Effect of the EMU accession on residential real estate prices in CEE countries

(1)

Effect of the EMU accession on residential real estate prices in

CEE countries

Master’s Thesis

MSc Finance, Real Estate track

Written by: Justinas Bernotas, 11823607

Thesis supervisor: dr. M. Constantinescu

August 2018

Amsterdam

(2)

2

Statement of Originality

This document is written by Student Justinas Bernotas who declares to take full responsibility for

the contents of this document.

I declare that the text and the work presented in this document are original and that no sources

other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of

the work, not for the contents.

(3)

3

Abstract

This paper examines the effect of the accession to European Monetary Union (EMU) for

residential real estate prices in Central and Eastern European (CEE) countries. The paper

discusses possible transmission channels through which the change in monetary policy and the

adoption of the Euro currency could affect housing markets and dynamics of house prices.

Empirical research was carried out using data of Estonia, Latvia, Lithuania and Slovakia from

2015:Q1 to 2017:Q1 and estimated using Panel OLS fixed effects and System-GMM estimators.

The paper finds that the development of dwelling prices in the CEE countries have been affected

negatively by the changes in monetary policy, yet only estimations obtained by System-GMM

exhibit statistically significant relationship. The results suggest that the EMU accession effect on

house prices could have been offset due to a currency change anticipation after countries joined

the EU and pegged their currencies to Euro, as well as due to a stronger regulation of housing

markets by National Central Banks.

Keywords: EMU, monetary policy, transmission channels, house prices, CEE countries, fixed

effects, System-GMM

(4)

4

In 1957 the political and economic union of Europe – the European Union (EU) was created. The main objective of the creation of the EU, was to create a joint market to trade for all the members of the EU. However, as time passed, the need for closer collaboration in terms of economic and monetary policies, so the market could continue to develop and flourish, became obvious.

As a measure, in 1991 EU members approved the Maastricht Treaty, which was a stepping stone for the creation of the common currency economy – the Eurozone. EU members are only allowed to join the Eurozone if their public finances are within the criteria set by Stability and Growth Pact - which objective is to maintain inflation and long-term interest rates within certain values. In 1999 11 EU members, which met the euro convergence criteria, joined the Eurozone. In the years to follow, several more countries joined the common currency economy. The influx of 10 new members, which were mostly CEE countries, saw one condition for their integration to the EU – new members agreed to join the Eurozone in the future. Three of these countries – Poland, Hungary and Czech Republic – have yet to adopt the Euro, as they are still in process of meeting the Maastricht criteria, while seven countries – Slovenia, Malta, Cyprus, Slovakia, Estonia, Latvia and Lithuania have already joined the Eurozone.

The residential real estate boom and bust, seen in recent years, has refueled the discussion regarding the most important factors of dwelling prices, the importance of the housing market on the general economy and the regulatory measures of the monetary policy.

The case of CEE countries is of special interest, as in the 21st century these countries experienced rapid expansion of the economy and the housing market, fueled by a steady growth of residential investment, consumption and credit availability. As the aftermath of 2008 financial crisis revealed, CEE countries were one of the most affected by the housing crash both regarding the decline of the real estate market and the slump of the general economy. As most of the CEE countries recently joined the European Monetary Union, it raises concerns about the loss of monetary autonomy. In countries with their own national currency and monetary policy, the central bank can increase interest rates to slow down the growth (even though the aftermath of the financial crisis indicated that Central Banks have failed to control the rapidly heating housing market) or to respond adequately to a sudden dip of housing prices. Whereas in the EMU, the European Central Bank is responsible for setting rates that are applied for all members of the Euro area despite their size and the peculiar nature of the economy. Therefore, the monetary policy of the ECB cannot be the first measure to counter the negative sector or country-specific shocks.

(6)

6 The benefits of the Eurozone for its members countries are seen in many areas, but mainly it is prominent in economic growth and stability, broader possibilities for markets and businesses, more significant financial integration and overall a better access to the EU market, as well as the global market.

Naturally, the adoption of Euro also brings certain risks, especially to the smaller economies. Due to better financial integration with Eurozone members, the financial convergence may lead to a higher inflation. Such situation was widely reported in the countries which recently adopted the Euro, as prices in these countries were catching up with price levels of the Euro area. As the Euro adoption, at least in the short run, affects the inflation of the newly joined countries, this maybe the case for housing prices, as well. As already discussed, the effect of monetary policy change through the adoption of the common currency and the extent of it is especially important for the more vulnerable and less economically developed CEE countries, where the fluctuations of house prices were more substantial and impactful to the general health of the economy in the context of the financial crisis of 2008. Thus, it is important to evaluate the significance of the Euro accession for the residential real estate prices in CEE countries, as it could also impact the economic development of these countries, either fueling or restricting the growth.

Therefore, this paper will seek to address the question whether the accession to the EMU was a significant factor to influence the direction of growth for residential real estate prices in CEE countries. The contribution of this paper to the existing literature base can be seen from few different angles. Firstly, according to the author’s knowledge, the specific literature on the effects of a common currency adoption on the dynamics of house prices is non-existent, not only for CEE countries, but for other economies as well. Also, this paper will use the most recent data, which extends to the last quarter of 2017. At last, this paper will provide more insights on the importance of the monetary convergence on small economies, and help to answer the question, whether country specific characteristics are more important for house price dynamics, than the economic and financial linkages of the Eurozone.

This paper is structured as follows – in the following chapter the evidence from the literature on the effects of monetary and financial integration to housing markets, as well as on several possible transmission channels are presented and discussed. Then, the methodology and the data which are used to carry out the empirical research are examined thoroughly. Finally, the findings are concluded using the results obtained through empirical analysis and the literature review. The findings and the implications of them are then discussed, in addition, the limitations of the research are presented as well.

(7)

7

2. Literature review

In the subsequent section, the literature evidence concerning the channels through which the change of monetary policy could affect housing prices, as well as past researches concerning the effects of monetary and financial integration on the housing markets will be presented and discussed.

2.1. Transmission channels of EMU accession effects

The importance of monetary convergence on house prices is widely reviewed in the literature, as most scholars recognize that recent house price developments in the Euro area are to a large extent driven by coordination of monetary policy, financial liberalization, integration, accession to wider international financial markets, as well as interrelationship of global business cycles. (Kim & Renaud, 2009; Tsatsaronis & Zhu, 2004; Scanlon et al., 2008; Helbling & Terrones, 2003).

The necessity to research the effect of the EMU accession is credited not only to its impact for the housing market, but to overall economic health as well, as economists agree, that fluctuation of housing prices is an important factor for the development of the economy. For example, a stable housing market helped the economy to recover after the stock crash of 2001, whilst the strong decline of housing markets around the world caused a recession of the economy in 2007. Not surprisingly, literature also argues that house prices have a significant impact on the economic development, in terms of inter-relation and predictability power. (Miller, 2011; Xu, 2016). In addition, literature evidence shows, that there is a substantial degree of correlation between changes in house prices and changes in the consumption through so called “wealth effect”. (Campbell & Cocco, 2007). As consumption is one of the main drivers of GDP growth, thus the correlation between house prices and consumption means that the development of housing prices is also an important factor in determining the overall economic development. The strong linkage between house prices and economic health is especially important for the smaller economies, researched in this paper. Naturally, the positive effect of the EMU accession on house prices, could be fueling economic growth or vice versa.

As for the synchronization of the monetary policy, the condition and cycles of housing markets are the areas of interest and major concern for policy makers. According to Mishkin (2007), monetary policy and changes in the interest rate set by policy makers can affect housing markets and, therefore the economy of the country, through six different transmission channels. Firstly, the driving forces in the housing market could be affected by through the balance sheet channel, as it could manifest into changes in consumer spending on durables, as well as housing. Also, it can cause “wealth effect” through changes in

(8)

8 house prices, influence the user cost of capital and affect expectations of house prices in the future, and consequently, the supply of the housing market.

The role of credit channel as one of the most important financial transmission mechanisms for financial integration effects is widely reviewed in the literature. Theory indicates, that the adoption of a common currency in 1990s led to a decline of sovereign interest rate spreads in peripheral countries, most evidently Spain and Portugal, compared to core economies (Sufi et al., 2017). Consequently, lower interest rates resulted in the surge of credit availability and, in turn, strongly visible growth of employment in the construction sector. Latter evidence is also strongly supported by the recent research, which finds that accession to the EMU leads to a reduction in long-term interest rate spreads in countries which joined the EU in 2004 or later (Subacius, 2018)

However, earlier researches indicate, that the convergence of interest rates in the EU members and accession countries is evident even before the countries joined the EU. For instance, in the early 2000s Estonia and Lithuania showed a consistent relationship between domestic and Euro area interest rates, whilst Latvia, Czech Republic and Slovakia exhibited of slightly weaker cointegration, yet still showing prospect for the monetary integration in the upcoming years (Holtemöller, 2005). Thesis findings are supported by a later research, which find, that real interest rate in most of the new EU members, except for Latvia, Hungary and Poland, began converging to the EMU average before the accession to the EU (Arghyrou et al., 2009). Thus, this continuous cointegration of interest rates could possibly reduce the shocks of the EMU accession related financial integration.

In case of CEE countries, the benefits of the Euro adoption which would potentially affect real estate prices in both long and short terms include lower transaction and administrative costs, receding risk of the exchange rate, capital inflows from increasing trade volumes and more accessible international capital markets, leading to growth in foreign investments (Ganev, 2009). In addition, a more stable and credible outlook of the monetary policy within the EMU would enhance consumer confidence and increase consumption, including housing expenditure.

Finally, Schadler (2005) argued that the regulative financial institutions of CEE countries which would join the Eurozone, should be equipped with necessary instruments to amortize economic shocks coming from monetary convergence, mainly rapid ascension in credit growth, as interest rates will begin to recede, slowly aligning to the levels of the Eurozone. Consequently, the sudden boom in bank lending may potentially lead to the hazard of a surge in real estate prices and overheating of the housing sector.

(9)

9 However, the countries of interest in this research – Estonia, Latvia, Lithuania and Slovakia – pegged their currencies to Euro several years before they joined EMU – around 2005. As the countries joined the ERM II mechanism, the National Central Banks started adjusting to the monetary policy of the ECB, therefore this, naturally, led to the convergence of inflation and interest rates and lower currency risk (Lamine, 2006).

In addition, in this research the effects of the Euro adoption on credit growth and the reduction in interest rates will be reflected in the control variables. Thus, the estimated effect of the EMU adoption on residential real estate prices may come from the increase in foreign investment, lower transaction and administrative costs or changes in consumer and investor confidence, as a result of more stable currency and monetary policy.

2.3. Convergence trends of housing markets

Even though the theoretical knowledge of the monetary convergence linkage to the dynamics of house prices is well established and discussed in the literature, the literature base concerning the effects of the EMU accession on residential real estate prices is rather scarce. However, there are papers which researched the spillover effects and the convergence of residential real estate prices in the EU, in addition, several papers have tried to estimate the convergence between the real estate equities on the countries that recently joined the EMU.

Looking at the spillover effects and convergence of housing prices, the literature evidence is rather equivocal. However, there is a consensus in the literature, that the country-specific effects play an important role in determining house prices on a national level, or for a given country group, therefore the convergence effects are not eminently evident, especially for smaller economies. For instance, even though the integration of financial and trade markets within the common European Monetary Union would suggest that house prices tend to integrate as well, Vansteenkiste & Hiebert (2011) finds that co-movement of house prices within the Euro area is limited, showing that house prices are affected mainly by country-specific changes. The findings by the latter paper is supported by A. Merikas et al. (2012), who argues that the differences in country-specific living conditions, the general tenure attitude towards residential real estate, regulation of the housing market and the degree of intervention by government institutions limits the importance of the underlying convergence effect on house prices.

However, the opposing opinion in the literature suggests, that there is some degree of convergence in house prices, even within the Euro area. Seltzer et al. (2011) argues, that the integration of the money

(10)

10 market influences housing market integration as well. Likewise, the convergence within OECD countries is even more significant, suggesting that house prices within given sample tend to level out in the long term (Demir & Yildirim, 2017). However, the latter paper again emphasizes the importance of country -specific factors. The findings of substantial convergence within OECD countries is not surprising, as the countries in this group are all highly developed in terms of economy, thus the divergence of economic factors within OECD countries is less substantial than in EU countries, for instance.

Several papers investigating the possible convergence of real estate equities within the Euro area, provided similar findings as the literature on the convergence of house prices, supporting the limited effect of the European monetary integration for the co-movements of real estate equities returns, For instance, McAllister et al. (2003, 2006) argues, that the integration of real estate equities is slow and less visible, than for general equities. Later, the author also found that monetary integration is an insignificant factor to cause convergence in European real estate equities, as non-EU and non-Euro area countries are converging to the same extent as members of the EMU. Moreover, the similar research focusing on EMU establishment effects by Yang et al. (2005), showed, that the integration of real estate equities is visible in the larger and more developed economies, while the smaller economies are not affected by the convergence effects, possibly more influenced by country-specific factors, like changes in demographics and unemployment.

2.4. Similar researches and factors affecting house prices

Most of literature that focuses on the EMU accession impact on economic factors, investigates the effect of Euro adoption for trade, or financial stability (Micco et al., 2003; Faruqee, 2004; Rose, 2000; Evaldsson, 2012). To capture the effect of the Euro adoption on the dynamics of trade volumes or debt levels, the authors used panel data of European countries within the Eurozone and included the dummy variable indicating the accession to EMU for a given country. Most of the papers used dynamic panel data models, like system - GMM proposed by Arellano-Bover and Blundell-Bond to carry out estimations. The common finding throughout the papers, was that the effect of the EMU accession on trade volumes was larger for larger economies, than for the smaller ones.

Finally, looking at the factors which affect house prices, there is a variety of literature evidence on the effects of macroeconomic indicators on residential real estate prices. Naturally, the most important articles are those which examine the countries in question of this research – CEE countries, or Baltic countries and Slovakia, in particular. The study of house price determinants in OECD and CEE countries by Égert & Mihaljek (2007) suggests that variables of interest – GDP, interest rate, availability of credit,

(11)

11 population growth and changes in labor force – have expected signs and are significant. However, the same paper emphasizes, that the effect of above mentioned variables is more significant in CEE countries. Moreover, contrary to most of the literature which focuses on the fundamental factors affecting house prices, the paper found that the change in unemployment is not a statistically significant factor in explaining house prices for some CEE countries. Moreover, Kulikauskas (2016) emphasized income, population, mortgage credit stocks and mortgage interest rates as main factors affecting house prices in the Baltic countries, in the paper which investigated the imbalances of the Baltic residential real estate prices. In addition, Andrews (2010) found, that residential real estate prices in OECD countries are moving in proportion to real income, while the decline in real interest rate and unemployment tend to result in higher house prices. Finally, several more papers came to similar conclusions and emphasized the same most important factors driving house prices (Tsatsaronis & Zhou, 2004; Baffoe-Bonnie, 1998; Grum & Govekarb, 2016).

Evidently, most of the literature suggests, that the main macroeconomic drivers of house prices are GDP, interest rate, unemployment, availability of credit, mortgage interest rates, demographics and construction input.

3. Empirical research

In the following section, the methodology and the data used to perform empirical research and answer the question of this paper is discussed. Later, the results obtained from empirical research are presented and discussed.

3.1. Methodology

Giving the existing literature evidence and the aim of this paper, the main hypothesis of this paper is as follows:

Hypothesis:

• The effect of the EMU accession of real residential real estate prices in CEE countries is significant and positive.

Ideally, the data from all CEE countries which joined the Eurozone following the EU inclusion class of 2004 should be included in the estimation. However, the estimation that brings robust results requires for data at least 3-4 years before and after the Euro adoption. Therefore, the data from four countries which met this requirement – Slovakia, Estonia, Latvia and Lithuania -, is used in the estimation.

(12)

12 The data from latter countries are pooled into a panel dataset, as they share similar economic and housing market characteristics. First, all four countries have a similar structure of the housing market. Looking at the tenure status, all Baltic countries and Slovakia are among the leaders in Europe in the rate of ownership (Table 1).

Table 1. General statistics of the housing markets of the researched countries.

Source: Eurostat (2016)

In Lithuania and Slovakia as many as 90% percent of the population lives in dwellings they own, while in Estonia and Latvia this figure is slightly above 80%. (Eurostat, 2016). In addition, the amount of household debt – which is comprised mostly of mortgage loans – in proportion to GDP is also very similar throughout the countries. It varies from 25% in Latvia, to 43% in Estonia, however, in all countries, this ratio is much lower than the EU average. Finally, the overburden rate of housing costs, which shows the proportion of the population which spends more than 40% of the disposable income on housing costs, indicates all countries of interest are far below the EU average. Only around 7% of the households in these countries (the only exception being Estonia) spend more than 40% of their income to cover housing costs.

In addition, the CEE countries, including Lithuania, Latvia, Slovakia and Estonia have experienced similar and substantial growth of credit lending to residents by commercial banks (ECB, 2009). This growth was supported by the increase in income, growing consumer confidence, favorable tax treatment and housing subsidies in some countries, including Lithuania, Slovakia and Estonia. Also, the growing competition of the banking sector led to the introduction of various credit instruments, which resulted in lower costs of borrowing and more flexible mortgage terms. However, the mortgage market structure in these countries is reliant on the behavior of foreign banks, which comprise most of the financial sector (Aydin, 2008). Given that foreign-owned banks act differently than domestic banks, this could impact the developments and the nature of the lending structure. Therefore, looking at the general statistics of the housing market and the characteristics of the mortgage market, the idea to pool them together into a panel dataset is justified and seems reasonable.

Estonia Latvia Lithuania Slovakia EU average

Home ownership rate 81.4% 80.9% 90.3% 89.5% 69.2%

Debt of households, % of GDP 42.8% 25.8% 29.1% 40.2% 65.4%

Housing cost overburden, % of total

(13)

13 As the main objective of this paper is to estimate the effect of the EMU accession on residential real estate prices, the dependent variable in the regression will be the real HPI – the index, which reflects dynamics of dwelling prices in the countries of interest, deflated by the HCPI. In addition, following the evidence from the literature which concerns the most important factors affecting residential real estate prices, several control variables will be included in the model to capture the fluctuations of house prices. In accordance to the literature, the expected signs of the control variables are as follows:

Positive sign – GDP, Credit, Population, Construction level;

Negative sign – Interest rate on mortgages, Unemployment, Crisis dummy variable.

Furthermore, to mitigate the influence of the financial crisis on the fluctuation of house prices, another dummy variable will be introduced, taking the value of 1 during the slump of the prices in the crisis period and 0 otherwise. Finally, the key variable of interest will be represented by the country’s adoption of the Euro. The EMU dummy variable will equal 1, from the year the country joined the EMU and 0 otherwise. Alternatively, a dummy denoting the day of the official announcement by European Commission, which confirmed that the countries have met convergence criteria for joining the single currency union will be introduced as well. The intention of this is to compare the estimates produced by using either a dummy marking an official adoption of the currency or a dummy denoting the first official announcement by the EU Commission, which was approximately half a year earlier.

The dependent and independent variables will also be transformed into logs, to avoid bias of skewed data and to normalize the variance of variables (Kulikauskas, 2016), except for interest rate and unemployment, which are in percentage values.

Before running the GMM models, several procedures will be performed to test the condition of the dataset. First, panel diagnostics will be applied to the dataset, to test for multicollinearity, autocorrelation, heteroscedasticity and normality of residuals. If these issues are found in the dataset, the standard errors will be clustered, as this solves autocorrelation and heteroscedasticity. Also, the Hausman specification test will help to determine whether Panel OLS should be estimated using fixed or random effects. Then, the modified, if needed, dataset will be estimated using Panel OLS method, to compare the results and robustness with the results estimated via System-GMM model, as proposed by Demir, C., & Yildirim, M. O. (2017 If the coefficients do not change significantly, the model and estimated results can be considered as “robust”. (Lu & White, 2014).

(14)

14 As indicated by diagnostics tests, there are several issues with the data, thus particular measures must be taken before the estimation, to arrive at consistent estimates. Firstly, Hausman specification test suggested that fixed effects method is a preferred specification when estimating the model via Panel OLS. Using STATA statistical software package, the fixed effects estimator will be applied using fe command, which fits fixed effects model using the within regression estimator. Moreover, Kernel density estimator indicated that the distribution of residuals in the sample is non-normal, therefore this should be considered when assessing the results of the analysis. In addition, modified Wald statistic suggested that heteroscedasticity is present in the regression model, while Wooldridge autocorrelation test indicated that autocorrelation is present in the sample. Finally, Variance Inflation Factor (VIF), suggested that there is multicollinearity present between GDP and Population. Therefore, alternative variable for GDP – GDP per capita – is used in alternative models, as VIF indicates, that there is no multicollinearity between GDP per capita and population variables. The results of above mentioned tests can be found in the Appendix A.

The main regression for the estimation of fixed effects Panel OLS is as follows:

𝑙𝑛𝐻𝑃𝐼𝑖𝑡= 𝛼𝑖+ 𝛽1𝑙𝑛𝐺𝐷𝑃𝑖𝑡+ 𝛽2𝑙𝑛𝐶𝑅𝐷𝑖𝑡 + 𝛽3𝐼𝑖𝑡+ 𝛽4𝑈𝑁𝑖𝑡+ 𝛽5𝑙𝑛𝑃𝑂𝑃𝑖𝑡+ 𝛽6𝑙𝑛𝐶𝑂𝑁𝑖𝑡+ 𝛽7𝐸𝑀𝑈𝑡+

𝛽8𝐶𝑅𝑆𝑡+ 𝜀𝑖𝑡 (1)

where:

HPI – Deflated House Price Index; GDP – GDP per capita;

CRD – stock of loans for house purchase; I – interest rate on mortgages;

UN - level of unemployment; POP – level of population (20-64y); CON – volumes of construction;

EMU – dummy variable denoting adoption of the Euro;

CRS – dummy variable denoting period of the housing market crash due to the financial crisis; ε = the error term.

In accordance with the literature, which used panel data and the dummy variable to estimate the effect of the EMU accession, the dynamic panel data estimators will be used to assess the significance of the EMU accession. The dynamic panel data estimation will be performed using Arellano-Bover/Blundell-Bond

(15)

15 one-step estimator based upon System Generalized Method of Moments (System GMM). This method is widely used in the research (Demir & Yildirim, 2017; Evaldson, 2012), focusing on similar questions to this paper. It is used when explanatory variables are not exactly exogenous, which indicates that variables are correlated with past and potentially current values of the residual. In this case, the growth of house prices is correlated with each other on a quarterly basis, and the control variables used in the model are dependent on the dynamics of one another. Thus, System-GMM is well suited to deal with possible endogeneity issues. It is also based on the assumption, that there is no correlation of differenced instruments with fixed effects.

System GMM method uses system of equations, mainly level and differenced equations. In addition, this model uses instruments of the variables in the estimation, which substantially increases efficiency. In this case, the instruments which are introduced are added automatically by STATA program, therefore adding instruments manually is not required. Both level and differenced equations include lagged dependent variable – in this case Real HPI - as an explanatory variable. However, the differenced equation uses first differenced variables as instruments, whereas the level equation uses first-differenced lagged dependent variable as an instrument, while other variables stay at levels.

Therefore, the general regression of System - GMM estimation method is presented below: 𝑙𝑛𝐻𝑃𝐼𝑖𝑡= 𝛼 + 𝛽1𝑙𝑛𝐻𝑃𝐼𝑖𝑡−1+ 𝛽2𝑙𝑛𝐺𝐷𝑃𝑖𝑡+ 𝛽3𝑙𝑛𝐶𝑅𝐷_𝑖𝑡+ 𝛽4𝐼𝑖𝑡+ 𝛽5𝑈𝑁𝑖𝑡+ 𝛽6𝑙𝑛𝑃𝑂𝑃𝑖𝑡+

𝛽7𝑙𝑛𝐶𝑂𝑁𝑖𝑡+ 𝛽8𝐸𝑀𝑈𝑡+ 𝛽9𝐶𝑅𝑆𝑡+ 𝜀𝑖𝑡 (2)

In addition to fixed effects method, the model will also be estimated using time fixed effects, to mitigate for risks of common macroeconomic shocks and cyclical effects affecting the housing market, or in other words, controlling for year effects. This will be done as sometimes panel regressions, which do not control for yearly effects, may be influenced by general trends, which do not reflect causal relationships. Therefore, panel OLS time fixed effects estimations using year dummies (excluding the first year) will help to validate statistical inferences obtained using fixed effects estimations.

Finally, estimations will be compared by excluding credit availability and interest rate variables. As the evidence from literature review indicated, the effect of financial market integration can be transferred either through the credit or interest rate channels. The basic model specification estimated in the empirical research will be controlling for changes in credit availability and interest rates dynamics, thus the possible effect of financial integration cannot be observed through credit channel, unless it is affecting

(16)

16 house prices through other channels. Therefore, it is important to estimate models excluding credit availability and interest rate variables to compare the estimations and attain robust results.

3.2. Data and descriptive statistics

As already mentioned, to carry out the empirical research, data of four countries – Slovakia, Estonia, Latvia and Lithuania – will be used to construct a panel dataset. The data used iwill be from will be from 2005:Q1 to 2017:Q4, as data for the dependent variable - House Price Index (HPI) - is mostly available from that date. The only exception is Latvia, for which the data of HPI is available from 2006Q1, however, the data for HPI was extrapolated for 2005, using the average apartment prices by Oberhaus (2018) and the estimated annual growth of dwelling prices by Eurostat. All the necessary data was freely accessible and collected through various databases, namely Eurostat, ECB Statistical Data Warehouse and the databases of National Central Banks of the countries in question.

As already mentioned, the main variable of interest – EMU Dummy – takes up the value of 1 starting from the day country joined the EMU. In the case of Estonia, Latvia, Lithuania and Slovakia, these dates are January 1st_{of 2011, 2014, 2015 and 2009, respectively.}

However, the first official announcements by the European Commission regarding countries meeting convergence criteria for joining single currency union were made approximately half a year before the official adoption of the Euro currency. To be exact, such announcement in case of Slovakia was made on May 7th_{of 2008, for Estonia on May 12}th_{of 2010, for Latvia on June 4}th_{of 2013 and on the same day of}

2014 for Lithuania. Taking this into consideration, the alternative EMU accession dummy, marking the official announcement by European Commission, will be introduced starting from the Q3 of the year prior to official adoption of the Euro.

To assess the relationship between real changes in RRE prices, real HPI will be used instead of the nominal House Price Index. This will be done by deflating HPI to the corresponding inflation values, measured by Harmonized Consumer Price Index (HCPI), also compiled by Eurostat for European countries. As literature suggests, the control variables used in this paper will be as follows:

Stock of mortgage loans, Mortgage interest rate, GDP, Unemployment, Population and Construction volumes.

(17)

17

Table 2. Information and sources of the variables used in the paper.

Variable Span Description Sources

House Price Index 2005Q1-2017Q4

The HPI shows the changes in dwelling prices bought by households, including

newly-built and existing dwellings.

Eurostat BIS

Real HPI 2005Q1-2017Q4 HPI deflated by the quarterly HICP growth rate.

Eurostat BIS Stock of loans for

house purchase

2005Q1-2017Q4 Monthly outstanding amounts of loans for house purchase averaged by a quarter.

Databases of National Central Banks Interest rate on loans

for house purchase

2005Q1-2017Q4 Weighted average Interest rate on outstanding loans for house purchase

Databases of National Central Banks

GDP, GDP per capita 2005Q1-2017Q4

Quarterly GDP at chain linked prices (2010), adjusted to calendar and seasonal fluctuations or ratio of real GDP to the

average population

Eurostat

HICP 2005Q1-2017Q4

Indicator of inflation for the ECB. It is compiled according to a methodology that has been harmonized across EU countries.

Eurostat

Unemployment rate 2005Q1-2017Q4 Quarterly average of unemployment rate

within active population. Eurostat Population 2005Q1-2017Q4 Quarterly average of the population from

20 to 64 years. Eurostat Construction volumes 2005Q1-2017Q4 Quarterly volume index of construction

production, 2015=100. Eurostat Note: compiled by the author of this paper.

Tables below show the descriptive statistics of two samples of variables used in the empirical research – before (Table 3) and after the countries joined the Euro area (Table 4). The descriptive statistics show the mean, minimum and maximum values for natural variables, as well as the standard deviations. Even though logarithmically transformed variables will be used in the estimation, the descriptive statistics of natural variables helps to better understand the variation and the nature of the factors affecting house

(18)

18 prices. The descriptive statistics for the full sample and logarithmic variables can be found in the

Appendix B.

Table 3. Summary statistics of non-transformed variables before Euro adoption.

Variable (N=116) Mean Std. Dev. Min Max

Real House Price Index 107.1 26.1 65.43 172.7

GDP 7045.4 3645.6 3535.5 17301.3

GDP per capita 2550.9 386.4 1700 3400

Mortgage credit 5040.7 1683.4 1083.8 8406.9

Credit to GDP .216 .097 .061 .425

Index of construction volumes 102.08 25.96 53.7 162.7

Population 1695.9 811.8 806.9 3509.2

interest rate .042 .012 .0181 .0624

Unemployment rate .111 .046 .041 .204

Note: compiled by the author of this paper

It can be observed, that even though the average house prices after EMU accession in the countries are below pre-EMU levels, the variation, as indicated by the lower standard deviation, is substantially less extreme. It could be explained by the fact, that countries joined EMU after financial crisis of 2008, therefore house prices after the Euro adoption were more stable.

Table 4. Summary statistics of non-transformed variables after Euro adoption.

Variable (N=92) Mean Std. Dev. Min Max

Real House Price Index 99.6 10.41 73.2 115.2

GDP 10495.5 6396.0 3870.4 20695

GDP per capita 3181.5 348.3 2300 3900

Mortgage credit 9519.5 5641.1 4366.5 25123.2

Credit to GDP .229 .0675 .137 .367

Index of construction volumes 98.82 9.96 80.8 131.4

Population 2059.76 1225.9 784 3560.3

Interest rate .0305 .0144 .016 .0627

Unemployment rate .101 .0270 .055 .149

Note: compiled by the author of this paper

It is also noticeable, that countries of interest have developed their economies substantially, as the average quarterly GDP grew by almost 50% after the Euro adoption. Also, the stock of mortgages almost

(19)

19 doubled, mainly because of lower interest rates and better economic conditions, including income. However, it must be noted, that these changes in macroeconomic and credit factors have not been the consequence of countries joining the Euro area, but more the outcome of continuing economic development of CEE countries and global financial markets in general.

Before estimating the effect of EMU accession on house prices, it is useful to compare the behavior of house prices before and after the countries joined the Euro area, to examine and capture the possible cases of convergence, once the country joined the common monetary area. Looking at the correlation matrix (Table 5) between real and nominal house prices in CEE countries and the averages of the Euro area, it is worth noting, that the results vary significantly among countries. It must be mentioned that averages of Euro area house prices have been calculated using weighted average of GDP for member countries, excluding Estonia, Latvia, Lithuania and Slovakia.

First, nominal house prices seem to be correlating to a higher extent then deflated house prices in all the countries in almost every case. This suggests the substantial importance of the inflation in the convergence of the house prices. What also stands out, is that correlation of Slovakia nominal house prices both before and after EMU accession is extremely high (0.93 and 0.85).

Table 5. Correlation of house prices between countries of interest and EU averages before and after the adoption of the Euro.

Type of prices Period Estonia Latvia Lithuania Slovakia

Nominal house prices Before EMU accession 0.33 0.46 0.55 0.93

After EMU accession 0.50 0.77 0.98 0.85

Real house prices Before EMU accession 0.15 0.15 0.34 0.92

After EMU accession 0.41 0.81 0.92 0.41

Note: Euro area house prices are based on weighted average of all the Euro area members, excluding Estonia, Latvia, Lithuania and Slovakia.

Note2: compiled by the author of this paper.

However, correlation of real house prices after the Euro adoption is noticeably lower – it dropped from 0.92 to 0.41. Looking at relatively smaller economies of Latvia, Lithuania and Estonia and less noticeable correlation in these countries, it could be said that bigger economies are more perceptible to house price convergence than less developed economies, as this is also evident from the literature. Lastly, the strongest evidence from the correlation matrix is that the relationship of both nominal and real house prices between all Baltic countries and the averages of Euro area tends to be much closer after the EMU accession. It must be noted, that such outcome might not only rely on the change in currency, but also in the development of the housing markets in these countries after the crisis of 2008. Following the

(20)

20 aftermath of the financial crisis, housing markets became more strictly controlled and thus being less susceptible to shocks and extreme deviations. Therefore, there are some things that could be taken from the correlation matrix, however the results are too ambiguous to come to conclusions about the effect of monetary integration and possible convergence of house prices.

3.3. Results

As already mentioned, initially the model will be tested using Panel OLS fixed effects estimation method. Later, estimations obtained via Panel OLS fixed effects method will be compared to those of System -GMM, to shed more light on the robustness of the results. Also, it is important to note, that all variables are transformed into logarithms, except for unemployment, interest rate, dummy variables and the differenced variables, which in this case is the quarterly difference of Credit stock to GDP ratio. In the table below (Table 6), results of six models can be seen. Model 1 is the framework model, taking all the control variables into regression and adding the lagged dependent variable as an explanatory variable. Model 2 represents the same basic framework regression; however the lagged dependent variable is not included. In regressions estimated in models 3 to 6, LDV is not included as well, in addition, these models use alternative variables for GDP and Credit stock, as noted at the top of every model. Also, some variables are excluded from models 5 and 6, mainly mortgage interest rate and the dummy variable denoting crisis period.

Even if significant, mortgage interest rate takes up the positive sign, contrary to what is expected from the literature evidence, therefore it is assumed that the effect of interest rate is already reflected in the Credit stock variable. Similarly, the negative effects of the crisis may be also reflected in changes of other variables such as GDP, Credit stock, Unemployment and levels of Construction, which have been affected extremely negatively during the crisis period, as the economic activity was suspended by the market crash. Presented estimations of six models exhibit the gradual build up to the Model 6, in which all variables have expected signs, as suggested by the literature. Comparing the 1st_{model with the 2}nd_{, the change in}

significance and signs of the variables is clearly noticeable and raises questions about the robustness of the model. For instance, when lagged dependent variable is excluded from the right side of the regression, most of the variables change signs, except for GDP variable and dummy variable denoting the adoption of the Euro.

(21)

21

Table 6. Results of different models using simple Panel OLS fixed effects estimator.

Dependent variable: Ln (Real HPI)

(1)

(2)

(3)

(4)

(5)

(6)

Fixed effects estimator Basic model W/out LDV Alternative credit Alternative GDP W/out Mortgage interest W/out Crisis Dummy

Lagged real HPI 0.938*** (0.0549) GDP 0.504** 0.532 0.816** (0.127) (0.497) (0.172) Credit stock -0.112** 0.0641 (0.0242) (0.0639) Population 0.560** -0.0400 0.242 0.502*** 0.991*** 1.164** (0.163) (0.349) (0.257) (0.0761) (0.146) (0.278) Construction level -0.0217 0.110 0.0724 0.0786 0.191** 0.243** (0.0259) (0.0796) (0.0714) (0.0709) (0.0394) (0.0668) Mortgage interest -0.135 5.571** 5.357** 5.025** (0.909) (1.607) (1.300) (1.384) Unemployment 0.374 -2.663** -2.359*** -2.805*** -2.807*** -2.442** (0.357) (0.736) (0.208) (0.423) (0.278) (0.455) EMU Dummy -0.00327 -0.0641 -0.0638 -0.0368 -0.0823* -0.0815 (0.00382) (0.0363) (0.0308) (0.0255) (0.0336) (0.0493) Crisis Dummy -0.0540** 0.00642 0.0196 0.0170 0.0656** (0.0107) (0.0123) (0.0229) (0.0221) (0.0178) Credit/GDP -0.240 0.504 0.372 0.750** (0.130) (0.231) (0.209) (0.211) GDP per capita 0.442*** 0.362** 0.387** (0.0501) (0.0832) (0.0905) Constant -7.265** -0.772 -4.675 -2.822* -6.111*** -7.854** (2.078) (6.188) (3.176) (1.179) (0.852) (1.760) Observations 204 208 204 204 204 204 R-squared 0.968 0.864 0.868 0.841 0.810 0.791 Countries 4 4 4 4 4 4 Adjusted R-sq. 0.967 0.858 0.862 0.835 0.803 0.785

Note: Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

In addition, the significance of the variables also changes dramatically – GDP, population, credit stock and crisis dummy become insignificant in the 2nd_{model, while mortgage interest rate and unemployment are}

the only significant variables in the 2nd_{model. It may seem that the 1}st_{model with lagged dependent}

variable is even more effective than the 1st_{one, however, subsequent experiments with the 1}st_model

indicated, that the model with the lagged dependent variable is not a viable option to arrive at expected results. However, the extreme shift of signs between 1st and 2nd models can be explained theoretically. The 1st model is so-called dynamic model, taking lagged dependent variable as an explanatory variable. Whereas the 2nd model shows the outcome of the model without the lagged dependent variable, thus

(22)

22 the model is static. The literature recognizes the possible bias of the dynamic model by the inclusion of lagged dependent variable. Since the model is continuous, the error term, 𝜀𝑖𝑡 has a direct influence

on 𝐻𝑃𝐼𝑖(𝑡−1). However, in that case, the error term cannot be statistically independent of 𝐻𝑃𝐼𝑖(𝑡−1).

Therefore, the existence of residual correlation leads to the bias of both lagged dependent variable and other explanatory variables. Mainly, the lagged dependent variable can cause the coefficients of control variables to be shifted downward, while the coefficient of lagged dependent variable can be inflated. (Kelly & Keele, 2005; Allison, 2015). Eventually, in some cases the signs of control variables can even take the false sign (Achen, 2001). Looking at the drastic change in coefficient signs of the variables, it is evident, that the case of above mentioned bias happened in the dynamic model estimated by including the LDV. The subsequent models from 3 to 6 shows the adjustments of the model, by taking alternative variables into regression, or by excluding variables with unexpected signs, under assumption that the effect of these variables is already reflected in other control variables. As it can be seen from the Model 3, the inclusion of alternative Credit variable – Credit/GDP – which reflects changes in the quarterly ratio of outstanding mortgage stock divided by quarterly GDP at market prices, already improves the model. Construction level takes up positive sign, which seems reasonable, as the production of new buildings is more active in the setting of heated housing market and rising house prices. Also, GDP variable becomes significant, which is also expected, as the literature recognizes substantially close relationship between the developments of the economy and the housing markets. However, contrary to the literature evidence, credit/GDP ratio remains negative, possibly indicating weak robustness of the model. However, later adjustments to the model, such as introduction of alternative GDP variable – GDP per capita and the exclusion of mortgage interest rate and crisis dummy variables, yielded positive results. In the last model, all the main variables discussed in the literature takes up expected signs, thus indicating the correct specification of the model. Explanatory variables as GDP, changes in Credit to GDP ratio, construction levels and the dynamics of populations leads to higher house prices, while the increase in unemployment is associated with the decline of house prices, as evidenced by the literature.

The most important findings of the models above is the sign and significance of the main variable of interest in this research - dummy of EMU accession. Interestingly, despite the adjustments of the models and changes in signs and significance in almost every explanatory variable, EMU dummy has a constant negative sign and insignificant coefficient throughout all the models. Consequently, the estimated negative effect on residential real estate prices means that the Euro adoption has led to the decrease in house prices in researched countries. Such result could be explained by the economic and financial

(23)

23 insecurity felt by the investors, leading to reduced investment levels into real estate. Also, it can be assumed, that the more stable monetary policy led to a reduction of shocks influenced by economic development and credit growth. As the researched countries adopted the Euro after the financial crisis, when the house prices were recovering from the housing market crash and were gradually increasing, the latter explanation seems to be more viable. Especially knowing, that house prices were growing more steadily in the Euro area members, compared to non-Euro area countries. However, none of the models estimated above suggest that EMU accession was statistically significant (at 95% confidence level) in explaining house price dynamics. Therefore, it is difficult to draw substantial conclusions from the results below, as GMM might provide more information.

Table below (Table 7) provides estimation results via system-GMM for four models. The specifications used in the models are the same as used in the panel OLS estimations, as all variables are transformed into logarithmic values, except unemployment, mortgage interest rate and changes in Credit/GDP ratio. Also, the 1st_{model is based on the framework specification, whilst the subsequent models are introducing}

alternative variables or excluding the variables which provide equivocal information about the variable. Finally, as system-GMM method estimates a dynamic model, it automatically includes lagged dependent variable as explanatory variable in all models.

Similarly to estimations of the Model 1 via Panel OLS fixed effects method, the results of System-GMM driven models indicate ambiguous as well as non-consistent signs and significance of explanatory variables. The previous value of house prices in all models is positive and substantially significant, indicating that current house prices are heavily affected by the past values. However, other explanatory variables demonstrate unexpected outcomes. For instance, the models suggest that economic development, even though positive, is not a significant factor to affect house prices.

In addition, one of the main factors that should affect house prices positively – Credit – is negative in all models even after introducing alternative variable – changes in Credit/GDP ratio. Other explanatory variables, such as construction levels, unemployment and population also vary in signs and significance throughout the models, suggesting the inconsistency of the specification used in the estimation. As with the Model 1 of the Panel OLS fixed effects estimator, the results of the estimation performed via System-GMM method proves that the inclusion of lagged dependent variable introduces bias to the model, putting too much weight on the effect of the past value, whilst pushing downward the coefficients of other control variables, even forcing some variables to take up wrong signs.

(24)

24

Table 7. Results of different models using Generalized Methods of Moments (GMM).

Dependent variable: Ln (Real HPI)

(1) (2) (3) (4)

System-GMM estimator Basic model W/out Mortgage interest rate

Alternative credit Alternative GDP

Lagged real HPI 0.883*** 0.915*** 0.878*** 0.733***

(0.0294) (0.0261) (0.0231) (0.0187) GDP 0.0531 0.0216 -0.123*** (0.0429) (0.0418) (0.0301) Credit stock -0.0352*** -0.0382*** (0.00963) (0.00961) Population -0.00562 0.0313 0.167*** 0.0831*** (0.0430) (0.0412) (0.0308) (0.00986)

Construction level -0.0168 -4.17e-05 0.00972 0.00666

(0.0212) (0.0200) (0.0197) (0.0193) Mortgage interest 0.783** (0.336) Unemployment -0.723*** -0.613*** -1.071*** -1.536*** (0.159) (0.153) (0.124) (0.113) EMU dummy -0.0110 -0.0143** -0.0192*** -0.0285*** (0.00699) (0.00691) (0.00682) (0.00588) Crisis dummy -0.0818*** -0.0783*** -0.0802*** (0.00719) (0.00707) (0.00702) Credit to GDP -0.616*** -0.890*** (0.132) (0.141) GDP per capita -0.0681*** (0.0229) Constant 0.567*** 0.396*** 0.528*** 1.316*** (0.146) (0.126) (0.125) (0.207) Observations 204 204 204 204 Countries 4 4 4 4 R-squared - - - -

Note2: System-GMM estimator does not calculate R-Squared indicator.

However, as in the case of Panel OLS fixed effects estimator, the signs for the variable of interest – EMU Dummy – are negative throughout the models. This suggests, that the introduction of Euro have pushed residential real estate prices in Estonia, Latvia, Lithuania and Slovakia downward. It must be noted that differently to results of Panel OLS fixed effects estimator, three out of four models in GMM estimation suggest that EMU accession was a significant factor in affecting housing prices.

As discussed in the methodology part, the model is also estimated using slightly different specifications. Mainly, regression is estimated using time fixed effects, introduction of announcement date dummy and excluding credit channel, by removing credit availability and mortgage interest rate variables. Results of these estimations can be found in the Appendix C (table 15 & table 16).

(25)

25 Results obtained using time fixed effects panel OLS estimation method indicates that even when controlling for year effects, the adoption of the Euro has no significant effect on house prices. Same outcome is observed by using either official Euro adoption date or the date of first official announcement. Likewise, the estimations via fixed effects panel OLS method also indicate that the use of official announcement dummy generates the same results.

By excluding variables which represent the credit channel, the significance of EMU accession on house prices is expected to increase and take up positive effect in estimations, since the literature evidence indicated that credit channel should absorb most of the effect from the financial integration. However, results indicate otherwise, as both official Euro adoption and announcement date dummies stay negative and insignificant. Moreover, estimations obtained via System - GMM method, indicate that EMU accession is negatively significant when excluding credit channel.

To shortly summarize the results, all models estimated both by Panel OLS and System-GMM estimator suggest, that house prices in CEE countries were affected negatively by the accession to the EMU. However, whereas Panel OLS fixed effects estimator suggests that this effect was not significant, results obtained via System-GMM indicate otherwise. However, the estimation of the dynamic model using lagged dependent variable as an explanatory variable could have presented biased coefficients of other control variables, as described in the literature. Consequently, the coefficient of EMU dummy may be pushed downward and inflated, thus increasing the significance of the coefficient. Moreover, results obtained using alternative specifications also confirmed, that the effect of EMU accession on real residential real estate prices in CEE countries has been negative, and in most case, insignificant. The subsequent part of the paper will present the robustness checks based on several assumptions, to test the robustness of the results estimated above.

3.4. Robustness checks

For the robustness checks, several assumptions which lead to adjusted specifications of the model are tested. First, as the main estimations were done using variables transformed into logarithmic values, consequently the same models will be tested using non-transformed variables, to see whether there are significant changes in the results. In addition, as Baltic countries share similar traits in terms of economy size and population, the model will be tested only these three countries, thus excluding Slovakia, to see if the results differ for Baltic countries only. Finally, as already discussed in the previous part, the results of the estimations using lagged dependent variable as an explanatory variable seem to be biased, therefore

(26)

26 the robustness will be checked using only the lagged dependent variable and the EMU Dummy, excluding all other control variables.

These different specifications are tested for Panel OLS fixed effects and System-GMM estimators. In addition, the model including only the variables which obtained expected signs using System-GMM is tested, excluding those variables which have been estimated to have different signs than expected from the literature evidence.

Table 8. Robustness checks using panel fixed effects estimator.

Dependent variable: Real HPI Ln (Real HPI) Ln (Real HPI)

(1) (2) (3)

Fixed effects estimator Non-transformed variables

Baltic countries With LDV & w/out control variables GDP per capita 0.0145** (0.00297) Credit to GDP 65.57** 0.856*** (18.96) (0.0637) Population 0.0705*** (0.00750) Construction level 0.465*** (0.0604) Unemployment -175.6** -2.749** (42.88) (0.500) EMU dummy -9.577 -0.106 -0.00365 (4.916) (0.0624) (0.0106) Ln (GDP per capita) 0.440** (0.0498) Ln (Population) 1.111 (0.520) Ln (Construction) 0.175* (0.0426)

Ln (Lagged real HPI) 0.941***

(0.00914) Constant -92.04*** -7.239 0.280*** (13.94) (3.456) (0.0468) Observations 204 153 204 R-squared 0.821 0.809 0.919 Countries 4 3 4 Adjusted R-squared 0.815 0.801 0.919

The results obtained by adjusting model specifications (Table 8) are in consensus with the outcome seen in main models. Using non-transformed variables, the same expected signs are obtained, only differing in the weights of the coefficients. Also, the EMU dummy stays negative and is statistically insignificant.

(27)

27 Furthermore, using the data from Baltic countries only yields the same results, as expected signs for control variables stay the same, as well as for the EMU dummy. Likewise, the outcome using only LDV as an explanatory variable indicates, that the effect of the EMU accession on real residential real estate prices had no significant effect.

Table 9. Robustness checks using System-GMM estimator.

Dependent variable: Ln (Real HPI) Ln (Real HPI) Real HPI

(1) (2) (3)

System-GMM estimator Variables with

expected signs only

With LDV & w/out control variables

Non-transformed variables

Lagged real HPI 0.887***

(0.0289)

Ln (Lagged real HPI) 0.815*** 0.944***

(0.0253) (0.00839) GDP per capita -0.00146* (0.000871) Ln (GDP per capita) 0.0143 (0.0229) Construction level 0.0789*** (0.0158) Ln (Construction level) 0.0588** (0.0282) Unemployment -0.920*** -43.33** (0.156) (18.92) EMU dummy -0.0343*** -0.00410 -2.027** (0.00272) (0.00822) (0.894) Crisis dummy -0.0920*** -7.990*** (0.0179) (1.811) Constant 0.601* 0.266*** 15.75** (0.336) (0.0419) (7.350) Observations 204 204 204 Countries 4 4 4 R-squared . . -

Note2: System-GMM estimator does not calculate R-Squared indicator.

The estimations via System-GMM estimator (Table 9) paint a similar picture: using LDV and control variables, the EMU dummy is negative and significant. As for the model which includes only LDV as an explanatory variable, excluding all other control variables, the effect of EMU dummy becomes insignificant, yet still negative. As already discussed, inclusion of both LDV and control variables in this case may be resulting in the bias of other variables, as indicated by variables taking false signs. Thus, it also may lead to the coefficient of EMU dummy being pushed downward, resulting in higher significance of the variable.

(28)

28 To shortly summarize the robustness checks, the results obtained via both estimators indicate, that EMU dummy stay negative even if models are adjusted to certain assumptions. It must be noted, that estimations obtained via Panel OLS fixed effects indicate that the effect of the EMU accession was insignificant, while System-GMM method exhibit the significant effect. However, as discussed earlier, this may be due to possible bias of the dynamic model.

4. Conclusion

This paper examined the effect of the EMU accession for real prices of the residential real estate in CEE countries. For the empirical research, data from Estonia, Latvia, Lithuania and Slovakia and covering the time span from 2005:Q1 to 2017:Q1 was used.

The paper covered the existing literature evidence on theories concerning effects of monetary and financial integration effects on housing markets, past researches related to the convergence of housing markets as well as possible transmission channels, through which the effects of the Euro adoption could influence changes in the housing market. It was found, that residential real estate prices should be affected positively after the CEE countries joined the common monetary policy of the Euro area, through the obtained accession to wider financial markets, lower currency risk, more stable monetary policy and reduced administration costs. In addition, the evidence from the literature indicates, that the effects of monetary and financial convergence of housing markets are less noticeable in smaller countries, as housing prices are more affected by country-specific shocks.

The empirical research was carried out using Panel OLS fixed effects and System-GMM estimator, with the purpose to compare the results in terms of robustness. The results obtained via fixed effects Panel OLS indicate, that the effect of the EMU accession for real residential real estate prices in the researched countries is statistically insignificant. Whereas estimations via System-GMM showed significant effect of the EMU accession, yet still negative. However, the results obtained using System-GMM method could have been biased, thus inflating the coefficients of the variables. The findings of the empirical research contradict the main hypothesis of this paper, which states that the effect of the EMU accession on real residential real estate prices in CEE countries should be positive.

However, the findings can be supported by the evidence from the literature, which states that housing markets of the less developed economies are more perceptible to country-specific factors. In addition, the impact of the EMU accession could have been weaker, since the countries pegged their currencies to

(29)

29 Euro years before the adoption of the Euro, thus reducing the currency risk and establishing anticipation for the adoption of the common currency. Finally, housing markets in these countries have already been influenced by dynamics of global and European financial markets and monetary policy prior to the accession to the EMU, as majority of the mortgage lending in researched countries are originating from subsidiaries of foreign banks operating in the EU.

The limitations of the empirical research must also be noted. First, the empirical analysis has included only four CEE countries, due to insufficient data for other countries. Also, the span of the data covers only 14 years of quarterly data, in addition, extreme fluctuations of housing prices and control variables are observed in this period, which might have influenced the results. Ideally, the empirical research should have been performed using all CEE countries that adopted the Euro and more observations before and after the EMU accession. Finally, the methodology used in the research proved to be prone to certain biases, as the system-GMM estimator using lagged dependent variable provided ambiguous results. Moreover, normality of residuals test indicated, that the residuals are not normally distributed, which may have led to incorrect or biased statistical inferences. Thus, more precise and reliable methodology, as well as broader approach to data used could produce different results, than those presented in this paper.

Effect of the EMU accession on residential real estate prices in CEE countries