1# RECORD 755578f10943061d8e5adfd02909b14166725d762# TITLE: Influence of the Urban Green Spaces of Seville (Spain) on Housing Prices through the Hedonic Assessment Methodology and Geospatial Analysis3# YEAR: 2022.0 VENUE: Sustainability DOI: 10.3390/su1424166134# AUTHORS: E. Ramírez-Juidías; J. Amaro-Mellado; J. Leiva-Piedra5# PAGES: 15 CANDIDATE TABLE PAGES (1-based): [8, 9]6# ABSTRACT: The city of Seville (Spain) is made up of a historical network of pre-existing city overlaps, which increase the economic and heritage value of certain urban areas. To date, green spaces are one of the most important variables in determining the economic value of housing. Thus, this paper uses the hedonic technique and geostatistical analysis with GIS as a methodological approach to infer the economic influence of urban green spaces on housing prices. Along with the traditional variables used to explain dwelling prices, the size of the green space has also been taken into account as an environmental variable affecting prices. The sample consists of 1000 observations collected from Seville. According to the findings, the most relevant variables depend on the hedonic model. Still, in general terms, a dwelling’s selling price is related to basic explanatory variables such as living area, number of rooms, age, and number of baths. The green area per inhabitant present in a dwelling’s district is also included as part of these basic explanatory variables. In conclusion, the hedonic linear model is the model that best fits housing prices where the values are similar to those obtained by kriging regardless of the district. Based on this research, each square meter of green space per inhabitant in a district raises the housing value by 120.19 €/m2.789===== PAGE 1 (score 20, numeric-paren 0, table-words 1, var-words 18) =====1011Citation: Ramírez-Juidías, E.;12Amaro-Mellado, J.-L.; Leiva-Piedra,13J.L. Influence of the Urban Green14Spaces of Seville (Spain) on Housing15Prices through the Hedonic16Assessment Methodology and17Geospatial Analysis. Sustainability182022, 14, 16613. https://doi.org/1910.3390/su14241661320Academic Editors: Marco Devecchi,21Fabrizio Aimar and Matteo Caser22Received: 6 November 202223Accepted: 9 December 202224Published: 12 December 202225Publisher’s Note: MDPI stays neutral26with regard to jurisdictional claims in27published maps and institutional affil-28iations.29Copyright:30© 2022 by the authors.31Licensee MDPI, Basel, Switzerland.32This article is an open access article33distributed34under35the36terms37and38conditions of the Creative Commons39Attribution (CC BY) license (https://40creativecommons.org/licenses/by/414.0/).42sustainability43Article44Influence of the Urban Green Spaces of Seville (Spain) on45Housing Prices through the Hedonic Assessment Methodology46and Geospatial Analysis47Emilio Ramírez-Juidías 148, José-Lázaro Amaro-Mellado 2,3,*49and Jorge Luis Leiva-Piedra 450151Instituto Universitario de Arquitectura y Ciencias de la Construcción (IUACC), Universidad de Sevilla,522 Reina Mercedes Avenue, 41012 Seville, Spain53254Departamento de Ingeniería Gráfica, Universidad de Sevilla, 41092 Seville, Spain55356Instituto Geográfico Nacional-Servicio Regional en Andalucía, 41013 Seville, Spain57458Laboratorio de Investigación en Teledetección, Universidad Tecnológica del Perú, Intersección Avenida El59Progreso s/n-Vía de Evitamiento, Chiclayo 14001, Peru60*61Correspondence: jamaro@us.es62Abstract: The city of Seville (Spain) is made up of a historical network of pre-existing city overlaps,63which increase the economic and heritage value of certain urban areas. To date, green spaces are64one of the most important variables in determining the economic value of housing. Thus, this paper65uses the hedonic technique and geostatistical analysis with GIS as a methodological approach to66infer the economic influence of urban green spaces on housing prices. Along with the traditional67variables used to explain dwelling prices, the size of the green space has also been taken into account68as an environmental variable affecting prices. The sample consists of 1000 observations collected69from Seville. According to the findings, the most relevant variables depend on the hedonic model.70Still, in general terms, a dwelling’s selling price is related to basic explanatory variables such as71living area, number of rooms, age, and number of baths. The green area per inhabitant present in a72dwelling’s district is also included as part of these basic explanatory variables. In conclusion, the73hedonic linear model is the model that best fits housing prices where the values are similar to those74obtained by kriging regardless of the district. Based on this research, each square meter of green75space per inhabitant in a district raises the housing value by 120.19 €/m2.76Keywords: hedonic method; GIS; geospatial analysis; urban green space; housing prices771. Introduction78Green spaces serve essential environmental and recreational functions [1]. Forest and79urban vegetation, in edaphoclimatic terms, improve the environment in which they are80found and serve as the foundation for the conservation of fauna and flora. Recreational81activities with a basis in nature are becoming more popular and, while the effects are on82a smaller scale, urban parks serve the same environmental and recreational functions as83larger forests and green spaces.84Gardens and city parks play the primary environmental function of absorbing carbon85dioxide (CO2) emissions. These emissions are primarily generated through the use of86private vehicles in urban transportation. Moreover, emissions have increased significantly87in recent decades. It is worth noting that, in Spain, each person in a large city emits884.62 tonnes of CO2 per year [2]. Given that one hectare of Mediterranean forest can absorb89approximately four tonnes of CO2 per year, a simple calculation leads us to conclude that90we require more than one hectare of green space per inhabitant (inh) to offset the pollution91generated by these emissions. The preservation of urban green spaces is thereby becoming92increasingly important in order to combat the growing contamination of our cities.93Additional factors, such as acoustic isolation, should be included in the list of the94environmental functions of urban green parks as some gardens serve as acoustic screens95Sustainability 2022, 14, 16613. https://doi.org/10.3390/su14241661396https://www.mdpi.com/journal/sustainability979899===== PAGE 8 (score 27, numeric-paren 0, table-words 7, var-words 9) =====100101Sustainability 2022, 14, 166131028 of 15103the average price per m2 and was taken into account (Table 4) in accordance with Solano-104Sánchez et al. [39]. Under the predefined hypothesis, a higher value per m2 in a district105implies a higher dwelling value. This index was created by assigning one value to the106city’s most expensive district and then assigning a proportional value to the remaining107zones. In order to be able to test and validate this study, a comparison of real price108(www.idealista.com) versus estimated price for each HPM model will be conducted.109Table 4. District Index (DI) value for each set of districts (Own elaboration).110Set of Districts111€/m2112DI113Triana1142715.691150.865116Torreblanca-Sevilla Este-Parque Alcosa1171207.941180.385119Santa Justa-Miraflores-Cruz Roja-San Pablo-Santa Justa1201919.31210.612122San Jerónimo-Pino Montano1231188.241240.379125Prado de San Sebastián-Felipe II-Bueno Monrreal-La126Palmera-Los Bermejales-Bellavista-Jardines de Hércules1272497.181280.796129Nervión1302642.491310.842132Macarena1331368.731340.436135Los Remedios1362899.171370.924138Cerro-Amate1391017.511400.324141Centro1423137.891431144Finally, a literature review was conducted to analyze the results obtained in this study145and to carry out a coherent discussion concerning the works published to date.1465. Results and Discussion147Following the data analysis, it was discovered that including the green area (m2/inh)148variable in the price equation allowed for the estimation of the influence of environmental149factors on dwelling market value. In order to find out if the HPM presents multicollinearity,150it was decided that obtaining the Variance Inflation Factor (VIF) (Table 5) would be required.151As is well known, if the VIF value for each independent variable is greater than ten then152the multicollinearity is considered to be high.153Table 5. Independent variables and coefficients of linear HPM model (r = 0.85, R2 = 0.721, p ≤0.001)154(Own elaboration).155Variables156Coefficients157Stand. Error158Student’s t159Prob.160VIF161Constant162−120.44316314.795164−0.0751650.000166-167Size (m2)16830.0321690.0391701.71710.0001721.622173Baths (n◦)17498.7251750.0421760.141770.0001781.467179Age (years)180−53.311811.42182−1.421830.0001841.535185Green area186(m2/inh)18711.1451880.871890.8721900.0001911.020192Regarding the HPM model selected as the most satisfactory, it should be mentioned193that the linear function form was found to be the most appropriate after performing the194data analysis. Like the VIF, Table 4 also displays the independent variables’ coefficients.195As for the logarithmic model, in addition to the VIF value, Table 6 provides the196differences with respect to the linear model. Thus, the main difference is the dependence197on the number of rooms (Rooms) variable in the logarithmic model. In contrast, under198the linear model, the property’s value depends on the dwelling area (Size) variable. This199may be because of the relationship between both variables which causes the predominant200variable to cancel the slave variable in the requisite model depending on the selected model.201202203===== PAGE 9 (score 17, numeric-paren 0, table-words 4, var-words 6) =====204205Sustainability 2022, 14, 166132069 of 15207Table 6. Independent variables and coefficients of the logarithmic HPM model (r = 0.816, R2 = 0.667,208p ≤0.001) (Own elaboration).209Variables210Coefficients211Stand. Error212Student’s t213Prob.214VIF215Constant2166.562171.492184.362190.000220-221Room2221.392231.082241.292250.0002261.705227Baths (n◦)2280.262290.62300.4322310.0002321.523233Age (years)234−0.3652350.205236−1.782370.0002381.380239Green area240(m2/inh)241−0.0132420.13243−0.12440.0002451.000246If the reciprocal model is analyzed, Equation (8) is obtained (r = 0.647, R2 = 0.42,247p ≤0.05):248P = 5581.151 −344, 864.49·249 1250S251252(8)253After the reciprocal model’s price estimation (Figure 3) is represented, it can be ob-254served that the price function shape in the limit, “P” trends towards “α = 5581.151”. The255term “−(β/α)” refers to the smallest amount of surface available in a dwelling (61.79 m2 in256this study).257Figure 3. Reciprocal model representation when “α > 0” and “β < 0”. (Own elaboration).258If the three models are compared based on the coefficient of determination (R2), it259can be said that a higher value of this coefficient implies a better estimate of the housing260price. For this reason, the linear model explains 5.4% more of the predicted price than261the logarithmic model and 30.1% more than the reciprocal model. On the other hand, the262logarithmic model explains 24.7% more of the estimated price than the reciprocal model.263Additionally, the Chow test [40] was used to ensure the models’ stability. It should264be noted though that the Chow test has its origin in economics. Nevertheless, it has also265been applied to other fields such as renewable energies [41], flood risk assessment [42], and266spatiotemporal disease diffusion [43]. In this study, the results revealed that no structural267changes occurred exclusively in the parameters of either the linear or logarithmic models.268This can be seen in Figure 4, which compares the real price (from www.idealista.com269(accessed on 18 October 2022)) to the estimated price for each model used. As can be seen,270Figure 4 allows for the simulation of hypotheses that facilitate the validity of the study to271be tested.272273274===== PAGE 10 (score 10, numeric-paren 0, table-words 1, var-words 8) =====275276Sustainability 2022, 14, 1661327710 of 15278Figure 4. Comparison of real (www.idealista.com (accessed on 17 October 2022)) versus estimated279prices for each HPM model. (Own elaboration).280When the models’ degree of fit is perfect, they should appear as point clouds in a281diagonal line starting at the origin (1017.51; 1119.94 in this study), as shown in Figure 4. In282this case, the estimated values for linear and logarithmic models indicate that the linear283form is a good fit. However, the reciprocal model presents a trend line that does not coincide284with the indicated origin. For this reason, the model does not have as much stability as the285linear and logarithmic ones.286Furthermore, with regard to the kriging, Figure 5 shows the geostatistical price interpo-287lation for the city of Seville. As is well known, in order to perform predictions using kriging288the forecasting system must be geometrically defined at some point. This means that we289must establish the neighbourhood of the data that will intervene in the approximation. It is290possible to use all the dwelling prices (assuming a global neighbourhood) and a set of data291(considering a local neighbourhood) in the estimation. In practice, only house prices within292a predefined circumference or ellipse centred on the estimated point are used. Obviously,293the proximity of the data for each point to be valued will differ, thereby necessitating294a different kriging equations system for each point on the map at which a prediction is295desired. On the other hand, this method allows for the synthetic representation of housing296price estimates in the form of a colour map. This representation is generated by performing297predictions on the nodes of a regular mesh. In this manuscript the housing price has been298estimated at each node of a 10-m-sided regular mesh.299The map in Figure 5 displays how the more valuable areas coincide with those closest300to the city centre. The least valued areas, on the other hand, are primarily in the districts of301San Jerónimo-Pino Montano, Torreblanca-Sevilla Este-Parque Alcosa, and Cerro-Amate.302This is primarily due to a shift in buyer preferences following the COVID-19 pandemic.303Following confinement the percentage increase in the preference for housing (new or used)304with a balcony was around 80% [44]. As a result, the average percentage decrease in housing305prices was 5% across the districts, except for the Centro district which is dominated by306homes with small balconies and a communal roof terrace. In order to compare the kriging307values in terms of €/m2 with those of the analyzed models, Table 7 lists the average values308per district corresponding to the centroid established in the geospatial analysis.309