Initial signs of post-covid-19 physical structures of cities in Israel

Abstract

The physical structure of cities is the result of self-organization processes in which profit-maximizing developers are key players. The recent Covid-19 pandemic was a natural experiment by means of which it is possible to gain insights into shifts in the spatial structure of cities by studying developers’ behavior. Behavioral changes of urbanites triggered by the quarantine and lockdown periods, such as home-based work and online shopping on scales that were unthinkable heretofore, are expected to persist. These are likely to induce changes in the demand for housing, for work, and for retail space, impacting developers’ decisions. Associated changes in the land values at different locations are occurring faster than changes of the physical shape of urban landscapes. It is possible that current changes in dwelling preferences will result in significant future shifts in the locational incidence of the urban intensities. We test this hypothesis by examining changes in land values during the last two years by means of a land value model calibrated with vast Geo-referenced data of the major metropolitan area in Israel. Data concerning all real estate transactions include information about the assets and the price of the exchanges. In parallel, built densities are calculated using detailed building data. Based on these data, we estimate the changes of land values for different types of dwellings before and during the pandemic. The result allows us to highlight possible initial signs of post-Covid-19 urban structures, driven by shifting behavior of developers.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12076-023-00346-8.

Introduction

Cities are three-dimensional objects in a constant state of flux, with poorly defined and fuzzy boundaries (Barthelemy 2016). Their physical structures are the result of self-organization processes in which profit-maximizing developers are key players (Broitman and Czamanski 2012, 2015; Czamanski and Broitman 2017). This conception of urban dynamics makes possible identification of the location of future development pressures by the analysis of the evolution of land values (Buda et al. 2021, 2022). The approach is useful for forecasting trends in the evolution of urban spatial structures, including in times of sudden and unexpected changes.

The Covid-19 pandemic that spread worldwide during the last three years is such an unexpected event. It represents a great natural experiment that provides a basis for the analysis of possible future spatial structure of cities (Florida 2020; Kang et al. 2020, 2021). Even if the pandemic is controlled, and ultimately eliminated, during the coming months, there were several behavioral changes triggered during the quarantine and lockdown periods that are expected to persist. The pandemic forced large groups of people to experience home-based work and online shopping on scales that were unthinkable heretofore. These behavioral changes imply possible changes in the demand for office space (Martínez and Short 2021), housing (Mouratidis 2021), and retail space (Nanda et al. 2021a, b). Following the new work and shopping patterns, we should expect changes in rents and in the demand for income-generating real-estate at various locations. The demand for housing, especially for high-rise apartments and for single family housing units is expected to experience modifications as a result of changes in access to recreational facilities, amenities, and services (Eltarabily and Elghezanwy 2020). While reliable and sufficiently extended data concerning these trends are still unavailable, in our opinion, spatial changes will start to be evident only during the coming years.

However, changes in the willingness to pay for locations are already evident. One plausible hypothesis is that dwelling preferences today, will be followed by significant shifts in urban structures in the future. The assumption is that, because of the experience of the last years, some households will prefer less crowded spaces, even if this implies living in peripheral locations. Other households may still prefer dense urban centers, but different types of dwellings that are more easily adaptable to the changing working, shopping, and leisure conditions (Glaeser et al. 2021; Celbiş et al. 2022; Östh et al. 2023). Combined with the expected decline of demand for large office or retail buildings, one possible consequence will be that future urban developments will be less dense and more spread out. All these trends are likely to impact the future preferences of developers for locations. Since this is the main driver of the urban physical growth, changes in developers’ choices could lead to urban development patterns that are different from those observed in pre-pandemic times (Buda et al. 2022). We test this working hypothesis by means of changes in land values during the last two years.

We present results of a land value model calibrated with vast Geo-referenced data of the major metropolitan area in Israel - data concerning all real estate transactions includes information about the assets, the price of the exchanges, and their location. Urban built densities are assessed using detailed data, that include footprint and height of all the built urban structures. We estimate the changes of land values for different types of dwellings before and during the pandemics. The choice of land values, instead of housing values, is anchored in the concept of developers as main agents of urban changes: This is one of the most important parameters of a real-estate investment (Broitman and Czamanski 2012, 2015). The result allows us to highlight possible initial signs of post-Covid-19 urban structures. The rest of this paper consists of 5 sections. Section 2 includes a literature review. In Sect. 3 we present the data used in the analysis. The methodology is described in Sect. 4. In Sect. 5 we present the results of our analysis. Some conclusions are presented in Sect. 6.

Literature review

The pandemic outbreak triggered changes in the behavior and mobility of people in cities (Morita et al. 2020; Przybylowski et al. 2021). While some of these changes may not necessarily affect the spatial structure and morphology of urban areas, other changes in preference shifts are likely to affect, at least to a certain extent, the physical structure of cities. During the first year of the pandemics, there were predictions about imminent massive migration away from urban cores (Nathan and Overman 2020; Gallent 2020). Although these early forecasts seem to have been exaggerated, there is some evidence of ongoing, albeit humbler, modifications in dwelling preferences of certain social groups. For example, changes in working and commuting patterns are catalysts of a shift from dense city centers to the suburbs in US cities (Ramani & Bloom 2021), having the potential to hollow out dense city cores (Toger et al. 2021).

Some argue that contradictory processes can happen in parallel, as pulling force from the suburbs, perhaps relevant for certain age groups, with back-to-urban-centers trends among other, younger populations (Florida 2020). A metropolitan level view of urban real estate prices indicates that, at least during the pandemic, real estate prices declined in urban centers and increased in the suburbs (Gupta et al. 2021). This support arguments regarding an expected flattening bid–rent curve in case of a widespread urban land use reshaping during the pandemic’s aftermath (Nanda et al. 2021a, b). Local services and transportation may be deeply influenced in the future due to these ongoing trends (Nathan 2021). Other analyses indicate that preferences for residences in low population density areas with outdoor facilities seems to be on the rise (Guglielminetti et al. 2021). These observations are in line with increasing preferences for locations away from dense urban centers (Ferreira and Wong 2022). In particular, there is evidence of the willingness to pay premium prices for locations adjacent to open spaces and beaches, and also a drift toward places further away from the city center, compared with pre-Covid-19 observations (Cheung and Fernandez 2021).

It is perhaps too early to discern whether these changes are short-lived and reversible, or long lasting (Florida et al. 2021). Others argue that in the long run the agglomeration forces that had shaped cities since their beginning will ultimately prevail (Reades and Crookston 2021). In any case, crowd-avoiding behaviors, the possibilities of teleworking and the search for nearby amenities, seem to have impacted on the locational choices of certain population segments (Florida et al. 2021). Thoughts and reconsiderations about the most appropriate residential area, seem to have been widespread during the successive Covid-19 waves (Kang et al. 2021). Besides these first and limited empirical case studies, theoretical urban growth models predict significant changes in the future spatial structure if the impact of the pandemic results to be long lasting (Buda et al. 2022). In a recent paper, there is evidence about adjustments of residential and commercial prices in urban areas during the pandemic and since its aftermaths. An extrapolation of the observed changes can lead to flattening urban gradients (Duranton and Handbury 2023).

Summarizing, the combination of the relatively short time elapsed since urban areas returned to a seemingly normal functioning in the aftermath of Covid-19, with the variated and sometimes contradictory results obtained by different studies in a wide range of locations, emphasizes the need of more test case studies. The goal is to figure out, first, if the pandemic is likely to leave a visible imprint in the urban structure. If the answer is positive, the following question would be what that imprint will look like. Moreover, there is increasing evidence of preferences shifts of real-estate developers (Uchehara et al. 2020; Balemi et al. 2021). Therefore, it is important to speculate about the future behavior of developers and its possible impact on the spatial and physical configuration of the city. This study aims to contribute to this literature, using the test case of the urban structure of Israel.

Data

Two different datasets were used to perform this analysis. The first one includes all real estate transactions performed in Israel from 2007 to 2021, as recorded by the Israeli Tax Authority. Each observation in the dataset represents a single transaction and its characteristics, such as the property price (in New Israeli Shekel), the dwelling size, number of rooms, the property age, the floor (if the property was not a detached house), etc. Each transaction also includes geographical characteristics that allow us to locate them on a map. After cleanup of outlying property price values and records with missing data, the resulting dataset included 1,338,455 real-estate transactions.

The second dataset is a polygon layer of the Israeli built structures and their main characteristics, such as footprint, height, and precise geographical location. This dataset is available from the Survey of Israel, the Geoinformatics governmental agency. Since we focus only on urban buildings (whether residential, as houses and apartment buildings, or industrial facilities, offices, and warehouses) we excluded other structures (as greenhouses or temporary structures). The resulting layer comprises 1,496,698 buildings.

Methodology

The existing built area density is calculated using the polygon layer of built structures. We transform each building polygon into a point, assigning it its built volume (footprint by height, in cubic meters). Then we create a continuous kernel density (KD) surface of the built environment, using a kernel bandwidth of 500 m and pixels of 50 × 50 m size. Finally, we normalize the KD to a range between 0 and 100. The resulting KD reflects the actual built densities in each location of the Israeli urban areas. In other words, it reflects where there is barely any room for further urban development (very dense locations in which further construction is difficult), compared with less dense places where future densification is physically easier.

The economic analysis of the real-estate transactions is based on the land value (LV), that is basically the difference between the market price of the property and its construction costs (Buda et al. 2022). The construction costs for each type of residence in Israel is compiled once in several years and was purchased from the surveyors (Dekel 2020). The property price was divided by the property size (in square meters), subtracting the construction price (also, per square meter). The result is the LV (NIS per square meter).

Then, we split the whole LV data set into two mutually exclusive subsets: Pre and Covid periods’ transactions (corresponding to 2007–2019 and 2020–2021 years, respectively). We fit each subset to a linear regression model based on the yearly means of land value levels. Therefore, the yearly calculated average of LV levels is the independent variable, while the periods served as the independent variable. In the next step, using the obtained model for the pre-Covid period, we made predictions of the expected average LV level for the two Covid years. Together with the expected average, we also construct prediction intervals, with a 95% likelihood. Finally, we compare confidence intervals of the actual mean LVs in the Covid period with the corresponding prediction intervals. Our hypothesis is that, if significant differences are found between the pre-Covid and Covid periods, these divergences are likely to be attributed, at least partially, to the impact of the pandemics on the real-estate market.

It is important to highlight that the transaction data is treated as a pooled cross-sectional data, with no assumptions of dependence between the panels, neither between multiple transactions of the same property. The suggested method is an implementation of a conventional annual averaging of real estate prices, typically used for long-term trends explorations or as starting point of multivariate analyses (Quan and Titman 1997; Gyourko et al. 2013; Wang et al. 2020). In other words, we look for the simplest and non-sophisticated cross-sectional analysis able to shed light on the basic research question: Is it possible to find traces of post-Covid-19 impacts on land values?

This analysis was applied both to the entire set of transactions and subsets of transactions with different characteristics, such as type of property (apartments or private houses), age (new or old properties), location (big cities, small towns, rural areas), etc. Part of the results do not show significative differences between both periods. But the subsets of transactions for which sensitive inter-period divergences were identified, suggest a possible initial shift towards a different post-Covid urban structure.

Finally, since the real-estate transactions dataset is geo-located, we use it also for the analysis of the location of transactions with specific characteristics. If a specific type of properties shows a marked difference between both periods, we can also identify the places where transactions of these types of properties are more likely to occur. We do this creating additional KD surfaces, using a kernel bandwidth of 500 m with pixels of 50 × 50 m size, and normalizing them to a range between 0 and 100. These KD surfaces reflect the transaction’s densities of each type of relevant real-estate properties.

Results

The most significant differences between the pre-Covid land values and the trends observed during the pandemics are related to two specific residential sub-sectors: New private houses, and aged apartments (over 50 years old) in high rise buildings. Table 1 reports the number of transactions of the corresponding real estate types:

Table 1.

Number of transactions by real estate types by years in Israel

Time period	The whole set of transactions	Transactions with new ( < = 1 year-old) private houses	Transactions with aged ( > = 50-year-old) apartments
2007	73,092	2,271	4,745
2008	67,793	2,145	4,394
2009	77,486	2,382	5,348
2010	80,922	2,203	9,549
2011	67,483	1,807	8,974
2012	82,889	2,151	10,867
2013	91,834	2,515	11,530
2014	81,621	1,836	9,339
2015	94,234	2,202	9,618
2016	92,876	2,086	9,921
2017	74,609	1,281	8,415
2018	79,714	778	9,517
2019	90,606	995	9,849
2020	82,197	1,266	16,931
2021	121,739	1,568	26,419

Figure 1 summarizes these main findings. Figure 1(A) shows the dynamics of the LV levels for the whole set of transactions (that is, transactions regarding all types of real estate) in Israel. The p-value of F-statistics versus the constant model is 9.76^10^− 10; adjusted R² = 0.97. The regression model reports that the increase of the average LV is 499 ± 52 per year. From the figure, the actual average levels of the LVs in the COVID period (dark red circles on the error bars for the years 2020 and 2021) are slightly lower than their corresponding predictions (on the dashed black line). These predictions are the extrapolation of the pre-Covid trend (black solid line), built to fit pre-Covid average LVs (black dots). The confidence intervals for the actual Covid LV levels (dark red error bars for the years 2020 and 2021) lie completely within the prediction intervals of the estimated LV levels. This means that, according to our method, we cannot reject the hypothesis that the effect of Covid was insignificant on the whole set of transactions.

Fig. 1.

Land value (LV) dynamics: The whole set of transactions (A), transactions of new private houses (B), and transactions of aged apartments in high-rise buildings (C). The solid black lines stand for the linear regression built upon the pre-Covid yearly averaged LVs (black dots). The black dotted lines stand for the extrapolation of the regression line onto the Covid period. The color stripes represent 95% prediction interval for the regression lines for pre-Covid (gray strips) and Covid (light red strips) periods

However, there are significant deviations for certain types of real estate transactions when the Covid period is compared to the pre-Covid levels. In Fig. 1 (B and C) we report two opposite cases: Transactions of new (less than 1 year old) private houses and transactions of apartments in old (more than 50 years old) buildings. As can be seen from Fig. 1(B), the new private houses’ mean LVs significantly exceeds its prediction in both 2020 and 2021: The confidence intervals of actual LV averages, do not overlap with the prediction intervals of the regression line. At the same time, aged apartments’ actual mean LVs during the Covid period are statistically significantly lower than their predictions both for 2020 and 2021 (Fig. 1(C)). Regression models for mean LVs of both new private houses and aged apartments emerged significant compared to corresponding constant models (F > 147.00; p < 0.001). They both fitted pre-Covid data well (R²-adj.=0.92 for aged apartments and R²-adj.=0.98 for new private houses).

Interestingly, the slope of the regression for pre-Covid means of LVs of aged apartments is markedly steeper than the slope for means of LVs of new private houses: The average yearly increase in LVs of aged apartments in the pre-Covid period is about twice the increase in LVs of new private houses (725 ± 117 NIS vs. 335 ± 25 NIS). High means of LVs and their average yearly increase in the group of aged apartments in the pre-Covid period is caused by the high price levels of aged apartments in large cities (of 909 ± 173 NIS per year – not shown). Relatively low LV means and their average increase in the group of new private houses might be influenced by the location of this type of real estate property type: Private houses are predominantly located in relatively small settlements (~ 85% of all transactions) or at the outskirts of relatively large cities.

The results of our analysis till now, shown in Fig. 1, describe the trends of the observed real-estate market and its segmentation into differentiated subsectors. However, it does not capture the spatial implications of these trends regarding the structure of urban areas. Figure 2 shows several density maps created according to the methods described in the methodology, applied to the Tel Aviv metropolitan area (the most dynamic and populated urban area in Israel): The built structures density as a general background of the urban form, and the transaction’s densities of the relevant real-estate properties: New private houses and aged departments. For the purposes of this assessment, the most remarkable and clearly visible feature are the very different location patterns of both real-estate subsectors.

Fig. 2.

Densities in the Tel Aviv metropolitan area. The gray spots in both maps represent the densities of built structures in general. The transaction density of new private houses appears in blue in the left map (A) while the transaction density of old apartments is shown in red in the right map (B)

The analysis aims to reveal a long-term trend in the yearly means of land values. Although the number of observations is indeed modest, each represents tens of thousands raw data points. We proceed from a plausible assumption that averaging over a large amount of data would level out the weaker effects of other factors responsible for the within-year heterogeneity. We use a 95% prediction interval to forecast the range of the yearly mean land values in the Covid period provided the preservation of the long-term pre-Covid trend. In those cases when the confidence interval for the actual covid-time yearly mean does not intersect with the corresponding prediction interval, we conclude that the pre-covid long-term trend has changed. Importantly, to reduce the within-year heterogeneity, we tested splitting the raw data points based on various real estate characteristics (such as area, number of rooms, type (house/apartment), age, city size, etc.) as well as all possible two- and three-factorial combinations of these characteristics. The two chosen subsets (new private houses and aged apartments in high-rise buildings) illustrate the most evident structural breaks observed when comparing both periods.

Conclusions

The study reported herein relies on the assumption that the physical structure of cities is the result of self-organization processes in which profit-maximizing developers are key players. We utilized the recent Covid-19 pandemic as a natural experiment by means of which it is possible to gain insights into shifts into the spatial structure of cities by studying developers’ preferences.

We find that during the Covid period the mean LV of private new houses increased and that of old apartments decreased. The coexistence of both trends in parallel supports the hypothesis of an ongoing change of preferences in the Israeli residential real estate market: On one hand, old apartments, located generally in dense urban areas, have become less attractive during the Covid period. On the other, the appealing of new private houses seems to be on the rise. These houses, offer more space in general, and particularly larger leisure and working related amenities. In addition, although there are stocks of private houses in inner cities, they are located generally in more peripheral areas, with better access to open spaces.

However, the main objective of the present study is not to predict future demand for urban lands, but to anticipate possible spatial changes in the supply side of the dwelling market. Real estate developers are in constant search for locations with high land values that are also expected to further increase (McAllister et al. 2018; Ratcliffe et al. 2021). But in addition, if the built environment in these places is not too dense, there are higher chances of finding suitable and undeveloped parcels (Buda et al. 2022). Observed through this “supply framework”, the spatial patterns observed in Fig. 2 suggest that, if the trends observed by the data continue, the LV in the urban outskirts is expected to rise while the LV in the city centers will decline. These dynamics support the hypothesis of a possible flattening bid-rent curve driven by future real estate developers’ behavior, ultimately being one of the long-term Covid impact on urban areas.

The obtained results allow us to highlight possible initial signs of post-Covid-19 urban structures. Whether or not post-Covid cities will be significantly different from their pre-pandemic versions, is a research question to be undertaken in the future. A central and obvious limitation of this study is the short time elapsed since the Covid-19 outbreak. This makes difficult to discern, whether the results reflect processes that will be long lasting, or momentary changes caused by an unexpected event that will be corrected once things return to their normal path. Additional limitations stem from the deliberately simple cross-sectional analysis chosen to shed light on the research question: Future follow-up studies should incorporate more detailed data about key factors that are likely to influence on the pandemic impact on the urban structure. For example, the ability work from home and its spatial variation, or possible changes in land supply and the regulatory environment during the aftermath of the pandemic that could have influenced on developer’s behavior. Nevertheless, we are able to draw a few relevant conclusions as an educated guess exercise drawing extrapolations from current trends. We believe that real estate developers will continue to be one of the driving forces of urban development also in the post-Covid-19 world: Their future preferences, interacting with planning policies, will reveal to which extent the outlook described in this paper will realize or not.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Declarations

Conflict of Interest

The authors have no conflicts of interest to declare. All co-authors have seen and agree with the contents of the manuscript and there is no financial interest to report. We certify that the submission is original work and is not under review at any other publication.