Discovering the determinants of house prices dynamics in poland using bayesian model averaging

Размещено на http://www.allbest.ru/

Размещено на http://www.allbest.ru/

Poznan university of economics and business

Cracow university of economics

Calisia university

Discovering the determinants of house prices dynamics in poland using bayesian model averaging

Torun

Kalisy

Abstract

We examine the price drivers in 18 capital housing markets in Poland from 2004 to 2021. Using the Bayesian Model Averaging method, we discovered that some financial elements explain house price changes more consistently than conventional economic and demographic variables. The dynamics of house prices in most cities were defined explicitly by financial factors - mortgage market activity, mortgage interest rate, and terms and conditions of granting mortgages. Additionally, the stock market played a vital role in house price movement in selected cities. Economic (regional GDP, salaries, unemployment) and demographic (migration) factors were significant predictors of housing price dynamics only for particular cities. Our study closes the knowledge gap on factors influencing home price dynamics in Poland and other developing nations that have gone through systemic change. In the case of Eastern European countries, we have limited knowledge of those linkages.

Main part

Recent economic, pandemic and geopolitical turbulences reignited the debate about demand and supply drivers contributing to house price dynamics. At least since the 1970s, economists have tried to capture several demographic, socioeconomic, institutional, and financial factors' roles in explaining housing price movements on a regional level. Those empirical affords were accompanied by significant theoretical contributions regarding the linkages between various mechanisms of the housing market. The housing markets have also evolved, thanks to lifestyle and cultural changes, the maturing mortgage industry and institutional shifts in many countries. The well-established relations found in classic papers do not seem to fit the current state of the housing market economy.

Moreover, the economic evidence is geographically unbalanced. The housing markets in developing countries or emerging economies have been under-researched compared to their mature markets in Western Europe and the United States. In many cases, the empirical evidence concerning the house price dynamics is relatively limited, primarily due to data availability issues. For example, in the case of Poland (or other countries that have transformed from socialist to market economies in Central and Eastern Europe, for that matter), the housing market had not fully operated before 1990, and only in the late 1990s and early 2000s the market-related data allowed for empirical investigation concerning house price dynamics and its drivers.

The paper aims to identify and evaluate the impact of selected supply and demand drivers on house price dynamics in Poland. Additionally, we compare the importance of house price predictors in 18 regional housing markets to find common and idiosyncratic factors that explain house price movements in given cities.

Our research narrows the gap in the body of knowledge on determinants of house prices in Poland and other emerging economies that have undergone system transformation. The paper contributes to the ongoing debate on determinants of house price dynamics using the Bayesian Model Averaging atheoretical approach (Koop, 2003). It allows us to account for uncertainty and accommodate a broad range of possible econometric model specifications, as discussed in Steel (2020). The empirical investigation is based on the novel database that contains over three million apartment offers in 18 Polish regional housing markets from 2004 to 2021. This allows us to cover a more extended period than studies that used standard house price information from the National Bank of Poland.

The rest of the paper is organised straightforwardly. Section 2 highlights the economic literature investigating the drivers of house prices. This synthetic overview aims to identify explanatory variables and discover potential research gaps. We focus primarily on the recent empirical studies that used a similar methodology. Section 3 discusses the novel house price dataset used in the practical part of the paper and the Bayesian Model Averaging Approach applied. Section 4 presents econometric results and compares them with the relevant literature. Last but not least, we summarise the findings and pinpoint their limitations in the Conclusion part of the paper.

The economic literature on factors affecting house prices distinguishes between two categories of drivers. The first category comprises various structural, locational, and environmental factors that capitalise on house prices and are helpful when explaining the values of properties within the local housing market. Within this category, we find property characteristics, neighbourhood effects and locational accessibility. The second category encompasses a broad selection of fundamentals that explain house price dynamics on the national and regional levels. Various economic, demographic and financial variables and exogenous institutional and regulatory changes are within this category. The empirical research on house prices typically focuses on either of two types and rarely overlaps.

The economic fundamentals of house price dynamics have been carefully investigated in empirical research. The list of factors considered in previous papers is immense. Based on prior literature, house prices are positively related to household incomes, industrial production, economic activity, and GDP and are negatively linked to unemployment. The economic drivers of house price house dynamics include disposable incomes (Abelson et al., 2005; Hamid et al., 2022; Mikhed & Zemcik, 2009; Oikarinen et al., 2018), gross domestic product (Alpha Kabine, 2022; Jarocinski & Smets, 2008; Nyakabawo et al., 2015) or industrial production (Bork & M0ller, 2015; Rapach & Strauss, 2009). Furthermore, according to various economic research, the unemployment rate negatively affects house prices (Engerstam, 2021; Mikhed & Zemcik, 2009; Mohan et al., 2019a; Plakandaras et al., 2015). In general, the evidence from developed economies reveals that house prices rise in expansion and fall in recessions (Hirata et al., 2013). Previous research in Poland has reported the links between economic growth and house prices (Belej & Cellmer, 2014; Tomal, 2019; Trojanek, 2010).

Economic literature recognises housing as an asset class, similar to stocks or bonds (Jorda et al., 2019). Therefore, from an investment perspective, the housing market is an alternative to the stock market. To reflect this fact, many empirical papers seek to understand the house price dynamics by considering the stock indices (Bagliano & Morana, 2012; Hirata et al., 2013; Mikhed & Zemcik, 2009; Zhang et al., 2012).

Extensive empirical evidence shows that inflation is vital in explaining house price movements. Housing has long been considered a hedge against expectable and unexpected inflation. The literature addressed the role of consumer price inflation when predicting house price fluctuations (Alpha Kabine, 2022; Bagliano & Morana, 2012; Melnychenko et al., 2022; Mohan et al., 2019b; Plakandaras et al., 2015).

The literature addressed the role of a wide range of socio-demographic fundamentals. For example, labour force and population growth have been used to predict house price dynamics (Borowiecki, 2009; Capozza et al., 2002; Plakandaras et al., 2015). In addition, some paper has addressed more subtle demographic facets, such as the role of migration (Han et al., 2018; Laurinavicius et al., 2022; Taltavull de la Paz & White, 2012).

The geographic constraints are instrumental to house price elasticity (Oikarinen et al., 2018). Land supply influenced by physical and regulatory conditions is essential to house prices (Saiz, 2010). Land supply (Glaeser et al., 2008; Han et al., 2018). Empirical studies on housing supply also focused on construction costs and their short-term and long-run relations with house prices (Capozza et al., 2002; Oikarinen, 2010). The recent papers addressed the impact of effective exchange rates on house prices. The former may be linked to housing dynamics due to currency depreciation and inflationary effects on construction materials. The effect is positively related to the dependency of a particular economy on necessary construction materials (Bahmani-Oskooee & Wu, 2018). Prior studies in Poland addressed the linkages between construction costs and house prices (Brzezicka et al., 2018).

The house is a significant asset of a typical household and often requires extensive external financing (Campbell & Cocco, 2003; Rouwendal, 2009). Therefore, financial factors were always considered a key to understanding house price dynamics. The interest in investigating complex economic and housing market relations has only increased since the Financial Crisis 2008. Borrowing constraints are essential to comprehend housing investment on a household level but are also instrumental for housing market dynamics (Khan & Rouillard, 2018). In general, dependence on mortgage financing makes housing investments very sensitive to interest rates and terms and conditions of credit lending. There is a strong relationship between the effective cost of mortgage and house price dynamics (Bork & M0ller, 2015; Cuestas, 2017; Jarocinski & Smets, 2008; Lin & Tsai, 2021; Mian & Sufi, 2011; Mikhed & Zemcik, 2009; Mukayev et al., 2022; Oikarinen, 2009; Tsatsaronis & Zhu, 2004). The prior research focused on the role of credit supply (Gimeno & Martmez-Carrascal, 2010). Still, it was also concerned with the impact of mortgage industry practices, subprime lending, extensive securitisation, and fraudulent mortgage origination that helped understand the housing crisis of 2008 (Griffin et al., 2021). The Spanish evidence suggests a strong link between foreign capital flows, credit supply and house prices (Cuestas, 2017).

Although fundamental factors tend to explain house price dynamics, there is substantial evidence that in the case of some local housing markets, the connection is relatively weaker (Rapach & Strauss, 2009). It is worth noting that although the selection of variables was often similar, the choice depended on the level of analysis (national vs regional) and data availability (country-specific).

Aside from many instrumental factors determining house prices, they are not driven by market forces alone. The housing market is highly regulated and subject to various policy instruments that generate exogenous demand and supply shocks. The impact of different housing policy tools on house price dynamics has been addressed empirically (Liu et al., 2020).

As predictable as it may seem, the housing market dynamics are not entirely driven by traditional economic, financial and socio-demographic variables that explain the supply and demand side of the market. Aside from those fundamentals, the housing market is likely driven by human sentiments, beliefs and expectations about future prices (Griffin et al., 2021). At least since the seminal work of Shiller, there has been consensus about the role of speculation in house price formation, especially in the booming market (Glaeser & Nathanson, 2017; Shiller, 2014).

In recent years, much research has investigated the impact of the COVID-19 pandemic on house prices (Qian et al., 2021; Trojanek et al., 2021), but that also applies to a larger category of epidemic events that had exogenously affected the housing market - Spanish Flu and Cholera Outbreaks being the most prominent examples (Meen et al., 2016). Last but not least, the economic literature explored the role of exogenous shocks caused by natural catastrophes like earthquakes, cyclones, tsunamis, volcano eruptions (Gluszak, 2018) and anthropogenic ones like industrial accidents (Fink & Stratmann, 2015) or wars (Trojanek & Gluszak, 2022) - mostly on local housing markets.

To date, few studies have attempted to explain the role of latent factors in house price dynamics using the Bayesian Model Averaging framework. The mainstream economic evidence comes from various geographical locations.

The Pan-European BMA study explored the role of macroeconomic, monetary, and demographic house price fundamentals (Risse & Kern, 2016). In particular, they studied the role of household incomes, GDP, industrial production, stock market dynamics, interest rates, exchange rates, money supply, inflation, labour force and unemployment rate. The study revealed that no single variable was a fundamental driver of house prices in all countries during a study period. Macroeconomic variables explained house prices better in Belgium, France, Germany, and Italy, whereas financial fundamentals were more strongly linked to housing price dynamics in the Netherlands and Spain.

Table 1. Recent BMA studies on factors that affect house prices

Another BMA used Chinese data to evaluate the forecasting performance of several indicators when predicting house price dynamics (Wei & Cao, 2017). The study used national and city-level economic and financial variables and focused on the market's behavioural side. The results revealed that internet searches are beneficial when predicting house price dynamics. Moreover, the results suggest that they outperformed several macroeconomic variables traditionally used to understand housing market fluctuations.

Swiss BMA study investigated the factors influencing house prices in 169 communities from 1998 to 2004 in the Zurich metropolitan area (Stadelmann, 2010). The study examined the impact of 33 location-specific, fiscal, economic, and socio-demographic variables. The study revealed that apart from several location-specific variables (distance to the urban centre, proximity to shops and schools and air pollution), property prices are positively related to expenditure for culture, health and social well-being and negatively associated with municipal debts. Furthermore, the results suggest that important drivers of house prices are also population incomes and the share of elderly, foreigners and commuters in the population.

Most of the prior studies focused on house price dynamics and performed the analysis at the aggregate level - municipal or national. Alas, successful attempts have been made to narrow the gap in knowledge and bridge the gap between macro and micro determinants of house prices. For example, the Chinese hedonic BMA study, based on individual property data, used individual and aggregate data to investigate variations in property prices. The results show that the availability of education and recreational facilities positively and significantly affects house prices and rents. Additionally, house prices are influenced by the availability of healthcare facilities (Ouyang et al., 2022).

The BMA framework considers the model uncertainty and can be treated as an atheoretical approach (Koop, 2003). A comprehensive and exhaustive presentation of the application of BMA can be found in Steel (2020). At a glance, the idea of the BMA is as follows. This approach deals with the whole space of possible specifications (including and excluding variables from a model). Assuming that we have кlikely regressors (determinants) that can explain the dependent variable, i.e., housing price, we have 2^k possible models. In this study, we were concerned with 41 variables (including lagged variables). It gives us the total number of possible models equal 2⁴¹=2 199 023 255 552 for each of the 18 cities. We are considering the following linear model: у = Xjfy + e, wherey is the vector of observations, Xj is the matrix of explanatory variables, is the vector of parameters, while єis the error term with a normal distribution (Hoeting et al., 1999).

The averaging technique employs the Bayes theorem, where the posterior odds ratio for supported by data over model Mn when the posterior odds ratio is greater than 1. As the model prior, we use the binominal prior with parameter 0.5, which indicates the uniform prior distribution, and in other words, all the variables are equally probable (Fernandez et al., 2001). Further considerations follow us to the posterior probability of the model Mi which is given by the formula:

The posterior density of vector parameters вis equalrp (fy\y) = ZP (M_r\y) E(p_r\y, M_r), which is the average of posterior densities p (Pr\y, M_r). The conditional mean is as follows:Ј(fy\y) = Z P (M_r\y) E(p_r\y, M_r) and the conditional variance is:

are the mean and variance of fir conditional on model Mr, respectively. The additional and beneficial measure is the posterior inclusion probability (PIP), which can be treated as the importance of the variable in explaining the modelled phenomena. The PIP for variable Xi is in the range between 0 and 1 (as a probability) and obtained as a sum of posterior probabilities of models, which includes Xi. In this research, we use the value of PIP to determine the explanatory significance and assume that if PIP>0.66, then the variable is highly probable, 0.33<PIP<0.66 then is medium probable, while for PIP<0.33, we say that variable is lowly probable (Blazejowski et al., 2020).

The results of the BMA procedure have been calculated in Gretl software using the BMA package. The total number of iterations for each city was 3,000,000, with 10% of burn - in draws. The specification of the prior variance matrix was based on the benchmark prior, which is recommended by Fernandez et al. (Fernandez et al., 2001), and the prior average model size is equal to 20.5, which is half of the number of explanatory variables and gives an equal probability for each variable. The total time of estimation and averaging of the models was between 1199 sec and 2007 sec for each city, depending on the town. The whole calculation time was 24468 sec. (using a machine with four real processors and four virtual processors). The Results of PIP values with means, variances, conditional means, and conditional variances obtained in the BMA approach for all investigated cities are in Tables A1 - A18.

This research also uses the widely known procedure X-13ARIMA-SEATS for seasonal adjustment (Gomez & Maravall, 2001). In addition, we exploit the ADF test to evaluate the unit root in the studied time series (Said & Dickey, 1984). The results of the ADF test are presented in Table 2 in the Results and Discussion section.

This study used house price indices built on a unique database of over 3 million housing offers in 18 Polish provincial capital cities from 2004 to 2021 (Trojanek, 2021). Previous research on regional house price dynamics relied on data from the National Bank of Poland (NBP) or the Central Statistical Office (CSO). Since 2013, the NBP house price indices have been part of the Public Statistics Statistical Research Program. They are likely Poland's most established source of information on residential price dynamics, published quarterly since 2010 (data from 3rd quarter 2006).

Nonetheless, we discovered that NBP data did not meet the study's objectives for two reasons. Firstly, there are differences in recorded transaction volume and average house prices between NBP data and complete housing transaction information in the beginning period of the indexes (Gluszak et al., 2018; Hill & Trojanek, 2022; Konawalczuk, 2014). Secondly, the NBP dataset does not include the early 2000s, which we believe is an especially interesting period to investigate. The alternative CSO dataset on house prices was unsuitable for our research owing to its short time series. It covers the dynamics of residential house prices in regional areas since 2015. Furthermore, it only contains data on ownership. Despite the relatively high share of market transactions, sales of cooperative ownership rights to housing units are excluded.

This study used house price indices based on asking about house price dynamics and considering some NBP and CSO information limitations. The idea of asking prices as a source of information for computing housing price indexes is not new (Pollakowski, 1995). There are not many research articles that compare asking and transaction price indexes. One reason for this lack of studies is that such research requires two different datasets, which various institutions often gather; hence, scientific researchers often do not have access to both. A few studies, however, indicate that the offer data are a good reflection of the changes taking place in the property market and offers, which may be an adequate substitute when transaction data are not available (Anenberg & Laufer, 2017; Kolbe et al., 2021; Lyons, 2019; Shimizu et al., 2016).

This study employed house price indices (PRICE) built on a unique database of over 3 million apartment listings in 18 Polish provincial cities from 2004 to 2021. The list of cities included Bialystok (BIA), Bydgoszcz (BYD), Gdansk (GDA), Gorzow Wielkopolski (GOW), Katowice (KAT), Kielce (KIE), Krakow (KRA), Lublin (LUB), Lodz (LOD), Olsztyn (OLS), Opole (OPO), Poznan (POZ), Rzeszow (RZE), Szczecin (SZC), Torun (TOR), Warsaw (WAR), Wroclaw (WRO), Zielona Gora (ZGO). Trojanek (2021; 2022) provides a detailed description of dataset formation.

Table 3. The p-values and integration level for explanatory variables based on the ADF test for all cities

Table 4. The p-values and integration level based on the ADF test (with trend included) for explanatory variables common for all cities (at national aggregation)

Figure 1. Nominal House Price Indices for provincial capital cities in Poland for Q1 2004 - Q4 2021 (Q1 2004 = 1)

price stock market estate

The results of the unit root evaluation provide the following remarks. First, all cities' dependent variables (square meter prices) have a unit root and must be differentiated, meaning they are interpreted as growth rates. A similar finding is for salaries, birthrate, and cost of building for all cities, while 17 of 18 cities (except Poznan) have unit roots for unemployment.

Explanatory variables used in the research were selected based on the literature and availability in the Polish context. Variables available for the entire analysis period were adopted for further analysis, and there was no change in the methodology for their determination. We incorporated three groups of variables. Economic variables potentially affecting housing demand and supply were Gross Domestic Product (GDP), unemployment rate (UNEMPL), gross salaries (SAL), new housing construction (NSUPPLY), and construction costs (CONSTR). Financial predictors were mortgage interest rate (MTR), number of active mortgages (A_MORTGAGE), terms and conditions related to issuing of mortgages (MORTGAGE), the margin of mortgage rate (MARG), and stock exchange index (WIG). Finally, we also accounted for demographic variables: migration rate (MIG) and birthrate (BIRTH). In principle, we tried to collect data on the city level, but the variables were sometimes aggregated on a country level. Data sources and detailed descriptions are presented in Table 2.

Table 2. Variables definitions and data sources

The first step of the research was seasonally adjusting the used time series with the X - 13-ARIMA procedure. Next, we checked the unit root for all variables and cities using the ADF test. If the null was not rejected, what meant that the data containing the unit root, or in other words, are integrated, the time series have been differentiated. The p-values and the integration level are presented in Tables 3 and 4.

Almost all cities are also integrated in the first order for migration. For new houses, only five cities needed differentiation. We obtained unit roots for GDP and WIG for variables at national levels. The ADF test did not indicate the significant unit root for credit interest rate, credit margin, mortgage, and active mortgage. Variables, which are integrated, were differentiated and are noted with the prefix «d_». All variables were transformed using a logarithm. The model for each city has a unique specification, which includes the set of explanatory variables related to this city (in levels or first differences according to Table 2) and a set of variables that are on a national level and are common for all specifications (the differentiation according to Table 3). All variables are included with their lags up to the fourth order.

In Bialystok, a medium-large city in the east part of Poland, we obtained two highly probable variables: salary and mortgage. The medium possible variables are the stock exchange index and an active mortgage. For Bydgoszcz, the highly probable variable is a mortgage (which is a lagged variable and is also medium probable), while birthrate, unemployment, and interest rate are medium probable. Close to Bydgoszcz is Torun, where a highly possible determinant is a mortgage. For this city, medium probable variables are related to the cost of the credit: interest rate and margin. In Gdansk, the city at the seaside, mortgages and active mortgages are highly possible determinants of explaining the price of houses. The stock index is medium probable in Gorzow Wielkopolski, where we obtained no highly possible variables. Similar to Torun, the cost of credit-related variables is medium probable. We have a similar situation for Zielona Gora, which is close to Gorzow Wielkopolski and of comparable size. The margin of credit is a highly possible variable, and the interest rate is medium probable in explaining the house prices. Katowice is a very industrial city in Poland. The mortgage is highly probable in this city, while the interest rate is medium probable. There are no highly probable variables in Kielce, but we have five medium probable determinants: margin, stock exchange index, cost of building, mortgage, and interest rate. The remaining variables are lowly probable. In the second largest city in Poland - Krakow - high probable are active mortgage and cost of building. In contrast, medium probable are new houses and other lags of the active mortgage and interest rate. Lublin is a city with only one variable - the highly probable mortgage. The variable with a medium PIP value is also mortgage (with another lag). There are no highly probable variables in the third largest city in Poland - Lodz in the centre of Poland. Mortgage, interest rate, and margin are medium probable determinants. The prices of houses in Olsztyn depend on the active mortgage, stock exchange index, and mortgage (highly probable). We do not obtain medium probable variables in this city. In Opole, the most probable determinant is the stock exchange index, while active mortgages with different lags are medium probable explanatory variables. The house prices in Poznan, a city above half a million inhabitants, can be modelled by the variable mortgage, which is a highly probable variable, and a lagged mortgage is medium probable. Rzeszow is a city in the southeast part of Poland. The only highly possible variable is a mortgage, with no medium probable variables. The mortgage is a highly probable variable in Szczecin, a northwest city close to the German border and seaside, but not at the seaside. The new houses, birthrate, interest rate, and active mortgage are medium probable variables. In Warsaw, the capital of Poland, the largest and wealthiest city in Poland, the highly probable variables are birthrate, mortgage, new houses, migration, and interest rate. The medium probable determinants are margin, cost of building, GDP, and unemployment rate. We obtained large posterior model sizes for this city. Similar posterior model sizes are for Wroclaw, a fast-developing city in Poland. We received many high and medium probable variables. The active mortgage, new houses, migration, unemployment, and the stock exchange index are highly probable determinants of house prices. Medium possible variables are mortgage, birthrate, cost of building, interest rate, and GDP. Please notice that the variables mentioned above appear with different lags.

The value of particular variables as valid predictors of house price dynamics differed significantly in the study period. Figures 2 to 5 present the posterior inclusion probability values for considered variables. For each city, the highest value of PIP of all lags is shown.

Размещено на http://www.allbest.ru/

Размещено на http://www.allbest.ru/

Figure 2. Posterior inclusion probability (PIP) for macroeconomic variables - Warsaw stock exchange index (WIG), salary (SAL), unemployment rate (UNEMPL) and gross domestic product (GDP) Source: own compilation

Generally, economic variables affect house price dynamics only in selected cities. For example, the Gross Domestic Product, measured on the national level aggregate, was a medium probable predictor of house price dynamics only in Warsaw and Wroclaw. On the other hand, aggregate GDP did not seem to impact house prices in different cities strongly. Variable salary (SAL) is very high probable for Bialystok, and for the remaining towns, this variable is not likely to be a house price determinant. Another economic variable - the employment rate (UNEMPL) - is a highly probable house price predictor in Wroclaw and a medium probable in Bydgoszcz and Warsaw. However, it does not seem to impact house price dynamics in other cities. The stock exchange index (WIG) is classified as a highly probable variable for three cities (Olsztyn, Opole, and Wroclaw). This determinant is medium probable for the following three cities, while the other twelve are lowly probable explanatory. Compared to macroeconomic factors, financial determinants seemed to have a more universal and generally stronger effect on house prices in the study period (Figure 3).

Figure 3. Posterior inclusion probability (PIP) for credit-related variables - interest rate (MTR), active mortgage (A_MORTGAGE), mortgage (MORTGAGE) and credit margin (MARG) Source: own compilation

The cost of granting credit is an essential determinant of tenure choice and strongly impacts house prices. Considering the interest rate and credit margin, we obtained a high inclusion probability for Warsaw (interest rate) and Zielona Gora (margin). The medium probable values are for most of the cities. The detailed values of PIPs for mortgage interest rate (MTR) and credit margin (MARG) are presented in Figure 3. Our findings suggest that the most important explanatory variables are the conditions of granting mortgages - MORTGAGE and the number of active mortgages - A_MORTGAGE. The MORTGAGE determinant is classified as highly probable or medium probable for fourteen of eighteen cities. It is a low probability for only four cities (Gorzow, Krakow, Opole, and Zielona Gora). On the other hand, the number of active mortgages was essential for Gdansk, Krakow, Olsztyn, and Wroclaw. Considering those variables jointly, we conduct the mortgage and active mortgage as the most probable determinants in describing house prices.

The cost of building new houses (CONSTR), measured per square meter, is a highly possible determinant in Krakow, medium possible in Kielce, Warszawa, and Wroclaw. However, for all remaining cities, the cost of building is not a likely house price determinant. In Warszawa and Wroclaw, the number of new houses built (NSUPPLY) is a high probable determinant, and for Krakow and Szczecin, it is close to a high probable threshold. However, for other cities, this variable is an unlikely explanatory variable of house prices (Figure 4).

Figure 4. Posterior inclusion probability (PIP) for supply variables - construction cost (CONSTR) and new houses (NSUPPLY)

In the paper, we evaluated the role of demographic factors, and the results generally suggest that their importance as house price determinants is minor (Figure 5).

Figure 5. Posterior inclusion probability (PIP) for demographic variables - birthrate (BIRTH) and migration (MIG)

The birthrate (BIRTH) variable is an important determinant for house price dynamics only in Warsaw. In Wroclaw and Bydgoszcz, the birthrate PIP is a medium inclusion probability. A similar variable type is migration, which is highly probable for Warszawa and Wroclaw. These two cities are the fastest developing and increasing in Poland. Therefore, from the model averaging point of view, these cities have the highest average model sizes (more variables are included).

The empirical findings shed new light on the dynamics of house prices in regional markets in Poland. The changes in Polish cities were linked to several factors. However, only some seem to have a more general effect in most markets investigated. Additionally, no single variable was a fundamental driver of house prices in all cities. This finding is in line with Risse and Kern (2016). The summary of the importance of selected predictors of city-level house price dynamics in Poland is presented in Table 4.

Generally, financial variables outperformed fundamental economic and demographic variables when predicting house price movements). Similar evidence was found in multiple papers (Bork & M0ller, 2015; Cuestas, 2017; Mian & Sufi, 2011). As expected, we found solid evidence of the linkage between the mortgage cost (interest rate - MTR and mortgage margin - MARG) and city-level house prices in the study period (probable predictor in 11 and 7 cities, respectively). Moreover, costs and terms and conditions of granting mortgages (MORTGAGE) were important predictors of house price dynamics in eleven cities. We also found that active mortgages (A_MORTGAGE) were vital in predicting house price movements in seven out of eighteen cities examined. We also found that the stock exchange index (WIG) was the strong determinant of house price dynamics in several regional housing markets in Poland (6 cities).

Economic factors also helped to understand house price movement, but the effect varied and was observed only in selected cities. Nevertheless, the results suggest that GDP, gross salaries (SAL), and unemployment rate (UNEMPL) in some cities affect house prices. The relatively weaker explanatory role of traditional fundamental factors in predicting house price dynamics was recently observed by Wei & Cao (Wei & Cao, 2017) and Rapach and Strauss (Rapach & Strauss, 2009). The findings extend the understanding of the relationship between macroeconomic variables and house price dynamics found in other Polish studies (Belej & Cellmer, 2014; Tomal, 2019). The supply side was also instrumental in predicting house price movements in some of the regional markets in the sample, which is in line with prior economic research (Capozza et al., 2002; Oikarinen, 2010). We found that the supply of new housing units (NSUPPLY) and construction costs (CONSTR) affect house price dynamics in at least four cities.

Demographic factors seem to play a minor role, which may contrast with prior literature (Han et al., 2018; Plakandaras et al., 2015). Furthermore, only in two cities (Warsaw and Wroclaw), which experienced a significant migration, did MIG impact house price dynamics strongly in the study period. On the other hand, the birthrate (BIRTH) helped to explain the house price movements in four out of eighteen cities examined.

Table 5. The relative importance of selected house price dynamics predictors

The research narrows the gap in the knowledge of fundamental drivers of house price dynamics, especially when emerging markets are concerned. Moreover, the findings extend the understanding of the determinants of house price dynamics found in other Polish studies (Belej & Cellmer, 2014; Tomal, 2019). Nonetheless, it is the first systematic attempt to address the supply and demand factors affecting house prices in Poland. It also adds to the relatively limited number of papers using the Bayesian Model Averaging approach to investigate house price dynamics.

The paper investigated house price dynamics in 18 regional housing markets in Poland from 2004 to 2021. Using the Bayesian Model Averaging approach, we found that selected financial factors are more critical in explaining house price dynamics than traditional economic and demographic variables. In particular, the mortgage interest rates and terms and conditions of granting mortgages helped predict the house price movements in most cities. On the other hand, economic factors (GDP, salaries, unemployment, construction cost and construction activity) and demographic factors (migrations, birthrate) were instrumental predictors of house price dynamics only for selected cities.

The study has some limitations. Due to data availability, we could not track several interesting predictors of house price movements in the literature. Future research could focus on market search behaviour (using Google search indices) for future house price developments. Recently, more empirical attention has been focused on the role of exogenous factors in house price dynamics. Using recent data could help to understand the impact of pandemics and war on house prices. Most certainly, adding the data from other cities in Central and Eastern European countries would add more generality to the research.

References

1. Abelson, P., Joyeux, R., Milunovich, G., & Chung, D. (2005). Explaining House Prices in Australia: 1970-2003. Economic Record, 81 (s1), S96-S103.

2. Alpha Kabine, C. (2022). Determinants of house prices in Malaysia. International Journal of Housing Markets and Analysis, 16 (1), 85-99. https://doi.org/10.1108/IJHMA-09-2021 - 0105

3. Anenberg, E., & Laufer, S. (2017). A More Timely House Price Index. Review of Economics and Statistics, 99 (4), 722-734. https://doi.org/10.1162/REST

4. Bagliano, F.C., & Morana, C. (2012). The Great Recession: US dynamics and spillovers to the world economy. Journal of Banking & Finance, 36 (1), 1-13.

5. Bahmani-Oskooee, M., & Wu, T.-P. (2018). Housing prices and real effective exchange rates in 18 OECD countries: A bootstrap multivariate panel Granger causality. Economic Analysis and Policy, 60, 119-126. https://doi.org/10.1016Zj.eap.2018.09.005

6. Belej, M., & Cellmer, R. (2014). the Effect of Macroeconomic Factors on Changes in Real Estate Prices-Response and Interaction. Oeconomia, 13 (2), 5-16.

7. Blazejowski, M., Kwiatkowski, J., & Kufel, P. (2020). BACE and BMA variable selection and forecasting for UK money demand and inflation with gretl. Econometrics, 8 (2). https://doi.org/10.3390/econometrics8020021

8. Bork, L., & M0ller, S.V. (2015). Forecasting house prices in the 50 states using Dynamic Model Averaging and Dynamic Model Selection. International Journal of Forecasting, 31 (1), 63-78

9. Borowiecki, K.J. (2009). The Determinants of House Prices and Construction: An Empirical Investigation of the Swiss Housing Economy 1. International Real Estate Review, 12 (3), 193-220.

10. Brzezicka, J., Wisniewski, R., & Figurska, M. (2018). Disequilibrium in the real estate market: Evidence from Poland. Land Use Policy, 78 (June), 515-531.

11. Campbell, J.Y., & Cocco, J.F. (2003). Household Risk Management and Optimal Mortgage Choice. Quarterly Journal of Economics, 118 (November), 1449-1494.

12. Capozza, D.R., Hendershott, P.H., Mack, C., & Mayer, C.J. (2002). Determinants of Real House Price Dynamics. NBER Working Paper Series, October, 1-35.

13. Cuestas, J.C. (2017). House prices and capital inflows in Spain during the boom: Evidence from a cointegrated VAR and a structural Bayesian VAR. Journal of Housing Economics, 37, 22-28. https://doi.org/10.1016/jjhe.2017.04.002

14. Engerstam, S. (2021). Long run apartment price dynamics in Swedish and German cities. Journal of European Real Estate Research, 14 (3), 309-330.

15. Fernandez, C., Ley, E., & Steel, M.F.J. (2001). Benchmark priors for Bayesian model averaging. Journal of Econometrics, 100, 381-427.

16. Fink, A., & Stratmann, T. (2015). U.S. housing prices and the Fukushima nuclear accident. Journal of Economic Behavior & Organization, 117, 309-326.

17. Gimeno, R., & Martmez-Carrascal, C. (2010). The relationship between house prices and house purchase loans: The Spanish case. Journal of Banking & Finance, 34 (8), 1849-1855. https://doi.org/10.1016/jjbankfin.2009.12.011

18. Glaeser, E.L., Gyourko, J., & Saiz, A. (2008). Housing supply and housing bubbles. Journal of Urban Economics, 64 (2), 198-217.

19. Glaeser, E.L., & Nathanson, C.G. (2017). An extrapolative model of house price dynamics. Journal of Financial Economics, 126 (1), 147-170.

20. Gluszak, M. (2018). Natural Disasters, Externalities, and Real Estate Prices. World of Real Estate Journal, 106, 65-72. https://doi.org/10.14659/WOREJ.2018.106.009

21. Gluszak, M., Czerski, J., & Zygmunt, M. (2018). Repeat sales index for residential real estate in Krakow. Oeconomia Copernicana, 9 (1), 29.

22. Gomez, V., & Maravall, A. (2001). Automatic Methods for Univariate Series (No. 9808; Servicio de Estudios).

23. Griffin, J.M., Kruger, S., & Maturana, G. (2021). What drove the 2003-2006 house price boom and subsequent collapse? Disentangling competing explanations. Journal of Financial Economics, 141 (3), 1007-1035. https://doi.org/10.1016/jjfineco.2020.06.014

24. Hamid, N., Razali, M.N., Azmi, F.A., Daud, S.Z., & Yunus, N.M. (2022). Prospecting Housing Bubbles in Malaysia. Real Estate Management and Valuation, 30 (4), 74-88. https://doi.org/10.2478/remav-2022-0030

25. Han, B., Han, L., & Zhu, G. (2018). Housing Price and Fundamentals in a Transition Economy: The Case of the Beijing Market. International Economic Review, 59 (3), 1653-1677. https://doi.org/10.1111/iere.12316

26. Hill, R., & Trojanek, R. (2022). An Evaluation of Competing Methods for Constructing House Price Indexes: The Case of Warsaw. Land Use Policy, 120, 106226.

27. Hirata, H., Kose, M.A., Otrok, C., & Terrones, M.E. (2013). Global House Price Fluctuations: Synchronization and Determinants. NBER International Seminar on Macroeconomics, 9 (1), 119-166. https://doi.org/10.1086/669585

28. Hoeting, J., Madigan, D., Raftery, A., & Volinsky, C. (1999). Bayesian Model Averaging: A Tutorial. Statistical Science, 14 (4), 382-417.

29. Jarocinski, M., & Smets, F. (2008). House prices and the stance of monetary policy by Marek Jarocinski and Frank Smets. ECB Working Paper Series, 891.

30. Jorda, O., Knoll, K., Kuvshinov, D., Schularick, M., & Taylor, A.M. (2019). The Rate of Return on Everything, 1870-2015. The Quarterly Journal of Economics, 134 (3), 12251298. https://doi.org/10.1093/qje/qjz012

31. Khan, H., & Rouillard, J.-F. (2018). Household borrowing constraints and residential investment dynamics. Journal of Economic Dynamics and Control, 95, 1-18. https://doi.org/10.1016/jjedc.2018.07.007

32. Kolbe, J., Schulz, R., Wersing, M., & Werwatz, A. (2021). Real estate listings and their usefulness for hedonic regressions. Empirical Economics.

33. Konawalczuk, J. (2014). Ocena wybranych baz danych prywatnych i publicznych wykorzystywanych do sporzqdzania w latach 2006-2012 raportow rynkowych dotyczqcych transakcji sprzedazy lokali mieszkalnych - na przykladzie Katowic. Studia Ekonomiczne, 204, 80-95.

34. Koop, G. (2003). Bayesian Econometrics. John Wiley & Sons Ltd.

35. Laurinavicius, A., Laurinavicius, A., & Laurinavicius, A. (2022). Macroeconomic Variables Influencing Housing Prices in Vilnius. International Journal of Strategic Property Management, 26 (1), 24-34. https://doi.org/10.3846/ijspm.2021.15961

36. Lin, C.C., & Tsai, I.C. (2021). House prices, rental costs, and mortgage interest rates. International Journal of Strategic Property Management, 25 (5), 356-368.

37. Liu, J., Dai, W., Li, X., Dong, Z., & Dong, J. (2020). The Differentiated Effect of Administrative Policy in China's Housing Market - Based on the Heterogeneous Households Multi-Agent Model. Journal of Systems Science and Complexity, 33 (1), 167195. https://doi.org/10.1007/s11424-019-8228-7

38. Lyons, R.C. (2019). Can list prices accurately capture housing price trends? Insights from extreme markets conditions. Finance Research Letters, 30, 228-232.

39. Meen, G., Gibb, K., Leishman, C., & Nygaard, C. (2016). Wars, Epidemics and Early Housing Policy: The Long-Run Effects of Temporary Disturbances. In G. Meen, K. Gibb, C. Leishman, & C.A. Nygaard (Eds.), Housing Economics: A Historical Approach (pp. 111-135). Palgrave Macmillan UK. https://doi.org/10.1057/978-1-137-47271-7_5

40. Melnychenko, O., Osadcha, T., Kovalyov, A., & Matskul, V. (2022). Dependence of Housing Real Estate Prices on Inflation as One of the Most Important Factors: Poland's Case. Real Estate Management and Valuation, 30 (4), 25-41. https://doi.org/10.2478/remav-2022 - 0027

41. Mian, A., & Sufi, A. (2011). House Prices, Home Equity-Based Borrowing, and the US Household Leverage Crisis. American Economic Review, 101 (5), 2132-2156.

42. Mikhed, V., & Zemcik, P. (2009). Do house prices reflect fundamentals? Aggregate and panel data evidence. Journal of Housing Economics, 18 (2), 140-149.

43. Mohan, S., Hutson, A., MacDonald, I., & Lin, C.C. (2019a). Impact of macroeconomic indicators on housing prices. International Journal of Housing Markets and Analysis, 12 (6), 1055-1071. https://doi.org/10.1108/IJHMA-09-2018-0070

44. Mohan, S., Hutson, A., MacDonald, I., & Lin, C.C. (2019b). Impact of macroeconomic indicators on housing prices. International Journal of Housing Markets and Analysis, 12 (6), 1055-1071. https://doi.org/10.1108/IJHMA-09-2018-0070

45. Mukayev, D., Shalbolova, U., & Tanas, J. (2022). Housing affordability analysis in Kazakhstan based on D-HAQ model. Journal of International Studies, 15 (4), 28-42. https://doi. Org/10.14254/2071-8330.2022/15-4/2 Nyakabawo, W., Miller, S.M., Balcilar, M., Das, S., & Gupta, R. (2015). Temporal causality between house prices and output in the US: A bootstrap rolling-window approach. The North American Journal of Economics and Finance, 33, 55-73.

46. https://doi.org/10.1016/_j.najef.2015.03.001

47. Oikarinen, E. (2009). Household borrowing and metropolitan housing price dynamics - Empirical evidence from Helsinki. Journal of Housing Economics, 18 (2), 126-139. https://doi.org/10.1016/jjhe.2009.04.001

48. Oikarinen, E. (2010). An Econometric Examination on the Share of Land Value of Singlefamily Housing Prices in Helsinki. Finnish Journal of Social Research, 3 (SE-Articles), 7-18. https://doi.org/10.51815/fjsr.110696

49. Oikarinen, E., Bourassa, S.C., Hoesli, M., & Engblom, J. (2018). U.S. metropolitan house price dynamics. Journal of Urban Economics, 105, 54-69.

50. Ouyang, Y., Cai, H., Yu, X., & Li, Z. (2022). Capitalization of social infrastructure into China's urban and rural housing values: Empirical evidence from Bayesian Model Averaging. Economic Modelling, 107 (April 2021), 105706.

51. https://doi.org/10.1016/j.econmod.2021.105706 Plakandaras, V., Gupta, R., Gogas, P., & Papadimitriou, T. (2015). Forecasting the U.S. real house price index. Economic Modelling, 45, 259-267.

52. Pollakowski, H.O. (1995). Data Sources for Measuring House Price Changes. Journal of Housing Research, 6 (3), 377-387.