1 Introduction

Houses and apartments serve multiple purposes, acting as durable consumption goods, investment assets, and sometimes speculative assets. Purchasing a house as an investment typically involves the aim of generating income through future rents or capital gain. Speculative investment motives drive demand when capital gains hinge solely on anticipated short-term increases in resale price (Tirole, 1985). While demand for housing for pure consumption remains relatively stable in the short run or gradually changes in a predictable manner, investment or speculative demand often leads to pronounced swings in housing price trends. Urban areas and agglomerations are particularly prone to these fluctuations compared to suburban and rural areas, given the concentration of investment and speculative demand in larger cities. Consequently, housing market dynamics are intricately intertwined with spatial and urban economics, in addition to real estate economics, due to the crucial significance of macro and micro locations for housing prices.

The impact of real estate markets on the economy is well-documented. It can significantly contribute to economic output while also posing substantial threats to the environment and economic stability (Reinhart & Rogoff, 2008). However, there remains a significant knowledge gap regarding the drivers of real estate booms and declines, particularly in the context of developing countries (Phatudi & Okoro, 2023). The study addresses this gap and provides valuable insights into the specific causes of real estate price fluctuations in a developing context, choosing Serbia as a developing country setting. This research is important as it not only explains the dynamics of the housing market in Serbia but also serves as a model for developing countries facing similar challenges. By offering insights into the unique factors influencing real estate dynamics in developing countries, the study aims to inform policymakers, investors, and stakeholders in the real estate sector, ultimately contributing to economic growth and development.

Rising housing prices not only pose risks to price stability (Bernanke & Gertler, 2001), but also contribute to deepening wealth (Wind et al. 2019) and consequently income inequalities. The ‘wealth effect’ (Yin & Su, 2022) leads to income growth and enhanced collateral positions for homeowners, further widening the wealth gap (Iyke, 2018). Given the pronounced disparities in income, education, healthcare, and household wealth in Serbia, understanding the forces governing the housing market is of particular significance.

In the last five years, Serbia's housing market has witnessed significant growth, with turnover value nearly tripling from the first quarter of 2018 to the last quarter of 2022. The growth experienced notable acceleration in the last two years, with increases of 43% and 26% respectively. However, this growth has been accompanied by regional inequality, as evidenced by variations in turnover value across different regions. The share of the capital city’s greater area (Belgrade region) consistently accounted for 60 – 67 percent of the total national turnover value. The heightened market activity coincided with a surge in new construction, with the number of newly built houses and apartments steadily increasing since 2015, reaching 2.7 times the initial figure by 2021. Intensified new construction further underscores spatial disparities in housing market activity, with noticeable hot spots in certain municipalities. Significant disparities were observed across cities and municipalities within statistical regions, evident from the spatial distribution of housing construction. In 2021, the highest number of new houses and apartments per 1,000 inhabitants was recorded in the tourist municipality Čajetina (98.5), followed by two Belgrade metropolitan area municipalities, Savski venac (Belgrade Waterfront) (25.3) and Vračar (15.8), as well as tourist municipalities Vrnjačka Banja (14.0), Sokobanja (13.3), and the regional capital city of Novi Sad (13.2). The national average for Serbia in the same year stood at 4.2 houses and apartments per thousand inhabitants.

This study evaluates the impact of specific supply and demand variables on housing prices at the subnational level in Serbia, aiming to inform policy decisions for equitable and sustainable urban development. By analysing a unique panel dataset covering macroeconomic, demographic, and urban infrastructure variables for 24 Serbian cities, we contribute to the literature on housing price determinants, particularly in developing country contexts. The novelty of our empirical approach is that it combines the foundational two-equation stock-flow model of the housing market with an analysis of urban infrastructural attributes. In addition to demand and supply dynamics, we assess how the availability and quality of urban infrastructure impact property values. By combining these two aspects, we aim to provide a more comprehensive understanding of the determinants of housing prices, offering insights into both market fundamentals and the influence of public services and amenities on real estate markets. This is of relevance for developing countries with pronounced infrastructure disparities. Overall, this study represents one of the few attempts to explore the fundamental determinants of housing prices at the urban level in the Serbian real estate market, serving as a model for similar contexts.

The following section lays the theoretical groundwork for the empirical analysis conducted in this article, examining the fundamental determinants of house prices as identified in prior research. Following this, Section 3 conducts a review of relevant empirical literature on housing price determinants. Section 4 elaborates on the empirical approach, detailing the data sources and selection of variables. Section 5 provides a summary of the results obtained from the model estimations. Lastly, Section 6 presents the concluding remarks of the study.

2 Economic theory on housing market and housing prices

The factors affecting urban housing prices have been extensively analyzed from many different perspectives, in terms of macroeconomics, policies, market forces and location properties (Harter-Dreiman, 2004; Hilber, 2015; Martins et al., 2021; Gao et al., 2022). Economic research focuses mainly on macroeconomic and demographic factors, as well as policy regulations in explaining temporal dynamics of housing costs. On the other hand, the hedonic price theoretical framework (Rosen, 1974) is most often used in studies that estimate the impact of structural, accessibility and neighbourhood attributes on housing prices. Those attributes are specific for individual dwellings or a group of dwellings on the same location, making micro-locational features the significant determinants of housing prices. Our approach builds on the studies that estimate the effects of aggregate national, regional and local level variables as price determinants in a panel setup.

The basic building block of modelling housing price is a two-equation stock-flow model of the housing market (Poterba, 1984; Meen, 2002). The model depicts a housing demand and a housing supply function. Both functions are presumably determined by housing price, so that demand ultimately equals supply, and the housing market settles equilibrium with a clearing housing price. Housing demand will drive the prices up while supply will drive them down. According to the model (rewritten from DiPasquale & Wheaton, 1994), demand for housing will depend on some exogenous variables, denoted X1, price of housing, P, cost of financing the purchase of a house, U, and the alternative cost of renting a house, R. Within the set of exogenous variables researchers usually include various demographic features and the household income. From the purely theoretical point of view the basic unit of the housing demand is the household itself, instead of an individual. Thus, a sophisticated analysis should carefully consider the differences between the number of households within a population.

According to the model, households maximize an inter-temporal utility function with non-separability between housing and non-housing consumption. The model assumes an efficient market able to swiftly bring supply and demand into equilibrium (1).

$$D({X}_{1},\boldsymbol{ }P,\boldsymbol{ }U,\boldsymbol{ }R)=S$$
(1)

Supply equals housing stock, which itself is a sum of periodical (yearly) change in the number of houses. The change (ΔS) is a difference between new constructed houses and the depreciation of existing housing stock (δS).

$$\Delta S=C \left({X}_{2}, P\right)-\delta S$$
(2)

New construction (C) itself is a function that depends on housing price and a set of exogenous variables (X2), which normally include prices of house production factors (key material, labour, interest rates etc.). Some authors (DiPasquale & Wheaton, 1994) argue that the list of exogenous variables can be easily expanded to include general macroeconomic conditions. It is noteworthy that the housing price, that appears both in demand and supply equations, is not one defined unambiguously. It can be not only current, observed spot price, but also an expected, future price. The idea that exactly market expectations of future path of housing prices predominantly shape current market price is well researched and documented in financial literature (Granziera & Kozicki, 2015). However, there are also attempts to incorporate the expected housing price into the stock-flow model. The expected price is translated into expected capital gains from house appreciation and is included in the above-mentioned costs of financing the purchase of a house. The rationale for including the renting costs in the demand equation is straightforward, since renting a house is a clear alternative to homeownership, and in a standard asset-pricing model the fundamental housing price is determined exclusively by future rents and the discount rate. Unfortunately, despite being potentially useful, data on rents are often inaccessible and seldom included in empirical models (Chen & Chiang, 2021). The data issue arises with some other perspective determinants, when the research focus is on subnational level. Namely, some data are obtainable only in ten years frequency (the number of households), while annual series on some income proxies (GDP) have not been disaggregated by metropolitan areas within the country.

It is also noteworthy to mention that the housing demand is affected by different predictors, depending on whether it reflects the demand of owner occupiers (for personal use) or that of speculators and those that view housing as an investment asset, motivated by the search of profits based on expected rents or capital gain. Modelling the investment demand therefore accounts for the role of uncertainty and expectations of the future housing prices, that in turn determine the current and future consumption choices (Ngene & Gupta, 2023). Predicting the housing market behaviour should therefore not neglect the dynamics of the housing price itself.

Though our prime interest here is to study determinants of the demand and supply of housing units and therefore the housing price, an equally important idea intrinsic to the model is how demand and supply respond to the housing price. Housing supply elasticity to housing prices plays an important role in the speed of adjustment and ability of market to attain an equilibrium, and consequently for the evolution of housing prices. If supply gradually and partially responds to the price, the supply will stay protractedly unlined with the demand, pushing the market price out of the equilibrium. According to empirical research (Caldera Sánchez & Johansson, 2013) physical limitation on land for development as well as more restrictive regulation on land use, the provision of infrastructure and other public services complementary to housing, may make housing supply less responsive to prices in more densely populated areas/cities. The same source underlines that the degree of competition in the residential construction industry potentially affects supply responsiveness. Competition influences an average mark-up in the industry, with collusive behaviour in the construction sector hindering the competition. Housing price is the main determinant of developers’ profit. We would say that some peculiarities of the national construction industry may contribute to the way supply responds to the housing price. There is a small number of contractors able to manage large projects. The industry comprises a large number of relatively small firms with many of them not specialized for residential construction or even being temporarily active on the market. It is peculiar that some residential developers do not have any respectable prior experience but are attracted by strong market and high mark-up. In such conditions, housing supply will have strong elasticity to contemporaneous and expected housing prices.

A considerable body of research deals with the relationship between housing prices and the accessibility of urban infrastructure, such as transport, educational, health and recreational facilities, (Thompson, 2017; Gao et al., 2019) and how the availability of (or proximity to) a particular type of infrastructure is capitalized into housing prices or rents (Rondinelli & Veronese, 2011). It is argued that public services provision and accessibility significantly impact the housing market (Tiebout, 1956). These effects are explained by rational consumer behaviour, as individuals tend to consider the quality of public services and infrastructure when deciding on buying property (Oates, 1969). The provision of public services is assumed to affect individuals’ well-being and quality of life, which in turn enhances housing prices (Caragliu et al., 2011). Infrastructure (transport, education or health facilities) shapes the housing demand by enhancing the utility derived from real estate (Belke & Keil, 2017). Used as the measure of aggregate public service package (McMillan & Carlson, 1977), local public expenditures shape the main locational effects of residential property - quality of schools, public transport facilities, crime rates and environmental attributes (García et al., 2010). Municipal expenditures for education and schools appear to be of particular importance for the housing prices, since they constitute a significant share of local spending, while at the same time home buyers consider the school quality as one of the most visible benefits of the residential areas (Mathur, 2008). The notion of cities as providers of abundant education possibilities makes them attractive due to increased workforce education and growth prospects, leading to vibrant labour markets and higher wages that consequently raise housing prices (Eaton & Eckstein, 1997). In addition, neighbourhood attributes reflecting the socioeconomic status of communities (GDP per capita, educational background, population density) contribute to the explanations of housing prices (Duan et al., 2021).

However, from the perspective of housing as an investment asset, the increase of housing prices could be a result of speculative market activity or the failure to capture the willingness of property renters to pay for the infrastructural facilities. Consequently, in such cases public infrastructure may not prove to be a relevant predictor of housing prices. There is research evidence that housing markets can be affected by the economic structure and the share of various industries in the local economy. For example, it is documented that tourism development increases housing prices in the tourist destinations (Meleddu, 2014). This is explained by the process of switching capital investments to real estate, conversion of housing into rentals or land use restructuring.

3 Selective review of empirical literature

The empirical research on housing prices differs in methods used by researchers, but more radically it differs in the level of aggregation. Starting from the highest level of aggregation, there is abundant research aimed at modelling the average housing prices on the national market, most often in multi-country framework. The next level of disaggregation is related to the research that models an average price of dwellings on individual metropolitan housing markets. Depending on data availability and research objective, the cities as ultimate units can be grouped based on similarities or observed separately. Finally, the highest level of disaggregation relates to the explorations oriented to the price determinants of individual pieces of residential real-estate. The selected level of aggregation significantly influences the set of plausible determinants. At the national, regional, and city levels, socioeconomic factors typically play a dominant role. However, when analysing individual dwelling prices, a significant portion of the variability can be attributed to physical attributes of the dwelling, neighbourhood characteristics, and accessibility variables. Naturally, when working with an average price for a group of dwellings, individual differences become less relevant, much like macroeconomic data may not effectively explain price variability across individual dwellings. By concentrating research efforts on various subnational levels such as regions, provinces, or cities, one can still capture macroeconomic influences on housing sub-markets, while simultaneously expanding the scope of determinants to include local socioeconomic and environmental specificities. The intercity level appears to be well-suited for investigating environmental factors such as air quality, temperature, and precipitation (Ou et al., 2022; Miłuch & Kopczewska, 2024), as well as density of transportation infrastructure (Tan et al., 2019), and other factors that contribute value to housing stock, such as tourism activity, quality of education, and healthcare services provided.

All approaches, except the last one, are naturally inclined to macroeconomic and macro-financial predictors. The principal objective of those explorations is to model price dynamics, most likely in order to find reasons behind and prevent market disequilibria, or boom-bust behavior of the real-estate market, well known and documented as real-estate bubbles. In an early study, Brooks and Tsolacos (1999) found several macro-financial variables able to explain the return on real estate investment trust in UK, i.e., unexpected inflation, interest rate term spread (difference between long-term and short-term governmental bond yield), contrary to rate of unemployment, nominal short term interest rate and dividend yield. Interestingly, the lagged values of the real estate return also have had some explanatory power, indicating the existence of trends in the return series (so-called positive serial correlation). Overall, the choice of variables is obviously influenced by the research of stock market return predictors, implicitly assuming that real estate and stocks/equity (and equity linked instruments) are clear alternatives of household investments. The assumption is not overly restrictive, since the majority of other financial instruments (fixed income securities, bank and insurance instruments) do not serve as a good hedge against inflation risk.

The empirical research on housing price determinants is naturally inclined to more developed countries. Building on previous research, DiPasquale and Wheaton (1994) have modelled US market demand and supply determinants and found housing stock to household ratio, permanent household income, homeownership rate, rent index and interest rate relevant for the housing price. Similar to the previous paper, Meen (2002) compared determinants of housing prices of two developed countries (US and UK), using time-series regression for different periods. The author confirms predictive power of housing stock supply, population, real interest rate, real wealth, real per household income, and a lagged dependent variable, with some differences between markets. Some papers operate with even smaller set of explanatories. Malpezzi (1999) tested the influence of mortgage interest rate, real income and population expressed both in change in logarithm and logarithm itself, along with some institutional features.

Englund and Ioannides (1997), in dynamic panel framework, revealed a remarkable degree of similarity across fifteen member countries of Organisation for Economic Cooperation and Development (hereafter OECD). It appears that GDP growth rate and the rate of change in real interest rate exibit significant predictive power expressed both as one period lagged and contemporaneous variables, in addition to lagged house price changes. Interestingly, the population variable, expressed as rate of change of population in the house-buying ages did not perform well. However, the authors discarded the existence of international housing price cycle.

One of the most extensive empirical works (Égert & Mihaljek, 2007), in terms of coverage, comprising 27 European countries, confirms statistical significance of several conventional fundamentals, GDP per capita, real interest rates, the volume of mortgage credits, and some demographic and institutional determinants. Caldera Sánchez and Johansson (2011) explored demand related determinants of housing price for dozens OECD countries and found income (real GDP per capita, or real household disposable income) and population (the share of the 25-44 years old cohort in total population or the population in that age cohort) positively related to the average national house price, and interest rate and housing stock predominantly negatively related. At the same time, supply (residential investments expressed as gross fixed capital formation in the housing sector) is positively influenced by real housing prices, while the real construction costs and population manifested a mixed influence (both with one period lag). In a more recent paper, Geng (2018), using cross-country panel, found long-run trends in housing prices within twenty OECD countries determined by per capita household personal disposable income, mortgage rate (negatively), per capita household net financial wealth, housing stock per capita (negatively), including some institutional and structural factors (e.g. strictness of rent control regulation and long-run elasticity of real dwelling investment with respect to real housing prices).

A single country study of Zhang et al. (2012) found China housing market determined by some key monetary and price variables, i.e., mortgage rate, building costs, wider monetary aggregates and real effective exchange rate (REER). Interestingly, the research does not support the assumption that income variables have to be an independently significant determinant. In a more recent paper, Vogiazas and Alexiou (2017), based on dynamic panel framework, found real GDP, bank credit growth, long-term bond yield and REER promising in explaining price bubbles on residential property markets of seven developed countries during the period from 2002 to 2015. The statistical significance of REER can depict meaningful presence of foreign capital on the local housing markets.

Inchauste et al. (2018) in a voluminous work that covers EU countries found housing price index invariably positively responsive to GDP, and mixed to short-term interest rate, housing stock and population. The analysis was based on time regressions run separately for sampled countries. The authors also found productivity growth associated with housing price growth in metro areas. As productivity increases, labour demand should increase, leading to higher wages, employment and housing prices. The evidence suggests that productivity growth has been concentrated in the main cities and agglomeration centres. The paper also implies that those areas are more prone to extreme price movements. If housing prices grow on a national level, they will grow faster in major cities, the same as they will fall faster in case of a falling trend on national level.

The research of Abraham and Hendershott (1996) can be considered a step forward in exploring macroeconomic and financial determinants of housing prices. Although the set of predictors and the method is the classic one, this is one of the first attempts to look at the national housing market as fragmented one. The authors explored the drivers of housing market appreciation in thirty metropolitan areas across the US. The paper separates determinants into the group that explains changes in the equilibrium price (the growth in real income, real construction costs and the real interest rate), and the other that accounts for the adjustment mechanism, i.e., deviations from the equilibrium (lagged dependent variable and the difference between the actual and equilibrium real house prices). The same approach is followed by Zhang et al. (2016). The largest Chinese cities were grouped into three subsamples (tiers) and explored separately in order to delve into differences in the way their housing markets react to the set of macroeconomic and financial predictors. The tiers are assembled based on socio-economic importance and infrastructure developments. The results show both differences in initial response to some predictors and the way it changes over time. The first-tier cities have had the strongest effects of interest rate, inflation rate and economic growth, compared to that of less important cities (second- and third-tier). Moreover, opposite to interest rate, which stays negative, and the growth, which has clear positive influence over time, the inflation influences reversed, from initial positive to negative one. Such findings imply that housing demand function may strongly differ depending on overall importance of cities within a country of a region. The more important the city, the more likely it will be targeted for speculative housing demand.

The other line of research on housing markets, related to locational specifics, also provides important empirical evidence on the relationship between local amenities and housing price. Contemporary research using the hedonic pricing method (Wu et al., 2019) reveals that there are several value generators that originate from the ecosystem, e.g., traffic connectivity, waters and air quality, proximity to the educational, healthcare institutions, or main city (business) centre. García et al. (2010) estimate the effects of local public spending on housing prices in Barcelona and confirm a positive capitalization effect of per capita local spending on housing prices. Their findings indicate that local policies aimed at improving locational attributes of the residential units (street maintenance, waste management, building improvements, sports and entertainment areas, parks and garden conservation) have a positive impact on housing values. Ouyang et al. (2022) use the Bayesian model averaging to explore the variables that predict housing prices and rents in urban and rural areas in China. Their findings indicate that access to healthcare, education and recreational facilities significantly impacts home prices and rents. Yuan et al. (2018) indicate that the proximity to medical centres affects appears to be relevant in predicting the regional variations in housing prices, while the presence of high-quality schools can lead to increased housing prices (Chan et al., 2020). Shin et al. (2024) found that accessibility to various urban amenities significantly influences housing prices, for instance access to learning, care, education and leisure facilities. The impact of public transport infrastructure on housing prices has been confirmed in a number of recent studies (Efthymiou & Antoniou, 2013; Song et al., 2019; Li et al., 2023). Using proximity- and accessibility-based measures for estimating the impact of transport infrastructure on housing prices, the research indicates that proximity to the big transportation hubs, such as airports, bus terminals, and road intersections adversely impacts individual housing price, opposite to transportation accessibility measured by proximity to rapid transit or public bike stations (Soltani et al., 2024) The impact of tourism expansion on housing prices at municipal levels has been extensively researched and confirmed (Biagi et al., 2015; Paramati & Roca, 2019; Alola et al., 2020).

4 Data set and methodology

This research focuses on the Serbian housing market, situated in South-East Europe (Fig. 1). The region exhibits a lower level of economic development compared to the European average. In terms of housing market development, the country shares similarities with former socialist nations. Much of the existing housing stock was constructed during the socialist period and subsequently privatized. In recent decades, private investors have played a significant role in housing supply. Construction activity is market-driven, directed towards sub-markets experiencing the highest demand. Choosing the most appropriate spatial scale for housing market analysis appears to be a challenging issue. The idea to explore it on subnational level means that spatial delineations are determined by administrative boundaries. However, it is already recognized (Meen, 2002) that cities/municipalities as territorial units do not correspond well to any recognizable housing market area, which should be defined on the basis of substitutability between properties.

Fig. 1
figure 1

Spatial distribution of Serbian cities. *Authors’ preview using Tableau Public

Full size image

According to legislation, an urban settlement in Serbia may have a legal status of a city if it satisfies several preconditions alternatively. A settlement can be granted the status if it has more than one hundred thousand inhabitants living on its territory, or if it has a special geographical position, i.e., represents a district centre, or there is spatial proximity to state borders. Some cities have been excluded from the data because of missing data. It is the case with the city of Kikinda, Bor, Pirot and Prokuplje. Among the total number of cities that have finally been included into the data set, there is a national capital (first-tier city), three regional capitals: Novi Sad, Niš and Kragujevac (second-tier cities) and twenty district capitals and other major cities (third-tier cities). The underlying idea that the average housing price recorded in those cities is regressed on a set of predictors which all represent quantities/values recorded within the city/municipality borders, rests on a doubtful assumption. We should reasonably assume that gravitational zone for Belgrade as a national capital city has to be much wider than the territory of Belgrade region (the Greater Belgrade). Hence, for the residential real estate price in Belgrade it is likely that income or population of even neighbouring countries will matter. Namely, based on some novel research (Vaz & Nijkamp, 2015) the socio-economic influence of cities on their neighbourhood can be measured with some concepts originally developed in Newtonian physics. For example, gravitational or attracting force between two objects (e.g., cities) is proportional to the mass of two objects and inversely proportional to the distance between them, where, based on the Second Law of Newtonian Gravity, the city mass is calculated as a ratio of a city force, understood as the influence of the possibility of change in the context of urban areas, and the city acceleration (in terms of urban sprawl). The force is conceptually defined as the temporal variation in population (which is assumed to be followed by equal change in supporting infrastructures), while the acceleration is assumed to represent the variation in urban areas. Therefore, the city mass will stay unchanged over time if the population density does not change. For our framework it means that assuming cities with significantly different population density to have more or less the same level of spatial influence will lead to fallacious results.

4.1 Choice of variables, definitions and data sources

For our econometric analysis, we utilized panel data with a yearly frequency, sourced at the city level. These data were sampled from official sources, relying on census data or official estimates regularly conducted on an annual basis (Table 1). In all model specifications dependent variable is taken to be unit housing price. It is the average price of residential property per square meter, expressed in local currency (RSD). The ultimate source of data on housing prices is the Republic Geodetic Authority, which runs the national Real Estate Cadastre—a public registry containing comprehensive information about properties in Serbia. This authority publishes average housing prices quarterly, utilizing real transaction data for the reporting period at the municipality level. However, since the majority of prospective candidates for explanatory variables are available only in yearly frequency, we have been restricted to operate with yearly data. The data are available for the time span 2011 to 2021. Such an approach in defining a dependent variable is strictly data limited. One should be aware that an average price corresponds to an average dwelling, which per se masks huge differences between individual housing units, in terms of micro-location, building quality, floor area, house orientation, other amenities, etc. Even based on census methodology definition, there is huge heterogeneity between housing units, which can be both inhabited or uninhabited, temporary used for leisure and recreational purposes (second or weekend houses) or for seasonal agricultural work. Unfortunately, the data on house prices, available on municipalities’ level, do not control for the physical, locational of neighbourhood attributes. Recently introduced official index of housing prices based on hedonic price method covers period from 2016 onwards and reports the time trend component. The data are collected from samples and reported for both primary (newly built) and secondary market (resale properties) but no data is available on city/municipality level.

Table 1 Variables, definitions and data sources
Full size table

The choice of explanatories in our baseline model (1) is inspired primarily by the stock-flow model, with our clear intention to cover the main inputs of supply and demand functions, as much as possible. It is elegant in a way that produces a rather good fit with a few variables, but it is also data limited.

Unbalanced economic growth has the potential to explain differences between housing prices in cities of different sizes and importance. Among the variables that represent household income our choice was average wage (tax and contributions excluded) per an employee in local currency, provided by Statistical Office of the Republic of Serbia. It is the monthly average for a reporting year. Taken together with employment data, the average wage may represent a major part of the household income, which is an exogenous variable in the housing demand equation. Income data has undisputable importance, to the extent that there is a branch of literature which explains housing market disequilibria based solely on house price-to-income ratio (Malpezzi 1999; Leung, 2014; Leung & Tang, 2023). Employment data enters specifications as level data, since this way of the variable construction proved superior to the number of employees per thousand of inhabitants. An obvious flaw of this option is that it ignores other sources of national income (e.g., capital income like interest, rents and profit). However, using average wage has another quality, since it represents the unit cost of labour that also enters housing supply equation, via costs of construction. Other potential candidates to be used as a proxy for household income, for example economic growth rate, personal or disposable income levels, were discarded, because the data were not available on municipalities’ level.

The next task was to find the suitable representative of demographic needs for housing, which will represent the so-called latent demand. One should note that a house is a durable consumption good purchased by households. Hence, the number of households and age distribution of households are the facts that should matter for housing demand, not population itself. Some preliminary calculations indicate that the four regions do not differ significantly in terms of number of houses per household. Somewhat surprisingly, more developed northern regions (Belgrade and Vojvodina) have slightly lower number of houses per household than two southern less developed regions, despite more intense construction activity in northern regions. A plausible explanation could be the change in population and household numbers due to migrations and negative natural increase of population. It is the point where the structure of the household should be considered. However, data on number and structure of households are available only in ten years frequency. It is the same with the age structure of the population, which, in addition, is reported with the break-out that does not allow recalculation of a house-buying age cohort. Data available from ten-years census show for example that there is no significant difference in average size of a household between regions and municipalities, what makes our task easier, since total population figure should be taken as a good approximation of demographic needs for housing. Although Serbia, as a whole, experiences a long-term trend of declining of total population, there are differences in the trends regarding the regional distribution. In the last decade the only region that recorded an increase in population is the Belgrade region. The other three regions faced around ten percent population decline. The difference should be largely attributed to internal migration flow, since natural increase is uniformly negative, the same as cross-border migration balance. Thus, what may make the difference is internal migration balance, which was positive for Belgrade region in all years in the past decade, and for the Region of Vojvodina after 2016, while the other two regions (South regions) recorded negative yearly balance in the entire period.

In order to portray the level of demographic needs, and more completely demand saturation, we added a variable which is constructed as a ratio of population to housing stock. It is inverse to DiPasquale and Wheaton (1994) and Geng (2018) choice of the housing stock to population ratio. The change in the ratio is jointly determined by change in numerator and denominator. Population change is largely determined by migration balance, i.e., relative attractiveness of cities for living, while the housing stock is changed by the difference in new constructions and the rate of depreciation (the decrease in number of units due to demolition and abandonment). This ratio had more explanatory power than other similar variables that we also tried (total population, population change represented by current index). Because of its importance, construction activity is represented by another variable: new houses per thousand inhabitants.

For above mentioned reasons (Vaz & Nijkamp, 2015), we have also tested the population density as a perspective explanatory variable. However, it failed to join the set of statistically significant variables in different model specifications. It was likely because cities’ territories do not correspond well to the concept of urban area. Namely, there is huge difference in territory among the municipalities which are rather close in terms of population and other determinants. It is especially the case with district capitals.

Inflation is one of most prominent choices of researchers that studied macroeconomic determinants of housing prices. Across empirical literature it is included either as an explanatory variable or used to deflate a nominal dependent variable and a set of explanatories. Papers that employ the latter approach operate with real rather than nominal figures. In our study, inflation is expressed in the form of a base index (2006=100) of consumer price level based on COICOP methodology of weighting. Inflation has a complex relationship with housing prices. The way that consumer price inflation impacts housing prices can go through changing proportion between disposable and total household income, in a way that an increase in consumer price level will decrease the disposable income. The relationship can be even more than proportional if the disposable income represents a smaller part of the total household income. Thus, lower income countries may expect a stronger negative impact of inflation on housing prices through this channel. From the life-cycle theory (Attanasio et al., 2012) we know that individuals delay purchasing their first home when real incomes are low or uncertain, so that consumer price increase may have negative wealth effect and substitution effect on housing demand. In aggregate, it will depend on the age structure of the population. The non-housing consumption of young and old homeowners is much more sensitive to house prices changes than that of middle-aged homeowners. While house price appreciation increases the net worth and consumption of all homeowners, it only improves the welfare of old ones (Li & Yao, 2007). However, there are some other channels. If housing prices are broadly perceived as to follow inflation path, investment in a house can be seen as a good inflation hedge. This is the place where market expectations get into calculus. However, inflation-driven housing prices are only one possible scenario, since housing prices can be driven more or less independent of inflation, influencing both housing supply and demand, for instance by increasing or decreasing the homeowner equity. Inflation also drives key interest rates. An increase in inflation will increase financial (and likely raw material) costs of construction, with a negative impact on housing supply, while at the same time will increase financial cost of buyers, with a negative influence on housing demand. Based on theoretical models (Wheaton, 1985), taking into account different impacts of inflation on mortgage rates, payments and homeowner equity, greater inflation creates an unambiguous increase in housing demand, provided that there are not credit constrains that prohibit all kinds of borrowing. Thus, the inflation–housing price nexus seems rather complex.

The user cost of homeownership is ideally captured by the interest rate for mortgage borrowing. It is expected that mortgage interest rate influences demand for mortgage loans and, consequently, also a part of housing demand which is credit financed. A household finances residential real estate purchases with net financial wealth and/or by borrowing from the credit market. In Serbia, the larger part of purchases are cash financed. Only a small fraction of the total value of transactions is financed by credit. Based on official reports (RGZ, 2023) in the last five years for multi-family residentials it was in the range of one third to one fifth, while solely for newly built ones it was notably less. The share of credit transactions in all real estate purchases (single-family houses, construction and agricultural land, and commercial real estate added) is barely ten percent. This leads to the conclusion that availability and price of credit cannot have a significant influence on housing demand. Nevertheless, we explored the influence of the cost of financing. In Serbian credit market long-term finance is granted with a practice of charging a variable interest rate. It is regularly Euribor rate plus fixed margin. Thus, the dynamics of mortgage rate is fully reflected in the dynamics of its variable part. Serbian banks favor Euribor 3m and Euribor 6m rate as reference rates. Those rates are rather close in levels and evolution, so we have freely chosen to operate with the rate of shorter maturity (three months). Based on the stock-flow model, the interest rate appears both in demand and supply equation, with uniform negative predicted influence on both quantities. Therefore, the ultimate, net effect of the rate is expected to be vague.

We include two additional model specifications to account for the dynamic effects of the housing prices, as well as to estimate the significance of the elements of urban infrastructure for the variations in prices. In these models, we include the lagged value of the outcome variable that could explain the dependence of housing prices on their past realizations. As suggested by DiPasquale and Wheaton (1994), contrary to the traditional stock-flow model that assumes rapid price adjustments, current real estate prices may be dependent on recent past prices, emphasizing the backward-looking expectations of actors at the housing market. The expectations of further increases in housing prices affect the expectations of future capital gains, which in turn further increases the (speculative) housing demand. We consider 1-year changes of the housing prices and estimate the impact of housing price changes in the model.

To capture the overall effect of the public infrastructure and services provision on the property values, we utilize the data on local expenditure. This variable is measured by the total public expenditures of the sampled cities per capita, serving as a proxy of the quality of infrastructure and services. We estimate the impact of total local expenditure on our response variable, since these data reflect the allocations to the largest beneficiaries (education, health services, social protection and art and culture). Developmental expenditures of the local governments are reflected by the investments in new fixed assets across different activities, not included in our set of explanatory variables due to the assumption of non-contemporaneous effects of local public investment that could not be captured by the time span of our panel data. The local public spending variable also serves as a proxy of economic performance of the city, jointly with other predictors used in our models that reflect the effective housing demand (wages, employment). Other things being equal, it would be expected that higher per capita expenditures are positively related to property values. We assume that local government spending leads to the higher quality of public services (health, environment protection, transport facilities and infrastructure, education, culture, etc.), thereby improving the location-specific characteristics of property. Increased property values further increase budget revenues, providing ground for further improvement of public services (García et al., 2010). Using this explanatory variable, we test if the local policies aimed at improving locational characteristics of dwellings can be capitalized in housing prices.

In addition to total municipality expenditures, we aimed to estimate the impact of local infrastructure amenities as potential predictors of housing prices. We tested a number of factors (transportation facilities, health care provision, educational infrastructure). After rigorous diagnostic tests and discarding the variables with potential multicollinearity issues, transportation infrastructure enters our extended model. We hypothesize a positive effect of this variable, measured as the total length of paved surface roads in sampled cities. We build on the evidence that transportation infrastructure, such as road networks or public transport availability to a large extent determines housing prices, by affecting proximity and accessibility attributes of residential units. The level of road network development affects the travel costs of inhabitants and their connections to other areas (work, school, entertainment), while the road network distribution depicts the functional distribution of residential areas and points to the urbanization level. These attributes are assumed to increase the utility derived from the property, influencing the increase in prices.

Finally, to investigate the relation between tourism and housing prices, we include data on tourism activity, as a potential predictor of property values. The impact of tourism on housing prices has been explored as a part of the wider efforts to estimate benefits but also negative social and environmental impacts of tourism on local inhabitants of tourism-dependent areas, one of them being increasing housing prices and consequent housing affordability issues. Tourism development is seen as a factor of switching capital investments toward real estate, as rational economic agents in search of profits tend to make best use of their capital (Mikulic et al., 2021). This variable is operationalized using data on total tourist overnight stays in the cities of our sample. We argue that housing prices in cities with more intensive tourism activity tend to have higher values.

The detailed descriptive statistics, encompassing average, minimum, and maximum values for the entire sample as well as segmented data for Belgrade, regional capitals, and third-tier cities concerning variables exhibiting regional differences, are provided in Table 4 located in the appendix of this manuscript.

4.2 Econometric model and estimation methods

In this study we employ a panel data analysis to empirically examine and quantify the relationship between residential property prices and a set of selected explanatory variables. The proposed econometric model, in its baseline form, is specified as:

$${Residential\;property\;price}_{i,t}={\beta\;}_{0}+{\beta\;}_{1}{Wage}_{i,t}+{\beta\;}_{2}{Employment}_{i,t}+{\beta\;}_{3}{Population/Stock}_{i,t}+{\beta\;}_{4}{Construction}_{i,t}+{\beta\;}_{5}{Inflation}_{i,t}+{\beta }_{6}{Interest\;rate}_{i,t}+{{\alpha\;}_{i}+u}_{i,t}$$
(3)

where \({Residential\;property\;price}_{it}\) represents the dependant variable, \({X}_{it}\) represents the set of the explanatory variables which differ over time (\(t\)) and across urban areas (\(i\)), \({\alpha }_{i}\) represents the unobserved individual effects, and \({u}_{it}\) represents the idiosyncratic term. After conducting model specification tests of the panel data model (F test for individual and/or time effects, Breusch-Pagan Lagrange Multiplier Test for random effects, and Hausman model specification test) it has been determined that the fixed effects regression model is the most appropriate choice (Table 2). To evaluate the validity and suitability of the results, a comprehensive set of tests were conducted to examine heteroskedasticity, cross-sectional dependence, and multicollinearity (Table 2). This rigorous testing methodology ensures robustness and reliability of the findings. Due to cross-sectional dependence and heteroskedasticity Driscoll and Kraay estimation procedure was applied.

Table 2 Model specification and diagnostics tests
Full size table

To tackle the issue of endogeneity, we utilized the dynamic panel approach through the System generalized method of moments (System GMM) estimator, which was pioneered by Blundell and Bond (1998). This method is highly suitable for datasets with a moderately short time dimension, as stated by Arellano and Bond (1991). We include the lagged variable of Residential property price (\({Residential\;property\;price}_{i,t-1}\)) and specify the extended model (2) as:

$${Residential\;property\;price}_{i,t}={\beta\;}_{0}+\rho\;{Residential\;property\;price}_{i,t-1}+{\beta\;}_{1}{Wage}_{i,t}+{\beta\;}_{2}{Employment}_{i,t}+{\beta\;}_{3}{Population/Stock}_{i,t}+{\beta\;}_{4}{Construction}_{i,t}+{\beta\;}_{5}{Inflation}_{i,t}+{\beta\;}_{6}{Interest\;rate}_{i,t}+{{\alpha\;}_{i}+u}_{i,t}$$
(4)

The utilization of the System GMM estimator has enabled us to circumvent various issues that commonly arise in econometric analysis, such as (Roodman, 2009; Kapil & Kumar, 2023): the presence of fixed effects, heteroskedasticity and autocorrelation within individuals, the endogeneity of control variables, as well as the possible omission of variables that persist over time. To ensure the reliability and consistency of our econometric analysis, we have employed two diagnostic tests to evaluate the System GMM estimation. Firstly, we utilized the Hansen test of overidentification to assess the validity of the instrument set, followed by the Arellano-Bond autocorrelation test, which is designed to test the absence of second-order autocorrelation for residuals (Konstantakopoulou, 2022). The results of the Hansen's overidentification test indicate no evidence of model misspecification, with no significant grounds to reject the null hypothesis of the joint validity of all instruments (Table 2). Additionally, the specifications outlined in Arellano and Bond's (1991) test for the absence of second-order serial correlation in first-differenced residuals have been satisfied.

Besides the extended model (2), an extended model (3) was formed in order to assess the impact of additional socio-economic factors on real estate prices:

$${Residential\;property\;price}_{i,t}=\beta_0+\rho\;{Residential\;property\;price}_{i,t-1}+\beta_1{\;Wage}_{i,t}+\beta_2{\;Employment}_{i,t}+\beta_3{\;Population/Stock}_{i,t}+\beta_4{\;Construction}_{i,t}+\beta_5{\;Inflation}_{i,t}+\beta_6{\;Interestrate}_{i,t}+\beta_7{\;Expenditure}_{i,t}+\beta_8{Transport}_{i,t}+\beta_9{Tourists}_{i,t}+{\alpha_i+u}_{i,t}$$
(5)

The extended model (3) also satisfies the conducted diagnostic tests (Table 2).

5 Results and discussion

We have tested numerous specifications and presented the best fitted ones. Table 3 reports on the estimated coefficients of three different model specifications, all of them with the same dependent variable. The baseline model (1) is generated based on a set of explanatories suggested by the stock-flow model, and that appears most frequently throughout the empirical literature. The baseline model explains more than sixty percent of variation of the housing price. The estimated impact of fundamental demand and supply factors are consistent with prior research and statistically significant. Differences in average wage (Wage) and employment (Employment) among cities and overtime, are proved significant in explaining cross-sectional and time dimension of housing price variability. The variables entered the dataset to jointly portray the influence of labor income differences across cities on housing price variability. Both variables proved the assumptive way of influence. The positive signs and high level of significance indicates that metropolitan areas with higher labor income will have comparably higher housing prices. It is obvious that the population income stimulates the housing market.

Table 3 Estimation results – dependent variable residential property price
Full size table

The housing prices respond positively to the ratio of population to housing stock (Population/Stock). The variable may represent an excess (unsaturated) demand proxy. Taking into account the construction of the ratio, the response may indicate the larger potential of cities in which population grows faster than housing stock (most likely because of internal migration) to establish a higher price housing market. Moreover, the positive influence of the construction activity (Construction) may provide additional explanations to the above regularity. It appears that the more intense construction activity leads to higher housing prices. It may mean that the construction activity cannot catch up with the population change and consequently the housing demand change, or simply that top-tier cities attract demand from the (wider) area that does not correspond to one that is delineated by administrative boundaries. It is probably because of the demand for houses for investment or even speculative purposes.

The undisputable statistical significance as well as positive sign of the inflation variable (Inflation), signals possible importance of residential real estate as an investment vehicle which is well suitable as a hedge against inflation. A mechanism that brings housing prices close to (anticipated) inflation can be multiple. Firstly, sellers will strive to transfer inflation-driven rise of costs of building to buyers, assuming that those costs are correlated to consumer prices increase. On the other hand, buyers will be ready to accept housing prices increased at the rate of inflation, if there is no better alternative to protect purchasing power of monetary assets. The Serbian case fits rather well to this regularity, having in mind that stock market has been experiencing poor performance since the recent Global financial crisis (2007-2008). Based on such causality, inflation should have a positive influence on housing prices. However, consumer price inflation–housing price nexus is more complex than that. If accompanied by mortgage rate increase, inflation will have less predictable influence on housing prices. This increase can be a consequence of monetary policy reaction, and usually would come with a time lag. This is why some researchers (Zhang et al., 2016) found a reversal of inflation influence in long term. Since the inflation burst in our case happened at the very end of the period of analysis there is still a possibility of having a reversal effect in the years to come.

The influence of the Euribor rate as a mortgage rate proxy (Interest rate) is confirmed negative, but slightly less significant than other variables (10 % significance in the baseline model). Based on the theoretical model, it can indicate that the credit finance is more important for investment demand than for investment supply, or that the credit is more (prohibitively) costly for buyers than for developers (housing suppliers), despite the same dynamics of rates. Moreover, above mentioned minor share of credit-financed purchases in total market volume can be a false lead. The total market volume includes a number of related sale and purchase transactions, where after the credit enters the market in initial transaction, becomes cash for the next related transactions, what makes the importance of credit underrated based on officially reported share of credit in the total market volume.

The second model incorporates a lagged dependent variable to examine the dependency of housing prices on their past values. The results reveal a positive and statistically significant coefficient estimate without compromising the robustness of the baseline model. All the variables keep their significance and expected directions of impact, while significance of interest rates diminishes. This model indicates that interest rates lose significance in predicting housing prices as their past realizations enter the model. That could imply that expectations of price appreciation may overshadow the impact of credit costs on housing consumption decisions. This finding aligns with the previous research, as the persistence in the series house prices has been confirmed in many previous studies (Biagi et al., 2015; Caliman & Di Bella, 2010). These results indicate that the housing price appreciation in Serbian real estate market is to a certain extent determined by the backward-looking expectations of the actors.

The third model introduces additional explanatory variables to assess the influence of public service provision and infrastructural amenities on housing prices dynamics. All predictors from the baseline model retain their explanatory power in this extended specification, confirming the robustness of fundamental supply- and demand-side variables on property value. In addition, the goodness of fit indicates that this model adds to the explanation of the variations of housing prices. The variables reflecting local public expenditure, road network and tourism concentration enter with the expected signs and significance. We exclude the employment variable from this specification, as per capita public expenditures enter the model as an indicator of the general economic environment at the urban level. Including the public service provision and infrastructural variables causes the lagged response variable to lose its explanatory power in this model.

The positive relation between local government spending and housing prices suggests that cities with better public services exhibit higher housing demand and higher propensity to pay for real estate. In addition, this variable also reflects the economic performance at the municipality level, that accounts for the vibrancy of local labour markets, labour income of inhabitants and other economic factors that determine housing demand. These results are consistent with the prior research evidence that communities with enhanced public service provision and better infrastructure attract more residents and consequently lead to higher housing prices. This reflects stronger preferences for Serbian cities or districts that receive more public expenditure intended for education, social security, community development, culture, environmental protection. The same holds for the municipality level transportation infrastructure and its impact on housing submarket level. In our sample, well-developed road networks are associated with higher housing prices, likely due to enhanced accessibility.

This influence works through enhancing the locational desirability of housing – more kilometres of paved roads increase the overall accessibility of the transportation system within the municipality.

Finally, tourism activity positively impacts housing prices, indicating increased demand and improved urban infrastructure. We test for the effect of tourism activity using data on total tourist overnight stays measured in sampled cities and find evidence of a relatively strong influence on house prices. Although Serbian economy is not particularly tourism-driven, there are significant differences across Serbian cities in the intensity of tourist activities. Serbia’s capital, Belgrade, attracts the largest number of tourists, followed by a few large cities (tourist destinations, including those in the vicinity of mountains, lakes or spa centres are not included in the sample). Our analysis implies that tourism development affects the local housing markets through increasing pressure on housing demand for both existing stock and the interest of investors in new accommodation facilities. Also, property values could reflect the capitalization of tourist related amenities and more developed urban infrastructure. The results are to be interpreted with caution, since tourism does not need to be the only economic activity that enhances the growth of a local economy and translates into higher housing prices.

6 Conclusion

In this study we test the relevance of various demand and supply factors for explaining cross-sectional and time dimension of housing price variability. In addition, we examine the impact of public service provision and infrastructural amenities on housing prices. We perform a panel data analysis to empirically estimate the effects of potential drivers of real estate prices dynamics, controlling for the fixed effects at the city level. We analyze a unique panel data set covering subnational housing markets data and various economic, demographic and infrastructural factors for 24 Serbian cities.

We have identified a number of robust price determinants, our findings being consistent with previous research evidence and in line with theoretical assumptions. The analysis provides strong empirical evidence that labour income differences across Serbian cities, portrayed by average wages and registered employment, stimulate the real estate prices. Furthermore, demand-side determinants such as population/stock ratio and infrastructural amenities appear to be significant determinants of the square meter price of the property in a given municipality. Our estimations also confirm a positive relation between the construction activity and the property values, indicating that the housing supply cannot satisfy the excessive housing demand in top-tier cities. We argue that a significant share of the real estate price variability may be attributed to investment or speculative housing demand. This is confirmed by the positive inflation coefficients in the specified models, indicating that house buyers view housing as an investment asset, hedging against inflation. The negative impact of interest rates in models where this variable is significant could reflect the fact that the increasing finance costs act more prohibitively to housing demand than to supply. Economic activity of the given municipality, expressed through average wage growth, labour market activity rates, the size of local government budgets and the intensity of economic activity measured by tourism development appear to drive the housing demand and consequently, increase property value.

By providing evidence on the underlying drivers of property price dynamics, our analysis offers some important insights from the policy perspective. Given that booming real estate prices pose realistic threats not only to the housing market and macroeconomic trends but could potentially magnify the existing economic and social inequalities, it is of particular importance that the regulators monitor the dynamics of the housing market. Our findings indicate that the increasing property prices in Serbian cities are to a large extent determined by the demand-side pressures. Possible interventions could include adjusting monetary policies as long-term financial tools to control the demand driven price increases. In addition, as public service and infrastructure quality seem to be capitalized into housing prices in the cities in our sample, it can be reasonably assumed that the changes in local public expenditure structure could have an impact on sublevel housing markets. Public services should be upgraded in terms of providing high-quality facilities and infrastructure in underdeveloped urban areas, that would call for the redistribution and reallocation of the local governments’ functions.