**4. Results**

The yearly energy costs are used to calculate the present value of the different energy labels. In Figure 2, this is illustrated with an example from a 40 m<sup>2</sup> apartment. Note that there is a distinct difference between present values in 2009 and 2014. This difference is due to the discount rate being lower in 2014 than in 2009. This effect dominates, even if the energy price is slightly higher in 2009 than in 2014 (see Figure 1).

The expected price premiums of the energy labels in 2014 are presented in Table 5 and those for 2009 in Table 6. The price premiums are given per square meter. For instance, if we take an apartment of 40 m<sup>2</sup> with the energy label C, the expected price premium is NOK 682 per square meter, compared with a similar apartment with the energy label D. The same comparison in 2009 (Table 6) yields a price difference of NOK 502.

**A**

 **B**

**Figure 2.** Present value of the expected energy cost for different EPCs. The figure shows the present value of the expected energy cost for a 40 m<sup>2</sup> apartment in 2009 and 2014 for different EPC grades in Norwegian kroner (NOK). (NOK 1 = €0.11 (per 31.12.2014)).


**Table 5.** Expected price premium per m<sup>2</sup> from different EPCs in 2014 in NOK.

The difference between the expected and actual value added is interesting. First, for 2014, we find that the actual value added (Table 7) is higher than the expected price premium (Table 5). This implies that dwellings with better energy labels receive a higher premium than can be explained by the energy costs; that is, a value added beyond the cost savings expected from a more energy-efficient dwelling. The pattern is confirmed in the 2009 tables (Tables 7 and 8). This means that, even before the energy labels were available to buyers, there was a price premium beyond what could be explained by the energy cost. However, these results are dependent on the rate of discount and sensitivity analysis shows that if the rate of discount in 2014 was set at 4%, the difference between the actual and expected price premium is much lower.

 **D**  **E**

**FG**

 **C**


**Table 7.** Estimated (actual) price premium per m<sup>2</sup> from different EPCs in 2014 in NOK.



Table 9 presents the results from the hedonic models for the period after the introduction of EPCs (post-label), Model 1, and before the introduction of labels (pre-label), Model 2. All coefficients have the expected sign and are significant at the 1% level, except for the two location dummies for the districts of St. Hanshaugen and Outer Oslo West in Model 2, which are significant at the 5% level. The most interesting result in this analysis is the present value of the energy cost per square meter, which is positive and significant at the 1% level in both 2014 and 2009. The difference between the coefficients is rather small and not significantly different at the 1% level. Note that the 1% confidence intervals for the 2014 coefficient (0.061–0.119) and the 2009 coefficient (0.029–0.133) overlap significantly. Note also that the overall results do not change if we substitute the present value of the energy cost per square meter with the expected energy cost per square meter, or if we look at different dwelling types separately (these results are not reported in the paper).

**Table 9.** Energy costs and dwelling prices. Hedonic models, dependent variable: natural logarithm of transaction prices per m2.


Note: \*\*\*, \*\*, signal significance at the 1% and 5%levels, respectively. See Section 3.2 above for variable definitions.

House prices are in fixed 2014 prices, and every dwelling price is multiplied by the house price index value for 2014 divided by the house price index value of the year of the transaction. Model 1 hence consists of buildings sold in 2011–2014, in 2014 prices.

### *4.1. Robustness Check*

The nature of the potential causal relationship between energy labels and sales prices is crucial for our analysis. As a robustness check to test this relationship, we utilize the natural experiment that took place when the energy labels became mandatory in July 2010. The data allow us to compare the transaction prices of dwellings sold before and after the introduction of the EPC system in July 2010. If energy labels a ffect the sale prices, then two houses sold in, for example, 2008, for approximately the same price, should have approximately the same price as each other when resold after July 2010 if they were given the same energy label. On the other hand, if one of them received a higher energy label than the other, it should, ceteris paribus, have a higher resale price.

### 4.1.1. The Weighted Repeat Sales Method

The robustness check is performed with the weighted repeat sales method. The following model is applied [23,24]:

$$\ln(p\_n^t/p\_n^s) = \sum\_{t=0}^T \gamma \prime D\_n^t + \mu\_{n\prime}^t \tag{6}$$

where *Pt n* is the price at the time of the resale, *ps n* is the price of the previous sale, *D<sup>t</sup> n* is a dummy variable with the value 1 in the period in which the resale occurs, –1 in the period in which the previous sale occurs, and 0 otherwise. μ*t n* is the error term. To account for the possibility that the residual variance increases with increasing time intervals between sales, we apply the weighted repeat sales (WRS) method developed in [23].

The data does not contain enough observations before the introduction of the energy performance certificates in 2010 to create indices for energy label A and energy label B. To remove the house price trend, we divide the indices with a repeated sales index constructed based on all the dwellings in the dataset. We use a simple Dickey–Fuller test, to test whether variables are stationary (Table 10). All the variables have one unit root and are thus di fferentiated to make them stationary.


**Table 10.** Dickey–Fuller tests for unit root of all variables.

Note: The 5% interpolated Dickey–Fuller critical values are used. No lags are included in the test. Ln means that natural logarithms have been used. C = index with dwellings with energy label C; D = index with dwellings with energy label D; E = index with dwellings with energy label E; F = index with dwellings with energy label F; and G = index with dwellings with energy label G.

We use a Durbin Watson test and a Portmanteau test for white noise which shows indication of autocorrelation AR (1). To reduce the problem of autocorrelation, we apply a Prais–Winsten regression [25].

Our regression is:

$$\mathbf{Y}' = \beta\_0 (1 - \rho) + \sum \beta\_j \mathbf{x}'\_{j\mathbf{t}} + \sum \delta\_j \mathbf{s}\_{j\mathbf{t}} + \varepsilon\_{j\mathbf{t}} \tag{7}$$

where β*j* is the coefficient for the *j*th explanatory variable *x*, δ*j* is the coefficient for the *j*th dummy variable *s*, and <sup>ε</sup>*jt* is the error term. The symbol - indicates the transformation of the variables. The explanatory variable is the present value of the energy cost in the different categories. In addition, we use a dummy for the time when the energy labeling was made mandatory, from July to December 2010.

### 4.1.2. Repeat Sales Results

We start to explore the effect of introducing energy labels by constructing price indices for the different labels and let them all have a value of 100 in the year 2000 (Figure 3). The figure shows some price variations, but do not indicate a price effect from the energy performance certificates in July 2010. If energy labeling has the price effect found in the hedonic data, we should expect a kink with an increasing slope after July 2010 for the most energy efficient energy labels. However, it is difficult to ascertain any shift taking place in July 2010.

**Figure 3.** Dwelling price indices in different energy label categories. All of the indices start at 100 in year 2000 (Note: Fixed house price indices between 2000 and 2014, with trend removed. All indices start at 100 in 2000. As energy labeling was made mandatory on 1 July 2010, the year 2010 has been given two data points in the indices, one for January–June and one for July–December. The vertical line indicates when the energy labeling became mandatory. C = index for dwellings with energy label C; D = index for dwellings with energy label D; E = index for dwellings with energy label E; F = index for dwellings with energy label F; and G = index for dwellings with energy label G).

In Table 11 we test for the effect of introducing energy labels controlling for the present value of the energy cost. The dependent variable is the house price in the different energy label categories, and where we regress on the main index as well a dummy variable for the second part of 2010, when the energy label was made mandatory. The adjusted R-squares are all negative, while the Durbin Watson statistics, transformed after using the Prais–Winsten regression, range from 1.59 to 2.40, which means that we keep the null hypothesis of zero autocorrelation. (With *n* = 15 and *k* = 2, the retained H0 critical values range from 1.25 to 2.75.) If energy labeling has the price effect found in the post-label hedonic data, we should expect significant dummy coefficients in Table 11. However, none of the dummies are significant, nor the present value of energy cost. Hence, despite the strong label effect demonstrated in the hedonic post-label model (Model 1), just as in the pre-label hedonic regression (Model 2), we find

no evidence to support the price premium e ffect. We also find no price e ffect from the present value of energy cost.


**Table 11.** House price under di fferent energy labels.

Note: We compare how well the dummy for the period when energy labeling was made mandatory (July–December 2010) together with the PV of energy cost is able to explain the house prices indexes for different energy labels. Ln C = logarithmic house price index with dwellings with energy label C; Ln D = logarithmic house price index with dwellings with energy label D; Ln E = logarithmic house price index with dwellings with energy label E; Ln F = logarithmic house price index with dwellings with energy label F; and Ln G = logarithmic house price index with dwellings with energy label G. DW transf. refers to the Durbin–Watson statistic, transformed after using the Prais–Winsten regression.

### **5. Discussion and Concluding Remarks**

The energy performance certificate system was introduced in Europe to provide buyers with better information about the energy performance of dwellings. In part, the aim of this policy was to provide better valuations of dwellings when they are sold and to give buyers incentives to purchase energy-e fficient dwellings. Earlier studies in this area have yielded contradictory conclusions. Brounen and Kok found that there was a significant price premium associated with energy labels in the real estate market in the Netherlands [4], whereas other studies, such as Murphy, found little or no e ffect of energy labels in the same market [6]. The present paper follows up the study by Olaussen et al. of the Norwegian real estate market [11]. Replicating the hedonic model by Brounen and Kok for Norwegian data, Olaussen et al. found the same results as Brounen and Kok [4,11]. However, when running a fixed e ffect model with data before and after the introduction of energy labels in 2010, they found that something other than the energy label must explain the apparent price premium. One potential explanation for this is that the energy e fficiency of the dwelling was known to the buyers even before the labeling system was issued. To test for this, we use the energy price over time to see if the cost of energy may be the underlying explanation. By controlling for the present value of the expected energy consumption, we find no evidence of energy costs being important for the energy label premium.

By applying data for energy prices and the rate of discount, and the associated demands for the di fferent energy label categories, we calculate the expected price premium that dwellings with better energy labels should achieve compared with similar dwellings with lower energy labels. Then, these price premiums are compared with the actual price premiums estimated in the hedonic models. The analyses show that the actual price premiums are much higher than the expected price premiums based on the energy cost di fferences. Moreover, we find this di fference both before and after the energy label system was introduced. In addition, we find no significant di fferences in the actual price premium before and after the introduction of the energy labels in 2010. The same results are provided by the robustness check, in which we apply the repeated sales method; that is, we find that the present value of energy costs has no e ffect on the price of dwellings.

These results support previous studies that showed that the energy label does not a ffect the price of dwellings at the time of sale [11,26,27]. This is in line with the inferences of several survey studies, which indicate that when people buy a dwelling, they pay considerably less attention to its energy performance compared with other factors, such as the availability of garden and outdoor space, the location, the neighborhood, and the size of the property. Hence, there are reasons to believe that, when energy labels have been associated with price premiums in other studies, this results from factors other than the energy labels themselves. One explanation for our result may be that the buyers are well informed about the energy e fficiency of the dwellings even without the energy labels and, hence, were

already well informed before the energy label system was introduced. Another explanation may be that we have omitted explanatory variables in our models. Potential omitted variables may be the standard of the dwelling, for example, how recently it was renovated, or different amenities associated with the building. This explanation is in line with [26–28]. These omitted variables were visible to buyers before the energy label system was introduced, and it is quite likely that, e.g., the dwelling standard is closely correlated with the energy efficiency. Hence, it may be that the price premium associated with the energy label is explained by the standard of the dwelling. However, data regarding when dwellings have been renovated are not easily accessible. A detailed, in-depth study of potential omitted variables correlated with EPCs may be a fruitful path for future research.

**Author Contributions:** All authors have contributed equally, L.K. prepared an initial first draft, which was completed, corrected, reviewed and revised by J.O.O., A.O., and J.T.S.

**Funding:** This research received no external funding.

**Conflicts of Interest:** The authors declare no conflict of interests.
