Studies of Price Effects of Eco-Labels in Real Estate Markets: An off the record record. Pat McAllister

Transcription

1 Studies of Price Effects of Eco-Labels in Real Estate Markets: An off the record record Pat McAllister This document attempts to provide a list and summary of studies 1 on the effects of environmental performance on real estate prices. The particular focus is on the price effect of eco-labels in real estate markets. Many of the studies discussed below are still at a draft stage and are part of the grey literature in that they have not yet been subject to peer review, are at very early draft stages or are just about to be published, etc. Too often, the results of studies from this grey literature are cherry-picked and/or the (preliminary) nature of the results poorly understood. If you feel that this brief summary of a paper contains errors in terms the detail and/or evaluation or that the research has been misrepresented in any way, firstly I apologise and, secondly, please contact me at p.m.mcallister@rdg.ac.uk and set me straight. Clearly, this is not a static area. Studies are always been refined. I would be very grateful to receive copies of updated and/or new studies. I have found a few occasion that my version of the results has been superseded by an new version it is worth checking if you are going to rely on the information. It is hoped that the comments and observations made are helpful and constructive to interested readers of the papers and fellow researchers in terms of evaluating the contribution of the research. In terms of criteria for evaluation, a key issue for many parties interested in this area will be the robustness of the empirical results. As with many academic papers, the key information is often in the detail rather than in the abstract or conclusions. Working with my colleague Franz Fuerst, we think that there are a number of key questions that need to be asked when looking at any study. We have fallen into most of the pitfalls listed below. Given the volume and pace of academic work currently underway and changing market and regulatory conditions, this snapshot literature review is likely to become obsolete quite quickly. Below, we indicate the key criteria against which existing evidence has been evaluated. Evaluating Research Studies: Key Questions We identify eight questions against which studies can be evaluated. 1 What is the scope of the data? The feasibility and quality of empirical research into the price effects of environmental certification is dependent upon the availability of data on three main areas: 1. Market prices rents and sales 1 I have not looked at studies conducted on energy efficiency and home prices that were published in the 1980s, 1990s or before.

2 2. Environmental performance of real estate assets 3. Attributes of buildings leases, specification, size, location, quality etc. Outside of US office markets, adequate data in terms of scale (sample size) and scope (number of variables) has rarely been available and it is unlikely that any researcher has had all the information that they would have liked. The main problem tends to be obtaining data on all the potential price determinants. This issue is discussed in more depth below. 2 What is the sample size? It can be surprisingly difficult to find this basic information in some papers about how many green assets are being compared to a (usually) much larger sample of conventional assets. In studies involving commercial real estate assets, sample sizes have been small with typically hundreds of environmentally certified assets being compared to thousands of conventional assets. For studies involving residential properties, typically the sample size is much larger. 3 What econometric model specification has been selected? Due to differences between green and conventional buildings (e.g. green buildings tend to be newer, larger and located in better locations), it is usually not possible to use simple descriptive statistics such as simple averages to identify whether there are significant differences between green buildings and non- green buildings in terms of rents, prices etc. It can be difficult to measure the contribution of an individual attribute (such as eco-certification, energy performance, location, design etc) to identified single price paid for a bundle of attributes. Most studies use some variation of widely accepted econometric procedure (a hedonic regression) to estimate the contribution of each asset attribute (size, age, location, certification etc) on its price. However, hedonic model outputs can be sensitive to choice of model specification and scope of explanatory variables. Indeed, many studies apply a number of different techniques (e.g. Ordinary Least Squares, Generalised Least Squares, Propensity Score Matching) and include different independent variables. This can be extremely useful since it provides some indication of the robustness of results. Clearly, concerns are raised when results vary significantly between different models i.e. when using the same data, the findings are not consistent between different models. This is surprisingly common but usually glossed over. 4 How are data errors handled? When dealing with thousands of observations, we have found that the can be a significant number of errors in the data. For instance, in pricing studies, the price of a portfolio sale may be allocated to a single building or the price of a single building may be allocated to a portfolio. These outliers can bias the results of hedonic regression models. 5 How are differences in location handled in the hedonic model? Usually, the most important control in this type of study is controlling for locational differences. Essentially, the research question boils down to All else equal, does being green affect the price of a building. The problem for researchers lies in the three words - all else equal. In terms of location, what other buildings are the green buildings being compared against? Other buildings in the same city? Other buildings in the same sub-market? Other buildings within a certain distance? This issue can be

3 crucial because green buildings may be concentrated in the specific (e.g. high value) locations. If this factor is not controlled for, a location effect may be mis-attributed to a green effect. 6 Are there other potential omitted variables? This is similar to the last point but it is broader. Due to data constraints, researchers may omit a variable that is having an effect on the prices of green buildings. Perhaps, being green is only one element of a bundle of extras that a developer uses to create a superior product. For instance, homes with better energy performance may tend to be of a higher quality of construction. By omitting this variable from a model, all else will not be equal and a construction quality effect will be attributed as an energy efficiency effect. However, it is also possible that these unobserved variables may also be affected by environmental certification. For instance, higher quality tenants may be attracted to eco-certified buildings. Tenants may also take longer leases or accept fewer lease incentives in eco-certified buildings. In a recent paper, Atkin et al (2011) looked again at the evidence of apparent large price premium that REITs seemed to pay (not necessarily for ecocertified asset) in the US. They referred to a potential bias in the studies that REITs may have tended to buy the best properties within a quality category with the result that an unobserved explanatory variables (the premium property explanation) resulted in hedonic models producing an unrealistic price premiums. The supplemented their hedonic estimation with a analysis repeat sales transactions (a sample of properties that had sold twice). Using this type of approach, they estimate REIT premiums of 6% compared to the 20%-50% premiums estimated in their hedonic regression models. It is possible that eco-labelled buildings may be best-in-class assets with superior attributes that are not included in a data set (better construction, specification, tenants, leases?). 7 What are the dependent variables? Ideally, the dependent variables will be clean achieved transaction prices. For rental markets, in particular, this can be problematic. Contract rents may not be adjusted for leasing incentives such as abnormal rent free periods. Market rents will tend to be appraisal-based estimates rather based on actual transactions. Obviously, it is also important to distinguish between asking prices and actual prices. Frustratingly, definitions within databases may be unclear about the variable provided. 8 To what extent are results consistent with expectations? This can be a tricky one. At conferences, we have had members of the audience shake their heads and state that some of our estimates simply don t smell right. Whilst a reality check is useful, it can be difficult to reconcile this combination of intuition and market experience with statistical model outputs. For instance, one unpublished working paper reported a very large premium for eco-labelled apartments. But, all else being equal, is it credible that buyers would really pay nearly double the price because the apartments are more sustainable? In a later version of the study published in a highly rated journal, the premium was substantially smaller. An additional check is whether the results are consistent. For LEED, are platinum, gold etc in the expected order in terms of price effects? Are LEED and Energy Star in the expected order in terms of price premiums? Is the combined estimated price effect for LEED and Energy Star (modelled as a single dummy for an eco-label) roughly the (weighted) average of the individual effects of an LEED or Energy Star label? If not, there may be problems with the data and/or the model specification.

4 9 What are the results telling us? Most of the studies are cross-sectional providing a snapshot of price differentials for a specific sample in a specific time period. It is expected that price differentials for certified buildings should vary over time and between buildings. Ultimately, any price differentials are the outcome of historic supply and demand conditions refracted through local institutional and market structures. The studies tend to estimate what average price premiums were for at a broad market level. They don t tell us what the actual price premium is or should be for a specific building. This is the type of information that appraisers typically need to have. Author(s) and status 1 Miller. Spivey and Florance (2008) Published in Journal of Real Estate Portfolio Management 2 Wiley, Benefield and Johnson (2010) Published in Journal of Real Estate Finance and Economics Data Approach Findings on price differentials Other findings and potential limitations Filtered sample of Class A buildings (larger than 200,000 sq ft, multi-tenanted, over five stories, built after 1970) to compare to 643 ES buildings. 927 sale transactions between 2003 and Breakdown between LEED and ES sale price observations is unclear. Class A office buildings only. 46 metropolitan markets (25 markets for sales). Breakdown between LEED and ES is unclear. We estimate 30 LEED and 440 ES rental observations and 12 LEED and Hedonic OLS regression for sale prices only. Controls for major markets but none for quality. Hedonic OLS and 2SLS regressions for rental and occupancy rates. Control sample seems to be other buildings in same metropolitan area. No controls for micro- Finds no statistically significant sales price premium. Hedonic OLS and 2SLS find rental differentials of 15-17% for LEED and 7-9% for ES. Hedonic OLS model of sales prices in absolute form. Estimate sale price premiums of $130 psf and $30 psf for LEED and ES. Occupancy rate is 2-4% higher for ES compared to non-es filtered sample. Report 30% lower operating expenses based on energy costs. Hedonic OLS and 2SLS with occupancy rate as dependent variable finds occupancy rate differentials of 16-18% for LEED and 10-11% for ES compared to control group. Key issue is location control. Metropolitan area may be too large. May be confusing location with certification effects.

5 70 ES sales observations. location effects. Potential omitted variable bias linked to dual certified buildings not taken into account. 3 Eichholtz, Kok and Quigley (2010) To be published in American Economic Review Weighted average rents for 694 certified buildings. Sale prices for 199 certified buildings Breakdown between LEED and ES is unclear. Hedonic OLS regressions for rental and sales prices. Control sample is buildings within 0.25 miles of certified building. No statistically significant rental premium for LEED. 3% rental premium for Energy Star. No statistically significant sale price premium for LEED. 19% sale price premium for Energy Star. No systematic treatment of data errors. Find a positive relationship between energy efficiency measure and level of rental premium. Potential omitted variable bias due to dual certified buildings not taken into account. LEED result is puzzling and may be due to dual certification problem. Fuerst and McAllister (2011) Published in Real Estate Economics 4 Chegut, Eichholtz, Kok and Quigley (2010) Draft working paper Asking rents for 990 ES and 210 LEED certified buildings. Sale prices for 662 ES and 139 LEED certified buildings Sale prices for 78 office BREEAM offices in UK. Sales data obtained from Real Capital Analytics Achieved rents for 1011 offices obtained from CoStar Hedonic OLS regressions for rental and sales prices. Control sample is based on buildings within same CoStar submarkets. Hedonic OLS regressions and PSM for rental and sales prices. Location control is buildings with 500m radius. 6% rental premium for ES and LEED certified buildings. 35% and 31% price premium for LEED and ES. Estimate rental premium of 16-20% to eco-certification. Excellent 22%- 27% Very Good 18%-19% Good 8%-11% Pass 17%-18% In two out of three specifications no statistically significant sale price premium No systematic treatment of data errors. Potential omitted variable bias issue due to dual certified buildings. No systematic treatment of data errors. Controls for spatial auto-correlation. No discussion of controls for lease terms, lease incentives unexpired lease length, tenant quality, building quality. No age control reported in rental regression. Included? Small sample problems in the sales data? 5 Dermissi (2009) Appraisal estimates of Market Robust, MLE and fixed Provides detailed results for NC, EB Results do not seem plausible. It is not

6 Published in Journal of Sustainable Real Estate 6 Fuerst and McAllister (2009) Published in Journal of Sustainable Real Estate Values for 351 LEED-rated US office buildings. Study period is first half of Occupancy rates for 292 LEED and 1,291 ES office buildings from a sample taken in December effects regression models. OLS and quantile regression. and CS sub-samples. Results are inconsistent between specifications. For LEED NC, estimates a 36% discount for LEED Silver. LEED EB has a 118% premium. Estimate 8% occupancy premium for LEED offices and 3% premium for ES. Quantile regressions find that, for ES offices, increased occupancy is concentrated in first two deciles only. clear what the benchmark variable is for value differences - other LEED buildings? Seem to be an absence of location controls. Did not control for potential omitted variable bias due to dual certification. 7 Fisher and Pivo (2009) Draft working paper 8 Deng, Li and Quigley (2011) Regional Science and Urban Economics forthcoming Investment performance of RPI properties 209 Energy Star, 158 in regeneration areas, 669 near transit stations. Residential sales prices of 74,278 dwelling units in 1439 projects in Singapore between % of sample had Green Mark by Had transactions in 62 residential projects with Green Mark were matched against transactions. Hedonic OLS procedures. CBSAs are the location controls. Hedonic OLS and GLS procedures. Sample refined using PSM. Location control is 55 planning areas. They also use project fixed effects in one model specification. Estimate 12.5% premium on appraised capital value for Energy Star compared to buildings in same CBSA. Estimates 1% higher occupancy rates in Energy Star buildings. Significantly different results for different models PSM Regression Estimate average price premium for Green Mark of about 4-6%. Platinum 14% Gold Plus 2.3% Gold 5.5% Certified 0.1% Estimate that utility costs were 10% lower in Energy Star offices. Energy Star offices have significantly lower income and total returns. Location controls are unlikely to fully capture the effects of variations in location quality. No explanation is offered for the fairly dramatic differences in estimated premium for the two model specifications. It is a concern that the premium for Gold is higher than the Gold-Plus in the PSM model. 51% Gold Project Fixed Effects Regression

7 21% Gold-plus 19% Green Mark 3% Platinum Have data on type of sale and buyer type. Estimate average price premium for Green Mark of about 14-21%. Platinum 21% Gold Plus 15% Gold 15% Certified 10% 9 Shimizu (2010) Draft working paper A blend of residential asking prices (80207) and actual (2063) sale prices for Tokyo Appears that approx 14%-15% of the sample had eco-labels. Hedonic OLS procedures. Estimate a 5% asking price premium for Green Label. Some puzzling results. Effect of energy efficient equipment is significantly negative. Premium is lower for better environmental performance in some variables e.g. insulation and long life 10 Brounen and Kok (2011) Journal of Environmental Economics and Management residential sale prices in the Netherlands in for buildings with EPC rating. Hedonic OLS procedures Compared to buildings rated D, they estimate premiums of 10%, 5.5% and 2.5% for A, B and C respectively. For dwellings rated E, F and G, the estimate discounts of 0.5%, 2.5% and 5% respectively. The paper concludes a wide range of controls for quality and other fetures. All coefficients are statistically significant. 11 Fuerst and McAllister (2010) Ecological Economics Weighted Average rents for 1846 Energy Star, 268 LEED and 254 dual certified office buildings. Sale prices for 876 ES, 87 OLS and robust regression techniques. Control sample is based on buildings within same CoStar No statistically significant effect estimated for time on market. Estimate rental premium of 3%-4% for ES. 4%-5% for LEED and 9%- 10% for dual certified. Estimate sale price premium of 18% for ES, 25% for LEED and 28%-29% Attempt to control for the potential of data errors and the inclusion of dual certified buildings to produce biased results. No control for potential bias due to spatial auto-correlation.

8 12 Eichholtz, Kok and Quigley (2010) Working paper LEED and 123 dual certified office buildings Occupancy rates for 2111 ES, 313 LEED and 254 dual certified office buildings. Two samples 2009 and green office buildings in a total sample of observations in a sample of Sales 744 rated 5249 control Rents 1943 rated control Breakdown between LEED and ES is unclear. Mentions 209 LEED rental buildings later in paper. submarkets. Use panel data approach to compare 2007 and 2009 samples. WLS hedonic modelled with propensity score weighting of observations for dual certified buildings. Estimate 1%-3% higher occupancy rate for ES and 5%-6% lower occupancy rate for LEED. When comparing performance of 2007 sample, they find that estimated rental premium goes down to 1.2% For 2009 sample, estimate rent premium of 2% for ES and 6% for LEED. Estimate sale price premium of 13% for ES and 11% for LEED. The sale price premiums seem high. Is there an omitted variable problem? Does not control for potential omitted variable bias due to dual certification. No apparent controls for outliers or data errors. Why is aggregate green premium higher than LEED and ES? Why is ES sale price premium higher than LEED dual certification problem? Finds that LEED registration is associated with rental premium of 7.9%! Not clear how potential tendency for LEED to have higher levels of single occupancy is handled in effective rent estimations. Also analyse the effect of LEED scores on rental premiums but not on sales premiums. The summary statistics contain some puzzling numbers. Whilst the control buildings have higher average rents and sale prices than green labelled buildings, the control buildings are on average older, smaller and of lower quality (70% Class B or C compared to 25% of labelled

9 buildings). No control for potential bias due to spatial auto-correlation 13 Yoshida and Sugiura Presented in March buildings evaluated under Tokyo Green Building Program which is mandatory for buildings exceeding 5000 square metres. They obtain a total sample of sales of condominiums. It is not clear how many condominiums are TGBP rated. Use standard hedonic regression procedure. Find a statistically significant discount of 5.5% for labelled buildings. Discount is robust in a number of tests. When they estimate the price effects of the different components of the rating, they find mixed results. The individual drivers of the negative premium tend to be increased energy efficiency, water efficiency and planting. They suggest that perceived additional maintenance costs associated with these features may produce the discount. The authors spend a great deal of effort trying to falsify their finding of a discount. However, their robustness checks further confirmed their findings. When presented, it was argued that the finding was consistent with research in the 1980s that found that buyers tended to be sceptical of non-familiar environmental technologies when purchasing homes. Additional costs of maintenance are emphasised. 14 Jaffee, Stanton and Wallace (November 2010) Working paper transactions ( ) for office buildings located in 43 US metropolitan areas. 545 Energy Star properties (3.6% of total). Only 142 rated at the time of sale (0.93% of total). Have data of total expenses (1473), NOI (1532) and cap rates (2323) for a subset of the Use standard hedonic regression procedure. Find evidence of lower depreciation for eco-labelled condominiums. Estimate an Energy Star premium of 13.5% when total expenses and operating costs are excluded as confounding factors. When they are included in the model of a subset of the data, the ES premium becomes statistically insignificant. Although sample size for Energy Star is not stated, it is likely to be quite small. In a subset of 816 observations, they The location control seems broad given the fact that ES buildings tend to be concentrated in CBDs. If sample size of ES is small, weak effects may be missed. Not clear if authors have tested for multicollinearity. If there is a negative association between operating expenses and ES, it is possible that ES coefficient may be biased.

10 15 Zheng, Wu, Kahn and Deng Feb 2011 sample. (Presumably) Due to missing variables, this sample decreases substantially in the hedonic models. In the absence of an eco-label in China, an index of building greenness is estimated from marketing claims. Apply this index to 1992 residential building complexes constructed Uses standard hedonic procedure. find no significant ES effect on NOI, cap rate or operating expenses. However, again sample size of ES is not indicated and may be small. Estimates a green premium of 9.1%. It is not clear how the premium is estimated. The result implies that buildings were rated in a binary manner. However, the paper describes the creation of a relative score. It is not clear how or whether separate groups of green buildings were identified. If they were identified, it is not clear how many there were. Fuerst and McAllister Energy Policy, published 2011 Examines effect of EPC rating on yield, Market Value and Market Rent for 708 commercial property assets in IPD UK. Includes 23 BREEAM rated buildings. Date of data is Q Given the large geographical scope of the sample and the number of EPC and geographical segments, it is likely that there were small sample effects. Uses standard hedonic procedure. Finds no significant effect of EPC rating on Market Rent and Market Value. Very similar results for yield estimation. For one EPC rating (E compared to G for retail) was the coefficient significantly negative at the 10% level. The BREEAM results were similar. The coefficient on the BREEAM dummy was statistically significantly negative at the 10% level. Paper argues that a larger sample is needed to provide a robust estimation of whether weak effects are being missed. 16 Kok and Jennen Energy Policy (2012) Examines the effect of EPC rating and Energy Index on 1057 rental transactions in Netherlands for the period Uses standard hedonic procedure. Finds a rental premium of approximately 4.7% for buildings rated C or lower compared to buildings rated D and above. Compared to D rated buildings, find There may be a potential omitted variable problem. Buildings rated Class A, B and C may be better quality than buildings with inferior performance. The level of energy efficiency may be correlated with other

11 Available online. significant premiums of about 10% for C rated and 5% for B rated. Significant discount for E (0.8%), F (0.5) but G rated offices had a 2.3% premium. unobserved quality variables e.g. design, interior finishes etc The sample of transactions covers the period When were EPCs introduced in Netherlands? For the UK, it was In the UK, buildings may have multiple EPCs. This issue is not discussed. 17 Harrison and Seiler Published in Journal of Property Investment and Finance 2011 Investigates whether there are rental price and occupancy rate premia in LEED and Energy Star certified office buildings. Also looks at variations in nature of variations focussing on the hypothesis that political characteristics determine level of premium. Uses hedonic procedure to investigate rent determinants. Seem to omit a number of variables e.g. size, land area, stories. Does not report on price effects of eco-label per se. However, the beta for the non-certified variable (0.93) implies a rental premium of 6%-7%. Estimate that premiums are higher in Democrat voting areas. Much more detail required on sample and modelling to evaluate this paper. Using rents/vacancy rates of non-certified assets as independent variables in regression model produces high explanatory power but introduces issues of endogeneity and multi-collinearity which are not discussed in the paper. No information is provided on number of assets in sample nor number of eco-certified assets. The limited number of control variables may also be a problem. For instance, since there is a well-known positive correlation between educational attainment and Democratic voting, there may be an omitted variable bias. 18 Newell, McFarlane and Kok Report for Compares 206 NABERS rated office buildings with 160 non- NABERS rated buildings in Sydney and Canberra in March (4-6) Green Star rated Uses hedonic procedure to investigate rent and value ( a constructed proxy which incorporates gross rent, Providing no indication of statistical significance, the study finds a small rental and value premium for high NABERS-rated buildings 0.3% and 1.9% specifically. Finds a discount in The limited number of control variables may be a problem. The lack of any reporting of tests for statistical significance is a major omission.

12 Australian Property Institute September Das, Tidwell and Ziobrowski (2011) Published in Journal of Sustainable Real Estate buildings were also included. Covers 51% of the total office market. Looks at rental performance of 53 LEED buildings in San Francisco and Washington DC between 2007 and Control sample is 70 conventional buildings. vacancy, incentives, outgoings and yields ) determinants. Seem to omit a number of potential price determinants e.g. stories, age, unexpired lease terms, tenant quality etc Use panel data techniques to estimate the impacts on rents in rising and falling markets. value for offices with NABERS rating of 2.5 stars or less. Find positive effect of green label in falling market but no significant effect in rising market. Models have low explanatory power. There seems to be no controls for submarkets within San Francisco and Washington. It is not clear why the control sample is so small. No information is provided on the attributes of the LEED and non-leed offices size, age, class, height, occupancy rates etc. Given the extremely small sample size, the paper is rather unconvincing. 20 Fuerst, van de Wetering and Wyatt (2012) Looks at the relationship between achieved rent and EPC rating for a sample of 448 offices in the UK. Draws upon a database provided by CoStar for transactions in Includes potential price determining variables such as lease incentives and lease lengths. Uses a number of specifications of hedonic regression procedures. Finds that, compared to A rated buildings, there is a statistically significant discounts for properties rated D, E, F and G. This is confirmed in other model specifications. As expected, the discounts tend to increase as the energy performance deteriorates. In relation to A rated buildings, the discount for D rating is 30%, E is It is difficult to interpret the coefficients produced by the various specifications and to quantify a discount in percentage terms. The results do not seem to be stable between different model specifications. There are some typical potential omitted variable problems condition, design, internal specification etc.

13 34%, F is 100% (of its level compared to A) and G is 100% There are strong findings on the relationship between service charge and EPC rating. As expected, service charge levels decrease significantly with improvements in energy efficiency. 21 Reichardt, Fuerst, Rottke and Zietz (2011) Drawing upon the CoStar database, examines a time series of rents for a sample of 1584 Energy Star and 337 LEED certified buildings in 10 largest metropolitan areas of US over the period Uses a difference-indifferences approach and panel regression to estimate the changing rental premiums over time. DID approach indicates variations in the effect of certification between years. For Energy Star, for buildings certified between 2004 and 2007, there is an approximate increase in rent of 3-5%. For LEED rated buildings certified in 2008 and 2009, no significant effect on rent is identified. This is the first paper to address robustly the potential time varying effects of environmental certification on rental levels. It does not take into account buildings that have dual certification. This omission tends to bias downwards the effect of LEED certification. 22 Chegut, Eichholtz and Kok (2012) Published by RICS Draw upon CoStar, EGi and RCA databases to create a database of office leasing and sales transactions lease transactions including 64 BREEAM-certified buildings Uses hedonic procedure to investigate rent and sale price determinants. Apply propensity weighted OLS regression. For the panel regression fixed effects model, average rental premiums of 2.5% and 2.9% are estimated over the sample period. However, the premiums tend to vary. They emerge in 2005, increase in the period and decrease after this point. Estimate rental premiums of 25-30%. Different model specifications estimate quite different sales price premiums 17%, 26% and 38%. Paper is very transparent about the limited information on building quality. It is careful to acknowledge that premiums may be biased upwards by this omitted variable. The paper provides some interesting analysis of the effects of supply (on rents) and investor types on sale prices sale transactions including 69 BREEAM-certified buildings

14 23 Aroul and Hansz (2012) Published in Journal of Real Estate Research 24 Australian Bureau of Statistics (2007) Report for Department of Environment, Water, Heritage and the Arts Draws upon NTREIS transaction database of two centres in Texas, USA (Frisco and McKinney) From a sample of residential transactions, 6781 were tagged as green transactions. Draws upon a database of residential sales in the Australian Capital Territory for the period 2005 (2385 transactions) and 2006 (2719 transactions) Since a four star rating was a mandatory requirement, houses less than 10 years old are excluded. Uses standard hedonic procedure. Uses standard hedonic procedures. Estimates five different models. Estimates price premium of 2% for green transactions. When disaggregated into mandatory and voluntary green transactions, the respective premiums are 5% and 1%. For 2005 sample, estimate an (approximately) 1% premium for every 0.5 increase in rating (EER ranges from 0-5). For 2006 sample, estimate an approximately 2% premium for every 0.5 increase in EER. For pooled sample, relative to zero rating house estimates premiums of 1.6% (EER 1), 3% (EER 2), 5.9% (EER 3), 6.3% ((EER 4) and 6.1% (EER 5) It is unclear (albeit to me) how green and non-green buildings were differentiated given that there seems to have been a mandatory program. Overall the results look plausible. A potential issue is location and quality. Is it possible that green transactions may be associated with better quality neighbourhoods? Alternatively, meeting the green standards may have been associated with higher specification homes? An overlooked (at least by most researchers in this area) study recently brought to my attention (thanks, Chris Nicholls).. It is, however, one of the first studies of the price effects of mandatory eco-labelling in real estate markets. Find interesting result that the marginal addition to the premium declines as rating increases. The independent variables seem to be different between the 2005 and 2006 models. No rationale is provided. Explanatory power of model is high and there is a large range of quality controls. 25 Fuerst, Gabrieli and McAllister Sale prices for 4,591 transactions involving Class A Uses standard hedonic procedures. Estimate sale price premium of 14% for Energy Star, no premium for Study also examines whether investors with above average levels of investment in

15 (2012) Non-refereed paper presented at ARES conference offices sold in US Energy Star certified 108 LEED certified 355 LEED and Energy Star certified LEED and 24% for dual certified buildings. eco-certified assets tend to pay different levels of premium. Finds (unexpectedly) that it is non eco-investors who seem to pay more for eco-certified assets. Study is prone to the omitted variable problem affecting all hedonic studies. Ecocertified buildings may be best-in-class assets i.e there is an unobserved quality effect 26 Kok and Kahn (2012) Publication produced by UCLA Sale prices for 4231 ecocertified dwellings and 1,600,558 control dwellings in California sold between % ES label 47% Green Point 3% LEED for Homes 17% had multiple certifications Use standard hedonic procedures. Estimate sale price premium of 14.5% for Energy Star, no significant premium for Green Point and no premium for LEED for Homes When restricted to recently constructed homes, Energy Star premium reduced to 11%. No change for other ratings Find that level of premium is positively associated with eco-conscious consumers and severity of climatic conditions. Also find that premium becomes significant in 2010 and 2011 and not in 2008 and Paper acknowledges potential omitted variable problem due to unobserved differences between control and rated buildings. The paper does not seem to take into account dwellings that have multiple certification. This omission tends to bias the results. 27 Hyland, Lyons and Lyons University of Oxford, Working Paper Where information is provided on energy efficiency score, the sample has asking sale prices for approximately 20,000 dwellings in Republic of Ireland. Asking rental rates for approximately Use Heckman procedures to control for selection bias. Apply standard hedonic procedures. Find substantial asking price premiums related to D-rated properties for A (11%), B (6%) and E!! (2%). Find a discount of 6% for F/G. Whilst the study is interesting, the major limitation of this research is the lack of control for age. Since age and (quality of) location are likely to be significantly positively correlated with EPC rating, it is unlikely that the price effects are caused by

16 40,000 dwellings over the period Jan 2008-Mar For rental rates, find rental rate premiums related to D-rated properties for A (2%), B (4%). Find discounts for E (2%) and 3% for F/G. the energy labels. A more plausible explanation is that depreciation due to age is also included in the apparent discounts for energy efficiency. Version: October 2012