The Inﬂuence of Energy Certiﬁcation on Housing Sales Prices in the Province of Alicante (Spain)

: This work examines the implementation of energy labelling by the residential real estate sector. First, it considers the interest by real estate sellers in not publishing energy certiﬁcation information, and then, it quantiﬁes the impact of the housing’s energy certiﬁcation on the asking price. The results are compared with those obtained from other studies conducted in distinct European countries. The study’s ﬁnal sample was collected, including information from 52,939 multi-family homes placed on the real estate market in the province of Alicante (Spain). One-way analysis of variance (ANOVA) was used, as well as an ordinary least squares regression model. This study highlights the fact that, in the current market, owners and sellers have no incentive to reveal the energy certiﬁcation, since this permits them to sell homes with low energy ratings at prices similar to those of more energy e ﬃ cient homes. In addition, it was found that homes with better energy ratings (letters A and B) are not sold at higher prices than homes with other rating letters, unlike the case of other European countries that were examined.

The European directives [2,3] establish a mandatory certification system, the so-called "ABCDEFG qualification", which rates buildings based on their energy efficiency, similar to the classification used for household appliances. In addition, these directives require the publication of an energy performance certificate (EPC), to be included in the documentation supplied by owners to purchasers or renters. These policies are an attempt to offer increased transparency and information to consumers, to assist in decision-making related to property purchase or rental.
In Spain, these directives result from the enforcement of a series of decrees [14,15] requiring energy certification and the presentation of an energy efficiency label in properties placed on the rental or sales market. On the other hand, the level of compliance with the directive varies depending on the specific case. In the case of real estate sales, compliance is significant, given that certification documentation is required by the notary public when formalizing a deed of sale. In the case of publishing energy In terms of energy and climate ( Figure 2), Spain's climatic diversity should be considered (CTE-HE 2013) [35], since it has always, and continues to, significantly condition the energy characteristics of the building stock. In Spain, within the same region, it is possible to find mild climates in Mediterranean coastal areas and continental climates with more extreme temperatures, as occurring in the Alicante province. This characteristic may even be found in cases in which the separation between climate zones is barely 70 km, as occurs in the coast and the interior of the province, where mountain chains exceeding 1000 m in height are found. This climatic diversity leads to a distinct energy allocation of the buildings. Therefore, electricity is frequently used to heat and cool buildings on the Mediterranean coast, via heating pumps, whereas in the interior, it is more common to find buildings that are heated with community or individual heating systems using natural gas. As for the heating system used in the Valencian community [36], centralized or collective systems represent only 4.6% (10.56% on a national level), while heating in the entire housing represents only 27.4% (as compared to 46.30% across Spain). In this province, it is common to use small heating devices for single rooms (54.2%) or even, to have no heating devices at all (13.8%). (see Table A1 of the Appendix A). In terms of energy and climate ( Figure 2), Spain's climatic diversity should be considered (CTE-HE 2013) [35], since it has always, and continues to, significantly condition the energy characteristics of the building stock. In Spain, within the same region, it is possible to find mild climates in Mediterranean coastal areas and continental climates with more extreme temperatures, as occurring in the Alicante province. This characteristic may even be found in cases in which the separation between climate zones is barely 70 km, as occurs in the coast and the interior of the province, where mountain chains exceeding 1000 m in height are found. This climatic diversity leads to a distinct energy allocation of the buildings. Therefore, electricity is frequently used to heat and cool buildings on the Mediterranean coast, via heating pumps, whereas in the interior, it is more common to find buildings that are heated with community or individual heating systems using natural gas. As for the heating system used in the Valencian community [36], centralized or collective systems represent only 4.6% (10.56% on a national level), while heating in the entire housing represents only 27.4% (as compared to 46.30% across Spain). In this province, it is common to use small heating devices for single rooms (54.2%) or even, to have no heating devices at all (13.8%). (see Table A1 of the Appendix A). In terms of energy and climate (Figure 2), Spain's climatic diversity should be considered (CTE-HE 2013) [35], since it has always, and continues to, significantly condition the energy characteristics of the building stock. In Spain, within the same region, it is possible to find mild climates in Mediterranean coastal areas and continental climates with more extreme temperatures, as occurring in the Alicante province. This characteristic may even be found in cases in which the separation between climate zones is barely 70 km, as occurs in the coast and the interior of the province, where mountain chains exceeding 1000 m in height are found. This climatic diversity leads to a distinct energy allocation of the buildings. Therefore, electricity is frequently used to heat and cool buildings on the Mediterranean coast, via heating pumps, whereas in the interior, it is more common to find buildings that are heated with community or individual heating systems using natural gas. As for the heating system used in the Valencian community [36], centralized or collective systems represent only 4.6% (10.56% on a national level), while heating in the entire housing represents only 27.4% (as compared to 46.30% across Spain). In this province, it is common to use small heating devices for single rooms (54.2%) or even, to have no heating devices at all (13.8%). (see Table A1 of the Appendix A). The document is organized as follows: In the second section, the materials and methods are described, detailing the sources used and the database generated. The third section offers the results. The fourth section contains the discussion and finally, a summary of the conclusions obtained is presented. The document is organized as follows: In the second section, the materials and methods are described, detailing the sources used and the database generated. The third section offers the results. The fourth section contains the discussion and finally, a summary of the conclusions obtained is presented.

Materials and Methods
First, an analysis of variance (ANOVA) was proposed to determine whether or not differences exist in the offered prices, based on the published energy qualification. To examine the economic premiums of the housing based on energy qualification, an ordinary least squares regression model has been proposed. For this, various estimates have been made, based on the reference used in the housing's energy qualification (letter or group of letters).
Hedonic regression models have been used since the "New Approach to Consumer Theory" created by Lancaster [37]. Ridker and Henning [38] used ordinary least squares (OLS) for the first time in the context of the housing market. Authors such as Zietz, et al. [39] indicate that hedonic regression analysis is normally used to identify the marginal effect of a set of characteristics on the housing price. For the case of heterogeneous goods such as housing, hedonic methodology permits the estimation for the contribution of each characteristic on the price [40]. Currently, this methodology is the most commonly used to determine the economic premium generated by distinct characteristics. In Table A2 of the Appendix A, the variables that are the most commonly used by other authors to determine housing price are shown.

Population and Sample
The database consists of multi-family housing placed on the market in the province of Alicante (Valencian Community, Spain) see Figure 3. The interest and selection criteria were based on the significant activity of the construction sector in this area, which is the third province in the country in terms of having the largest number of property transactions (purchases), after Barcelona and Madrid. On the other hand, it is fourth in terms of number of unsubsidized housing sales initiated in 2017 [41].

Materials and Methods
First, an analysis of variance (ANOVA) was proposed to determine whether or not differences exist in the offered prices, based on the published energy qualification. To examine the economic premiums of the housing based on energy qualification, an ordinary least squares regression model has been proposed. For this, various estimates have been made, based on the reference used in the housing's energy qualification (letter or group of letters).
Hedonic regression models have been used since the "New Approach to Consumer Theory" created by Lancaster [37]. Ridker and Henning [38] used ordinary least squares (OLS) for the first time in the context of the housing market. Authors such as Zietz, et al. [39] indicate that hedonic regression analysis is normally used to identify the marginal effect of a set of characteristics on the housing price. For the case of heterogeneous goods such as housing, hedonic methodology permits the estimation for the contribution of each characteristic on the price [40]. Currently, this methodology is the most commonly used to determine the economic premium generated by distinct characteristics. In Table A2 of the Appendix A, the variables that are the most commonly used by other authors to determine housing price are shown.

Population and Sample
The database consists of multi-family housing placed on the market in the province of Alicante (Valencian Community, Spain) see Figure 3. The interest and selection criteria were based on the significant activity of the construction sector in this area, which is the third province in the country in terms of having the largest number of property transactions (purchases), after Barcelona and Madrid. On the other hand, it is fourth in terms of number of unsubsidized housing sales initiated in 2017 [41]. The study sample consists of housing properties that were placed on the market via the idealista.com real estate portal between June 2017 and May 2018. During this period, information was collected on 97,279 properties placed on the market, extracting data on the characteristics of the housing and buildings. Subsequently, via GIS, information was provided on the location, neighborhood and market, obtained from other information sources. The final database was subject to a univariate analysis of outliers, discarding properties that differed by more or less than three The study sample consists of housing properties that were placed on the market via the idealista.com real estate portal between June 2017 and May 2018. During this period, information was collected on 97,279 properties placed on the market, extracting data on the characteristics of the housing and buildings. Subsequently, via GIS, information was provided on the location, neighborhood and market, obtained from other information sources. The final database was subject to a univariate analysis of outliers, discarding properties that differed by more or less than three standard deviations in their Appl. Sci. 2020, 10, 7129 5 of 30 respective variables (Z scores). This process was performed on the following variables: natural log of the property price, age, height in stories, constructed surface area, number of bedrooms and bathrooms. To identify the multivariate atypical cases, the regression model was calibrated by calculating the Mahalanobis distance (MD) and its statistical significance, discarding any files in which the significance was less than 0.001, in accordance with Hair, et al. [42]. Finally, those properties having missing data on any of the variables that were subject of the analysis were discarded, obtaining a final sample of 52,939 observations, of which, 9194 included information on energy qualification.
The sample's representativeness was verified via the Equation (1), designed for large or infinite populations when the exact size of the units making it up is unknown [43].
where: z α/2 is the Z score corresponding to the selected level of confidence, p is the probability that event p takes place (when not having sufficient information the least favorable value is assigned, p = 0.50), and E is the maximum admissible error or the maximum error that is committed in the sample. Using a 95% confidence level (z α/2 = 1.96), a probability of p = 0.50 and samples sizes of n = 52,939 and n = 9194, clearing E, a maximum estimated error of 0.4% (0.004) and 1.0% (0.010) were obtained, respectively, ensuring the high statistical precision of the sample.

Sources of Information
The main source of information is the real estate portal idealista.com, which publishes the asking prices along with the characteristics of the housing and the building in which it is located. Other studies have also considered real estate portals with the same objective, given the lack of official information available [44][45][46][47][48], with the real estate asking prices being a suitable substitute for the transaction prices [49]. In Figure 4, the distribution of energy certificates in the Alicante province is shown, as well as details on the province's two largest cities (Alicante and Elche).
Appl. Sci. 2020, 10, x FOR PEER REVIEW 5 of 34 standard deviations in their respective variables (Z scores). This process was performed on the following variables: natural log of the property price, age, height in stories, constructed surface area, number of bedrooms and bathrooms. To identify the multivariate atypical cases, the regression model was calibrated by calculating the Mahalanobis distance (MD) and its statistical significance, discarding any files in which the significance was less than 0.001, in accordance with Hair, et al. [42]. Finally, those properties having missing data on any of the variables that were subject of the analysis were discarded, obtaining a final sample of 52,939 observations, of which, 9194 included information on energy qualification. The sample's representativeness was verified via the Equation (1), designed for large or infinite populations when the exact size of the units making it up is unknown [43].
where: / is the Z score corresponding to the selected level of confidence, p is the probability that event p takes place (when not having sufficient information the least favorable value is assigned, p = 0.50), and E is the maximum admissible error or the maximum error that is committed in the sample. Using a 95% confidence level ( / = 1.96), a probability of p = 0.50 and samples sizes of n = 52,939 and n = 9194, clearing E, a maximum estimated error of 0.4% (0.004) and 1.0% (0.010) were obtained, respectively, ensuring the high statistical precision of the sample.

Sources of Information
The main source of information is the real estate portal idealista.com, which publishes the asking prices along with the characteristics of the housing and the building in which it is located. Other studies have also considered real estate portals with the same objective, given the lack of official information available [44][45][46][47][48], with the real estate asking prices being a suitable substitute for the transaction prices [49]. In Figure 4, the distribution of energy certificates in the Alicante province is shown, as well as details on the province's two largest cities (Alicante and Elche).  Based on the alphanumeric and vectorial information from the DGC [50,51], a raster map was created to estimate the age of the housing (Figure 5a) and the ratio of the constructed surface area in the proximity of each building (150 m around the same). With population census and INE housing data [36] and IGN mapping, the type of occupation was collected for each census tract (vacant, main and secondary), as well as the type of tenancy regime (rented, mortgaged and owned) and the population's sociodemographic characteristics (dependency, ageing, foreign population and education level). With the information from the CECD and the DGOEAPCS, distances between the housing and the public services or the following points of interest were calculated: hospitals, health centers, pharmacies, schools (Figure 5b), as well as proximity to the coast. Distances have been calculated by network, that is, based on the length of origin and destination using a layout of pre-established streets and intersections, simulating the reality of the urban network. The CTE-DB-HE is used to determine the climatic zone of the town where the property is located (climatic severity of summer and winter seasons) [35,52].
Appl. Sci. 2020, 10, x FOR PEER REVIEW 6 of 34 the Health Department (DGOEAPCS) and the Basic Document on Energy Saving from the Technical Code for Buildings (CTE-DB-HE). Based on the alphanumeric and vectorial information from the DGC [50,51], a raster map was created to estimate the age of the housing (Figure 5a) and the ratio of the constructed surface area in the proximity of each building (150 m around the same). With population census and INE housing data [36] and IGN mapping, the type of occupation was collected for each census tract (vacant, main and secondary), as well as the type of tenancy regime (rented, mortgaged and owned) and the population's sociodemographic characteristics (dependency, ageing, foreign population and education level). With the information from the CECD and the DGOEAPCS, distances between the housing and the public services or the following points of interest were calculated: hospitals, health centers, pharmacies, schools (Figure 5b), as well as proximity to the coast. Distances have been calculated by network, that is, based on the length of origin and destination using a layout of preestablished streets and intersections, simulating the reality of the urban network. The CTE-DB-HE is used to determine the climatic zone of the town where the property is located (climatic severity of summer and winter seasons) [35,52].

Data
Variables were selected based on a literature review (see Table A2 of the Appendix A). Based on the information received, 63 variables were obtained, as summarized in Table 1. The variables are ordered based on five categories: Housing characteristics (A), Building characteristics (B), Location characteristics (C), Neighborhood characteristics (D), and Market characteristics (E). The unit with which each variable has been measured is also indicated, as well as a brief description of the same and verification as to whether or not it was used in model estimation.

Data
Variables were selected based on a literature review (see Table A2 of the Appendix A). Based on the information received, 63 variables were obtained, as summarized in Table 1. The variables are ordered based on five categories: Housing characteristics (A), Building characteristics (B), Location characteristics (C), Neighborhood characteristics (D), and Market characteristics (E). The unit with which each variable has been measured is also indicated, as well as a brief description of the same and verification as to whether or not it was used in model estimation.

Descriptive Statistics
The descriptive statistics of the variables are shown in Table 2.

Methodology
The analysis of variance allows for the contrasting of the null hypothesis that the means of K populations (K > 2) are equal, with the alternative hypothesis that at least one of the populations differs from the others in terms of its expected value (Equation (2)). The one-way analysis of variance consists of three parts. The first part of the analysis permits contrasting of the null hypothesis of equality of means in the groups through the F statistic. The second contrasts the equality of the variances of the dependent variable in the groups using Levene's test. The third and final part of the analysis determines which of the distinct levels of the factor differ from the others, based on different post hoc tests.
The regression model is estimated using ordinary least squares (OLS), and its specification is semilogarithmic, based on the following expression: where ln (P i ) is the natural logarithm of the advertised asking price for housing "i"; α is the fixed component, it does not depend on the market; β j is the parameter to estimate related to the characteristic "j"; X ij is the continuous variable that collects the characteristic "j" of the observation "i"; γ k is the parameter to estimate related to the characteristic "k"; D ik is the dummy variable that collects the characteristic "k" of the observation "i"; and ε i is the error term in the observation "i". The semilogarithmic functional form was selected, since according to [53,54], this form offers certain advantages. First, it facilitates the interpretation of the coefficients. That is, for each increase in unit of the explanatory variable (X j and D k ), the dependent variable (P)-in this case, the asking price-varies on average (100·β). And second, it minimizes the problem of heteroscedasticity, improving the goodness of fit of the estimates.
The model is estimated on distinct occasions, based on the energy qualification characteristic ( Table 3), such that the results obtained may be compared with other studies. For this analysis, the SPSS statistics package for Windows, version 24 was used [55], based on the method of "excluding cases listwise". This leads to the elimination of observations with missing data.

One-Way Analysis of Variance (ANOVA)
In the database created, from a sample of 52,939 homes, only 9194 published their energy qualifications (17.4%), despite the fact that Royal Decree 235/2013 [15] requires the publication of energy rating of homes that are being sold or rented. This low percentage leads us to believe that the failure to publish an energy qualification may have some sort of advantage for real estate sellers. In order to examine this supposition, a statistical test was created for a one-way analysis of variance (ANOVA), graphically revealing the data in Figure 6. By evaluating the homogeneity of the variance of each group using Levene's test (F (7, 52931) = 68.8, p = 0.000), the variance of the groups is found to differ. This result supports the use of robust tests of equality of means, specifically, those by Welch [56] (F (7, 2185) = 314.2, p = 0.000) and Brown-Forsythe [57] (F (7, 3835) = 237.6; p = 0.000), which confirm that the mean asking prices differ between energy qualification letters.

One-Way Analysis of Variance (ANOVA)
In the database created, from a sample of 52,939 homes, only 9194 published their energy qualifications (17.4%), despite the fact that Royal Decree 235/2013 [15] requires the publication of energy rating of homes that are being sold or rented. This low percentage leads us to believe that the failure to publish an energy qualification may have some sort of advantage for real estate sellers. In order to examine this supposition, a statistical test was created for a one-way analysis of variance (ANOVA), graphically revealing the data in Figure 6. By evaluating the homogeneity of the variance of each group using Levene's test (F(7, 52931) = 68.8, p = 0.000), the variance of the groups is found to differ. This result supports the use of robust tests of equality of means, specifically, those by Welch [56] (F(7, 2185) = 314.2, p = 0.000) and Brown-Forsythe [57] (F(7, 3835) = 237.6; p = 0.000), which confirm that the mean asking prices differ between energy qualification letters. To identify the relationships between the groups, a "post hoc" test was performed with Scheffé's method and a classification of the groupings (or homogenous subgroupings) based on the value of the means. For each subset, a test was carried out on the equality of means hypothesis, with To identify the relationships between the groups, a "post hoc" test was performed with Scheffé's method and a classification of the groupings (or homogenous subgroupings) based on the value of the means. For each subset, a test was carried out on the equality of means hypothesis, with significances of over 0.05 (no difference in means), in accordance with that observed in Table 4. Subset 1 is formed by housing with letters G and F, whose means do not differ significantly (p = 0.055). Subset 2 includes housing with letters F and E, whose means do not differ significantly (p = 0.123). Subset 3 is made up of homes with letters NT, A, D and B, whose means do not differ significantly (p = 0.053), and subset 4 consists of only those homes with letter C, which, obviously, do not differ from themselves (p = 1.0).

Regression Analysis
Upon introducing the variables in the regression model, problems of self-correction were observed between some of the same. Therefore, a total of eight variables have been discarded. Three correspond to the climate area (Zone_B4, Zone_C3, and Zone_D3), one is referred to the percentage of the population with primary and secondary school educations (D_students), two are referred to the percentages of vacant and main homes (E_vacant_dw and E_main_dw), and two more are referred to the percentages of mortgaged homes and properties with homeowners (E_mortgaged_dw and E_homeownership). Figure A1 in the Appendix A shows a graph with the most relevant correlations.
In order to determine if the estimates achieved suitable quality criteria, the following were examined: the normality of the population, the lack of problems of specification in the estimates (no multicollinearity, heteroscedasticity or autocorrelation), the statistical significance of the estimates, and finally, that the proportion of the estimated variance was high (R 2 ). The normality of the population is verified through a histogram (Figure 7a,d) and a graph of normality of the residuals (Figure 7b,e), revealing that the sample has a normal distribution. The multicollinearity was verified via the VIF statistic (Variance Inflation Factor), with various authors suggesting that there are collinearity problems if any VIF exceeds 10 [58,59]. In the new estimations made, the majority of the VIF values are between 1 and 4.6, therefore it is considered that there are no problems arising from multicollinearity. The heteroscedasticity was analyzed with a residual dispersion plot (Figure 7c,f) and there was no evidence of serious problems of heteroscedasticity, given the random distribution of the residuals. The existence of autocorrelation was verified using the Durbin-Watson statistic, obtaining values close to two in all of the estimations, which suggests the absence of autocorrelation in the residuals [60,61]. The significance of each estimation is measured with Snedecor's F-test, being found to be statistically significant. The coefficient of determination (adjusted R 2 ) of the estimates is indicated in Table 5 and all of these have an explanatory power approaching 71%. In summary, the estimations have a sufficient level of robustness and significance, making them acceptable for purposes of inference making. To control the fixed effects due to the spatial location of the data, the comarcas location variables are used. A positive spatial autocorrelation is detected with the Moran's I test [62,63] (residuals from estimation 3, I = 0.274, z = 99.89, p < 0.001; inverse distance squared, bandwidth 500 m), a common result in global regression models [64].
obtaining values close to two in all of the estimations, which suggests the absence of autocorrelation in the residuals [60,61]. The significance of each estimation is measured with Snedecor's F-test, being found to be statistically significant. The coefficient of determination (adjusted R 2 ) of the estimates is indicated in Table 5 and all of these have an explanatory power approaching 71%. In summary, the estimations have a sufficient level of robustness and significance, making them acceptable for purposes of inference making. To control the fixed effects due to the spatial location of the data, the comarcas location variables are used. A positive spatial autocorrelation is detected with the Moran's I test [62,63] (residuals from estimation 3, I = 0.274, z = 99.89, p < 0.001; inverse distance squared, bandwidth 500 m), a common result in global regression models [64].     Currently, studies carried out have been based on distinct scenarios and the literature has revealed a certain diversity in terms of determining the letter (or set of letters) of reference for measuring and comparing the impact of energy qualifications on housing prices (Figure 8). This circumstance hinders comparisons between the premium resulting from going from one value to another within the ABCDEFG qualification scale. To facilitate comparison between studies, some authors have recommended that letters not be grouped and that D be considered the letter of reference [32,65], since it is in the middle of the scale. Based on this background, the results of the model obtained for the estimate 4 are presented, since it is the one that complies with the recommendations of using letter D as a reference. As for the characteristics of the housing, the model estimates that for each additional year (in terms of age of the housing), the asking price will be reduced a mean of 0.29%. As for size, the estimated impact implies that an increase of one square meter in surface area results in a 0.58% price increase, whereas the addition of another bedroom leads to a reduction of 2.09%. However, an additional bathroom represents a mean price increase of 22.44%. If the property has extras, such as built-in wardrobe, air conditioning or a terrace, the mean impact on prices estimated by the model is 0.25%, 8.42% and 2.81%, respectively. The results reveal that a home situated on an additional story has a price increase of 0.02%. Using as a reference a second-hand home in good state, the model estimates a mean discount in the asking price of 9.49% for a second-hand home that needs renovation. On the other hand, if the home is a new construction, the results reveal a price increase of 20.78%. Within the typology of homes and using apartments as the reference, a duplex or attic apartment has a price increase of 1.30% and 9.45%, respectively, whereas studio apartments have a discount of 23.56%.
The values obtained in the estimation of the parameters related to the building characteristics, such as having an elevator, garage, storage space or swimming pool, imply a mean price increase of 19.48%, 10.18%, 4.88% and 9.38%, respectively. On the other hand, having a garden has a contrary effect, leading to a mean price reduction of 0.16% (not significant).
As for characteristics related to location, for properties situated in neighborhoods with a higher gross development (those in which there are more homes per sector surface area), the model estimates a mean reduction in price of approximately 1.80%. As for the geographic distances, they are all statistically significant, except for the distance to pharmacies. The results reveal that for each kilometer that the housing is distanced from pharmacies or level 2 schools (secondary and high schools), the price decreases by 0.50% and 1.18%. The opposite occurs when the housing is distanced from health centers, hospitals and level 1 schools (infant and primary schools). Homes that are in coastal towns have a price increase of 16.57%. Finally, the estimated impact on prices of homes situated in the Marina Baja district have an increase of 1.81% with regard to the reference district (Alicante). As for the rest of the districts, the effect that is estimated by the model implies a reduction in asking prices, reaching discounts of between 15% and 20% in interior and southern districts of the province [66,67].
As for the neighborhood characteristics, an increase of 1% was found for dependency and ageing, implying an increase in sales price of 0.16% and 0.01%, respectively. As for the percentage of foreigners or the percentage of individuals with university studies, an increase of 1% for these variables implies a 0.12% and 0.85% price increase, respectively. On the other hand, with a 1% increase in the percentage of the population without an education, there is a price reduction of 0.54%. Based on this background, the results of the model obtained for the estimate 4 are presented, since it is the one that complies with the recommendations of using letter D as a reference. As for the characteristics of the housing, the model estimates that for each additional year (in terms of age of the housing), the asking price will be reduced a mean of 0.29%. As for size, the estimated impact implies that an increase of one square meter in surface area results in a 0.58% price increase, whereas the addition of another bedroom leads to a reduction of 2.09%. However, an additional bathroom represents a mean price increase of 22.44%. If the property has extras, such as built-in wardrobe, air conditioning or a terrace, the mean impact on prices estimated by the model is 0.25%, 8.42% and 2.81%, respectively. The results reveal that a home situated on an additional story has a price increase of 0.02%. Using as a reference a second-hand home in good state, the model estimates a mean discount in the asking price of 9.49% for a second-hand home that needs renovation. On the other hand, if the home is a new construction, the results reveal a price increase of 20.78%. Within the typology of homes and using apartments as the reference, a duplex or attic apartment has a price increase of 1.30% and 9.45%, respectively, whereas studio apartments have a discount of 23.56%.
The values obtained in the estimation of the parameters related to the building characteristics, such as having an elevator, garage, storage space or swimming pool, imply a mean price increase of 19.48%, 10.18%, 4.88% and 9.38%, respectively. On the other hand, having a garden has a contrary effect, leading to a mean price reduction of 0.16% (not significant).
As for characteristics related to location, for properties situated in neighborhoods with a higher gross development (those in which there are more homes per sector surface area), the model estimates a mean reduction in price of approximately 1.80%. As for the geographic distances, they are all statistically significant, except for the distance to pharmacies. The results reveal that for each kilometer that the housing is distanced from pharmacies or level 2 schools (secondary and high schools), the price decreases by 0.50% and 1.18%. The opposite occurs when the housing is distanced from health centers, hospitals and level 1 schools (infant and primary schools). Homes that are in coastal towns have a price increase of 16.57%. Finally, the estimated impact on prices of homes situated in the Marina Baja district have an increase of 1.81% with regard to the reference district (Alicante). As for the rest of the districts, the effect that is estimated by the model implies a reduction in asking prices, reaching discounts of between 15% and 20% in interior and southern districts of the province [66,67].
As for the neighborhood characteristics, an increase of 1% was found for dependency and ageing, implying an increase in sales price of 0.16% and 0.01%, respectively. As for the percentage of foreigners or the percentage of individuals with university studies, an increase of 1% for these variables implies a 0.12% and 0.85% price increase, respectively. On the other hand, with a 1% increase in the percentage of the population without an education, there is a price reduction of 0.54%.
As for market characteristics, the model estimations reveal increases in prices in areas having a higher percentage of homes in rent and secondary homes, at 0.32% and 0.21%, respectively. The sale of homes indicates that when the properties are sold by professionals or banks, the price is reduced by a mean of 0.63% and 0.62%, respectively, with these values not being statistically significant.
As for the characteristics having a greater impact on asking prices, the five variables from the estimates having the greatest explanatory power, according to the standardized beta coefficients (not included due to problems of extension) are: (A) housing characteristics-constructed surface area and number of bathrooms; (B) building characteristics-having an elevator; and (C) location characteristics-percentage of individuals with university studies and being situated in a coastal town.
As for the energy qualification, the results for the entire sample are summarized in Figure 9, where it is observed that the housing with any qualification type (ABCDEFG grouping) and homes with high energy qualifications (letter A) had lower prices, respectively, 3.22% and 0.30% lower. Estimate 3 reveals that housing with high qualifications (letters A and B) do not have better economic premiums than other homes with lower qualifications or those that have not published their qualifications. This suggests that by not publishing the energy qualification, sellers may ask for higher prices than those asked for other homes with lower qualifications (E, F or G).
Appl. Sci. 2020, 10, x FOR PEER REVIEW 21 of 34 As for the characteristics having a greater impact on asking prices, the five variables from the estimates having the greatest explanatory power, according to the standardized beta coefficients (not included due to problems of extension) are: (A) housing characteristics-constructed surface area and number of bathrooms; (B) building characteristics-having an elevator; and (C) location characteristics-percentage of individuals with university studies and being situated in a coastal town.
As for the energy qualification, the results for the entire sample are summarized in Figure 9, where it is observed that the housing with any qualification type (ABCDEFG grouping) and homes with high energy qualifications (letter A) had lower prices, respectively, 3.22% and 0.30% lower. Estimate 3 reveals that housing with high qualifications (letters A and B) do not have better economic premiums than other homes with lower qualifications or those that have not published their qualifications. This suggests that by not publishing the energy qualification, sellers may ask for higher prices than those asked for other homes with lower qualifications (E, F or G). Estimations 3 to 9 are carried out with the sample of homes that published their energy qualifications (Figures 9 and 10). If letter D is used as the reference (estimations 4, 5 and 6), it is observed that letters A, B (and the AB grouping) do not have better premiums than those of letters C and D. In the case of homes qualified as E, F or G, they have very similar negative premiums, a decrease of approximately 8%. If adopting the letter G as a reference (estimations 7, 8 and 9), the positive price premiums for letters C and D are of special note, as well as the similarity of prices for Estimations 3 to 9 are carried out with the sample of homes that published their energy qualifications (Figures 9 and 10). If letter D is used as the reference (estimations 4, 5 and 6), it is observed that letters A, B (and the AB grouping) do not have better premiums than those of letters C and D. In the case of homes qualified as E, F or G, they have very similar negative premiums, a decrease of approximately 8%. If adopting the letter G as a reference (estimations 7, 8 and 9), the positive price premiums for letters C and D are of special note, as well as the similarity of prices for the lower qualifications (E, F and G).

Discussion
The results of the one-way analysis of variance (ANOVA) support the first hypothesis (H 1 ), since it reveals that the mean of the asking prices for the properties that do not publish their energy qualification (NT) are similar to those of homes with higher ratings, letters A, B or D (subset 3). Therefore, the sellers of these properties have no interest in publishing the qualification letter. These results are in line with other works [16,31]. It is very likely that homes hiding the energy qualification have letters E, F or G, since this segment represents 86% of the labelled building stock of the autonomous community [68].
The results of the regression model obtained from estimations 1, 2 and 3 are contradictory to the second hypothesis (H 2 ) proposed in this document. Homes with an energy qualification-ABCDEFG grouping (estimation 1)-or a high qualification-letter A (estimation 2)-have a discount in price of 3.22% or 0.30%, respectively, with respect to the homes that do not publish their energy qualification (NT). In addition, if comparing estimations 1 and 2 with the results obtained by Cespedes-Lopez, et al. [32], it is observed that having an energy qualification, as compared to not having one, does not have a positive effect on the asking price, as it does in Europe, in general (2.32%). In estimation 3, it is observed that homes that do not publish their energy qualifications have higher prices than those qualified with letters B, E, F and G. This estimation may be compared with [22] (see Table 6), where a positive impact was found on prices for the high qualifications (A and B), and a decrease of between 0.8 and 1.6% for the poorer qualified (D, E, F and G).  Note: * indicates that the coefficient is statistically significant.
Results of the regression model for estimations 4 to 9 ( Figure 10) are contrary to the third hypothesis (H 3 ). For estimations 4, 5 and 6 (ref. D) a negative sign was anticipated for letters below the reference letter, and a positive impact for letters above the reference letters (Figure 10a). Estimations 7, 8 and 9 (ref. G) anticipated that the premium of letter A would be positive and the sign of the subsequent letters would also be positive and with a decreasing impact on the prices until reaching the reference letter ( Figure 10e). However, upon comparing the expected and obtained results, it can be seen that they do not comply with the initially proposed hypothesis (H 3 ), since homes with better qualifications do not have better premiums than those with poorer qualifications. For housing with high qualifications-letters A, B and the AB and ABC groupings-a discount was obtained with respect to the reference housing -letter D-(estimations 4, 5 and 6). For estimations 7, 8 and 9, it is seen that housing qualified as C and D are the best valued of the market segment, with the highest premiums as compared to the reference G.
If comparing these results with those obtained from other studies (Table 6), it may be observed that, in general, housing with higher qualifications have higher sales prices. For example, estimation 6 from this study may be compared to the results of [18,25,26], where high qualifications (the AB grouping) obtained a positive price premium, and for the housing with low qualifications, the premiums are negative and decrease as the qualification decreases.

Conclusions
This work seeks to examine the effect of energy qualifications on the asking price of housing located in the Alicante (Spain) real estate market. To do so, a database was constructed based on 52,939 observations, of which 9194 offered information on energy qualifications (17.4%). The information contained in the database has permitted the creation of 63 variables that are used to estimate the regression model. In order to compare the results of this work with those of other studies, the model has been estimated 9 times.
The first objective attempts to determine if an interest exists in not publishing energy qualification information for homes being sold. Two initial hypotheses are proposed-H 1 and H 2 . The one-way analysis of variance (ANOVA) reveals that the H 1 hypothesis is supported, since hiding the qualification may lead to higher asking prices. The second proposed hypothesis-H 2 -attempts to contrast whether or not energy qualification is a determining characteristic of the asking price. Estimations 1 and 2 reveal that this hypothesis is rejected, since homes with an energy qualification-ABCDEFG grouping-or those having a high qualification-letter A-as compared to those that did not publish their rating -NT-, have a negative premium of 3.2% (significant) and of 0.3% (not sig.) respectively.
The second objective proposes quantifying the economic impact of energy qualification on the asking price, offering the hypothesis-H 3 . Estimations 4 to 9 suggest the contrary, finding that housing qualified with letters C and D have higher premiums than housing with higher qualifications (A or B).
This study reveals that, in multi-family housing sold in the Alicante (Spain) province, a positive relationship does not exist between the energy certification system and the housing's asking price. This is due to a variety of reasons.
-First, real estate sellers and owners who do not publish energy qualifications offer their homes at prices that are similar to those having high qualifications. -Second, there is the lack of sanctions placed by the public administration on companies, owners and real estate portals that do not publish the energy qualifications of the housing that is for sale or rent, motivating owners to not publish the letter and generating distorted asking prices for the housing. Therefore, it is important for the administration to closely supervise compliance with regulations and assign the necessary resources to local authorities to ensure said compliance, and if needed, to impose sanctions. -Third, owners are not interested in improving energy qualifications, since, according to [71,72] there is no compensation for the additional investment needed to improve this qualification. And fourth and finally, the current regulations for housing only require that these homes obtain energy qualification if they are going to sell, rent or publish. However, there is no obligation to obtain a minimum qualification, so the improved energy performance of the homes is not encouraged [73,74]. -Fifth and finally, society's perception of EPC is negative, as revealed by several studies relying on surveys completed by professional real estate agents [74,75] or energy certifiers [73]. Regardless, these studies suggest that the main criteria used to select a home is price and location [74][75][76].
Currently, both nationally and regionally, economic incentives exist in order to offer value to housing with higher qualifications and to promote renovation. On a national level, the PAREER II (2014-2020) [77] program was financed with 204 million euros. Regionally speaking (Valencian community), there are distinct plans such as RENHATA [78], which intends to offer 4.95 million euros between 2020-2021 to improve the preservation of housing, accessibility and energy rehabilitation. Given that in Spain, there are over 9.5 million buildings, it is unlikely that a country's building stock will be renewed thanks to public budgets. Therefore, it will be necessary to rely on private initiatives, based on market incentives (higher sales prices, higher rents) that encourage investments in energy renewal of buildings. In this way, not only would property owners directly benefit from these renewals, the entire population would also receive benefits. This would ensure a more sustainable and environmentally respectful building stock, helping to create cleaner cities and an improved quality of life.   Table A2. Variables used by other authors for the determination of the price of housing. Own elaboration from [66]. Table A2. Cont.

Price
In all studies this is the dependent variable Use of the dwelling [17,28,96,118] Housing tenure [26] Appl. Sci. 2020, 10, x FOR PEER REVIEW 28 of 34 Figure A1. Correlation between the characteristics of the properties-independent variables-and the asking prices. Only correlations greater than 0.35 (in absolute value) are shown.