Real Estate Price Formation: Energy Performance Certificates and the Role of Real Estate Agents

Abstract

Improving energy efficiency in buildings is a major priority of industrialized countries. By eliminating market asymmetries, Energy Performance Certificates (EPCs) is a potential policy instrument when it comes to promoting energy efficiency of real estate. Real estate agents have an important role in providing information about dwellings for sale on the market. The aim of this paper is to study whether the introduction of EPCs changes the asking price setting of real estate agents. We take advantage of the fact that the introduction of a mandatory energy certification system represents a quasi-natural experiment, where we have data on house price and asking price. Based on the analysis, both of a hedonic model and a fixed effect model, we provide evidence that the implementation of EPCs did not affect the price setting of real estate agents. This indicates that real estate agents either disregard EPCs as providers of new information or believe that the market is indifferent to this kind of information. Our results also indicate that there are large similarities between the effects of energy labels on the asking prices and the transaction prices.

Keywords:

• Energy Performance Certificates
• energy savings
• EPC impact
• real estate agents
• sustainable housing policy

Introduction

Real estate agents play an important role in providing information about properties for sale to potential buyers. They assist the seller with the planning and execution of the sale, set the asking price, advertise the dwelling in newspapers and on the Internet, and arrange home viewings. Moreover, as in Norway, real estate agents conduct the sales negotiations (arranging the English auction), write the sale contract, and do the financial settlement between buyer and seller.

Being this much involved in the transaction process, real estate agents possess comprehensive information about the key components explaining the value of real estates. Typically, they appropriate information from the technical reports and value assessments of appraisers, gain knowledge about the energy labeling of buildings, familiarize themselves with the various characteristics of the properties through their close contact with the seller, and have firsthand knowledge about market conditions through their own professional experience. Their saying and doing may thus be a reliable source for understanding the role of Energy Performance Certificates (EPCs) in real estate price formation.

In a recent inquiry about the impact and reliability of EPCs in real estate markets,¹ Pascuas et al. (2017) conclude that real estate agents perceive EPCs to be unreliable, and their impact to be low. Indeed, according to the opinion of real estate agents, the energy performance of buildings is on the 10th position on the list of factors rendered most important for households when selecting properties to buy or sell (Pascuas et al., 2017). When asked about potential problems with the current energy certification, the real estate agents report things like additional costs for owners, insufficient knowledge of the customers, the practice of unreliable certification, and unnecessary paperwork, as the most important ones (Pascuas et al., 2017).

Olaussen et al. (2017) also conclude negatively when it comes to the expediency of the EPC system. Applying a dataset in which dwellings were sold multiple times in Oslo, Norway, they find no price premium of the EPCs and indicate that this may be so either because buyers ignore the EPCs at the purchasing moment, or because they are informed about the energy performance of buildings through other channels, thus making the EPCs redundant. Acting as communicators in the transaction process, real estate agents may represent such a channel. Accordingly, in the present paper, we test the hypotheses that real estate agents provide adequate information about the energy performance of buildings without the aid of EPCs.

Real estate agents partly provide information to potential buyers by capacity of setting the asking price. One hypothesis may be that price signal from the real estate agent through the asking price is stronger than the price signal from the EPC. Hence, we test the hypotheses by analyzing the relationship between the asking price and the EPCs. Considering the introduction of the mandatory energy certification system in Norway the 1. July 2010 as a quasi-natural experiment, we apply a panel dataset where we can follow the same dwellings sold in Oslo between 2000 and 2014.² The dataset contains dwelling characteristics and the asking price set by real estate agents. Moreover, it includes the energy labeling (EPCs) after its implementation in 2010. This allows us to study the asking price for the same dwellings both before and after the introduction of the EPCs. Hence, we can test whether real estate agents were able to assess the energy performance of buildings even before the introduction of the EPCs.

In Section “The Introduction of EPCs in Europe and Their Impact on Transaction Prices,” we discuss the introduction of EPCs in Europe and review the empirical literature on the impacts of EPCs on transaction prices in real estate markets. Section “The Real Estate Market in Norway” provides facts about the real estate market in Norway. Here, we describe the transaction process, the role of real estate agents, and the EPC system. We describe our data and present our empirical methods and results in Section “Data, Methods and Results.” Here, we formulate hedonic regression models for the asking price, both before and after the introduction of the energy labels. Then we address the omitted variable bias by estimating fixed effects models. Finally, we discuss our results and conclude in Section “Discussion and Conclusions.”

The Introduction of EPCs in Europe and Their Impact on Transaction Prices

Improving energy efficiency in buildings is a major priority of industrialized countries. As a response, the EU countries gradually introduced a system of Energy Performance Certificates (EPCs) in the period 2006–2010. The EPCs, or the energy labels, intend to provide reliable information about the energy performance of buildings to tenants and buyers. As improved energy performance of buildings may increase rents and sales prices, the EPC is supposed to generate incentives among owners to make investments in order to improve the energy efficiency (Bio Intelligence Service et al., 2013).

In the aftermath of the EU implementation, several studies have addressed the expediency of EPCs empirically. One of the pioneering statistical studies, Brounen and Kok (2011), analyze the impact of EPCs on residential prices in the Netherlands by using a hedonic regression model. They find a positive correlation between the best-rated dwellings and the sales prices, indicating that the EPCs have the intended positive price effect. Fuerst et al. (2015) and Jensen et al. (2016) support this finding for the cases of United Kingdom and Denmark, respectively. Moreover, for countries without available transaction prices, Hyland et al. (2013), the Bio Intelligence Service et al. (2013), and Marmolejo-Duarte and Chen (2018) find similar results by using listing prices. In addition, Devine et al. (2017) decompose the effects of sustainable investment on the value and performance of listed real estate investments and compare UK and the US with and without mandatory environmental reporting on investments properties. In a sense, the present study follows up on Devine et al. (2017) by focusing on another type of professional actors within real estate as the real estate agents are the professional party in the private dwelling transaction process, in the same way as real estate investors are in the REIT market.

Opinion-based research, however, comes to the opposite conclusion. Surveys carried out in the Netherlands (Murphy, 2014), the UK (Laine, 2011), and Germany (Amecke, 2012) all conclude that only a few householders use the energy labels during the transaction process, and maintain that the energy labels do not have the intended impact. Likewise, based on in-depth interviews with homeowners in ten European countries, as well as a large survey among homeowners in five European countries, Backhaus et al. (2011) conclude that the EPCs have a small or negligible impact on the investment decisions of homeowners. Moreover, in a more recent survey applied to real estate agents in eight countries, and based on in total 618 interviews (of which 90 was in Norway), Pascuas et al. (2017) provide supplementary evidence, reporting a disbelief among real estate agents when it comes to the positive influence of energy labels on rents and prices.

Applying a dataset in which dwellings were sold multiple times in Oslo, Norway, Olaussen et al. (2017) explain the contrasting conclusions of the statistical studies and the opinion-based research. Based on a similar hedonic regression analysis, they provide results in accordance with the above statistical studies, indicating a price premium associated with the energy labels. However, after controlling for fixed effects, they find no evidence of a price premium. That is, when improving the specification of the regression model, the result of the statistical analysis falls in line with those of the opinion-based research.

The present investigation follows up the work of Olaussen et al. (2017). A reasonable interpretation of their finding is that real estate agents act as communicators of the energy performance of buildings quite independently of the EPCs. Hence, our hypothesis to be tested is that real estate agents are capable to adjust the asking price according to the energy performance of buildings without the aid of the EPCs. If all the information captured by the EPCs was already known information to the real estate agents before the introduction of the EPC system was introduced, we would expect the introduction of EPCs to have no effect on the real estate prices before and after the EPC introduction in 2010.

The Real Estate Market in Norway

Most dwellings are sold through real estate agencies in Norway.³ Just companies with a special permit and lawyers are allowed doing business as a real estate agency. Moreover, in order to be called a real estate agent, you need a real estate agent bachelor’s degree, and two years of working experience from a real estate agency. The education includes subjects in business, marketing and law.

The real estate agencies sell dwellings primarily by arranging English auctions (Khazal et al. 2020; Olaussen et al. 2018). The buyers compete with open bids, and the highest bid wins the auction. Moreover, the seller cannot accept the first bid until 12 noon the first day after the showing of the dwelling (Sønstebø et al. 2021). Furthermore, the seller is committed to accept bids that are equal to, or larger, than the asking price. Because of this, the seller set the asking price officially. As a matter of fact, however, the real estate agents set the asking price.⁴

The real estate agents base their setting of asking price on the sale comparison approach. This appraisal method compares a specific dwelling to other dwellings with similar characteristics. Based on characteristics such as type, size, age, location, and standard, the real estate agents identify the price level they believe the market is willing to pay for a specific dwelling. Real estate agents normally have no technical competence. Hence, they frequently use appraisers as building experts. The appraiser typically generates a document where the technical information about the property appears. This includes surveying of the property, deception of the property, and a quality assessment of the main housing elements, preferably with a description of the condition measured against life expectancy. In several Norwegian cities, Oslo included, the appraiser will use this information to conduct a property valuation.

The real estate agents are well acquainted with the price mechanism and highly trusted by the market (Olaussen et al (2018). Indeed, in Oslo for the period of 2000–2014, the correlation between the asking price set by real estate agents and the actual sales price was 0.95 – where about 80% of asking prices fell within a range of 90–110% of the corresponding sales prices (Table 1). Hence, it is evident that real estate agents play an important role in providing information to potential buyers through their setting of the asking price.

Table 1. Correlation matrix, asking price and sale price.

	Asking price 2000–2014	Asking price 2012–2014	Asking price 2000–2009
Price 2000–2014	0.954
Price 2012–2014		0.890
Price 2000–2009			0.966

The table shows the correlation between the natural logarithms of price per square meter and asking price per for the three time periods, 2000 to 2014, 2000 to 2009 and 2012 to 2014.

The EPC system was implemented in the real estate market in Norway in July 2010. The energy performance certification was mandatory from the beginning; that is, since July 2010 all transactions must be accompanied by an EPC. The EPC is a legal document and it is required that it be shown to the buyer (Isachsen et al., 2010). The owner of the dwelling may hire an expert for carrying out the energy performance certification. However, in Norway there is also a self-assessment option for the owner. In most cases, the certificates issued by the owners themselves are more general than those carried out by experts. Table 2 reports the number of energy labels issued for dwellings in the Norwegian market in the period from 2009 to 2014. Note that the number of certificates issued in 2010 is about half that of the succeeding years since the system was made mandatory for sales from July 2010.

Table 2. Number of energy performance certificates issued.

	Year
	2009	2010	2011	2012	2013	2014
Number of dwellings	258	50,183	85,591	104,587	102,587	98,909

The table shows the number of Energy Performance Certificates issued in Norway between 2009 and 2014. Energy labeling was made mandatory on 1 July 2010. Source: Energimerking.no.

Data, Methods and Results

We have compiled the real estate transaction data from the property register of Oslo. The data are collected from the source eiendomsverdi.no. The transactions were registered in December 2014. We registered sales price, asking price, date, address, city district, size, type of housing, and the year of construction for all transactions. Moreover, we registered the energy label ranging from A to G in the post-label period of 2011–2014.

Table 3 reports the number of dwellings in our dataset with an asking price in the post-label period 2011 to 2014 and the pre-label period 2000 to 2009.

Table 3. Energy labels for dwellings traded with asking price 2011 to 2014 and 2000 to 2009.

	Total	A	B	C	D	E	F	G
2011–2014	2619	2	47	264	477	380	621	828
2000–2009	1225	0	2	51	205	201	351	415

The Hedonic Model

Housing assets possess several attributes (including historic preservation) and are as such composite products. In such cases, a hedonic model serves well for estimating the overall value of a dwelling. In order to mitigate some statistical problems, we apply the log-linear functional form (Malpezzi, 2003), which also makes the interpretation of coefficients easier. Thus, we apply a time dummy equation of the form: (1) $$\ln \left(P_{it}\right)={\gamma }_0+{\delta }_t+\sum _k{\alpha }_k{c}_{\mathit{kit}}+e_{it},$$(1) where $P_{it}$ is the asking price per square meter for dwelling i and period t ($t=1,\dots ,T$), ${\gamma }_0$ is the base year intercept, ${\delta }_t$ represents the time dummy coefficient for period t, defined as changes with respect to the base year intercept, where ${\delta }_t={\sum }_{s=1}^S{\delta }_s{d}_{\mathit{sit}},$ in which d_sit takes the value 1 when s = t and 0 otherwise, $c_{\mathit{kit}}$ is a set of explanatory variables for the presence of certain characteristics k, dwelling i, and period t, respectively. The explanatory variables are age, dummy variables for location (based on the different city districts),⁵ dwelling type, advertised energy labeling (from A to G), year dummies, and dummies for different size categories (measured in square meters). Note that the age variable is limited to 20 years. When the difference between the year of sale and construction is larger, the construction year seems to be of negligible importance. Probably, this is due to other circumstances, such as refurbishment and reconstruction, which are more likely to be important when the building is older. As we lack information about renovation, we have accordingly chosen to limit the age variable. It is constructed by taking 1/(sale year – construction year), the variable is set to 0 if the dwelling is more than 20 years and to 1 if the dwelling is sold in its construction year.

Hedonic Result

The result from our hedonic model are presented in Table 4. The logarithm of the asking price per square meter is explained by traditional explanatory variables comparable to those Olaussen et al. (2017) and Brounen and Kok (2011) use to explain the logarithm of transaction prices (see Table A1 in the Appendix), such as the age of the building, the neighborhood characteristics identified by the address, the dwelling type, year dummies, the energy label dummies, and three dummies for different size categories (in square meters).

Table 4. Energy labels and asking price (dependent variable: natural logarithm of asking prices per square meter).

	Pre-label 2000–2009		Post-label 2011–2014
	Coef.	(Robust std. err.)	Coef.	(Robust std. err.)
Age	0.226***	0.039	0.185***	0.038
Dummy St. Hanshaugen	−0.108***	0.032	−0.156***	0.036
Dummy Gamle Oslo	−0.267***	0.035	−0.357***	0.038
Dummy Grynerløkka and Sagene	−0.206***	0.028	−0.301***	0.037
Dummy Outer Oslo West	−0.135***	0.025	−0.224***	0.037
Dummy Outer Oslo East	−0.415***	0.025	−0.540***	0.036
Dummy Single-family houses	0.092***	0.022	0.044**	0.018
Dummy Townhouses	0.049**	0.020	0.022	0.016
Dummy Semi-detached houses	0.055**	0.023	0.050**	0.021
A	–	–	0.142	0.120
B	0.483***	0.133	0.203***	0.034
C	0.139***	0.029	0.143***	0.020
D	0.098***	0.018	0.114***	0.013
E	0.026	0.017	0.015	0.015
G	0.036***	0.014	0.046***	0.011
2013			−0.023	0.015
2012			−0.085***	0.020
2011			−0.173***	0.017
2009	0.634***	0.047
2008	0.620***	0.047
2007	0.570***	0.047
2006	0.490***	0.047
2005	0.386***	0.047
2004	0.288***	0.048
2003	0.189***	0.048
2002	0.194***	0.048
2001	0.193***	0.050
Small	−0.107***	0.020	−0.120***	0.011
Medium	−0.154***	0.022	−0.136***	0.014
Large	−0.276***	0.027	−0.243***	0.018
Constant	10.126***	0.052	11.149***	0.035
Adj R-square	0.66		0.47
Number of observations	1225		2619

***Significant at the 1% level. **Significant at 5% level. *Significant at 10% level.

Note: The Age variable is measured as Age is constructed by taking 1/(sale year – construction year), the variable is set to 0 if the dwelling is more than 20 years and to 1 if the dwelling is sold in its construction year. The dummies St. Hanshaugen, Gamle Oslo, Grynerløkka og Sagene, Outer Oslo West, and Outer Oslo East are dummies for different parts of Oslo (districts), and where the district Frogner is the baseline. The dummies Single-family house, townhouse and semi-detached houses are dummies for different housing types apartments as the baseline. The energy label dummies are A, B, C, D, E, and G with F as the baseline. The year dummies in the pre-label hedonic time dummy model from 2001 to 2009 have a baseline in year 2000. The post-label model has year dummies from 2011 to 2013 and have a baseline in year 2014. The dummies Small, Medium, and Large allow square meter prices to be different at different square meter levels. Small is dummy for Square meters between 51 and 80, Medium is 81–120 square meters, and Large is above 120 square meters. The baseline size is hence below 50 square meters.

Age, location, year, and size variables are all significant at the 1% level and with the expected sign, both for the data from the post-label period from 2011 to 2014, and for the data from the pre-label period from 2000 to 2009, with the exception of the year dummy for 2013 that is too similar in price compared with 2014 to get significant values. The house types are also significant at 1% and 5% in the data from the pre-label period, while only the dummy for semi-detached houses was significant in the post-label period data (1% level).

The energy label dummies are not very different in the post- and pre-label period. The energy labels in both datasets have the expected sign and are significant at the 1% level, for the B, C, and D labels. F is the reference label, and may thus explain that E is not significant since the difference between E and F is not too pronounced. The A label is not significant for the post-label period because of the very low n, and there are no dwellings with an A label in the pre-label period. Finally, we must acknowledge that the G category may be of a different type than the others. Residences where the owner does not take any action with respect to energy labeling are put in the G group. Hence, if the owner for some reason neglects to fill in the energy forms, the dwelling will end up in this category. We may therefore have buildings with a high energy performance in this group. Accordingly, the “wrong” sign for the G category is not surprising.

In the same way as Olaussen et al. (2017), Brounen and Kok (2011), and several others, find that higher energy labels are associated with higher sales prices, we find that higher energy labels are associated with higher asking prices. To test whether real estate agents have changed their valuation because of the new information from the energy labeling, we include a time dimension similar to that used by Olaussen et al. (2017). In addition to looking at the asking prices after the energy labels were made mandatory in 2010, we look at the asking price for the same dwellings before 2010, i.e., before they had an energy label. This applies to dwellings that are transacted both in the pre-label and the post-label period. Before 2010, the energy label of these dwellings was not known to real estate agents setting the asking price. But since we are able to follow a dwelling through both the pre- and post-label period, thus knowing which energy label a pre-label transaction was assigned in the post-label transaction, we may assess whether the asking price set by real estate agents in the pre-label period reflects the energy label of the post-label period.⁶ Indeed, it seems to be a price premium in the asking price associated with energy labels of the post-label period also before 2010. This indicates that the introduction of energy labels does not explain the price premium observed in the 2014 data. In fact, the significant label dummies in the data before labels were introduced, more than indicate that they capture something different from an effect of the labels themselves. As we are not able to find that the introduction of the energy label changed the asking price, real estate agents seem not to put much weight, if any, to the energy labels when they valuate dwellings.

The adjusted R-square is 0.47 for the post label data, while it is as high as 0.66 in the pre-label data for the asking price, while adjusted R-square for the appraisal value is 0.52 in the post label data, and 0.64 in the pre-label data. In appendix Table A2, we have included a robustness check where we categoricating A–C as Grenn, and E–G as non-green, D has been treated as neutral (baseline). The results are very similar to the results in Table 2 and do not change the intuition.

Fixed Effects Model

We consider the event of the introduction of the Energy Labeling System in Norway as a quasi-experiment. Recall that Energy labeling was introduced in July 2010 and that the use of energy labels went from almost non-existing to fully implemented within one year (Table 2). We have an unbalanced panel dataset covering the period from 2000 to 2014, where the same dwellings typically are sold both before and after the introduction of energy labels.

We apply a two-way fixed effects model (Difference in difference), making use of the panel structure of the data, to account for unobserved effects. During the fixed effects transformation, the variables are time-demeaned for each unit, which makes the estimator explore the relation between transaction price per square meter and the presence of different energy labels within a unit. When including a dummy for events of energy label, its coefficient reports how much the mean value of the transaction price per square meter changes when dwellings change from non-labelled to labelled, which is made possible when the transaction price per square meter from before and after the energy label is known. We thus assess the price effect from the new information provided by the event of energy label itself. The energy labels came as new information about the dwellings after 2010. The independent variables have the value 1 for its given energy label after 2010 (B–G) and 0 otherwise. There are no dwellings sold before 2010 with energy label A.

We also need to control for the general development in the house prices over time. These price changes represent changes in macroeconomic variables and are not related to the introduction of energy labels. Because of time trends, house price data are non-stationary, and this can cause problems in our statistical inference, and may give misleading results due to spurious relationships. To make our dependent variable (price per square meter) stationary, we divide it with an index value from the observation year. Similar to Olaussen et al. (2017), we use a house price index that is weighted average of a hedonic index for Oslo from the Norwegian central bank (Eitrheim and Erlandsen, 2004) and a Case Shiller repeated sales index for Oslo by Oust (2015). This gives us stationary dependent variables, but there seems to be some signs of heteroscedasticity, and hence we report robust standard errors.

We estimate a fixed effect equation of the form: $$Y_{it}=\mathrm{ }{\alpha }_i+{\beta }_1{X}_{it1}+\dots +{\beta }_k{X}_{\mathit{itk}}+{\beta }_a{X}_{\mathit{ita}}+u_{it},\mathrm{ }(i=1,2,\dots ,\mathrm{\hspace{0.17em}}N,\mathrm{ }t=1,\mathrm{\hspace{0.17em}}2,\mathrm{\hspace{0.17em}}\dots ,\mathrm{ }T,\mathrm{ }k=A,\mathrm{\hspace{0.17em}}B,\mathrm{\hspace{0.17em}}\dots ,G),$$ where $Y_{it}$ is the natural logarithm of the dwelling price per square meter for dwelling i, deflated by the value of the house price index for transaction period t, ${\alpha }_i$ is the unknown intercept for dwelling i (the fixed effects), $X_k$ is the independent dummy variable for the energy performance certificates A-G for characteristic k, while ${\beta }_k$ is the accompanying coefficient for characteristic k. $X_a$ is a variable for age, Age is constructed by taking 1/(sale year – construction year), the variable is set to 0 if the dwelling is more than 20 years and to 1 if the dwelling is sold in its construction year, while ${\beta }_a$ is the accompanying coefficient for characteristic a. We have an error term $u_{it}$ for dwelling i and period t.

Fixed Effects Result

The results from the fixed effects models are presented in Table 5 and confirms the results from the hedonic model above. If real estate agents base their valuation on the energy label, we would expect that dwellings with better energy labels should be associated with positive and increasing coefficients, and those of poorer energy labels with negative and decreasing coefficients. However, quite the opposite, it seems that dwellings with better energy labels are given lower asking prices as compared to those with poorer energy labels. In Table A2 in the Appendix, we report the results from the adjusted Wald test. Again, we find no evidence for the case that better energy labels are associated with higher asking prices. Note that the R-square is low for the regression, 0.022. At the same time, the intraclass correlation coefficient, rho, which reports the correlation among the observations within each group, is quite high, 0.840.

Table 5. Energy labels and asking prices, fixed effect model (dependent variable: natural logarithm of asking prices per square meter, adjusted with price index).

	Coef.	Robust std. error	t-Value	95% coef. interval
B	−0.077***	0.023	−3.33	−0.123	−0.032
C	−0.012	0.025	−0.50	−0.061	−0.036
D	−0.021	0.015	−1.37	−0.051	−0.009
E	−0.046***	0.127	−3.65	−0.071	−0.021
F	−0.006	0.011	−0.59	−0.027	0.014
G	0.009	0.010	0.92	−0.010	0.028
Age	0.136***	0.023	5.82	0.090	0.182
Constant	7.493***	0.005	1399.16	7.482	7.504

R-square: within = 0.035, between = 0.028, overall = 0.028.[AQ]

Number of obs. = 3846, groups = 2219.

Obs. per group: min = 1, average = 1.7, max = 7.

Prob>F = 0.000, rho = 0.830.

***Significant at the 1% level. **Significant at 5% level.

Note: Following Olaussen et al. (2017) the dependent variable is divided by an index value from the observation year to make it stationary. The index is a weighted price index from Eitrheim and Erlandsen (2004) and Oust (2015). The energy label dummies are A, B, C, D, E, F, and G, and where A is omitted due to collinearity. Age is constructed by taking 1/(sale year – construction year), the variable is set to 0 if the dwelling is more than 20 years and to 1 if the dwelling is sold in its construction year. The dataset is unbalanced with observations from 2000 to 2014, and where observations from 2010 are excluded (energy certificates were introduced in July 2010).

Discussion and Conclusions

For environmental and energy dependency reasons, improving energy efficiency in buildings is a major priority of industrialized countries. Energy Performance Certificates (EPCs) are allegedly an important policy instrument aimed to promote energy efficiency of real estate by eliminating market asymmetries. Several studies have addressed the expediency of EPCs empirically. Statistical studies, like Brounen and Kok (2011), Fuerst et al. (2015), and Jensen et al. (2016), indicate a positive correlation between energy label of dwellings and the transaction prices. Moreover, Hyland et al. (2013), the Bio Intelligence Service et al. (2013), and Marmolejo-Duarte and Chen (2018) find similar results statistically by using listing prices rather than transaction prices. Opinion-based research, however, like Murphy (2014), Amecke (2012), Laine (2011), and Backhaus et al. (2011), comes to the opposite conclusion. Olaussen et al. (2017) bridge the gap between the statistical studies and the opinion-based research. By improving the specification of the hedonic regression analysis, applying a fixed effect model, they show that the results of the statistical analysis fall in line with those of the opinion-based research. Their study indicates that buyers either ignore the EPCs at the purchasing moment or that they are informed about the energy performance of dwellings through other channels, in both cases making the EPCs redundant. Moreover, Olaussen et al. (2019) investigate whether changes in energy prices have an impact on the valuation of EPCs in Norway. They found that in the hedonic models, EPCs seem to have much higher valuation than what should be expected when accounting for the theoretically calculated energy expenses. They also found that changes in the energy price did not change the empirically estimated value of EPCs. Both of these findings indicate that a higher pricing of homes with high EPC scores, is due to some other properties of dwellings rather than the energy savings as indicated by the EPCs.

The present study provides evidence that real estate agents may represent such a channel. We show that real estate agents have been able to adjust the asking price to the genuine energy standard of dwellings both after and before the implementation of EPCs. That is, by the capacity of setting the asking price, real estate agents provide buyers with information about the energy efficiency of dwellings quite independently of the energy label. It seems that the EPCs do not provide additional information to the real estate market. So even if the energy performance of buildings matters in the purchasing decision, our study provides evidence that real estate agents were able to pick up and mediate information about the energy performance before the EPC introduction.

In case buyers are not ignorant about the energy performance at the moment of purchase, we believe our findings may reflect the important role played by real estate agents as providers of reliable information in the transaction process. The appropriate important information about the energy performance of dwellings through their close contact with the seller, through their access to information from the technical reports of appraisers, and finally through their own professional experience. Moreover, the high correlation between the asking price and the actual sales price indicates that real estate agents are well trusted by the market.

The results also fall in line with the work of Pascuas et al. (2017), reporting a belief among real estate agents that energy labels have low or no impact on transaction prices. Based on this, we believe that the price signal buyers receive from real estate agents through the asking price is more important than the signal received from the energy label. In so far real estate agents do not change their valuation after the introduction of energy labels, it is unlikely that transaction prices will change. It should be noted that the results of Pascuas et al. (2017) also indicate that real estate agents in Norway gain distinction. The share of real estate agents finding the energy performance of buildings to be important, useful, and reliable is clearly lower in Norway as compared to other countries. This may indicate that real estate agents have a more pronounced role as a provider of information during the transaction process in the Norwegian real estate market. Moreover, if real estate agents are equally informed in other countries where there is no auction, but a fixed price set by the real estate agent, there should be no reason to expect that the EPC system contributes to more than a redundant extra cost. It should nevertheless be mentioned that based on the short time span from the introduction of the EPC system in 2010 until our data ends in 2014, we cannot be certain that there is no development over time in how real estate agents apply the information from the EPC system.

Notes

1 The survey was carried out in eight European countries: Austria, France, Germany, Italy, Norway, Poland, Romania, and Spain (Pascuas et al., 2017).

2 The data were collected by hand, 2014 was chosen as the main year for the data collection. In 2014 the time passed since the EPC was made mandatory should be more than sufficient to see the full effect of the introduction. At the same time 2014 is close enough in time that it is easy to keep track of and check for events that could interfere with the causality of the results.

3 In Oslo, 91% of all dwellings sold in 2011 were sold through real estate agencies (Stamsø, 2012).

4 The seller will sometimes take part in the final discussion about the level of the asking price, but in most cases the seller trust the judgement of the real estate agent.

5 We divide the city into six districts, Frogner (that is the baseline in our model), St. Hanshaugen, Gamle Oslo, Grymerlølla and Sagene, Outer Oslo West and Outer Oslo East.

6 Here we assume that the energy status of the dwellings is constant through the two periods.

Appendix

Table A1. Energy labels and price (dependent variable: natural logarithm of prices per square meter).

	Pre-label 2000–2009		Post-label 2011–2014
	Coef.	(Robust std. err.)	Coef.	(Robust std. err.)
Age	.260***	0.055	0.263***	0.046
Dummy St. Hanshaugen	−0.085***	0.027	−0.105***	0.025
Dummy Gamle Oslo	−0.219***	0.032	−0.282***	0.026
Dummy Grynerløkka and Sagene	−0.143***	0.025	−0.238***	0.022
Dummy Outer Oslo West	−0.093***	0.023	−0.183***	0.020
Dummy Outer Oslo East	−0.368***	0.023	−0.488***	0.020
Dummy Single-family houses	0.050**	0.020	0.046***	0.015
Dummy Townhouses	0.035*	0.020	0.022	0.015
Dummy Semi-detached houses	0.046**	0.018	0.035**	0.017
A	–	–	0.019	0.118
B	0.246***	0.092	0.147***	0.027
C	0.115***	0.021	0.107***	0.016
D	0.097***	0.015	0.072***	0.013
E	0.030*	0.016	0.005	0.013
G	0.027***	0.055	0.053***	0.011
2013			−0.037***	0.013
2012			−0.079***	0.015
2011			−0.171***	0.014
2009	0.601***	0.023
2008	0.555***	0.023
2007	0.566***	0.023
2006	0.497**	0.023
2005	0.359***	0.023
2004	0.281***	0.024
2003	0.152***	0.024
2002	0.179***	0.024
2001	0.117***	0.025
Small	−0.089***	0.014	−0.144***	0.012
Medium	−0.137***	0.016	−0.164***	0.014
Large	−0.252***	0.021	−0.289***	0.017
Constant	10.126***	0.031	11.178***	0.023
Adj R-square	0.66		0.49
Number of observations	1887		2781

***Significant at the 1% level. **Significant at 5% level. *Significant at 10% level.

The Age variable is measured as Age = 1/(sale year-construction year) for all buildings where difference between sale and construction year is less than 20 years. The dummies St. Hanshaugen, Gamle Oslo, Grynerløkka og Sagene, Outer Oslo West, and Outer Oslo East are dummies for different parts of Oslo (districts), and where the district Frogner is the baseline. The dummies Single-family house, townhouse and semi-detached houses are dummies for different housing types apartments as the baseline. The energy label dummies are A, B, C, D, E, and G with F as the baseline. The year dummies in the pre-label hedonic time dummy model from 2001 to 2009 have a baseline in year 2000. The post-label model have year dummies from 2011 to 2013 and have a baseline in year 2014. The dummies Small, Medium, and Large allow square meter prices to be different at different square meter levels. Small is dummy for Square meters between 51 and 80, Medium is 81–120 square meters, and Large is above 120 square meters. The baseline size is hence below 50 square meters.

Table A2. Energy labels and asking price, green vs non-green (dependent variable: natural logarithm of asking prices per square meter).

	Pre-label 2000–2009		Post-label 2011–2014
	Coef.	(Robust std. err.)	Coef.	(Robust std. err.)
Age	0.220***	0.039	0.189***	0.038
Green	0.057*	0.029	0.036**	0.018
Non-green	−0.080***	0.016	−0.091***	0.011
Attributes FE	Yes		Yes
Time FE	Yes		Yes
Constant	10.224***	0.060	11.263***	0.036
Adj R-square	0.66		0.47
Number of observations	1225		2619

***Significant at the 1% level. **Significant at 5% level. *Significant at 10% level.

Note: We group A–C as Green, and E–G as non-green, D is the baseline. The Age variable is measured as Age = 1/(sale year-construction year) for all buildings where difference between sale and construction year is less than 20 years the variable is set to 0 if the dwelling is more than 20 years and to 1 if the dwelling is sold in its construction year. We use time fixed effect for 2001–2009 in the pre-label regression, baseline in year 2000. For the post-label model we use time fixed effect for 2011–2013 and have a baseline in year 2014. Both models have the following attributes fixed effects: Locations: St. Hanshaugen, Gamle Oslo, Grynerløkka og Sagene, Outer Oslo West, and Outer Oslo East, where the district Frogner is the baseline. Dwelling type: Single-family house, townhouse and apartments as the baseline. Dwelling size: Small, Medium, and Large allow square meter prices to be different at different square meter levels. Small is dummy for Square meters between 51 and 80, Medium is 81–120 square meters, and Large is above 120 square meters. The baseline size is hence below 50 square meters.

Table A3. Adjusted Wald test and asking price.

	Asking price
$H_0$	$F\left({df}_1,{df}_2\right)$	Rejection $H_0$	Better energy label gives higher asking price
B = C	F(1, 2218) = 5.53	Yes	No
B = D	F(1, 2218) = 6.01	Yes	No
B = E	F(1, 2218) = 1.44	No	No
B = F	F(1, 2218) = 7.96	Yes	No
B = G	F(1, 2218) = 11.93	Yes	No
C = D	F(1, 2218) = 0.12	No	No
C = E	F(1, 2218) = 1.57	No	No
C = F	F(1, 2218) = 0.05	No	No
C = G	F(1, 2218) = 0.66	No	No
D = E	F(1, 2218) = 1.66	No	No
D = F	F(1, 2218) = 0.66	No	No
D = G	F(1, 2218) = 2.77	No	No
E = F	F(1, 2218) = 5.99	Yes	No
E = G	F(1, 2218) = 12.08	Yes	No
F = G	F(1, 2218) = 1.11	No	No

Note: The Wald statistics test in STATA show whether the coefficients are equal by testing if their difference is zero, e.g. ${\beta }_2-{\beta }_3=0$ is run to test if B = C. The third column conclude on the H₀ (critical value at 5% level, F(1,2218)=3.84). Note that the results for label B is based on two observations. Concludes whether better certificates gives a price premium based on a transformation from the F-statistics in the Wald test to a one sided t-test, which is straightforward as the F distribution is a squared t-distribution when the F statistics has only one degree of freedom for the numerator. Then, we utilize that the t-distribution is symmetric, and hence the p-value in the one sided case is simply the half of the p-value obtained from the F-test.