1. Introduction

Between 2012 and 2022, the supply of social housing in Mexico decreased by 54.1%. This decline was even more pronounced in specific segments: economical housing fell by 94.6% and low-income housing by 73.5%, according to data from the ^{Single Housing Registry (RUV) (2024)}. One of the main causes of this contraction is the limited availability of land in areas with the public services and urban infrastructure necessary for the population. The ^{Sectoral Business Office (2024)} reports that the supply of land has shifted to peripheral or semi-urban areas, which have limited access to basic services and fewer formal job opportunities. In this context, analyzing the factors that determine land prices is important for understanding the structural challenges of the real estate market in Mexico and its link to the availability of affordable housing.

Land costs are an important component of the final price of housing. In residential housing, land costs represent 13.4% of the total cost. In social housing, land costs amount to 40.1%, according to estimates based on prices published on real estate websites and financing data provided by the ^{National Housing Commission (CONAVI) (2024)}. Despite its importance, little research has been conducted on the land market in Mexico due to a lack of public data, which limits the ability to design appropriate public policies to address the housing deficit and meet the needs of lower-income groups.

This study helps to fill this gap by using modern data collection and analysis techniques. A database was created using data obtained through web scraping of real estate websites, supplemented with public sources relating to the physical and social conditions of the environment. The initial database contained 33,375 records for the state of Yucatán, 22,820 of which were located in the city of Mérida (CDM). After applying filtering criteria to eliminate outliers, data entry errors and duplicates, a final sample of 17,274 observations was obtained, allowing for consistent statistical analysis. Previous studies in the region have worked with smaller samples, for example, ^{Sandoval Escalante and Becerril García (2025)} used only 228 observations.

In addition to using traditional hedonic regressions, this study incorporates a spatial component in order to correct autocorrelation in the data and identify relationships between prices and environmental characteristics. This approach is relevant in Yucatán, where the land market is fragmented by geographical differences, accessibility and urban structure among municipalities. The results demonstrate that prices depend not only on the physical attributes of the land, but also on the characteristics of the neighborhood or surrounding neighborhoods.

Analyzing the market for undeveloped land provides insight into the urbanization challenges facing Yucatán. From 2000 to 2022, the state recorded an annual population growth rate of 1.6%, compared to 1.2% nationwide. Recent migration has accentuated this demographic dynamism. For example, between 2015 and 2020, 100,200 individuals arrived in the state, which is double the number recorded between 2010 and 2015. According to data from the National Institute of Statistics and Geography ^{(INEGI) (2021)}, by 2020, 4.3% of the total population of the state was migrant. This population increase puts pressure on the land market and on the availability of land suitable for urbanization.

Studying land in Yucatán is important because population growth, both local and due to migration, together with recent urban expansion, has intensified demand for land, driving up prices and putting pressure on the availability of suitable housing space.

Mérida, located in southeastern Mexico is a tourist and cultural hub of national importance that faces a territorial dynamic marked by differentiated patterns of urbanization. These conditions raise questions about how land prices are established, in order to guide urban development and housing access policies.

This study is organized into six sections following this introduction. The second section presents a review of the literature on hedonic pricing from theoretical and empirical perspectives. The third section describes the data used and the process of constructing and refining the database. The fourth section presents descriptive statistics and initial market patterns. The fifth section details the econometric methodology, followed by a discussion of the results and their implications. The paper concludes with conclusions and suggestions for future research.

2. Literature review

The literature review focuses on the hedonic approach applied to the study of real estate markets, emphasizing the importance of considering the intrinsic attributes of land and environmental characteristics. ^{Palmquist (1984)} points out that the sociodemographic attributes of the resident population, such as level of education, income and household structure, influence the demand for certain spatial attributes. These variables influence location preferences and affect the spatial distribution of prices. From this perspective, the hedonic price function reflects not only the physical characteristics of the asset, but also the social and economic dynamics that structure urban areas. This evidence justifies using models that incorporate spatial heterogeneity and market segmentation, particularly in regions with recent urban expansion, such as the CDM.

The theoretical basis for the hedonic approach stems from the work of ^{Rosen (1974)}, which argues that the price of a heterogeneous asset, such as housing or land, stems from the overall valuation of its attributes. These attributes include surface area, location, accessibility, availability of services and environmental conditions. Each attribute contributes differently to the total price. This approach assumes that consumers choose the combination of attributes that maximize their well-being subject to budget constraints, which simultaneously determines their willingness to pay and the price structure in the market.

The theoretical and empirical literature reviewed in this section supports the importance of integrating physical, social and spatial variables into hedonic models and analyzes the most suitable methodologies for capturing the complexity of urban land value. These contributions form the conceptual basis for the analysis developed in this study, wherePirepresents the price function of the asset, which is determined by the set of its attributes:

Consumers choose housing that provides the greatest utility, represented as:

wherexare observable consumer characteristics, ε reflects unobserved preferences and z are the attributes of the asset.

Market equilibrium is achieved when the marginal valuation of each attribute by the consumer is equal to the price increase associated with that attribute:

This allows us to measure the value that consumers place on each attribute. The gradients of the pricing function show the marginal utility associated with each characteristic: how much the price increases with a unit change in each attribute.

The literature on hedonic pricing recognizes that identifying the implicit prices of attributes requires additional assumptions. ^{Heckmanet al.(2010)} demonstrate that this identification is limited within a single homogeneous urban market because exogenous variation in attributes is often insufficient. The authors point out that the hedonic function is not necessarily linear or continuous, as residential choice processes generate discontinuities derived from heterogeneity of preferences, market segmentation and budget constraints.

Nonlinearity is related to the process of spatial ordering, whereby households differ in terms of income, preferences and sociodemographic composition, and are distributed in neighborhoods with heterogeneous amenities, generating segmented submarkets. This process of self-selection produces differentiated price function curves or segments, as documented by ^{Nesheim (2004)} and ^{Dayet al.(2004)}. According to these authors, the socioeconomic composition, the quality of schools, age distribution and ethnic diversity influence the shape of the price function. Similarly, ^{Goodman and Thibodeau (1998)} suggest that urban markets are structured into submarkets where households are grouped by income, race and preferences for services, revealing patterns.

In this context, incorporating variables that capture the attributes of an asset and its surroundings improves the identification of hedonic parameters. Explicitly considering the immediate surroundings, including amenities, services, social composition and quality of education, helps to reflect the segmented structure of the market and to avoid biased estimates resulting omitting relevant attributes.

At the international level, spatial components have been incorporated into hedonic models. ^{Glaesener and Caruso (2015)} and ^{Munshi (2020)} demonstrate that accessibility, urban amenities and neighborhood characteristics significantly influence land and housing prices. Techniques such as geographically weighted regression, employed by ^{Soler and Gemar (2018)} and ^{Wanget al.(2022)}, capture spatial heterogeneity using functions that vary according to location. ^{Glumacet al.(2019)} applied the Spatial Durbin Error Model (SDEM) to estimate a land price index in Belgium, which highlights the importance of considering spatial autocorrelation and indirect effects. The authors categorize hedonic variables into eight groups:1)accessibility,2)proximity,3)physical attributes,4)environmental,5)legal,6)social,7)socioeconomic, and8)economic. This classification, which we use in this study, makes it easier to systematically include attributes of the asset and its surroundings.

In the field of spatial parametric models, ^{Elhorst (2010)} summarizes advances in models designed for spatially dependent data. These models capture direct and indirect effects between geographically close observations, such as the SDEM itself. This approach is particularly suitable for real estate markets, where property prices are often influenced by the value of neighboring properties and the structure of the immediate environment.

In the case of Mexico, empirical evidence is limited due to data restrictions and is mainly concentrated in Mexico City. ^{Sobrino (2014)} identifies submarkets defined by sociodemographic characteristics in the Metropolitan Area. ^{Atuestaet al.(2018)} found that labor informality, level of education and distance to employment centers influence housing prices. Meanwhile, ^{Martínez-Jiménezet al.(2022)} demonstrate that neighborhood characteristics and accessibility carry more weight than proximity to green or protected areas in informal peripheral markets. The increased use of web scraping techniques has enabled the creation of extensive, up-to-date databases, prompting more recent studies. Institutions such as the ^{International Monetary Fund (IMF) (2022)} and ^{Savioet al.(2018)} recognize the potential of these methodologies to generate timely, low-cost real estate indicators.

In the case of the CDM, ^{Aguilaret al.(2025)} document a pattern of diffuse expansion with variations in amenities and land prices. Evidence shows internal differentiation: the south and the surrounding municipalities such as Kanasín, Umán and Caucel, offer economical and low-income housing, while the north and Conkal are home to middle-class residential developments (^{Pérez Medina, 2024}). This spatial heterogeneity means that prices cannot be assumed to be uniform, as contextual attributes are not randomly distributed. This gives rise to the need to incorporate spatial dependency structures into hedonic models. Therefore, this study proposes a hedonic model with a spatial component to capture the spatial heterogeneity and segmentation of the land market in the CDM. This model uses digital data and robust statistical techniques.

3. Data

4. Descriptive statistics

Table 1 presents the descriptive statistics of the physical, social and spatial characteristics of the analyzed land, as well as the variables of accessibility and legal status. The total price shows wide dispersion, with an average of $1,500,000 Mexican pesos (MXN) and a standard deviation of $1,800,000 MXN, indicating variability associated with structural and environmental differences between plots of land. The average surface area is 597.1 m², with a deviation of 494.8 m², reflecting an uneven distribution in plot size. The price per m² also varies significantly: the average price is $2,698.20 MXN and the standard deviation is $1,900.40 MXN, indicating differences even within nearby geographical areas.

In terms of the immediate surroundings, the average distance to the nearest economic center is 11.5 km, with an internal variation of 4.4 km. Taking into account education, health and finance, the average number of services available is 2.7 services per plot of land, although there is high dispersion (8.2), associated with plots of land located in both urbanized areas and in areas with limited infrastructure. The median indicates that 50% of plots of land do not have services in their electoral section, suggesting differences in basic accessibility between areas.

In sociodemographic terms, the average years of schooling for the population in the immediate surroundings is 11.1, with a deviation of 1.8. Regarding spatial distribution in relation to the coast, 1,555 plots of land (9%) are located less than 3 km from the coastline. The remaining 15,719 (90%) are located farther away, which allows for differentiation between plots of land with potential for coastal tourism or residential use and those located farther inland.

Legally, 7,845 plots (54.6%) are ejido land, while 9,429 plots (45.4%) are privately owned. In terms of type of area, 14,704 plots (46%) are in areas classified as human settlements, 2,467 (9.5%) are in other areas and only 103 are in agricultural areas. This distribution indicates a market dominated by areas geared toward urban and peri-urban development, with a marginal share of agricultural land.

The data is distributed across different quarters of the study period: 8,841 records correspond to the first quarter, 2,877 to the second, 4,613 to the third and 2,429 to the fourth. It is possible that the same plot of land may appear in more than one quarter, which is explained by repeated records. In total, 2,803 observations correspond to plots of land recorded on multiple occasions, while 14,471 correspond to unique plots of land.

The exploratory analysis in Appendix 1 enables us to identify preliminary patterns in land price formation based on graphical relationships and comparisons between variables. The scatter diagrams comparing total price and continuous variables such as surface area, distance to the economic center, schooling and number of services show relationships accompanied by correlation coefficients that help interpret their influence. Surface area has a positive correlation of 0.51, though there is high dispersion, indicating that there are small plots of land with high prices, suggesting the involvement of other attributes not represented by a single physical dimension. Distance to the economic center has a negative correlation of 0.13 and average prices remain similar when the sample is divided according to proximity. This indicates that the relationship with price is not strictly linear. Average schooling has a positive correlation of 0.22, which is possibly associated with higher incomes in areas with greater human capital, while the number of public services has a low correlation of 0.05.

Comparisons with categorical variables reinforce the role of territorial context. Plots of land located in human settlements have the highest average prices, close to one million pesos, while agricultural land and land classified as other types have lower values. In terms of ownership, private land tends to fetch higher prices than ejido land, consistent with the legal restrictions on ejido land.

Territorial accessibility also shows associations with price. Distance to the coast has limited effects, as plots of land located less than 3 km from the coastline have similar average prices to those located farther away. Taken together, these findings demonstrate the heterogeneity of the land market in the CDM.

Figure 1 provides two maps that complement these results. The first map shows the spatial concentration of land offered for sale and highlights high-density areas on the northern and northeastern outskirts of Mérida, particularly in municipalities such as Conkal, as well as areas near Cholul. Smaller pockets of available land can also be seen in the south. The second map shows the spatial distribution of prices, where we can see that the highest values coincide with these corridors of urban expansion. Prices gradually decrease as the distance from the metropolitan center increases or in locations with less urban consolidation.

Source: Prepared by the author using data from the Internet and the ^{INE (2024)}.

Figure 1 Number and total price of plots of land in Mexico City 

Both maps show that, although land concentration is not uniform, there is a spatial pattern where higher prices tend to be found in regions close to the city center and in emerging areas of real estate expansion. This pattern is consistent with a dynamic of rising values linked not only to proximity to the historic center, but also to new nodes of urban growth.

5. Methodology

Spatial data analysis requires econometric techniques capable of capturing the dependence and autocorrelation present in geographically distributed variables. When real estate prices, infrastructure or socioeconomic characteristics are not randomly distributed across the territory, the assumptions of the Ordinary Least Squares model are not suitable. In this context, ^{Elhorst (2010)} assembles a variety of spatial models and describes their uses and differences. A significant advantage of these models is the interpretability of their parameters, which is essential for analyzing the determinants of land prices. This criterion was central to selecting the model to be used in this study.

To identify the factors affecting land value and recognize possible spatial spillovers, we selected the SDEM model. The SDEM model corrects spatial autocorrelation and incorporates spatial effects into the explanatory variables, providing a more accurate representation of the price formation process.

According to ^{LeSage (2009)}, spatial econometric analysis should begin with an Ordinary Least Squares model and should evaluate the residuals using diagnostic tests. The first recommended tool is the Moran I statistic, which allows us to determine whether neighboring units have similar values. A significantly positive value indicates the presence of autocorrelation and evidence that the basic model omits relevant spatial relationships.

^{Elhorst (2010)} subsequently suggests applying Lagrange multiplier tests and likelihood ratio tests to confirm autocorrelation in the dependent variable or errors and compare nested models to verify the presence of spatial processes.

While spatial models solve the autocorrelation problem, the construction of the spatial weight matrixWis often questioned, as there is no single, universal method for defining it. The choice depends on the phenomenon being analyzed and is the responsibility of the researcher. TheWmatrix is an n × n square matrix where each elementw_ij represents the intensity of spatial interaction between unitsiandj:

The most common approach for constructing theWmatrix is binary contiguity:

1 ifIandjshare a frontier

0 if they do not

This study uses an inverse distance matrix, as it captures the spatial influence continuously and on a decreasing scale, which aligns with the logic of the real estate market.

TheWmatrix formalizes the idea that the characteristics of neighboring areas influence local prices. For instance, if a school or hospital is located in a neighboring unit, we can quantify its impact on the price of land in the focal unit, even if said service is not located within the unit itself.

The SDEM addresses two different sources of spatial dependence: dependence on explanatory variables and dependence on errors:

y: vector of the dependent variable.

X: matrix of explanatory variables.

β: vector of coefficients of the explanatory variables (direct effects).

W: matrix of spatial weights.

WX: spatial term of the explanatory variables (effects of the variables of neighboring units).

θ: vector of coefficients associated withWX(indirect effects).

u: error term with spatial autocorrelation.

The error term can be broken down as follows:

λ: spatial autocorrelation coefficient in the errors.

W_u: spatial term of the errors.

ϵ: random error term (white noise), with ϵ~N(0,σ2 I).

The SDEM interpretation is based on direct and indirect impacts rather than elasticities.

Direct impact: measures how a change inX_i affects the price in the same uniti.

Indirect impact or spillover: measures how changes inX_j affect the price in neighboring units.

These effects allow us to measure the interaction between submarkets, reflecting the fact that prices depend not only on internal characteristics, but also on the spatial environment. In short, the SDEM provides an appropriate framework for assessing the spatial heterogeneity of land prices and how the characteristics of neighboring areas influence their formation.

6. Results

The SDEM estimates the total price of land in the CDM by incorporating both the characteristics of the land itself and those of its closest neighbors. In Table 2, the results of the model show that the surface area variable has a significant negative effect, indicating that larger plots of land tend to have a lower price per m², which is a common trend in markets with a discount for scale or low liquidity. This relationship is also reflected in simple linear regression models,³where the surface area coefficient is positive for the total price (MXN$1,800 per additional meter) and negative for the price per m2 (MXN$0.4), confirming the a size penalty.

In terms of time, the first quarters show negative effects, indicating a progressive increase in land prices. Simple models show that prices in the first quarter are MXN$270.9 pesos per m² lower than in the last quarter, indicating an upward trend. Distance from the urban center shows negative effects in both the spatial model and the simple models (-MXN$35,000 and -MXN$44.7 per m²). This pattern is interpreted based on the polycentric structure of Yucatán, where Mérida and Progreso function as differentiated urban nodes and there is no single dominant center.

The methodological decision to incorporate a polycentric structure into the analysis allows us to evaluate land value based on its relative proximity to different centers of activity. However, the results show that distance from urban centers loses explanatory power when compared to social and spatial variables, even when this is taken into account. This suggests that the immediate environment, accessibility and available services influence land prices more directly than proximity to a particular urban node. This behavior is consistent with previous studies documenting processes of loss of centrality in historic centers in cities such as Culiacán and Guadalajara (^{Pérez Tamayoet al., 2017}; ^{Mcenulty and Mercado, 2019}).

The spatial approach also allows for the identification of dynamics associated with speculative processes. The model shows high prices in areas adjoining high-value land, even when the plots lack attributes that would explain this difference. This pattern is consistent with the existence of a speculative premium, a phenomenon documented in Latin American urban markets. Likewise, the average schooling in the surrounding area emerges as a significant variable in all specifications, indicating that the perceived human capital in the area affects land value.

Meanwhile, the number of services negatively affects the spatial model, as shown in Table 2. This result may be associated with oversupply or speculative anticipation processes in unconsolidated areas. Simple models also show that the presence of at least one basic service is related to lower prices, possibly because these plots are urbanized but have not yet seen an increase in value. Furthermore, ejido lands do not exhibit a significant direct effect; however, they appreciate in value when located close to areas with high prices, which may reflect expectations of regularization.

The model has some limitations due to the absence of relevant variables, such as proximity to cenotes or tourist developments. Without these variables, the model cannot fully explain certain observed patterns, such as land located on golf courses, where prices are above average. Finally, opportunities for methodological improvement were identified by incorporating advanced measures of accessibility, such as travel times or road connectivity indicators. While the methodology is replicable, interpreting the results requires adaptation to each context because land prices depend on individual attributes and the characteristics of the environment.

7. Conclusions

The analysis of the land market in the CDM, based on a spatial econometric approach, shows how physical, temporal and spatial attributes influence the pricing of land. The most stable result among the estimated models is the negative relationship between land area and price per m2. Larger plots of land have lower unit prices, suggesting a market structure in which size penalties consistently apply. This pattern appears in both spatial and non-spatial specifications, indicating that it is a persistent feature of the Yucatecan land market.

In terms of timing, the results reflect a sustained increase in land prices during the analyzed period. This behavior can be associated with gradual appreciation processes or investment dynamics aimed at obtaining capital gains. The observed upward trend in prices supports the need to analyze the interaction between urban expansion, appreciation expectations and the behavior of those involved in selling and purchasing land.

Regarding spatial factors, the results show that proximity to specific urban centers, such as Mérida or Progreso, alone does not explain price variation. Although these distances were incorporated from the outset to capture a polycentric logic, the model indicates that the pricing of land depends more on the immediate environment and effective accessibility. These findings suggest that the territorial structure operates as a network of interconnected nodes, where services, connectivity and relative position within the urban system carry more weight than simple proximity to a dominant center.

The model also identifies patterns associated with speculative investment processes. High prices are observed in areas adjacent to valued areas without a clear correspondence to visible attributes, indicating anticipation of capital gains. Meanwhile, the positive effect of the educational environment and the negative effect of the number of available services suggest purchasing behavior aimed at taking advantage of expectations of urban growth. These dynamics are present in municipalities such as Kanasín and Umán and reveal the importance of designing fiscal and institutional instruments that mitigate speculative land retention and guide land use toward productive or residential purposes.

From a public policy standpoint, the results of the model indicate that infrastructure and basic services influence land prices more because of their potential for future appreciation than for their immediate value. This underscores the need to prioritize public investment in areas with proven demand and to control urban expansion driven by speculation.

The variability of prices between similar areas reflects the existence of information imbalances. Consequently, it is advisable to create market monitoring bodies that incorporate land registers, public records and geospatial data. Finally, while the model is replicable, its interpretation must consider the restrictions of the available variables and the specific territorial characteristics of each region. One direction for future research is to extend this exercise to the national level, which would allow for the comparison of regional patterns and provide a more comprehensive view of how the land market functions in Mexico. Likewise, future research could strengthen the database by incorporating more attributes to improve the accuracy of the price function estimation.