Map of Chile and its regional and communal administrative boundaries 

Contexts in source publication

Context 1

... However, and as noted before, the main attractiveness of the data is that all Chil- ean regions are represented in the sample for which the 13-regions classification was chosen 7 . This is portrayed in Figure 1 along with the municipality classification. ...

Context 2

... begin the empirical analysis with the hedonic regressions and Table 2 shows the estimated coefficients for equation (10). As discussed in the previous section, we estimate this regression for each of the 13 regions portrayed in Figure 1. Given that housing attributes are considered consumption goods, we expect positive coefficients to be associated with each one of the four characteristics. ...

Context 3

... empirical reason to not rely on hedonic regressions alone is the fact that signifi- cance levels vary across regions, which is presumably due to different sample sizes. While the Metropolitan Region of Santiago consists of 42% of the total population, regions such as Aysén (XI) make up less than 1% (see Figure 1). This fact has direct consequences on sample sizes obtained from CASEN where the Aysén Region (XI) sample size ac- counts only for 10% of the Metropolitan Region sample size (930 versus 9204). ...

Context 4

... propose a fourth dimension called Housing Loca- tion (HL) but it must be carefully interpreted because of the lack of information about the exact address of each house in the CASEN survey. To overcome this limitation 4 This follows the recommendations of the Latin American and Caribbean Demographic Center (CELADE -Centro Lati- noamericano y Caribeño de Demografía in Spanish). 5 Please see in the Supplementary Annex A for complete details about the construction and result tables for both Housing Fea- tures definitions. Appendix A1 shows the HS computed trough a Factorial Analysis, while A2 shows the weighted average pro- posed by Parsons (1986). All our results consider the index built by Factorial Analysis, but we use the Parson's index to evaluate the robustness of them. this paper follows the approach used in Garcia and Raya (2011) who countered the same problem using an ingen- ious strategy. The authors discovered that the second best way of capturing locational effects when the actual loca- tion is not available is assuming that the level of schooling of the household head be a satisfactory proxy for a desir- able neighborhood. The intuition behind this assumption is that households with higher education levels will enjoy a higher income and hence be able to access houses located in more preferable areas. Garcia and Raya (2011) supports this assumption by including a set of fixed effects at the lowest level of spatial aggregation in the hedonic regres- sions, and then comparing those coefficients with the edu- cation level via a correlation coefficient. Even when this procedure does not explicitly highlight the individual local amenities affecting a particular housing price, these effects should be captured by the coefficients of each dummy. In fact, Garcia and Raya (2011) find a strong correlation (close to 0.9) (p. 5), which is expected since their data is only for Barcelona and hence the disaggregation level is also higher. In our case, we find a weaker correlation (close to 0.4), which is expected since the lowest disag- gregation level available is the municipality and more than one small housing market can be contained in a single municipality. Nonetheless, its positive sign is interpreted here as a signal to support our use of the variable "years of schooling" as a proxy for housing location. Due to changes in variable names and codifications over time, prior to conducting estimations it was neces- sary to achieve consistency as well as to select only rel- evant variables and observations. 6 However, and as noted before, the main attractiveness of the data is that all Chil- ean regions are represented in the sample for which the 13-regions classification was chosen 7 . This is portrayed in Figure 1 along with the municipality ...

Context 5

... begin the empirical analysis with the hedonic regressions and Table 2 shows the estimated coefficients for equation (10). As discussed in the previous section, we estimate this regression for each of the 13 regions portrayed in Figure 1. Given that housing attributes are considered consumption goods, we expect positive coefficients to be associated with each one of the four characteristics. Besides paying attention to the coefficient signs, we follow the literature on hedonic prices which suggest we take advantage of the log-linear form which is used here to discuss the coefficients as pseudo-elas- ticities (Cropper, Deck, & McConnell, 1988). Housing Size (HS) shows a positive and significant coefficient in each re- gion implying that larger houses have higher renting prices. In particular, HS ranges from 4.37% (I Region of Tarapacá) to 13.8% (IV Region de O'Higgins). This variability of pseudo- elasticity across regions can be understood as initial evidence to support our hypothesis of spatial heterogeneity. The rest of the housing attributes also show a positive coefficient, with their magnitudes differing among themselves and across re- gions. For example, Housing Quality (HQ) is the highest co- efficient for all regions, while our proxy of Housing Location (HL) is the lowest coefficient, with both housing attributes statistically significant in all regions. Nevertheless, as the previous analysis and literature highlight, real economic meaning must be derived from the estimation of a system of equations Rosen (1974) instead of only using hedonic regressions. An empirical reason to not rely on hedonic regressions alone is the fact that signifi- cance levels vary across regions, which is presumably due to different sample sizes. While the Metropolitan Region of Santiago consists of 42% of the total population, regions such as Aysén (XI) make up less than 1% (see Figure 1). This fact has direct consequences on sample sizes obtained from CASEN where the Aysén Region (XI) sample size ac- counts only for 10% of the Metropolitan Region sample size (930 versus 9204). Albeit these differences in sample sizes and significance levels, the estimated coefficients for housing attributes are significant for almost all ...

Context 6

... begin the empirical analysis with the hedonic regressions and Table 2 shows the estimated coefficients for equation (10). As discussed in the previous section, we estimate this regression for each of the 13 regions portrayed in Figure 1. Given that housing attributes are considered consumption goods, we expect positive coefficients to be associated with each one of the four characteristics. Besides paying attention to the coefficient signs, we follow the literature on hedonic prices which suggest we take advantage of the log-linear form which is used here to discuss the coefficients as pseudo-elas- ticities (Cropper, Deck, & McConnell, 1988). Housing Size (HS) shows a positive and significant coefficient in each re- gion implying that larger houses have higher renting prices. In particular, HS ranges from 4.37% (I Region of Tarapacá) to 13.8% (IV Region de O'Higgins). This variability of pseudo- elasticity across regions can be understood as initial evidence to support our hypothesis of spatial heterogeneity. The rest of the housing attributes also show a positive coefficient, with their magnitudes differing among themselves and across re- gions. For example, Housing Quality (HQ) is the highest co- efficient for all regions, while our proxy of Housing Location (HL) is the lowest coefficient, with both housing attributes statistically significant in all regions. Nevertheless, as the previous analysis and literature highlight, real economic meaning must be derived from the estimation of a system of equations Rosen (1974) instead of only using hedonic regressions. An empirical reason to not rely on hedonic regressions alone is the fact that signifi- cance levels vary across regions, which is presumably due to different sample sizes. While the Metropolitan Region of Santiago consists of 42% of the total population, regions such as Aysén (XI) make up less than 1% (see Figure 1). This fact has direct consequences on sample sizes obtained from CASEN where the Aysén Region (XI) sample size ac- counts only for 10% of the Metropolitan Region sample size (930 versus 9204). Albeit these differences in sample sizes and significance levels, the estimated coefficients for housing attributes are significant for almost all ...