KINDERGARTEN PROXIMITY AND THE HOUSING MARKET PRICE IN ITALY CESPIC WORKING PAPER

: This paper investigates the impact of kindergarten proximity on housing market prices in the eleven major Italian Municipalities over the period 2004-2017. For this purpose, we employ a hedonic property price model. We also differentiate the impact of kindergarten proximity on houses' market price between state and non-state premises. The findings highlight that (i) the level of housing price depends on kindergarten proximity; (ii) some quality school characteristics played a crucial role and (iii) the distinction between public and non-state kindergartens shows that the vicinity of the latter generates a more significant capitalization effect.


Introduction
Housing is a particular type of asset with a dual meaning as consumption and an investment good (Glindro et al., 2011). For this reason, the determinants of housing market price are a topic of great relevance. According to Gibbons and Machin (2003) the evaluation of housing prices by buyers are affected by several factors (i) the real estate characteristics (e.g. the number of rooms and house condition); (ii) neighbourhood characteristics (e.g. low crime rates and neighbourhood peers); and finally (iii) other amenities (e.g. proximity to workplace, parks and shops) (Gibbons and Machin, 2003).
Also, the quality (Brasington and Haurin, 2006;Gibbons et al., 2013;Wen et al., 2017) and especially the proximity to school represents a concern to home buyers (Theisen and Emblem, 2018). Regarding the proximity, the distance between the kindergartens and the houses is crucial, since it is supposed that their proximity to homes may affect housing market prices much more than the closeness of other school levels can do.
As it is mostly in early life that children need somebody (i.e. their parents) to take them to school daily, it is supposed that parents are inclined to live within walking distance of kindergarten. Therefore, it is expected that this preference influences the residential location and, in turn, property values. There is no research focusing on the relationship between accessibility to kindergartens and housing price in Italy to the best of our knowledge.
Therefore, this paper sheds some light on this debate, estimating the impact of kindergarten proximity on houses' price in the eleven major Italian Municipalities. To this end, we investigate the impact of kindergarten proximity on housing market prices employing the hedonic property price model. Then, we differentiate between state and non-state 1 premises the effect of the kindergartens' proximity on houses' market price.
To estimate the impact of kindergarten proximity on housing market prices in the eleven major Italian Municipalities, we use various sources. Data on housing market price, covering the period 2004-2017, is provided by the Osservatorio del Mercato Immobiliare (OMI) of the national Fiscal Agency (Agenzia delle Entrate). Data on the distance between kindergartens and the centre of neighbourhoods come from a personal dataset constructed in the following way. Using data on the addresses of childcare institutions provided by the Ministry of Education, Universities and Research (MIUR), the geo-codes of the centre of the districts of interest are collected exploiting the OMI internet map 2 that shows the boundaries of the neighbourhoods (as identified by OMI) of Italian cities. Besides, we also use the information on kindergarten (made available by the MIUR) and municipal characteristics (provided by the Ministry of Interior and the National Institute of Statistics). This analysis's main findings show that proximity to kindergarten is capitalised into housing market price and confirm that close location to kindergarten has a significant and positive effect on housing price, causing their capitalisation. Also, the estimated coefficients are stable across all specifications with a weak increase in intensity over time. Finally, we find that adding the variables that capture the quality of schools mitigates the proximity effect.
Besides, results are of particular interest when we divide our estimates between state and non-state kindergartens. We find that the degree of capitalisation depends mainly on the proximity to non-state kindergartens. This result is primarily due to the asymmetrical dislocation of private kindergarten/schools; on the contrary, public schools have a more uniform distribution.
The remainder of this paper is organised as follows. The next section provides an overview of the literature on the relationship between school proximity and housing market price. Section 3 describes the data and variables. Section 4 outlines the econometric strategy used to examine the questions of interest. Section 5 discusses the main results, and section 6 presents some alternative estimations. The last section summarises and concludes the paper.
Conversely, few studies have explored the relationship between housing market prices and school proximity despite this factor may affect house values since the attractiveness of a house increases with the proximity to a school, in particular with school-aged children due to commuting and safety worries in the district (Huang and Hess, 2018).
In what follows, we expound the existing literature on the linkage between school proximity and housing price classified as follows: (i) a substantial part of studies estimates the impact of school proximity on housing market prices through a pure hedonic model; (ii) other studies employ different techniques such as the spatial approach to improve the hedonic price model.
The first study investigating this topic employing the hedonic approach and measuring proximity to school with some specific ranges of distance was carried out by Des Rosiers et al. (2001). Their analysis focuses on the effect of distance of primary school on residential values in Quebec, Canada. They are using data covering the period January 1990 and December 1991 on a sample of 4,300 single-detached and 116 primary schools. They find that the proximity of primary schools strongly affects the market house price.
In line with the previous study, Chin and Foong (2006) exploit data on sales records of individual housing transactions (13,790) for 2000-2003. They observe the relationship between the housing prices and accessibility of both primary schools and junior high schools in Singapore and show that home buyers consider the proximity and school reputation in their home purchase decision. Findings suggest that accessibility to prestigious and primary schools is more important than access to junior high schools for households.
Yet, Owusu-Edusei et al. (2007) study the impact of school proximity and school quality on the house prices at the elementary, middle, and high school level. They use data on 3,732 house sales between 1994 and 2000 in the metropolitan area of Greenville, South Carolina, and measure the distance to schools following the criteria defined by Des Rosiers et al. (2001). Their empirical evidence confirms that proximity to schools at all levels and the quality of schools have a positive impact on housing prices.
Analysing the university's effect on house prices, Chen (2010)  Haung and Hess (2018) study the relationship between a residential property's price and the proximity to school using a continuous distance measure. They employ the quantile regression method of Koenker 3 (2005) on a sample of 1,075 residential property in Oshkosh, US, during the period January 2006 and July 2007 and find that the distance to all three school levels has a significant effect on housing prices.
As noted above, other studies use different approaches to investigate the relationship between housing market prices and school proximity.
Among these, the study carried out by Wen et al. (2014) apply both the hedonic price approach and the spatial econometric model to explore the relationship between housing price and educational resources from kindergarten to university level.
Exploiting data on the Chinese house market during May 2012 for a sample of 6 cities and 660 communities in Hangzhou (China), they find that kindergartens, high schools, and college improve the nearby housing prices through accessibility. Moreover, elementary and junior high schools have a significant school district effect. In particular, kindergarten's presence within one kilometre from residential communities leads to a substantial housing price increase. According to Theisen and Emblem (2018), the proximity to kindergartens is more important than school accessibility for children aged 1-5 years. They employ data of on properties value that cover the period 2010-2017 for a sample of 15,307 house transactions in Kristiansand (Norway) and explore the relationship between house prices and the distance to kindergarten. To calculate distance, they use the methodology developed by Weber and Péclat (2017) 4 and show that house price decreases when the distance to the kindergarten increases.
In line with previous literature, our analysis is the first to investigate the impact of kindergarten proximity on Italy's housing prices. This paper starts filling this gap by estimating the degree of capitalisation of kindergarten proximity on housing prices in the eleven major Italian Municipalities.

Data collection and variables
4 They use the geo-route command in Stata.
The data used for the empirical analysis refers to Italian Municipalities 5 . We focus our research on the eleven largest cities with more than 250,000 inhabitants where location choices and more relevant for households with school-aged children. We consider the following municipalities: Turin, Milan, Verona, Genoa, Bologna, Florence, Rome, Naples, Bari, Catania and Palermo 6 . Our dependent variable is the housing market price that equates to the average between the minimum and the maximum cost per square meter of residential real estate located in each micro-zone of the target Municipalities. Data on housing market prices are taken from OMI that provides several pieces of information on micro-zones. Under the OMI definition, these microzones are sections of Municipality with uniform partitions of the real estate market since they present real estate with the same socio-economic and urban characteristics.
To define micro-zones within a Municipality, the maximum deviation of the range of real estate market values in each micro-zone should be lower than 1.5. For each microzone, the dataset shows the minimum and maximum price per square meter. Moreover, prices are differentiated according to the following properties' characteristics: (i) the use (residential, commercial, offices, and productive activities); (ii) the condition (normal, historical, luxury, ruined, etc.) and finally (iii) the city area where it is located (city centre, mid-central zone, suburban zones, rural zones etc.). All the data have been Information on state and non-state kindergarten quality is taken from data collected by the MIUR for the school year 2010-2011. For each kindergarten, this data gives information on (i) pupils; (ii) teaching staff, and (iii) structure. The data contains 5 Municipalities are the lowest level of government in Italy 6 Among the Municipalities with more than 250,000 inhabitants only Venice has been excluded from the analysis because of its lagoonal structure.
information on sex, year of birth, nationality, religious orientation, disability status, and type of disability regarding the pupils. The second group of data refers only to support teachers, discerning them according to the child's kind of disability. Finally, concerning the structure, the data provides information on the number of classrooms, schooling time, special facilities (antemeridian sections, Saturday sections, etc.) and size (square meters per pupils) of covered and uncovered playgrounds. Based on this data, it is possible to compute some key indices of kindergarten quality, such as the average class size, and the average size in square meters of playgrounds. Moreover, exploiting the data classification into state and non-state kindergartens makes it possible to assess the extent to which the competition between state and non-state childcare institutions could affect housing prices.
It is essential to analyze in more details the time structure of our dataset. We focus on 2010/2011 school information to evaluate kindergarten proximity's impact on housing prices in the subsequent years up to 2017. In this way, we can measure the persistence of school localization on the house values. Moreover, we also exploit information on housing market quotations before 2010 to depurate house prices from the influence of local amenities, the so-called neighborhood effect.
Finally, as control variables, we use municipal characteristics. This information is taken from two different datasets carried out by: (i) the Ministry of Interior (Ministero dell'Interno) and (ii) the National Institute of Statistics (ISTAT).

Empirical strategy
The main purpose of this paper is to assess the impact of the proximity of kindergartens on housing prices. To this end, the empirical framework is based on the basic hedonic housing price model developed by Rosen (1974). Thus, the price at year t of the house in micro-zone (neighborhood) j of Municipality i is determined by our basic estimation model: finally, !"# is the error term. The coefficients β, λ, δ and measure the marginal purchaser's willingness to pay for house, municipal, kindergarten quality and kindergarten proximity, respectively. Given that the main focus of the analysis is on 8 , the regressor !"% must be isolated from the other vectors in the model (1). For this purpose, we perform a multi-step strategy. In the first step, by exploiting the classification of the Italian dataset on real estate market, we consider the information on houses that have the same state of preservation (i.e. standard houses) and the same use (i.e. residential real estate and parking), so !"# $ becomes a constant in our specification, we remove it from the model and equation (1) becomes: The product of variables !"%# $ and !"% yields the matrix Ω !"# $ that indicates, for each neighborhood, the sum of the quality, at time t, of all kindergartens of town i, weighted by the distances of all kindergartens to the center of the target micro-zone j, in town i.
Therefore, equation (2) becomes: 8 captures the degree of capitalization of kindergarten proximity on housing price Since kindergartens and municipalities' structural characteristics can change only over a long-time span, we can remove the subscript t from the independent variables of the model. The time dimension remains valid only for the dependent variables since we will test the persistence of capitalization considering the housing prices at time t+x where x goes from 2011 up to 2017. As a result, the model in equation (3) becomes: However, we cannot identify !" (neighborhood effect) separately from !" (idiosyncratic error term) since we have no specific information on neighborhood characteristics. The problem is that since neighborhood characteristics are probably correlated with schools' feature, the OLS estimator will produce biased estimates of . To circumvent this lack of information we exploit the long time series of housing prices and we perform a twostage approach to compute correct estimates of . In the first stage, we estimate the following model: where !"# are the i.i.d error term and t go from 2004 up to 2011, i.e. all the years before observing school characteristics. The model in (5) is estimated using the Within-the-Group estimator to obtain an estimate of 7 !" that works as a proxy of the neighborhood effect on the housing price.
The final specification of the second stage model is reported in the following equation (6): where the dependent variable 9999999 !"# correspond to : !"# − 7 !" < equal to the housing price of each neighborhood j depurated from the neighborhood effect. In this way, we can estimate (through the OLS) the unbiased impact of kindergarten proximity on housing prices and its persistency up to the sixth year after the evaluation of school's localization and quality and other municipal characteristics.
As a final step of our empirical strategy, since we are also interested in examining the impact of the presence of non-state kindergartens on the capitalization of kindergarten proximity, we add a dummy variable % in equation (6) to differentiate between non-state and state kindergartens. Hence, equation (6) takes the following form: where % is the non-state kindergarten dummy variable 9 . Multiplying % Ω !" $ with % !"% the equation (7) is expressed as: where captures the effect of non-state kindergarten proximity on housing price.

Empirical results
In our basic specification, we estimate each regression equation considering three dependent variables: the mean, maximum, and minimum housing price value.
Moreover, the housing price is taken either in its original or depurated neighborhood effect estimated in the first stage regression. Besides, since variables are expressed in different measurement units, we have standardized them imposing mean 0 and standard deviation equal to 1. In this way, we can compare the magnitude of the coefficient point estimates interpreting them in terms of standard deviation.
9 Note: among the regressors in equation (7), there is also a variable that defines the percentage of non-state kindergartens in a Municipality. For a sake of simplicity, it is not displayed explicitly in the model. 10 Proximity is measured as the inverse of distance. We use the inverse in order to interpret the coefficients' point estimates as the direct effect of proximity of kindergartens on housing prices. The main results confirm the capitalization of the house to the kindergarten proximity. In other words, the school proximity coefficient estimates suggest that, overall, close location to kindergarten has a significant and positive effect on housing price. As we expected, the capitalization effect becomes smaller after we depurate the housing price from the neighborhood effect but remains in most of the specification positive and statistically significant.
Yet, we focus on each dependent variable. When we consider the mean value of housing price as the dependent variable, the estimated coefficients are stable across all specifications with a weak increase in intensity over time. When taking the minimum value of housing price as reference, the proximity effect is lower concerning the medium and maximum value. It decreases more and more over time, especially in the years 2016-2017. Therefore, we can observe a persistency effect, although it decreases over time.
On the contrary, the maximum value of housing price appears to be stronger than the other two values even if the magnitude of proximity kindergarten decreases during the years considered.
Adding the variables that capture the quality of schools, we find that these mitigate the proximity effect. It follows that although proximity plays a key role, other variables affect the housing market price. Tables  To sum up, in line with other empirical findings (i.e. Owusu-Edusei, et al.,2007;Chin and Foong, 2006;Wen et al., 2014Wen et al., , 2017Huang and Hess, 2018), our results confirm that home buyers consider the proximity to schools in their home purchase decision. Moreover, this analysis shows that the degree of capitalisation of kindergarten proximity in housing price depends mainly on proximity and some quality school characteristics.

Alternative estimations and robustness check
In what follows, we describe the results of the alternative estimations (Table 3).
We have re-run the baseline specification dividing public and non-state kindergartens to investigate in more details the proximity impact on the housing price. Also, in this case, for readability reasons, the coefficients of other variables are not exhibited. The column sequence follows Table 2, Columns 1-4 present the empirical results not purified from the neighborhood characteristics for all kindergarten proximity. Specifically, findings in column 1 refer to a simple model which regresses kindergarten proximity on housing market prices without considering any control variable. Column 2 shows the results of the model that includes quality characteristics of kindergartens as control variables. Column 3 exhibits the results obtained, taking into account the municipal characteristics, and finally, column 4 refers to a model that considers both municipal and school aspects. On the other hand, columns 5-8 contain results obtained running OLS on the same models employed in columns 1 to 4 considering as a dependent variable the price of housing depurated from the neighborhood effect. In other words, if the kindergartens were all equidistant from the centroid of the micro-zone, the capitalisation effect could disappear. On the contrary, there is a capitalisation effect when the kindergartens are more concentrated in some areas with respect to other ones. The capitalisation effect seems to depend on private kindergartens that do not act like public institutions. The latter are located mainly in the same place as other types of educational institutes.
Based on the discussion above, the introduction in our analysis of the distinction between public and private kindergarten allows us to observe how the degree of capitalisation of kindergarten proximity in housing price depends mainly on non-state kindergartens' distance. In other words, house prices decrease as the distance to private kindergarten increases.

Final remarks
The paper aimed to investigate the impact of kindergarten proximity on housing market prices in Italy. In particular, we focused on the eleven major cities in the country.
Although several empirical studies investigate the capitalization of the quality and proximity of schools in the housing market, no research has, so far, focused on the Italian context. Therefore, this paper has started filling this gap by estimating the impact of kindergarten proximity on housing prices. To this end, we employed a hedonic property price model, exploiting the panel dimension of the Italian dataset to control for endogeneity. It has then been investigated whether non-state kindergartens' presence generates a different impact than state kindergarten proximity on the market price of houses. Empirical results have shown that homebuyers do consider the proximity to kindergartens in their home purchase decision. The main results confirm the capitalization of the house to the kindergarten proximity. In other words, the school proximity coefficient estimates suggest that, overall, close location to kindergarten has a significant and positive effect on housing price.
Moreover, findings have shown that non-state kindergarten is the main determinant of the capitalization of kindergarten proximity in housing price. These results can be interpreted as evidence of the higher quality of non-state kindergartens with respect to state institutions. In particular, it seems that households perceive a higher quality of non-state kindergartens run by local governments.
Broadly speaking, the unequal presence on the territory of private kindergartens leads to greater capitalization of real estate with respect to the public kindergartens that present a more uniform distribution.
To conclude, given this positive relationship between private kindergarten proximity and housing market price, our findings could be useful in letting real estate developers and urban planners decide where to locate kindergartens to develop a city more homogeneously. Our results could also support investors in valuing the education facilities in the investment return and families in the buying of property.
Finally, the crucial caveat to be highlighted descends from the limited number of Italian Municipalities even if they present homogeneous characteristics. Further research on this topic could be based on larger datasets to go beyond the limitations of this current work.