Economic Returns to Energy-Efficient Investments in the Housing Market: Evidence from Singapore Yongheng Deng National University of Singapore Zhiliang Li National University of Singapore John M. Quigley University of California Berkeley Abstract The Green Mark scheme was launched by the Building Construction Authority of Singapore in January 2005 as a key strategic program to help shape a greener and more sustainable built environment. Since then, 250 building projects have been awarded the Green Mark for sustainability, and about 90 of these are residential projects. This paper reports the first economic analysis of the private returns to these investments, evaluating the premium in rents or asset values they command in the market. We analyse some 62,434 transactions in the Singapore housing market to estimate the economic impact of the Green Mark scheme on Singapore's residential sector. We adopt a two-stage research design; in the first stage, a hedonic pricing model is estimated based on transactions data for green and non-green residential units. This hedonic model includes fixed effects for each of 656 individual projects. In the second stage, the fixed effects estimated for each project are regressed on the location attributes of the projects, as well as Green Mark controls. As noted below, this research design provides quite conservative estimates of the effect of Green Mark certification on asset prices. Our results suggest that the economic returns to green building are substantial. This is one of the first analyses of the economics of green building in the residential sector, and the only one analysing property markets in Asia. Our results provide insight about the operation of the housing market in one country, but the policy implications about the economic returns to sustainable investments in the property market may have broad application for emerging markets in Asia. 1

2 I. Introduction In the past decade, systems for rating and evaluating the sustainability and energy efficiency of buildings have proliferated. In part, this reflects the potential importance of real property in matters of environmental conservation. For example, buildings and their associated construction activities account for almost a third of world greenhouse gas emissions. The construction and operation of buildings account for about forty percent of worldwide consumption of raw materials and energy. Thus, small increases in the "sustainability" of buildings, or more specifically in the energy efficiency of their construction, can have large effects on their current use of energy and on their life-cycle energy consumption. Projected trends in the urbanization of developing economies, particularly in Asia, suggest that the importance of energy efficiency in building will increase further in the coming decades (Zheng, et al, 2009). In the U.S., two major programs evolved to encourage the development of energy-efficient and sustainable buildings through systems of ratings to designate and publicize exemplary buildings. The government-sponsored Energy Star program began as a voluntary labeling program intended to identify and promote energy-efficient products. The Energy Star label was extended to new homes in 1993, and this has been promoted as an efficient way for consumers to identify builders as well as buildings constructed using energy-efficient methods. The Energy Star label is marketed as an indication of lower ownership costs, better energy performance, and higher home resale values. Buildings can receive an Energy Star certification if the source energy use of the building, as certified by a professional engineer, achieves a specified benchmark level; the label is awarded to the top quarter of all comparable buildings, ranked in terms of source energy efficiency. 2

3 In a parallel effort, the U.S. Green Building Council (USGBC), a private nonprofit organization, has developed the LEED green building rating system to encourage the adoption of sustainable green building and development practices. The requirements for certification of LEED buildings are substantially more complex than those for the award of an Energy Star rating, and the certification process measures six distinct components of sustainability, one of which is energy performance. It is claimed that LEED-certified buildings have lower operating costs and increased asset values and that they provide healthier and safer environments for occupants. It is also noted that the award of a LEED designation demonstrate[s] an owner s commitment to environmental stewardship and social responsibility. In the short time since these rating systems for buildings were developed in the U.S., quite similar certification procedures have been codified in many other countries. For example, the BREEAM rating system is becoming widely diffused in the UK, and the Greenstar rating system for buildings has been adopted in Australia. Both the British and Australian rating systems have much in common with the LEED system in the U.S. A program to publicize exemplary buildings in Canada, called BOMA-Best, has been launched, and the European Union is currently negotiating a common system for the certification of commercial and residential buildings. In 2005 Singapore became the first Asian country to adopt a system of green labeling for newly constructed and rehabilitated buildings. The system, called Greenmark, has been widely publicized in the city-state, and the award of Platinum, Gold-plus, Gold, and Certified plaques are regularly reported in the newspapers. Despite the international diffusion of these rating systems, little is known about their impact on the choices of consumers and investors or about their impact on energy usage or 3

4 carbon emissions. By now, there are a few studies of rating systems for commercial office buildings in the U.S. (e.g., Eichholtz, et al, 2010, and Fuerst and McAllister, 2011), but there is no systematic body of evidence for other countries. There is also no evidence at all about the effect of these certification programs on the housing market. This paper analyzes the Greenmark program in Singapore, evaluating the effect of the program on the housing market, in particular, the effects on the asset values of dwellings in multifamily housing projects. In Section II below, we describe the salient features of the Greenmark program and its history. In Section III we present a detailed analysis of the sales of some 62,000 housing units in more than 650 projects. Of these projects, about ten percent have earned a Greenmark label. Ceteris paribus, we find that Greenmark-labelled dwellings command a substantial premium in the Singapore housing market. In Section IV, we summarize the evidence on economic premium for Greenmark projects. Section V is a brief conclusion. II. Singapore Green Mark scheme and Point-Scoring system. The Singapore Green Mark Scheme (GM), which evaluates buildings for their environmental impact and energy performance, was developed by Singapore s Building and Construction Authority (BCA) and National Environment Agency (NEA) in January As a benchmarking scheme that aims to achieve a sustainable built environment, it provides a comprehensive framework for assessing the overall environmental performance of new and existing buildings to promote sustainable design, construction and operations practices in buildings. The GM scheme covers a wide range of property sectors commercial, residential, retail, industrial, hotel, institutional, office, park and public housing. Typically, the regulations and building codes differ between residential and non-residential buildings. The scheme provides 4

5 incentives for developers and design teams to construct green, sustainable buildings which can promote energy savings, water savings, and healthier indoor environments, as well as the adoption of more extensive greenery for their projects. For existing buildings, the GM scheme encourages building owners and operators to meet sustainable operations goals and to reduce adverse impacts of their buildings on the environment and occupant health over the building life cycle. The label is marketed for its ability to reduce water and energy consumption, improve indoor environmental quality and reduce the potential negative impact on the environment. Importantly, the label also helps to recognize developers with strong commitments to corporate social responsibility. It also helps publicize achievements in environmental sustainability. A. Application and assessment process Developers, building owners and government agencies submit an application form to the BCA to register their interest in participating in the BCA Green Mark Scheme. Following that, the assessment process involves a briefing to the project team to understand better of BCA Green Mark requirements and the certification level sought. Actual assessment is carried out at later stage to verify that the building project meets the criteria and certification level. The assessment includes design and documentary reviews as well as site verification, which identifies the specific energy efficiency and the environmental practices incorporated in the project. Documentary evidence submitted as a part of the assessment. Upon completion of this assessment, a letter of award showing the certification level of the projects is sent to the team. 5

6 B. The rating system The assessment criteria cover the following key areas: Energy Efficiency Water Efficiency Environmental Protection Indoor Environmental Quality Other Green Features and Innovation The Green Mark rates the environmental attributes of a building based on a point-scoring approach. Points are awarded for incorporating conservation features which exceed standard practice. Depending on the score, the rating is categorized into four levels - Platinum, Gold Plus, Gold and Certified, which can be seen from the Table 1. - Insert Table 1. Here Detailed information on point-scoring is presented in the appendix. Specifically, up to 140 points are awarded in five categories (with an additional 20 bonus points for renewable energy). Up to 65 points can be awarded for energy efficiency, 13 for water efficiency, 29 for environmental protection, 6 for indoor environmental quality and 7 other green features. To qualify for a Green Mark required a minimum score of 50. After achieving certification, Green Mark buildings are required to be re-assessed every three years to maintain the Green Mark status. Newly constructed and certified buildings are subsequently re-assessed under the existing buildings criteria. Existing buildings are re-assessed under the existing buildings criteria. 6

7 III. The Analysis A. The Data As of June 2009, 250 building projects were awarded the Green Mark, of which 92 are residential housings. The Building Construction Authority has released lists of GM-rated buildings projects on an annual basis from 2005 onwards. Thus, the names and addresses of GM awarded projects are identified 1. As one residential project usually has multiple buildings, we matched the GM-rated residential project names and addresses to those maintained in the Real Estate Information System (REALIS) database as of June The REALIS database stored by Urban Redevelopment Authority (URA) is a comprehensive source of real estate information, for Singapore 2. In this way, some 65 GM-rated residential projects, including both new and existing properties were matched. Transactions for some of 15,787 dwelling units in those 65 GM-rated projects between January 2000 and June 2009 were identified. Besides price, the transactions records included unit price per square foot, floor area, floor number, tenure, property type, completion state, transaction type, postal code, transaction date. We also identified for control purposes some 591 projects with 46,647 dwelling unit transactions in projects that were not GM-rated. 1 In Singapore, each building corresponds to a unique postal code. 2 Realis provides information for residential, commercial and industry property market. Specifically, it includes Time Series - more than 1,300 time series; Project Database - integrated information on each project, such as the approval status and the number of units launched and sold; Stock Database - allows users to customize their own stock and vacancy statistics; Transaction Database - contains records of caveats lodged at the Singapore Land Registry since 1995 for the residential, commercial and industrial sectors. The Transaction Database is updated fortnightly. 7

8 The empirical analysis presented below is based on observations on the sales of some 62,434 multifamily dwelling units between January, 2000 and June, These units are in 656 different housing projects (residential estates) throughout the city of Singapore, of which 65 projects with 15,787 dwelling unit transactions are GM-rated while 591 with 46,647 dwelling unit transactions are not GM-rated (NGM). Figure 1 compares the annual average unit price per square foot in GM and NGM projects over the Insert Figure 1 here - It shows that, except for 2002 and 2008, sales of GM-rated dwellings commanded a higher sale price than NGM dwellings. Figures 2 and Figure 3 report the temporal and spatial variations in the fraction of GM-rated sales. - Insert Figure 2 here - As indicated in Figure 2, the proportion of GM sales reached a peak of nearly forty percent of all sales at Q Since Q1 2006, the fraction of GM-rated sales more or less fluctuates with the dynamics of the overall property market. - Insert Figure 3 here - 8

9 Figure 3 indicates that the sales of both GM and NGM-rated dwelling units follow a quite similar pattern; in particular, both GM and NGM sales are mostly concentrated in the central region. For each dwelling unit that has been sold, we gathered as much information as we could about its hedonic characteristics. Table 2 summarizes the hedonic characteristics frequently used in property value studies, as reported in two leading international journals. Some of the attributes measured in other studies are hardly relevant to the Singapore context. For example, all private housing projects in Singapore have air conditioning, a garage, and a swimming pool; the climate makes fireplaces less important. Some attributes (e.g., the number of rooms and bathrooms) are simply unavailable, but since the size of rooms in residential housing projects in Singapore is quite standardized, we can measure the total area of each unit. Some other attributes, not reported in Table 1, may be more important in the Singapore context. For example, there is empirical evidence that a good view is greatly valued among Singaporeans (Yu, Han and Chai, 2007). Accordingly, we expect the floor level to be positively related to sale price, other things being equal. - Insert Table 2 here - Apart from structural attributes, we are able to control for location by adding indicator variables for properties located in the Central, East, West, Northeast and North parts of the city. We also include indicator variables for the year of sale, from 2000 to 2009, to control for the macro environment. 9

10 Furthermore, we exploit information on the sale date, the property type (Condominium, Apartment, Semi-detached house, Terrace house and Executive house), the state of completion of each unit sold (completed or uncompleted as of June 2009), the type of transaction (New sale, Re-sale or Sub-sale), postal district, postal code, and the tenure type (freehold 3 or leasehold). In addition, we can also measure the type of purchaser in the REALIS data source -- a buyer who already owns a public housing unit (an HDB flat) and aims to upgrade to private housing or a buyer from the private economy (who currently lives in a private dwelling unit). Columns (1) to (3) in table 3 report the comparison mean value of the hedonic attributes sold in GM and NGM-rated residential projects. - Insert Table 3 here - On average, GM-rated projects are around four stories higher than NGM projects. In particular, a GM-rated dwelling unit is more likely to be located on a high floor level than NGMrated one sale. Also, GM housing is slightly larger in living area than NGM by four square meters. Clearly, there exists a substantial difference in the average total transaction prices and unit prices per square feet between GM and NGM-rated units, suggesting the possible existence of a price premium for GM housing. In terms of property type, both GM and NGM share a similar pattern: more than half of projects/dwelling units are condominiums, with other property types only taking up a minor proportion. For both groups, over seventy-three percent dwelling units are newly transacted. We simply observe that more than sixty percent of sales occurred in 3 We assume 99-year and 103-year as leasehold, while 999-year and all other inputs, as freehold property. 10

11 the central region, which is in line with its land scarcity and the fierce competition for land use in centre. The number of private buyers exceeds buyers trading up from public ( HDB ) flats. Our data show that up to sixty percent of GM-rated dwelling units were sold while they were still under construction, whereas NGM-rated dwelling units are relatively old. In general, freehold property yields more secure property rights and longer occupation terms to the owner than a leasehold, making buyers willing to pay a price premium (Tu and Bao, 2009). Most of the dwelling units sold in this sample are freehold in tenure, though its share in GM group is smaller than that in NGM groups. We control for this potential impact on housing prices in the regressions reported below. Within GM-rated dwelling units, more than 57 percent of them are awarded the Green Mark Gold, leaving 22 and 20 percent of total GM-rates sales to Green Mark Gold-plus award and Green Mark certified award, respectively. Housing sales vary over time between 2000 and Thirty percent of sales took place in 2007, which reflects the underlying property market cycle in Singapore. For each of the 656 projects in the sample, we gathered information on its location and amenity characteristics. For each of these housing estates, we define a public transit variable, MRT, with a value of one if the project is located within 500 meters of a subway station. We measure the distance in kilometres of each project to Orchard Road (the major shopping district), Dist2Orch. We also define an amenity variable with a value of one if the project is located within 300 meters of open space or park (Park) and another access variable with a value of one if the project is within 300 meters of a major road (MajRd). In line with previous findings, the coefficient of MajRd variable represents a net effect of closeness to major road on housing price due to its positive effect of accessibility and simultaneous negative effect of traffic noise on 11

12 property valuation (M.Wilhelmsson, 2000). Considering the fact that buyers from private economy may possess stronger purchasing power due to their financial profile compared to citizens residing in government public housing (HDB), the proportion of purchasers from private economy for individual project may crucially affect housing price on project level attributed to the self selection. 4 Thus, we construct the variable of PrivateRatio with value equal to 1 if the share of private buyers for each project is more than half and expect it to be positively associated with housing price. Beyond these access measures, we identify the local community containing each project by a set of 35 indicator variables, one for each community. We measure the Age of each project -- June 2009 minus the date of the earliest transaction in the project. The age variable proxies the obsolescence of the project and also the accumulated depreciation of each project. The research design thus recognizes the distinction between attributes measured at these two levels: dwelling units and the projects in which they are situated. Note that this research design may also lead to quite conservative estimates of the importance of green certification on asset values. By design, all the covariation between higher quality dwelling units and green certified properties is attributed to the dwelling units, not the environmental certification. As noted below, dwellings in Greenmark-certified properties tend to be larger, newer, and of higher quality than those in non-greenmark projects. Our method attributes all of this covariation to dwelling unit attributes. 4 In Singapore, those who are eligible for public housing (HDB) featured by substantial government housing subsidy and favorable housing mortgage are evaluated in terms of salary, family composition, age and so forth. By doing so, the government aims to fulfill the goal of enabling the majority to afford their own houses without bearing too much pressure. The vast majority (more than 80%) of Singaporeans live in public housing. The rest of the population, who are in general belong to upper end of the income distribution, live in private housing. As a result, the control for share of private buyers can be used as an instrument for household income measure, which is typically not available for public data set in Singapore. 12

13 We also address the positive correlation between dwelling quality and Greenmark certification using propensity score measures. This is also discussed below. Table 4 presents the descriptive statistics of those project-level accessibility attributes. It can be seen that on average, eighty percent of projects have a better access to a major road. Similarly, about 60 percent of the projects are within 300 m of a park/open space and 30 percent of the projects are served by a subway station within 500 m. The mean distance to Orchard road is 7.3 km. Almost 70 percent of projects are purchased by private buyers whose share is over half in a project. The average economic age of projects is about 3 years. - Insert Table 4 In addition, of course, for each project, we know its Greenmark rating category: unrated, Certified, Gold, Gold-plus, or Platinum award. B. Empirical Analysis Our empirical modelling strategy consists of a two-stage hedonic pricing equation, the most straightforward and conservative way to investigate the economic premium of Green Mark. In the first stage, we simply relate the logarithm of unit sale price per square foot to floor area, floor level, tenure, property type, purchaser type, transaction type, completion status, time-fixed effects (year and month) when the sales take place and region-fixed effects where the housing units are located, as well as the project fixed effect. The second stage regression considers the locational and amenity attributes measured at the project level. The environmental attributes and 13

14 the Greenmark rating category are measured at the project level, that is, all dwellings in a given project have the same locational and environmental attributes. 1 In equation (1), the dependent variable is the logarithm of the selling price per square foot unit transaction i. C is a constant (intercept) and is an error term. is a vector of hedonic characteristics of property i. To control for the regional difference among sales,, a region dummy, is added to represent the five planning regions in Singapore 5, where the projects and their attached dwelling units sold are located. The added time effect, T n, is intended to control for macro-economic attributes common to all 6.,, are estimated coefficients. Importantly, measures the potential price premium of Green Mark. 2 Equation (2) analyzes the four categories of the Green Mark premium: Platinum, Gold-plus, Gold and Certified. 3 The essential difference of equation (3) from equation (1) is the inclusion of project fixed effect dummies More importantly, the coefficient estimation of the variable PFE n,, is thus meant to capture the project fixed effect that is used as the dependent variable on the LHS at the second stage hedonic equation. 5 They are Central, East, West, North east and North region, of which north is taken as base group.. 6 Jan, 2000 is taken as base group. 14

15 Next, we proceed with the second stage hedonic equation. 4 In equation (4), the dependent variable,, derived from the first stage equation, is regressed on a set of accessibility variables on project level. Here, c is a constant (intercept) and is an error term in the equation. X i is the vector of hedonic attributes for project i, including tenure (Freehold), property type dummies, Age (the time difference between the earliest dwelling unit transaction of a project and June 2009), distance to orchard road (Dist2Orch), closeness to subway (MRT), access to major road (MajRd) and closeness to park or open space (Park). R n,, a region dummy variable, is used to control for the regional variation among projects. Likewise, N n is meant to control for the neighbourhood effect. The coefficient of primary interest is, the economic return of Green Mark at the project level. In equation (5), the effect of Green Mark on housing selling price may vary substantially across the four distinguished four GM awards: Platinum, Gold-Plus, Gold and Certified in the sample of 656 projects. Hence, the economic premium for each GM category is. 5 Table 5a presents the results of estimating the first stage hedonic model using 62,434 dwelling units transactions between 2000 and Insert Table 5a here 15

16 For each model, regional and month-year dummies are included, which are not reported separately in the table. The 656 project fixed effects are estimated in column (5). Overall, housing attributes at unit level have expected effect, specifically, the higher level the unit sold is locate, the more the selling price. Every one level higher yields 0.03 percent increase in unit price/psf, ceteris paribus. We confirm the appreciation of good view by looking at the negative sign of low level and positive sign of high level compared to medium level. Other housing characteristics, such as, condominium in dwelling type, new-sale, freehold in tenure, uncompleted projects and private buyers, all consistently have anticipated positive effects on unit price/psf through column(1)-(4), though vary in magnitude. Green Mark premium at the first stage regression is found to be significantly different from zero at 1 percent level, indicating that Green Mark will on average benefit dwelling units by commanding 9.7 percent premium relative to NGM-rated ones, ceteris paribus. The GM premium also significantly varies across different awards: Platinum earns the highest return of 56.9 percent increase in unit price/psf, gold-plus and gold earns 1.8 percent and 15.3 percent price premium, respectively. Interestingly, certified award is subject to a significant 4.3 percent unit price loss compared to NGM group. - Insert Table 5b here Table 5b reports the results of the second stage regression at the project level. Property type dummies and regional dummies are included in each model, while community dummies are also 16

17 added in column (3) and (4) to control for the inherent neighbourhood amenity effects. In general, if more than half of purchasers for a project are from private economy, the average selling price on project level is higher than otherwise. More than half are HDB upgrading buyers by 0.17 to We find that freehold in tenure and proximity to MRT is statistically insignificant across models. Surprisingly, Park is found to have a negative effect on selling price. The consistent negative sign of MajRd implies that the negative aspect of being near major road---traffic noise and pollution outweighs its enhanced accessibility. Also, the less proximity to Orchard road, the lower the selling price. By observing the estimation of completion year, it s consistent that price index from 2005 is quite similar across models. It soars before the financial crisis and declines thereafter, takes off again from 2009 onwards, mostly due to global economic rebound. More important, the Green Mark measure is statistically significant in column (2) at the five percent level, and column (3) at the ten percent level, even after controlling for community amenity fixed effect. In column (4), the return variation of GM award effects is reported. Except for that, Gold-plus is statistically insignificant, both Platinum and Gold have positive signs and are statistically different from zero at one percent level and five percent level, respectively. Also, Certified award still has a negative sign that is significant at five percent level, which is larger than Gold in absolute value. IV. Evidence on Economic Premium for Green Mark Table 3 provides the means and standard deviations for a set of hedonic characteristics of GM-rated dwelling units and NGM-rated ones, it reveals that the GM-rated dwelling units are of higher quality: they are larger in area and more likely to locate on the high levels. They are 17

18 substantially newer than their counterparts and more private buyers tend to buy GM-rated compared to NGM-rated ones. To control for the fact that the average characteristics of the GM and NGM samples are different, we employ Propensity Score Matching (PSM) techniques. PSM weights the observations in the NGM group (control group) to make the control and treatment groups more similar in terms of average characteristics. Dwelling units sold in NGM group are weighted corresponding to their propensity scores, that is, the probabilities that their hedonic characteristics are identical to those in the GM group. (Dan Black and Jeffrey Smith, 2004) We match on the basis of this scalar propensity score rather than matching on the basis of all housing characteristics (Rosenbaum and Rubin, 1983). Among the several specific matching methods 7, we find Nearest One-to-One Neighbour Matching is the best fit to our sample. It minimizes the difference in distribution among GM and NGM groups. The key idea of One-to-One Nearest Neighbour Matching (NNM) is that each unit sold in GM group attempts to choose individual in NGM group with the closest propensity score, in which we avoid the occurring of replacement of matched NGM-rated units sold in this study. Meanwhile, in order for matching to be valid and exact, we impose common support restriction that units in GM group whose propensity score are larger than the largest score in the NGM group are left unmatched. By doing so, we eventually manage to match 15,637 pairs of dwelling units, equivalent to 460 projects in total. 7 Apart from Nearest Neighbor matching, there are also Kernel based matching, radius matching, caliper matching, Mahalanobis matching and local linear regression matching and so on. For details about them, refer to (Dan Black and Jeffrey Smith, 2004) 18

19 Column (4) and (5) in Table 3 present the mean values for GM and NGM groups weighted by their propensity scores. After matching, it is clear that the average values of the hedonic attributes of the NGM group are far closer to those of the GM group. For instance, prior to PSM, around 21 percent, 71 percent and 70 percent of GM-rated dwelling units locate on the high level, are condominium in property type and are purchased by private buyers, respectively, while only 10 percent, 57 percent and 60 percent of NGM-rated units are on high level, condominium and purchased by private buyers. Yet, after nearest neighbour matching, the matched GM and NGMrated pairs yield a mostly similar average quality measures. The average sizes for GM and NGM are and square meters, respectively. Around 15 percent instead of 10 percent, 75 percent instead of 57 percent and 68 percent rather than 60 percent of NGM group are suited on high level, condominium and purchased by private buyers. First-Stage Regression Using the PSM weighted observations, we conduct similar two-stage hedonic regression. Table 6a presents the first-stage regression estimation results relating the logarithm of unit price/psf to a number of hedonic characteristics. The results are based on equation (1), (2) and (3). - Insert Table 6a here As compared to Table 5a, although the sample size is a lot smaller, the variation of housing quality between GM and NGM groups at unit level has been considerably diminished, leading to the more convincing empirical result. Specifically speaking, what are consistent results across 19

20 column (1)-(4) are: the positive effect of living area on unit price; price appreciation of high view and de-appreciation of low level compared to medium level; the positive impact of freehold property on unit price; the negative influence on unit price of resale compared to reference group of sub-sale, as well as the negative effect of uncompleted on unit price/psf. Noticeably, however, variables of condominium and new sale have a negative sign across column (1)-(3) with a bit difference in magnitude, while having a positive sign in column (4) after controlling for project fixed effects, all of which are significantly different from zero at 1 percent level. In contrast, private buyer is positively related to unit price/psf throughout the first three columns and statistically significant at 1 percent level, while having a negative sign in column (4) with 5 percent statistical significance. In column (2), the price premium of Green Mark is statistically significant at 1 percent level. On average, the overall Green Mark will yield percent price premium on unit price/psf, ceteris paribus, which is slightly larger in magnitude compared to its counterpart in Table 5a. Likewise, the premium variation within GM categories is significant at 1 percent level, of which ceteris paribus, Platinum are mostly valued with percent price premium, Gold-plus and Gold award earn 3.48 percent and percent price premium, respectively. Still, certified award is somehow depreciated by the public with a 5.04 percent unit price loss. As in column (4), after controlling for housing characteristics, the project fixed effects are thereafter estimated by adding 459 project indicators, explaining around 95 percent variation of unit price/psf. 20

21 Second-Stage Regression Table 7 presents summary statistics on selected hedonic characteristics of the PSM 460 projects. It shows that, almost 82 percent projects are close to major road, while only 30 percent have better access to subway station. More than 60 percent of projects have both proximity to park or open space and private buyers whose share are over half of total purchasers for a project. The average distance for the 460 projects away from Orchard road is 6.7 kilometres and average economic age is 3 years. It s also found that condominium and apartment in property type dominate. Besides, up to 50 percent of projects are still under construction by the time of June Insert Table 7 here Table 6b reports the estimation results of the second stage regression at project level. The estimations are based on equation (4) and (5). Compared to Table 5b, the variable Age is added to control for the effect of obsolescence and depreciation of each project on price. Similarly, property type dummies and regional dummies are included in each model, while community dummies are also included in column (3) and (4) to control for the inherent neighbourhood amenity effects. Clearly, the proportion of private buyers for each project indeed positively significantly affect project price at 1 percent level, which almost remains the same in magnitude across columns. MRT and MajRd are found to be statistically insignificantly associated with project price in the present sample. Both Freehold and Park are negatively related to project price with 21

22 significance at 1 percent level throughout the estimation. As expected, the further away to Orchard road, the less proximity to the city centre, the cheaper the project s price. Thus, Dist2Orch is found to be significantly negatively related to project price with magnitude varying from to in absolute term. As expected, Age has a negative sign across columns, representing the economic and physical depreciation of each project. It s statistically significant at 5 percent level in the first two columns and 1 percent level in the last two columns. Likewise, projects being constructed have relatively large positive effects on project price with significance at 1 percent level across columns. Most importantly, given the 460 matched projects by Nearest Neighbouring Matching, Green Mark appears statistically significant at 10 percent level in column (2) and (3) when controlling for neighbourhood fixed effect, in which the GM premium nearly remains 0.09 in both columns. In column (4), the variation of distinguished GM awards effects is reported. As in Table 5b, except for that Gold-plus is again statistically insignificant, both Platinum and Gold have positive signs and are statistically different from zero at 1 percent level and 5 percent level, respectively. The premium for Platinum is and that for Gold is Nonetheless, Certified award consistently has a negative value of 0.19 that is significant at 1 percent level based on the present sample. - Insert Table 6b here 22

23 Conclusion Our empirical analysis based on 656 individual projects and 62,434 transactions in the Singapore housing market suggest substantial economic returns to green building. The two-stage estimation shows that the Green Mark measure is statistically significant even after controlling for community amenity fixed effect. We report the return variation of GM award effects. Except for that, Gold-plus is statistically insignificant, both Platinum and Gold have positive signs and are statistically different from zero at one percent level and five percent level, respectively. Further analysis based on nearest one-to-one neighbour matching between control and treatment samples find Green Mark appears statistically significant at 10 percent level where the GM premium nearly remains The variation of distinguished GM awards effects suggests, except for that Gold-plus is again statistically insignificant, both Platinum and Gold have positive signs and are statistically different from zero at 1 percent level and 5 percent level, respectively. The premium for Platinum is and that for Gold is This is one of the first analyses of the economics of green building in the residential sector, and the only one analysing property markets in Asia. Our results provide insight about the operation of the housing market in one country, but the policy implications about the economic returns to sustainable investments in the property market may have broad application for emerging markets in Asia. 23

24 References Costa, Dora L. and Matthew E. Kahn, 2009, Towards a Greener California: An Analysis of Household Variation in Residential Electricity Purchases, UCLA working paper. Dan A. Black and J.A.Jeffrey A. Smith, 2004, How robust is the evidence on the effects of college quality? Evidence from matching, Journal of Econometrics, 121(1-2): Davis, Lucas W., 2009, Evaluating the Slow Adoption of Engergy Efficient Investments: Are Renters Less Likely to Have Engergy Efficient Appliances? UC Berkeley working paper. Eichholtz, Piet M.A, Nil Kok and John M. Quigley, 2009, Why do companies rent green?, UC Berkeley working paper. Eichholtz, Piet M.A, Nil Kok and John M. Quigley, J.M. 2010, Doing Well by Doing Good: Green Office Buildings, American Economic Review, forthcoming. Fuerst, Franz and Patrick McAllister, 2011, Green Noise or Green Value? Measuring the Price Effects of Environmental Certification in Commercial Buildings, Real Estate Economics, forthcoming. Greening, L. A., D. L. Greene, and C. Difiglio, Energy efficiency and consumption the rebound effect a survey. Energy Policy. 28 (6-7), Jacobsen, Grant D. and Matthew J. Kotchen, 2009, Are Building Codes Effective at Saving Energy? Evidence From Residential Billing Data in Florida, Conference on Green Building, the Economy and Public Policy, University of California, Berkeley. Morancho A., A Hedonic Valuation of Urban Green Areas. Landscape and Urban Planning, 66:

25 Paul R. Rosenbaum and Donald B. Rubin, 1998, The central role of the propensity score in observational studies for causal effects, Biometrika, 70(1):41-55 Tu Yong and Helen X.H. Bao, 2009, Property Rights and Housing Value: The impacts of political instability, Real Estate Economics, 37(2): Wilhelmsson M., 2000, The impact of traffic noise on the values of single-family house. Journal of Environmental Planning and Management, 43: p799 Yu Shi-Ming, Han Sun-Sheng, Chai Chee-Hian, 2007, Modelling the value of view in high-rise apartments: a 3D GIS approach, Environment and Planning B: Planning and Design, 34: Zheng, Siqi, Matthew E. Kahn, and Edward L. Glaeser, 2009, The Greenness of China: Household Carbon Dioxide Emissions and Urban Development, NBER working paper. 25

26 Figure 1.Annual Average Unit Price/psf for GM & NGM, ,400 Annual unit price/psf (S$) 1,200 1, GM NGM Year (Data source: Realis) Figure 2. Market Volumes & Fraction of GM housings, 2005Q1-2009Q2 45% 40% Market share of GM buildings Market volum 8,000 7,000 Market sahare (%) 35% 30% 25% 20% 15% 10% 5% 0% 6,000 5,000 4,000 3,000 2,000 1,000 0 Market volume (Data source: Realis) Quarter 26

27 Figure 3. Regional Allocation of Market Volume 35,000 30,000 29,428 Volume 25,000 20,000 15,000 GM NGM 10,000 9,611 6,800 6,782 5, ,680 1,682 1,899 3,073 1,479 0 Central East West North Northeast Region (Data source: Realis) 27

28 Table 1. Point-Scoring Rating Criteria Green Mark Score Green Mark Rating 90 and above Green Mark Platinum 85 to< 90 Green Mark Gold Plus 75 to< 85 Green Mark Gold (Source: BCA) 50 to< 75 Green Mark Certified Table 2. Attributes used in Empirical Works Published (percentage) Variable Journal of Real Estate Research Journal of Urban Economics Living area No.of bathrooms No.of bedrooms Garage Fireplace Pool Air conditioning Age Lot size Subjective judgment (Source: Wilhelmsson M. (2000)) 28

29 Table 3. Comparison of GM-rated and NGM-rated Dwelling Units (standard deviation in parenthesis) Total Sample GM-rated NGM-rated PSM GM-rated (1:1 Nearest) PSM NGM-rated (1:1 Nearest) (1) (2) (3) (4) (5) Sample size 62,434 15,787 46,647 15,637 15,637 Total price (S$) 1,321, ,643, ,212, ,624, ,446, (2,259,541.51) (2,651,709.30) (2,099,269.84) (2,414,026.85) (3,254,319.18) Unit price/psf (S$) , , (528.82) (603.75) (490.49) (606.23) (592.28) Size (sq.m) (109.20) (119.75) (105.37) (84.51) (117.38) Freehold(percent) (49.03) (49.20) (47.33) (49.26) (49.64) Stories (number) (9.46) (12.59) (7.81) (12.59) (8.43) Storey (percent) Low (49.34) (50.00) (48.75) (50.00) (50.00) Medium (45.23) (45.40) (45.16) (45.40) (47.94) High (33.86) (41.00) (30.46) (41.00) (35.73) Property type (percent) Condominium (48.73) (45.01) (49.42) (44.67) (43.16) Apartment (46.87) (44.54) (47.50) (44.67) (43.16) Terrace (15.46) (9.57) (16.96) (0) (0) Detached (7.71) (0.80) (8.89) (0) (0) Semi-detached (8.85) (1.13) (10.21) (0) (0) Executive Condo (15.25) (0) (17.57) (0) (0) Transaction type (percent) New sale (43.66) (44.09) (43.51) (44.21) (44.56) Sub sale* (36.71) (39.34) (35.72) (39.48) (38.52) Resale** (29.43) (25.98) (30.47) (26.06) (28.89) Purchaser type (percent) Private (48.33) (45.79) (48.92) (45.88) (46.29) HDB upgrade (48.32) (45.77) (48.91) (45.87) (46.29) Planning Region (percent) 29

30 Central (48.41) (47.93) (48.55) (48.03) (48.32) East (34.26) (32.99) (34.67) (33.12) (32.89) Table 3. Comparison of GM-rated and NGM-rated Dwelling Units (Continued) (standard deviation in parenthesis) West (34.23) (32.60) (34.75) (32.74) (39.71) North (17.17) (0) (19.76) (0) (0) Northeast (26.00) (31.58) (23.67) (31.71) (22.15) Transaction year (percent) (2.77) (4.50) (1.85) (4.52) (2.26) (5.93) (2.76) (6.66) (2.77) (4.00) (20.16) (8.58) (22.66) (8.62) (9.08) ,43 (22.15) (12.03) (24.52) (12.09) (18.19) (24.14) (23.69) (24.29) (23.79) (19.24) (33.06) (34.92) (32.39) (35.06) (33.01) (38.63) (37.63) (38.96) (37.75) (39.99) (46.01) (48.73) (44.70) (48.60) (48.42) (30.89) (29.98) (31.18) (30.10) (28.23) (32.59) (31.88) (32.82) (32.10) (33.63) Completion year (percent) (5.15) (10.20) (0) (10.25) (0) (37.61) (25.40) (40.36) (25.51) (31.11) (32.09) (26.01) (33.78) (26.12) (26.06) (30.78) (13.68) (34.21) (13.74) (18.13) (38.89) (41.40) (37.93) (41.54) (43.02) (18.77) (0.80) (21.57) (0.80) (1.39) uncompleted (48.59) (48.81) (46.06) (48.89) (49.85) Notes: *Sub-sale: The sale of a unit by one who has signed an agreement to purchase the unit from a developer or 30

31 a subsequent purchaser before the issuance of the Certificate of Statutory Completion and the Subsidiary Strata Certificates of Title or the Certificates of Title for all the units in the development; **Resale: The sale of a unit by a developer or subsequent purchaser after the issuance of the Certificate of Statutory Completion and the Subsidiary Strata Certificates of Title or the Certificates of Title for all the units in the development; Columns(1)-(3) report the comparison of mean values of original sample prior to adjustment of Propensity Score Matching (PSM); Columns(4) and (5) report the mean value comparison between treatment(gm) and control (NGM) groups after one to one Nearest Neighboring PSM, in which the sample size for matched groups is 15,

32 Table 4. Summary Statistics on Selected Hedonic Variables of 656 Projects Variables Mean Std.dev Minimum Maximum MajRd(percent) Park(percent) Dist2Orch(km) MRT(percent) Age(years) PrivateRatio(percent) Freehold(percent) Property type(percent) Condominium Apartment Detach Semi-detach Terrace Executive Condo Planning region(percent) Central East West North Northeast Completion year(percent) Uncompleted

33 Table 5a. OLS Regression Estimation of Unit Price on Housing Attributes (Dependent variable: Logarithm of Unit Price per square foot) (1) (2) (3) (4) (5) Constant *** *** *** *** *** (0.1143) (0.1151) (0.1141) (0.1119) (0.0554) Green Mark (1=Yes) *** (0.0030) Platinum *** (0.0134) Gold-Plus *** (0.0055) Gold *** (0.0036) Certified *** (0.0060) Size (sq.m) *** *** *** *** *** ( ) ( ) ( ) ( ) ( ) Stories (number) *** (0.0001) Stores: Low (1=Yes) *** *** *** (0.0027) (0.0027) (0.0027) High (1=Yes) *** *** *** (0.0040) (0.0039) (0.0039) Dwelling type: Condominium (1=Yes) *** *** *** *** (0.0082) (0.0083) (0.0082) (0.0081) (0.0337) Apartment (1=Yes) *** *** *** *** *** (0.0085) (0.0086) (0.0085) (0.0084) (0.0210) Detached (1=Yes) *** *** *** *** *** (0.0528) (0.0532) (0.0527) (0.0517) (0.0117) Semi-detached (1=Yes) * ** * * *** (0.0991) (0.0998) (0.0989) (0.0970) (0.0098) Terrace (1=Yes) *** *** *** *** (0.0230) (0.0231) (0.0229) (0.0225) Transaction type: New sale (1=Yes) *** *** *** *** (0.0036) (0.0037) (0.0036) (0.0036) (0.0018) Resale (1=Yes) *** *** *** *** *** (0.0050) (0.0050) (0.0049) (0.0048) (0.0025) Freehold (1=Yes) *** *** *** *** *** (0.0027) (0.0027) (0.0028) (0.0028) (0.0159) Uncompleted (1=Yes) *** *** *** (0.0034) (0.0034) (0.0035) (0.0035) (0.0130) Private Buyers (1=Yes) *** *** *** *** *** (0.0026) (0.0026) (0.0026) (0.0025) (0.0011) Region dummies Y Y Y Y Y Month-year dummies Y Y Y Y Y Project fixed effect dummies N N N N Y 33

34 Sample size 60,164 60,164 60,164 60,164 62,434 R-Squared 63.73% 63.26% 63.91% 65.29% 94.06% Adjusted R-Squared 63.65% 63.18% 63.83% 65.22% 93.99% Notes: Base dummy region is 'North regions' for all models; Base dummy purchaser type is 'HDB upgrading buyer'; Base dummy dwelling type is 'Executive condominium' for model (1)-(4); base dummy dwelling type is 'Executive condominium' and 'Terrace' for model (5) where project fixed effects can thus be estimated without causing multi-collinearity problem; Base dummy level is' Medium'; Base dummy transaction type is 'Sub sale'; Standard errors are reported in brackets and Significance at the 0.10, 0.05 and 0.01 levels indicated by *, ** and ***, respectively; For Model (1)-(4), sample size is 60,164 due to 2,270 missing information on individual units storey; for model (5), storey variable is exclude so that 656 project fixed effects may be estimated explicitly without raising multi-collinearity problem. 34

35 Table 5b. GLS Regression Results of Project Fixed Effects on Housing Attributes (Dependent variable: Project fixed effects) (1) (2) (3) (4) Constant *** *** *** *** (0.249) (0.258) (0.274) (0.216) Green Mark (1=Yes) ** * (0.057) (0.060) Platinum (1=Yes) *** (0.262) Gold-plus (1=Yes) (0.125) Gold (1=Yes) ** (0.067) Certified (1=Yes) *** (0.059) PrivateRatio (percent) *** *** *** *** (0.025) (0.025) (0.026) (0.025) Freehold (1=Yes) * (0.045) (0.046) (0.047) (0.045) Accessibility MRT ( 1 if a project located within 500m of a subway station (0.031) (0.031) (0.033) (0.032) Dist2Orch (kilometers) *** *** *** *** (0.005) (0.005) (0.005) (0.005) MajRd (1 if a project within 300m of a major road) ** ** *** ** (0.033) (0.033) (0.035) (0.005) Park (1 if a project within 300m of a park or open space) * ** ** (0.025) (0.025) (0.027) (0.026) Completion Year 2005 (1=Yes) * (0.232) (0.240) (0.227) (0.149) 2006 (1=Yes) * * ** (0.231) (0.239) (0.228) (0.150) 2007 (1=Yes) * * ** (0.231) (0.239) (0.228) (0.150) 2008 (1=Yes) * ** (0.231) (0.240) (0.228) (0.150) 2009 (1=Yes) * ** ** *** (0.233) (0.241) (0.228) (0.150) Uncompleted (1=Yes) * ** ** *** (0.230) (0.238) (0.225) (0.145) Property type dummies Y Y Y Y Region dummies Y Y Y Y Community dummies N N Y Y Sample size R-Squared 65.07% 65.41% 66.76% 68.76% Adjusted R-Squared 63.92% 64.21% 63.67% 65.68% Notes: Base dummy property type is 'Executive condominium'; Base dummy region is 'Northeast region'; Base completion year is 1997 year; 35

36 White Heteroskedasticity-Consistent Standard Errors reported in brackets; Significance at the 0.10, 0.05 and 0.01 levels indicated by *,** and ***, respectively. 36

37 Table 6a. OLS Regression Estimation of Unit Price on Housing Attributes (1 : 1 Nearest Neighboring Propensity Score Matching Observations) (1) (2) (3) (4) Constant *** *** *** *** (0.1167) (0.1152) (0.1117) (0.1124) Green Mark (1=Yes) *** (0.0034) Platinum *** (0.0139) ` Gold-Plus *** (0.0059) Gold *** (0.0039) Certified *** (0.0068) Size (sq.m) *** *** *** *** ( ) ( ) ( ) ( ) Stores: Low (1=Yes) *** *** *** *** (0.0038) (0.0038) (0.0037) (0.0016) High (1=Yes) *** *** *** *** (0.0050) (0.0050) (0.0049) (0.0021) Dwelling type: Condominium (1=Yes) *** *** *** ** (0.0040) (0.0040) (0.0039) (0.0385) Transaction type: New sale (1=Yes) *** *** *** *** (0.0049) (0.0048) (0.0047) (0.0023) Resale (1=Yes) *** *** *** *** (0.0074) (0.0073) (0.0071) (0.0037) Freehold (1=Yes) *** *** *** *** (0.0037) (0.0037) (0.0038) (0.0398) Uncompleted (1=Yes) *** *** *** (0.0050) (0.0049) (0.0050) (0.1052) Private Buyers (1=Yes) *** *** *** ** (0.0038) (0.0038) (0.0037) (0.0015) Region dummies Y Y Y Y Month-year dummies Y Y Y Y Project fixed effect dummies (460projects) N N N Y Sample size 31,274 31,274 31,274 31,274 R-Squared 64.98% 65.92% 67.98% 95.34% Adjusted R-Squared 64.84% 65.79% 67.85% 95.25% Notes: Base dummy region is 'North' and 'Northeast'; Base dummy purchaser type is 'HDB upgrading buyer'; Base dummy level is' Medium'; Base dummy transaction type is 'Sub sale'; Standard errors are reported in brackets and Significance at the 0.10, 0.05 and 0.01 levels indicated by *, ** and ***, respectively; 37

38 38

39 Table 6b. GLS Regression Results of Project Fixed Effects on Housing Attributes (Dependent variable: Project fixed effects) (1) (2) (3) (4) Constant *** *** *** *** (0.105) (0.104) (0.157) (0.148) Green Mark (1=Yes) * * ( (0.055) Platinum (1=Yes) *** (0.117) Gold-plus (1=Yes) (0.115) Gold (1=Yes) ** (0.064) Certified (1=Yes) *** (0.064) PrivateRatio (percent) *** *** *** *** (0.027) (0.027) (0.029) (0.029) Freehold (1=Yes) ** *** *** *** (0.052) (0.052) (0.053) (0.051) Accessibility MRT ( 1 if a project located within 500m of a subway station (0.033) (0.033) (0.035) (0.034) Dist2Orch (kilometers) *** *** *** *** (0.005) (0.005) (0.006) (0.005) MajRd (1 if a project within 300m of a major road) ( (0.038) (0.040) (0.039) Park (1 if a project within 300m of a park or open space) ** ** *** *** (0.029) (0.029) (0.031) (0.031) Age (yrs) ** ** *** *** ( (0.008) (0.008) (0.008) Uncompleted (1=Yes) *** *** *** *** (0.034) (0.036) (0.036) (0.036) Property type dummies Y Y Y Y Region dummies Y Y Y Y Community dummies N N Y Y Sample size R-Squared 54.64% 55.10% 58.51% 61.27% Adjusted R-Squared 53.42% 53.79% 54.00% 56.75% Notes: Base dummy region is 'Northeast' and 'North' region; White Heteroskedasticity-Consistent Standard Errors reported in brackets; Significance at the 0.10, 0.05 and 0.01levels indicated by *,** and ***, respectively. 39

40 Table 7. Summary Statistics on Selected Hedonic Variables of 460 Matched Projects Variables Mean Std.dev Minimum Maximum MajRd(percent) Park(percent) Dist2Orch(km) MRT(percent) Age(years) PrivateRatio(percent) Freehold(percent) Property type(percent) Condominium Apartment Detach Semi-detach Terrace Executive Condo Planning region(percent) Central East West North Northeast Completion year(percent) Uncompleted

41 Appendix 1. Project and Transaction Frequencies of Community Neighbourhood Project No. Transaction No. Ang Mo Kio Bedok Bishan Bukit Batok Bukit Merah Bukit Panjang Bukit Timah Choa Chu Kang Clementi Downtown Core Geylang Hougang Jurong West Kallang Mandai Marine Parade Museum Newton Novena Orchard Pasir Ris Punggol 1 9 Queenstown River Valley Rochor Sembawang Sengkang Serangoon Singapore River Southern Islands Tampines Tanglin Toa Payoh Woodlands Yishun

42 Appendix. Point Allocations - BCA Green Mark for Residential Buildings (Version RB/3.0) 42