Examining the Effect of Squatter Settlements in the Evolution of Spatial Fragmentation in the Housing Market of the City of Buenos Aires by Using Geographical Weighted Regression

Abstract

The spatial fragmentation in the housing market and the growth of squatter settlements are characteristic for the metropolitan areas in developing countries. Over the years, in large cities, these phenomena have been promoting an increase in the spatial concentration of poverty. Therefore, this study examined the relationship between the squatter settlement growth and spatial fragmentation in the housing market of Buenos Aires. By performing a spatiotemporal analysis using geographically weighted regression in the house prices for the years 2001, 2010, and 2018, the results showed that while squatter settlements had a strong negative effect on house prices, the affected areas shifted over time. Our findings indicate that it is not the growth of the squatter settlement that causes spatial fragmentation, but rather the widening income disparities and further segregation of low-income households. However, squatter settlements determined the spatial demarcation of fragmented housing market by attracting low-income households to surrounding low house price areas.

1. Introduction

In the past decades, many metropolitan areas have experienced the continuity of their traditional urbanization patterns, but in parallel, new trends associated with real estate speculation have emerged [1]. From the 1990s, the “new urban policy” approach has promoted deregulation of private investments into the real estate market. This deregulation led to an increased number of decisions in the private sector regarding urban change and development [2,3]. This policy is related to large-scale redevelopment featuring commodity housing, public services, and commercial and office space. It also includes the city center redevelopment in response to restructuring the processes associated with transforming of the production and demand in different scales combining physical upgrading with socio-economic development objectives by replacing the more traditional redistribution-driven approaches [3]. The consequences of socio-economic changes deepened social disparities and increased spatial fragmentation [4]. The term fragmentation is originally used in ecology to capture a loss of continuity in a particular kind of habitat regarding a loss of connectedness to the external environment [5]. In urban studies, spatial fragmentation addresses land use activities’ discordance and the physical properties of space. The spatial fragmentation in the urban landscape can undermine integration and limit interactions among urban dwellers. The fragmentation process is considered to be aligned with the increase in inequalities [6]. The spatial econometric estimates suggest that house prices are negatively impacted by spatial fragmentation at low fragmentation levels, yet there is a positive price relationship at high fragmentation levels [7]. The fragmentations expose the rise of inequalities, suggesting breakup of neighborhood types, which culminates in the spatial segregation of social groups.

One of the main reasons for this fragmentation is the perception of violence and crime inside the city, which is related to the deterioration of living conditions in certain parts of the city, typically described as squatter settlements and slums [8]. These settlements are multiple exclusion sites, both materially and socially, and they are stigmatized spaces, associated with criminal activities. Their association with illegal activities leads to negative perceptions of the squatter settlements as a no-go zone by a part of the population outside these enclaves and also being spatially segregated from the rest of the city. They are also vulnerable to communicative diseases including COVID-19 of the ongoing pandemic [9], posing a threat to public health. These perceptions influence the housing market, lowering the neighborhood’s property prices [10]. The consequence of spatial segregation leads to an uneven distribution of the city’s services, infrastructure, and investments.

Although the spatial fragmentation exposes the rise of inequalities, the interaction between neighborhoods continues. However, in spatial segregation, this interaction is almost absent, particularly in squatter settlements. In many metropolitan areas in developing countries, these kinds of settlements have increased in number and size. In 2010, a total of 872 million people lived in squatter settlements and slums [11]. That growth is linked with the increase of spatial fragmentation and segregation within the city. Understanding the relational geographies and revealing the causes of the gradual proliferation of the spatial fragmentation and the evolution of the phenomenon allows us to define the possible responses to reduce these disparities. This article aims to contribute to this knowledge base by assessing how squatter settlements are one of the causes of socio-spatial fragmentation and their dynamic effect on house prices over time.

This paper begins with a literature review concerning spatial fragmentation and segregation in the housing market, leading to our hypothesis that effects of squatter settlements on house prices promote gradual spatial fragmentation. Subsequently, we introduce our study area, the City of Buenos Aires (CBA), where rising disparities in house prices within the city exemplify spatial fragmentation. The methods section describes how geographical weighted regression (GWR) in multiple time slices is used to measure the different effects of squatter settlements across space and their changes over time. After presenting the results, the discussion section examines our hypothesis and identifies areas that the fragmentation has been concentrated concerning house prices.

2. Literature Review

Fragmentation is a multidimensional concept with different characteristics [12]. Different terms and concepts are used as synonyms or notions closely related to fragmentation, such as spatial separation, spatial polarization, social-spatial exclusion, and disconnected cities [13]. Although fragmentation is repeatedly emphasized in urban studies, the concept is still in its infancy. This deficiency is mainly due to the term “fragmentation” being associated with several socio-spatial phenomena without a specific definition. Terms such as urban fragmentation, dual city, disconnected city, illegal and legal city, city of walls, and divided city are also used to describe the fragmentation in urban areas [6,14,15]. They explain that dualizing space and society in the metropolis increases the polarization between wealthy and poor social sectors in urban areas.

The housing market is central in reproducing social inequalities [16]. This is strongly reflected in the housing and home-ownership concentration, which is becoming reserved for the wealthy social sectors [17]. In this context, housing markets are inherently spatial, and current housing-market developments reveal substantial variation between neighborhoods; these inequalities promote spatial fragmentation. The consequences of this fragmentation result in the proliferation of a dual urban space with well-equipped business and residential areas on the one hand and areas of the city that are ill integrated within the urban structure on the other. This spatial fragmentation is a dynamic process that involves a fracture, and social separation in space, which is reflected in the emergence of closed or similar neighborhoods located transversely in the city [18]. This process is outlined by increased social homogeneity in the local neighborhoods’ social composition and “fragmented” cities where slums surround high-income neighborhoods. Taubenböck et al. [19] gave an example of this division in urban areas by estimating the digital access of informal areas of the city and poor areas by using remote sensing and Twitter data. The results from eight cities show that residents of the slums are digitally left behind compared to residents of formal settlements.

Michael Janoschka [20] emphasized a break in the Latin American cities, which have been traditionally open and marked by public spaces, giving way to extremely segregated and divided cities. Although the concept of fragmentation has similarities with segregation, we should not confuse fragmentation with segregation. By socio-spatial segregation, we mean the geographical dimensions of social inequality, polarization, and exclusion [21,22], whereas with socio-spatial fragmentation, we mean the increase of the spatial inequalities, but this does not necessarily imply an exclusion or segregation of specific social groups.

The segregation shaping the isolation of neighborhoods and the processes shaping that isolation and the reactions deriving from it are inherently and spatially represented in that configuration. Consequentially, communities such as squatter settlements are less able to physically interact with different social groups, which stigmatizes their inhabitants. This segregation highlights the differences resulting from an overlapping of different spaces and increased visibility to the differences, creating an archipelago society [23]. The concept of segregation brings us to the duality concerning the formal and informal city institutional domains. Sabatini et al. [24] identified a change in their geographical scale in Chilean cities’ residential segregation patterns. The authors identified two main scales: large-scale poverty areas and a noticeable agglomeration of high-income groups in the periphery; and small-scale poverty areas, consisting of homogeneous neighborhoods arranged alternately in urban space. The geographic scale of segregation is decreasing in the areas of largest private real estate dynamism, whereas it is increasing in areas where new low-income families are settling. In other words, the intensity of segregation decreases on an aggregated geographic scale and is intensified on a smaller scale. This duality is also represented in the formal and informal city’s institutional domains.

Squatter settlements originate in certain segments of the population appropriating the space by squatting open lands, public or private, for personal use. From that moment, the inhabitants of squatter settlements re-signify the urban territory symbolically and physically. This re-signification enters into contradiction with the rest of the city. They become perceived as blighted areas, creating a gap within the city, as Hardoy and Satterthwaite [25,26] call them “legal and illegal city”. The illegal city is associated with a constructed territory that does not follow the rest of the city’s norms. Thus, the legal city promotes the segregation of the illegal city by exclusion and stigmatization. However, these two regions enter in contact, overlapping with different norms, where decisions are taken outside the formal regulation and are tolerated by the institution as a way of coexistence.

One of the main factors that promote socio-spatial fragmentation is crime perception, which is strong and negatively associated with the neighborhood quality and is detrimental to the property values [27]. Buonanno et al. [28] performed a hedonic analysis using data from the housing market and victimization survey in the city of Barcelona from 2004 to 2006 to estimate the crime perception effect on house prices. The authors found that crime exerted hidden costs beyond its direct costs. The perceived level of security in the district had a positive impact on the district’s hedonic price. Meanwhile, crime perception was negatively correlated with it. Geoghegan, Wainger, and Bockstael [29] suggested that increasing diversity might lower property values by introducing negative visual and noise externalities. Yet, diversity may also provide convenient access to work, shopping, and recreational activities. However, many recent works enhance the role of the mixed-income developments, highlighting the affluence of middle income by ensuring equitable access to various neighborhood amenities and opportunities for assisted households to promote poverty deconcentration [30,31,32].

Increased fragmentation results in a more checkered landscape with potentially conflicting neighboring land use activities. The fragmentation is not static but dynamic and evolving. Coy [33] makes an explanation of the expansion of gated communities in Brazilian cities. The author explained that the rise of spatial fragmentation was during four phases associated with increased city violence. In that research, security and crime perception were stronger in the higher quantiles of the district hedonic price distribution. Insecurity causes a larger price reduction in more highly valued areas.

Squatter settlements are major factors promoting socio-spatial fragmentation in the developing metropolitan areas. These types of settlements have a nuisance effect, which futher promotes this phenomenon. Hussain et al. [9] explored the negative impact of slum proximity on property prices and rentals and found that rent declined as one moved closer to the slums. Song and Zenou [34] investigated whether the proximity to urban villages in China negatively affected house prices and found that they increased as the nearest urban village’s distance increased. Chen and Jim [35] analyze the urban village influence on house price using the three-dimensional model (availability, accessibility, and visibility). Their results indicated that the nuisance effect differed depending on the dimension type, having a more substantial negative effect through visibility than through the availability of urban villages. Zhang and Zhao [36] showed the effect of the informal housing market in the urban villages in China, where the variances in ability between villages lead to heterogeneity of tenure security, thus creating price differentials in the market. In this context, higher tenure security means an increase in house prices.

Henderson, Regan, and Venables [37] developed a model of the city’s built environment growing in Nairobi by distinguishing between the formal and slum construction. The formal sector is more dynamic and involves investment decisions based on expected future rents. In contrast, in case of the informal construction, the buildings are built low, with high-land intensity. The slums volume increases through time, not by building taller houses, but by increasing crowding and already high cover-to-area ratios. In this context, the model’s dynamic predicts that as house prices rise, there should be the ongoing conversion of older slums to formal sector use; meanwhile, new slum areas will be expanded in the city’s edge.

Many empirical studies have theoretically suggested the effect of certain facilities in house price depreciation by conducting a spatiotemporal analysis [38,39,40]; however, there is a lack of studies that examine the spatiotemporal effect of squatter settlements on house prices and how to promote or discourage the spatial fragmentation in cities.

In this research, we consider two types of the spatial division process. Spatial segregation is the territorial intra-urban structure defined between the squatter settlements and the rest of the city. Interactions with different neighborhoods do not characterize this segregation. In contrast, socio-spatial fragmentation is the spatial intra-urban structure marked by spatial inequalities, although fragmentation is defined as an urban social fracture, still retaining the interaction between neighborhoods.

While theoretical debates over urban structures have existed in the literature review, no empirical studies have tested the dynamic role of squatter settlements in spatial fragmentation, to the best of our knowledge. Here, we examine the case of the CBA in the period 2001–2018 because the house price disparities increased tremendously during that period, with the rapid growth of squatter settlements. Our hypothesis is that the growth of squatter settlements increases the house price disparities. This research aimed to give a comprehensive description of the CBA’s socio-spatial fragmentation from 2001 to 2018 that associates house prices with the squatter settlements’ location.

3. Study Area

This paper considers the case of CBA, with a population of 2.89 million, which makes it one of South America’s biggest cities. The city itself is divided into 15 comunas (communes) (Figure 1). The CBA is characterized by de-industrialization and transformation into an administrative and business services center. This change in the economic organization has profoundly restructured the city’s labor market and urbanism projects that transform the city, changing the CBA’s social composition.

Squatter settlements started emerging in CBA during the 1930s. Most of the population then were internal immigrants, stigmatized for their rural background [41]. Those informal settlements were tolerated until the military regime, which came into power in 1976, initialized city’s socio-economic restructuring. This process includes the relocation of productive activities and squatter settlements outside the CBA [42]. This “cleansing” of the city forcedly evicted an estimated 208,783 squatter settlement dwellers [43]. In 1983, with the return of democracy, squatter settlements’ population resumed growing (

Table 1. Population in the squatter settlements in CBA, 1960–2010. Data from INDEC [44,45], DGEyC [46], and TECHO [47].

Year	Squatter Settlement (Persons)	Growth Rate (%)	CBA (Persons)	Share of Squatter Settlement Population to the Total Population in CBA (%)
1960	34,430	-	2,966,634	1.2
1970	101,000	193.3	2,972,453	3.4
1980	37,010	−66.2	2,922,829	0.9
1991	52,608	42.4	2,965,403	1.8
2001	107,422	104.1	2,776,138	3.9
2010	170,054	58.3	2,890,151	5.9
2018	292,732 *	72.1	3,068,043 *	9.5

* Estimate values.

).

Meanwhile, the policies implemented during the 1990s aggravated almost all socio-economic indices, such as income disparity, real incomes, unemployment, and underemployment [48]. During this period, the squatter settlement population increased by more than 104%, having a significant inflow of international migration from bordering countries. In 2001, Argentina experienced an intense economic and political crisis, which brought a total collapse of the economic and political system. Since 2001, the squatter settlement population has increased by approximately 10,000 inhabitants every year. In this crisis, the decrease in the banking system’s stability and the loss of trust in the monetary market opened the gates to real estate speculation.

After a fall in house prices in 2001, the housing market experienced a rapid increase in house prices due to the asset boom [49] and widening house price disparities within the city. From 2003, the country’s economy began to recover at an annual growth rate of 8% [45,47]. Along with economic growth, there was a boom in the real estate sector that sharpened prices (see

Table 2. Average land price in the CBA, 2001–2018. Data from DGEyC [48].

Year	U$S/M2	Annual Change Rate %	Index to the Price 2001 %
2001	891	-	100.0
2002	505	−43.3	−43.3
2003	602	19.2	−32.4
2004	813	35	−8.8
2005	915	12.5	2.7
2006	1117.0	22.1	25.4
2007	1300.0	16.4	45.9
2008	1599.1	23	79.5
2009	1692.4	5.8	89.9
2010	1783.8	5.4	100.2
2011	2168.2	21.5	143.3
2012	2322.4	7.1	160.7
2013	2213.5	−4.7	148.4
2014	2320.5	4.8	160.4
2015	2234.5	−3.7	150.8
2016	2487.7	11.5	179.2
2017	2795.3	12.3	213.7
2018	2560.7	−8.3	187.4

), impeding low-income sectors from accessing the formal housing market. In this context, the real estate market became attractive with potentially high rent, low construction cost, and low interest rates [50]. The investment has been concentrated in the CBD, and extended to the north and historic city center in the southeast. These new projects and revitalized neighborhoods in the often-degraded city centers also increased the land value. Urban redevelopment projects such as Puerto Madero built over the abandoned port area near the CBD are an example. In addition, blighted areas such as Barrio de la Boca and San Telmo were re-evaluated through urban redevelopment projects based on their historical heritage and centrality. This investment in the real estate market has increased rapidly, especially in neighborhoods in the city’s northern area where the real estate market is dynamic. These neighborhoods, occupied predominantly by high- and middle-income inhabitants, cover the area between two axes that advance from the center to the northwestern and northwestern areas of the city.

In contrast, the residential areas located outside this zone, particularly the southern area, are dominated by housing affordable to low- and middle-income residents. The southern area is historically associated with manufacturing, and accommodates a large proportion of squatter settlements. The decline of industrial activities and the worsening housing conditions compared to the rest of the city are the leading cause for the lack of investment, unlike the northern areas of the city, which are historically related to residential and commercial activities. Looking at the distribution of squatter settlements in CBA, it is evident that they are concentrated in the southern area. These urban environments describe a spatial pattern that presents a marked differentiation between northern and southern urban regions in land prices and income distribution (Figure 2).

The rising housing costs due to real estate speculation and the lack of housing policy affects the most vulnerable population in the city. The share of immigrants, mainly from neighboring countries, in squatter settlements have been increasing. Based on the national census [44], the share of immigrants was 22.0% in 1991, 40.9% in 2001, and 47.7% in the last census year, 2010. It is expected that the share will continue to rise toward the next census.

The growth of squatter settlements exemplify a process of spatial segregation. Simultaneously, these settlements may explain the increase of socio-spatial segregation between the northern and southern areas in the city in the last 20 years. This process is associated with an increase in the city segregation levels, thus establishing a social and spatial distance between one area of the city and the rest [51].

4. Method and Data

4.1. Hedonic Prices

The negative externalities emanating from nuisance facilities are typically measured using the hedonic price model. This model assumes that property comprises a bundle of individual components where each one has an implicit rate. The components are the inherent features of a property ( e.g., an area, age of the building, the number of rooms) and extrinsic aspects of a property (e.g., neighborhood characteristics, proximity to the CBD, stations, schools, nuisance facilities) [52]. A vast body of literature explores house prices’ determination using this model, highlighting the positive, negative, or both externalities [53,54,55]. However, the traditional hedonic price analysis has used global regression models, the conventional regression assuming the same coefficients everywhere in a study area, which do not consider spatial dependency and heterogeneity [56]. This traditional analysis assumes that the effect of explanatory variables is uniform across space.

4.2. Geographical Weighted Regression

We consider using a local regression model GWR to demonstrate the socio-spatial fragmentation. This model allows regression coefficients to vary over space [57]. This technique can identify spatial heterogeneity through geographically localized regression outputs, typically displayed on a map of estimated local coefficients [58]. Recent works on price determination applied GWR and have shown that it is necessary to consider the spatial heterogeneity [59,60]. However, time is also an essential dimension related to the housing market. A temporal analysis can provide valuable information concerning the consolidation of urban fragmentation in the city. Several spatiotemporal models have been developed to incorporate the spatial and temporal variation into hedonic house price analysis [61,62,63]. Bo Huang et al. [64] integrated temporal effects in the GWR model. Comparing geographically and temporally weighted regression (GTWR) outperforms both GWR and time-weighted regression (TWR) in the sample data’s model accuracy. The results showed that there were substantial benefits in modeling both spatial and temporal non-stationarity simultaneously. Yao and Fotheringham [65] applied GWR to investigate both global and local relationships between house prices and associated influencing factors and their variations over time. Using 10 years of house prices in Fife, Scotland, the study found that the relationships vary spatially and that spatially varied relationships changed over time.

For this study, as we could only retrieve the house price associated variables at several time slices, we conduct an independent GWR analysis for each time slice. Following the GWR model formulation, a house price model can be expressed as follows:

$$\log y_i={\beta }_{io} \left(u_i v_i\right)+{{\displaystyle \sum }}_j{\beta }_{ij}\left(u_i v_i\right)x_{ij}+{\epsilon }_i,$$

(1)

where $y_i$ is the price of the ith house; the intercept term ${\beta }_{io} \left(u_i v_i\right)$ and the coefficients ${\beta }_{ij}$ of the jth explanatory variables $x_{ij}$ are expected to be spatially varying, dependent on geographical coordinates $u_{i }$ and $v_i$ of each individual property. ${\epsilon }_i$ is the i.i.d. Gaussian error term for the ith observation.

In the global regression, the intercept and the coefficients are constants, that is, ${\beta }_{io} \left(u_i v_i\right)={\beta }_o$ and ${\beta }_{ij}\left(u_i v_i\right)={\beta }_j$, but with the GWR, each observation has different intercept and coefficients capturing the spatial dependency in the error term and the spatial heterogeneity in the effects of the explanatory variables, respectively. To estimate the varying parameters, the GWR model assumes that the parameters vary smoothly over space and applies locally weighted regression to estimate every single observation parameter in the sequence. Locally weighted regression estimates parameters with a geographically weighted kernel centered on each regression point. The weighting kernel assigns smaller weights to the observations away from the regression point following Tobler’s first law of geography [66]. Common weighting kernel functions are Gaussian and bi-square weights [58]. We chose the Gaussian kernel weighted function in our analysis because it generally produced a better fit. In the Gaussian weighting function, the weight of observation $j$ for a regression point $i$ $w_j\left(u_i,v_i\right)$ is represented by the following formula:

$$w_j\left(u_i,v_i\right)=\exp \left[-\frac{1}{2} {\left(\frac{d_{ij}}{h}\right)}^2\right]$$

(2)

where bandwidth $h$ controls the size of the weighting neighborhood to estimate the local coefficients, and $d_{ij}$ is the distance between the regression point and the observation.

The bandwidth of a kernel can either be fixed in terms of distance or adaptive to include the same number of observations within the bandwidth. In this research, we use fixed bandwidth to produce estimates comparable between periods. Initially, we optimize the bandwidth using the golden section search to minimize the second-order Akaike Information Criteria (AICc) for each time period. Then we choose a single bandwidth within the range of optimized bandwidth for direct comparison of the results over time. The bandwidth search, variable selection, and parameter estimation are conducted using GWR4 [67].

The GWR model provides improved explanatory power to explore the spatial differences in the accessibility to squatter settlements. We calculate the spatial fragmentation using cross-sectional data for 2001, 2010, and 2018 in the CBA. An increase in the accessibility to squatter settlements over time implies a process involving the breakup of contiguous neighborhood types and an increase of spatial fragmentation.

4.3. Data: Autonomous City of Buenos Aires Time Series Real Estate Price and Hedonic Data

Real estate data for September of 2001, 2010, and 2018 were collected by web scraping methods using the Python script from the largest real estate portals in Argentina ‘Mercado Libre’ and ‘Zonaprop.’ We used the data for 3-room apartments because they are well distributed across the city and are suitable for families. Furthermore, families are more likely to have more substantial concerns about the neighborhood quality. The sample sizes are summarized in

Table 3. Sample size and house prices in the CBA. Data from DGEyC [48] and compiled by the author of data from ‘Mercadolibre’ and ‘Zonaprop.’

Year	Number of Observations	House Price Average (US$)	Average Salary (US$)
2001	3654	52,945.97	1504.23
2010	1887	145,252.27	1250.50
2018	4334	234,511.34	1079.05

, along with the average house price and the average salary from the official data of the government of the CBA. The number of houses in the market in 2001 is similar to that in 2018. However, in 2010, the number of houses in the market declined substantially during an economic expansion, without having relation to the purchasing power.

The definition of the explanatory variables is summarized in

Table 4. Definitions of the explanatory variables.

Variable Names	Description	Units
Intrinsic Characteristics
Area_NL	The total area of the property	M2 In a natural log
Multi_Bathrooms	0 = 1 bathroom; 1 = more than 1 bathroom	Dummy
Parking	0 = none; 1 = one or more	Dummy
Extrinsic Characteristics
NL_CBD	Distance to the CBD	Meters in natural log
NL_Subway	Distance to the closest train station	Meters in natural log
NL_Police	Distance to the closest subway station	Meters in natural log
NL_Green	Distance to the closest green area	Meters in natural log
NL_Squatter	Distance to the closest squatter settlement	Meters in natural log
ACC_Squatter_A	Area-weighted accessibility to squatter settlements	See the text
ACC_Squatter_Pop	Population-weighted accessibility to squatter settlements	See the text

. Explanatory variables are divided into two groups. On the one hand, intrinsic characteristics such as the floor area and the existence of parking and of multiple bathrooms are included. The age of the properties was not available for 2001, and thus is not included in the analysis to allow for direct comparison. On the other hand, extrinsic characteristics consist of the distances from each house sample to the following facilities: CBD, police station, subway station, green areas (>1 hectare), and the proximity/separation to squatter settlements. The effects of smaller business centers would be captured by spatially varying intercepts as we have relatively dense data points for house prices.

The variables of intrinsic characteristics were included in the real estate data. The variables of the extrinsic characteristics were assigned to each property based on the geographic coordinates of the properties using ArcGIS 10.8 [68]. The Euclidean distance to the closest of these services in meters in the natural logarithm was used. The locations of subway stations, train stations, police stations, CBD, and squatter settlements in 2001 and 2010 were obtained from the General Directorate of Statistics, Surveys, and Censuses of CBA [69], and for the squatter settlements in the year 2018 from the nonprofit organization TECHO [47].

The proximity/separation to squatter settlements is the focus of our inquiry to measure squatter settlements’ effects on house prices. In addition to the log distance to the closest squatter settlement, we prepared two alternative measures using gravity-based accessibility [67] defined as follows:

$$A_i={\displaystyle \sum }{w}_j . d_{ij}{}^{-2},$$

(3)

where A_i is the accessibility to squatter settlements of the ith property,$d_{ij}$ is the distance (in meters) from the ith property to the jth squatter settlement, and $w_j$ is the weight that can be per area or population of a squatter settlement.

These alternative measurements consider the sizes of squatter settlements and the effects of multiple squatter settlements that would simultaneously affect house prices.

5. Results

The dependent variable is natural logarithm of the house price. Models 1, 2, and 3 use all the variables listed in Table 3, except for the three alternative measures of squatter settlement separation/proximity. In Model 1, the log distance to the closest squatter settlement is used. In Model 2, area-weighted accessibility to squatter settlements is used. Model 3 includes population-weighted accessibility. For each model, there are versions that use the generalized linear model (GLM), estimating the global (constant) parameters, and the GWR, estimating the local spatially varying parameters for each of the three time periods. The preliminary GWR analysis for each time slice in each model resulted in an optimized bandwidth ranging between 391 and 8808 m. We adopted the bandwidth of 1500 m for all GWR estimates for a direct comparison of the results. The model performances are summarized in

Table 5. Model summary.

	Model 1			Model 2			Model 3
	GLM (global)
	2001	2010	2018	2001	2010	2018	2001	2010	2018
N	3654	1887	4334	3654	1887	4334	3654	1887	4334
AIC	−1111.0	806.5	1468.2	−1164.3	859.9	1447.0	−1150.5	861.9	1429.3
AICc	−1111.0	806.8	1468.3	−1164.2	860.2	1447.1	−1153.4	862.2	1429.4
CV	0.042	0.099	0.080	0.043	0.098	1012.870	0.043	1.998	0.311
R-squared	0.224	0.664	0.609	0.224	0.675	0.610	0.043	0.672	0.609
Adj. R-squared	0.222	0.661	0.608	0.223	0.672	0.609	0.224	0.621	0.608
	GWR (local)
	2001	2010	2018	2001	2010	2018	2001	2010	2018
N	3654	1887	4334	3654	1887	4334	3654	1887	4334
AIC	−2106.9	−248.8	−1473.4	−2150.6	−257.5	−1492.6	−2009.6	−255.7	−1130.3
AICc	−2103.47	−231.5	−1462.3	−2147.2	−231.4	−1482.4	−2095.9	−229.0	−1180.0
CV	0.033	0.084	0.040	0.033	0.083	64.399	0.033	0.232	0.140
R-squared	0.428	0.840	0.812	0.429	0.844	0.814	0.406	0.826	0.812
Adj. R-squared	0.412	0.823	0.803	0.412	0.825	0.805	0.406	0.756	0.802
Bandwidth (Fixed)	1500 m

.

Comparing the GLM and the GWR versions of the model, the GWR versions outperformed the GLM versions in terms of AICc in all cases, suggesting that spatial heterogeneities exist in house price determination. When comparing the three models, the GWR version of Model 2 has the smallest AICc. During 2001–2018, squatter settlements have grown by 173% in terms of its population but have expanded only by 8% in terms of area. The model using area-weighted accessibility to squatter settlements (Model 2) outperforming population-weighted accessibility (Model 3) might be suggesting that squatter settlement growth is not causing the house price disparities. Nevertheless, we adopt Model 2 that uses area-weighted accessibility to squatter settlements and examine its coefficients.

The coefficient estimates of the GLM version of Model 2 have expected signs and are statistically significant at 5%, except for a few variables (

Table 6. Coefficient estimates of the global Model 2.

	2001		2010		2018
Variable	Estimate	T-Value	Estimate	T-Value	Estimate	T-Value
Intercept	8.224 **	60.114	8.789 **	74.533	10.061 **	100.542
Intrinsic Characteristics
Area_NL	0.689 **	21.855	0.739 **	40.068	0.696 **	45.423
Multi_Bathroom	0.139 **	9.826	0.214 **	14.877	0.252 **	25.949
Parking	0.211 **	15.310	0.264 **	14.414	0.263 **	18.292
Extrinsic Characteristics
NL_CBD	0.013 **	3.800	−0.034 **	−6.379	−0.051 **	−14.869
NL_Subway	−0.024 **	−6.534	−0.042 **	−5.208	−0.013 **	−2.660
NL_Green	−0.014 **	−4.067	0.001	0.102	−0.013 **	−2.707
NL_Police	−0.014 *	−2.434	−0.013	−1.116	0.003	0.421
ACC_Squatter_A	−0.008 *	−2.004	−0.066 **	−4.709	−0.003 **	−3.252

Notes: The dependent variable is house prices in the natural logarithm. Model summary is given in Table 5. *, ** Statistical significance at 5% and 1%, respectively.

). The house prices rise with floor area, but less than proportionately. Multiple bathrooms and parking space raise prices. House prices decline as distances to the CBD increase although not in 2001, possibly because of the financial crisis in 2001. The distances to other facilities generally lowered house prices, but the distance to the police station was statistically significant at 5% only in 2001, and the distance to green spaces was not statistically significant in 2010. The accessibility to a squatter settlement was negative for all three periods, which means a decrease in the house price with proximity to squatter settlements. Comparatively, the effect of squatter settlements was much greater in 2010 than in other years, and it was particularly small in 2018.

The coefficients for GWR vary over space; therefore the median, minimum, and maximum for each period are shown in

Table 7. Coefficient estimates of the GWR Model 2.

Variables	2001				2010				2018
	Median	Min	Max	Diff	Median	Min	Max	Diff	Median	Min	Max	Diff
Intercept	8.166	4.663	30.070	−3014.3	9.443	−8.102	28.08	−1339.9	9.963	−6.857	25.022	−6711.3
Intrinsic Characteristics
Area_NL	0.706	−1.414	0.863	−7848.7	0.693	0.470	1.042	−62.8	0.711	0.326	0.973	−2122.7
Multi_Bathroom	0.184	−0.004	0.261	3.7	0.217	0.023	0.280	7.3	0.194	−0.102	0.347	−35.6
Parking	0.124	−0.150	0.205	2.1	0.126	−0.698	0.224	−6.3	0.185	0.114	0.332	−37.9
Extrinsic Characteristics
NL_CBD	−0.001	−1.530	0.504	−524.5	−0.106	−2.813	2.761	−32.4	−0.092	−1.874	2.137	−2295.9
NL_Subway	−0.015	−0.266	0.245	−1901.0	0.016	−1.123	0.914	−1116.6	0.036	−0.336	0.412	−266.2
NL_Green	−0.008	−0.070	0.035	−141.7	0.002	−0.067	0.037	−13.6	−0.023	−0.073	0.078	−62.3
NL_Police	−0.018	−0.159	0.076	−942.8	−0.026	−0.139	0.106	−301.2	0.009	−0.343	0.086	−1430.5
ACC_Squatter_A	−0.010	−1.507	0.222	−43.6	−0.026	−2.710	0.030	−19.6	0.000	−15.768	0.080	−21.7

Notes: The dependent variable is house prices in the natural logarithm. Model summary is given in Table 5.

. The coefficient estimates will be shown later in the form of maps. Most parameter estimates range between positive and negative values, depending on the location; the accessibility to a squatter settlement is no exception. The table includes geographical variability test values to explore whether coefficients vary over space [70]. The Diff is a geographical variability test statistic that tests the change in AICc from the model that fixes the coefficient of a variable to a constant one by one. The coefficient is considered spatially varying if the test value is <−2, while values <−4 substantially improve the model [67]. The test statistic showed that all variables are spatially heterogeneous, except for multiple bathrooms in 2001 and 2010, and parking in 2001.

Before examining the squatter settlements’ influence on the housing market, it is necessary to confirm the widening house price disparities controlling for the houses’ structural characteristics. The controlled house price was estimated using Equation (4), which represents the house price of a house with an average floor area, one bathroom, and no parking, based on the GWR version of Model 2:

$${\widehat{y}}_i^{\prime }={\widehat{y}}_i-{\beta }_i\left(AreaNL\right)+{\beta }_i \left(\overline{AreaNL}\right),$$

(4)

Figure 3 shows an increase in house prices affecting the whole city, but with different intensities. The areas with higher house prices are clustered in the city’s northeast axis, and they expand there more than in the rest of the city. Meanwhile, the house prices in the southern areas are stable with a moderate or no increase in prices. Concerning the west–east axis, the house prices are more dispersed. However, the limitation of this analysis is that the house age is not controlled because there was no such variable for 2001.

Figure 4 shows changes in controlled average house prices by communes. This figure thus shows a particular positive polarization of the values, located near the CBD, and a negative one, situated in the city’s southern area. The communes in the south with multiple squatter settlements have the lowest increase in house values compared to the rest of the city.

Conversely, the coefficient values associated with the squatter settlements’ accessibility varied positively and negatively. Figure 5 shows the spatial variation of the local model’s coefficients and their t-values. Coefficients are considered to be statistically insignificant if the t-value falls between 2 and −2.

The coefficients of accessibility to squatter settlements exhibited considerably dif-ferent spatial heterogeneities across periods. In the first period, in 2001 (Figure 5a), the negative value is concentrated in the city’s southern area with small spots in the north. However, estimated values for 2010 and 2018 (Figure 5a,b) show a gradual transition and clustering of the city’s northern area’s negative effect. This change shows a transition of the negative effect of squatter settlements moving from an area with a high agglomeration of squatter settlements toward the north, where there are few squatter settlements. There were also locations having positive coefficients, which indicates an increase in the house price with the proximity to squatter settlements.

The interpretation of the squatter settlements’ effect is strongly related to the spatial extent of neighborhood effects on housing prices. These results might indicate three different housing regimens in the city, explaining the CBA’s spatial fragmentation, especially between the northern, southern, and central areas.

6. Discussion and Conclusions

We confirmed that the proximity to a squatter settlement lowered the house price, but locally in the areas surrounding the squatter settlements clustering in the south region in 2001 (Figure 5a). However, the geographical pattern changed substantially over the period. The negative effect significantly diminished in 2018 in the southern region, and it increased in the northern region (Figure 5c).

These observations refute our initial hypothesis that the squatter settlement growth resulted in greater house price disparities between the city’s north and the south. Had the squatter settlement growth been the cause, the areas affected by squatter settlements would have remained the same. The negative effect of squatter settlements appearing in the high house price areas in the north implies that squatter settlements have instead narrowed the house price disparity between the north and the south.

We did confirm with the controlled house price that house prices were higher in the north than in the south (Figure 3), and the disparity has grown over the study period (Figure 4). Yet, the areas affected by squatter settlements have changed in a way that narrows the house price disparities. We attempt to interpret this seemingly contradicting observation through logically understanding what we have seen so far.

The negative effect of squatter settlements on house prices in the southern area may have disappeared because of the high demand for low-cost housing induced by widened income disparity (Figure 2a,b). The asset boom rapidly increased property prices in the northern area, but the proximity to the squatter settlement agglomeration in the south kept house prices there from rising. The southern region became almost the only housing alternative for the low-income households within the city in the face of increased housing prices. The concentration of low-income households in the south could have lowered the house price there. At the same time, poverty forced households to trade off the higher perceived safety to access to the labor market, and therefore, the negative effect of squatter settlements disappeared in the south.

In contrast, the negative effect has appeared in the northern area since 2010 (Figure 5b), far away from the bulk of squatter settlements. Possibly, the negative coefficients appearing in the north might be associated with increased perception of insecurity pertaining to the squatter settlements in the south due to widened disparity between the squatter settlement dwellers and the citizens in the north. Alternatively, we suspect that the perceived nuisance of high-income citizens in the north may not be particularly arising from squatter settlements alone but also due to the whole low-income south region.

The explanation of the positive coefficients for the accessibility to squatter settlements can be explained from the speculation of specific squatter settlements to be redeveloped. In 2001, the positive estimates appeared around the squatter settlements near the CBD and one of the city’s central transport hub, Retiro Railway Terminus (Figure 5a). A location near the CBD and the transport hub is a likely target for gentrification. In 2008, the positive estimates value shifts to another location (Figure 5b), where urban policies promoted a technological district in the city from 2008 [71]. By 2018, the technological district’s prices waned, and the positive estimates shifted back to the locations to the north of the CBD (Figure 5c). Urban renewal policies during this period in CBA focused on promoting a polycentric model by converting run-down neighborhoods and slums into subcenters. Accordingly, residential areas adjacent to the squatter settlements that have potential to be redeveloped into subcenters are attractive to private real estate investors, neutralizing the squatter settlements’ nuisance effect.

We conclude that the squatter settlement growth is not the cause of spatial fragmentation in the housing market and the widening house price disparities. Widened income disparity segregated low-income households to the low-housing-cost area, which reinforced the existing house price disparities. Therefore, it is the further segregation of low-income households to the south that is causing the widened house price disparities between the south and the high-income north. In the case of CBA, the squatter settlement growth is the symptom of widening income disparity, which might be applicable to other cities experiencing squatter settlement growth and widening income disparity. Squatter settlements had some role in the process, however, as they have determined where the higher concentration of low-income households have occurred.

In this context, squatter settlements are at the core of low-housing areas, implying the spatial fragmentation between regions in the city. This fragmentation is a recurrent and frequent phenomenon in developing countries’ cities, resulting in an urban structure and conditions of exclusionary housing access for certain social groups. The consequence is the gradual loss of the classical city’s heterogeneity that has enabled interaction between different social groups. An alternative to this fragmentation and the price depreciation in areas near squatter settlements is promoting an approach based on mixed-income housing developments. Such a program can achieve social integration, reduce uneven housing and infrastructure investments, provide affordable housing, and reduce environmental inequalities, including the exposure to communicative disease such as COVID-19.

The major limitation to this research is that a GWR analysis using proximities to geographical features cannot distinguish the effects of spatially coincident features, in our case, between the effects of squatter settlements and the low-income areas that surround them. It is therefore inconclusive as to whether the effect that has appeared since 2010 as the negative effect of squatter settlements on the affluent north, at some distance away from the bulk of the squatter settlements in the south, is the effect of squatter settlements alone. It is also possible that all the low-income areas that surround the squatter settlements in the south are lowering the house prices in the north. Approaches incorporating complementary research, such as the contingent valuation method, may be useful.