Estimating the Impact of the Balassa-Samuelson Effect in Central and Eastern European Countries : A Revised Analysis of Panel Data Cointegration Tests

This paper aims to reassess the contribution of the BalassaSamuelson effect to the inflation and real exchange rate appreciation using panel data for nine CEECs covering the period ranging from the mid-1990s to the third quarter of 2010. The main idea of this analysis is to answer the question of whether the Global Economic Crisis had a significant impact on the efforts of CEECs to stay on the path of real convergence. The BalassaSamuelson effect explains less than 1.5 percentage points on average of inflation differential relative to the euro area and around 1 percentage point of the total domestic inflation. The results are robust across the model specification and estimation method. Most of the results point out that the BalassaSamuelson effect has not changed considerably during the crisis even though it is lower compared to that in the earlier stage of transition (for the period up to 2004).

Since the beginning of the transition process, most Central and Eastern European countries (CEECs) have experienced relatively high inflation rates and real appreciation of their currencies.The Balassa-Samuelson theory (the BS effect) provides a supply-side explanation for the relative price of tradables and non-tradables in an economy and is one of the prime explanations for the continuous real appreciation of CEECs against their western counterparts.The key idea behind the BS theory is that, in a given economy, a higher productivity growth in the tradable sector than that in the non-tradable pushes up wages in all sectors.This in turn increases the relative prices of non-tradable goods and increases the ratio of non-tradable to tradable prices.If the productivity growth in one country is higher than in another, the overall inflation will be higher in the former, and the real exchange rate is likely to appreciate in the long run.Analyzing the extent to which inflation is caused by the BS effect is important for at least three reasons.
Firstly, the inflation arising from the BS effect does not cause a loss of competitiveness.That is, although the prices of non-traded goods rise relative to the prices of traded goods, the tradable sector does not become less profitable due to its higher PANOECONOMICUS, 2012, 4, pp.475-499 rate of productivity growth.Since this component of real appreciation occurs with no loss to competitiveness, there is no risk to external accounts.
Secondly, this issue has implications on the conduct of monetary policy.Rising inflation or continuous real exchange appreciation resulting from a higher productivity growth does not require the same monetary policy response as when it is not productivity driven, and therefore, the monetary authority should consider the strength of the BS effect in setting its inflation target.Rising relative prices of nontraded goods would need the prices in the traded goods sector to drop in order to achieve the inflation target.Disinflation in the tradable goods sector requires the exchange rate to appreciate over time, and the central bank has to keep the interest rate high to offset the inflation caused by the non-traded goods sector.
Thirdly, as a consequence of the EU accession of most CEECs, the BS theory has become a popular framework for assessing the feasibility of meeting the Maastricht criteria (see Cristian Paun 2010).A problem might arise in the new European Union member countries if the inflation and exchange rate criteria are impossible to achieve in the presence of the BS effect.
There would be a risk of sacrificing real convergence for nominal because a monetary tightening, which might be necessary for the inflation or exchange rate to meet the Maastricht criteria, could suppress real growth.
The BS effect contains at least two propositions.First, it implies that the relative price of non-traded goods in each country should reflect the relative productivity of labor in the traded and non-traded goods sector.Second, it assumes that purchasing power parity (PPP) holds for traded goods.
The first generation of studies, from the late 1990s to the early 2000s, suggested that the BS effect is one of the major determinants of the high inflation in CEECs.Here, the contribution of the BS effect to inflation usually constitutes up to 3 percentage points.
A second generation of studies pointed out that the BS effect may not be as important for the new EU members as had previously been suggested.These studies usually concluded that the observed inflation and real exchange appreciation are explained by the BS effect only in small part.
The main difference between these two waves of studies is the finding of the latter that the relative PPP does not hold in the tradable sector.This does not imply that the BS effect has a small impact on overall real exchange rate movements because the BS effect is supposed to explain the difference between the overall inflation-deflated and real exchange rate based on prices of tradable goods.Hence, if the share of market-based non-tradable prices in the CPI is large enough, the BS effect can explain a large part of the overall exchange rate movements.Therefore, another reason for the limited explanation of the real exchange rate appreciation by the BS model is the small ratio of market-based non-tradable goods in the CPI (see Balasz Egert, Laszlo Halpern, and Ronald MacDonald 2006).One more reason why earlier studies had obtained higher estimates of the BS effect is that the estimates were produced with time series models; in the newer literature, panel models were used.Also, many earlier studies, such as Marco Cipriani (2001), Fabrizio Corricelli andBostjan Jazbec (2001), Halpern and Charles Wyplosz (2001), and had failed to consider the PANOECONOMICUS, 2012, 4, pp.475-499 impact of productivity differentials on inflation relative to the euro area, focusing only on the Baumol-Bowen (BB effect) effect, which was usually stronger (as noted in Dubravko Mihaljek and Marc Klau 2004, p. 11).
Only a handful studies (for instance, Egert 2005;Maria Machova 2008;Mihaljek and Klau 2008;Paun 2010) analyzed the importance of the BS effect in countries other than the eight EU member states mentioned, such as Southeastern European countries.Also, in the literature little attention has been paid to testing the importance of this factor in the period after 2004 (see Mihaljek and Klau 2008).This motivates us to reassess the contribution of the BS effect to inflation and real exchange rate appreciation using panel data for nine CEECs covering the period from the mid-1990s to the third quarter of 2010, and to contribute to the empirical literature on this topic.The analyzed period is relevant because, with the exception of Croatia and Macedonia, all sampled countries have since joined the European Union.Assessing the size of the BS effect for these countries, as well as for countries that have entered the exchange rate mechanism ERM II, is therefore of particular interest.
The key idea of this analysis is to answer the question of whether the Global Economic Crisis had a significant impact on the importance of the BS effect in CEECs.
Besides the size and the up-to-dateness of the sample, several features related to the estimation procedure distinguish this study from others.For testing the cointegration between more than two variables, we used the panel cointegration test proposed by Gangadharrao Maddala and Shaoven Wu (1999), which enables identifying the number of cointegrated vectors.For the robustness check of empirical results, we used several panel estimation techniques: fully modified OLS (FMOLS), dynamic OLS (DOLS), pooled mean group (PMGE), and mean group (MGE).
The remainder of the paper is organized as follows: The next two chapters lay out the analytical and empirical framework used in the paper.In the first chapter, a theoretical exposition of the BB and BS effects is presented, as well as the econometric methodology used for estimating and testing those hypotheses.The second chapter covers a review of the empirical literature, a brief discussion of the data sources, definitions of the variables included in the empirical analysis, and the empirical results of the BB and BS effect.In the third chapter, the main conclusions of the empirical analysis and the policy implications of these results are presented.

The Theoretical Concept of the Balassa-Samuelson Effect
Consider the case of a small open economy that produces traded (T) and non-traded goods (N), assuming a Cobb-Douglas production function with labor (L) and capital (К) as production factors: where T

 and N
 refer to elasticity of output with respect to labor in the traded and the non-traded goods sector, respectively, and A is the total factor productivity.
PANOECONOMICUS, 2012, 4, pp.475-499 Competition in the labor market assures that the wages in the two sectors will be equal to the marginal product of labor: where T W and N W represent wages in the traded and the non-traded goods sector, respectively.
Assuming that the wages are the same in the traded and the non-traded sector, the price ratio of tradable to non-tradable goods is defined by: If the degree of labor and capital intensity is the same in the traded and the non-traded sectors, i.e., N T    , after log-differencing transformation, the previous equation obtains the following form: . log log log where T a  and N a  are the labor productivity rates in the traded and the non-traded sectors, respectively and The last equation shows that the growth of the relative prices is equal to the difference between the average labor productivity growth in the traded and nontraded goods sector.Assuming that the productivity growth is higher in the traded goods sector, this equation suggests that the relative productivity growth determines the prices in the non-traded goods sector and, consequently, headlines inflation.Given that, firms in the non-traded goods sector will increase the prices of their goods to retain profitability.Consequently, the relative prices will increase, . This effect is known in the literature as the domestic Balassa-Samuelson or Baumol-Bowen effect.
If the wage rates in the traded and non-traded goods sectors are allowed to differ, the equation that describes the behavior of the price of non-traded goods relative to that of traded goods after log-differencing becomes: If the wages in the traded goods sector grow faster relative to the wages in the non-traded goods sector, this will offset the relative increase in productivity and lead to lower growth of the relative prices of non-traded goods.
Furthermore, the inflation rate could be expressed as a weighted average of traded and non-traded inflation: where  is the share of non-traded goods in the consumer basket.
The expression R log   , which is defined as the relative productivity growth, represents the contribution of the BS effect to domestic inflation.
The BS effect can also help explain the inflation differential between two countries: , log log log where S is the nominal exchange rate (see Cipriani 2001, p. 6).
Next, it can be seen how the relative productivity difference contributes to real appreciation, assuming that the productivity growth is faster in the home country than in a foreign country (see Jesus Crespo-Cuaresma, Jarko Fidrmuc, and MacDonald 2003, p. 140).
Replacing equation ( 6) in the real exchange rate change equation ( ) and assuming the same share of non-tradable goods in the consumer basket in the domestic and the foreign country, Because the prices in the domestic traded goods sector and abroad are equal, if they are expressed in the same currency, the real exchange rate change is determined by the changes in relative prices in the non-traded goods sector, or Based on equation ( 4), and replacing it in equation ( 9), the real exchange rate change can be computed as The last equation shows that a faster growth in relative productivity in the tradable sector than in the non-tradable sector results in real appreciation of the domestic currency.

Estimation Techniques
All empirical tests for verification of the BS hypothesis were conducted in a heterogenous dynamic panel framework using Peter Pedroni (1997) and Johansen-type panel cointegration tests, as proposed by Maddala and Wu (1999).Although we presented all seven Pedroni tests, the decision relating to cointegration was made due to group ADF, panel ADF, and panel  statistics, that is, if at least one of those statis- tics confirms it.Specifically, we had in mind the results of Pedroni (2004) that showed that for values of T larger than 100, all seven statistics that were proposed do fairly well and are quite stable, while for smaller samples (T is lower than 20), group ADF statistics are the most powerful, followed by panel ADF and panel  statistics.
We chose to use the non-weighted instead of the weighted panel Pedroni statistics due to their better performance in small samples.
However, Pedroni type cointegration tests do not allow testing of the number of cointegrated vectors between more than two variables.Therefore, for testing cointegration between more than two variables, we used the Johansen panel cointegration test proposed by Maddala and Wu (1999), which enables identifying the number of cointegrated vectors.This test is actually based on combining the p values from the Johansen trace and the maximum eigenvalues statistics for each panel member.Therefore, in this work we used Pedroni tests for testing cointegration between two variables and the Johansen panel cointegration test for more than two variables.
The long-run relationship between relevant macroeconomic variables was estimated by fully modified OLS (FMOLS), dynamic OLS (DOLS), pooled mean group (PMGE), and mean group (MGE) estimator techniques.
FMOLS estimation allows for serial correlation in the residuals and for endogeneity of regressors in the cointegrating regression, and results in an asymptotically efficient estimation of the cointegrating vector.The pooled FMOLS coefficients can be computed in two different ways: within a dimension and between dimensions.Here, we will present only the between-dimension group FMOLS estimator of the mean panel cointegration parameter, which is given as: where: x is the m-dimensional vector of explanation variables, and  i L is the lower triangular decomposition of a consistent estimator of the idiosyncratic asymptotic covariance matrix . The FMOLS estimator is distributed normally (see Pedroni 1997, p. 103).
The expression following the summation over i is identical to the conventional time series FMOLS estimator, and the between-dimension estimator can be constructed simply as the average FMOLS estimator for each panel member.Likewise, the associated t statistics for the between-dimension estimator can be constructed as: Contrary to FMOLS, the DOLS estimator recommended by Chihwa Kao and Min-Hsien Chiang (2000) employs a parametric correction for endogeneity achieved by augmenting the regression with leads and lags of the first difference of it x , as below: where the estimated coefficient  is given by: , ,..., , _ is the vector of regressors.
The pooled mean group (PMGE) estimator involves pooling and averaging, and allows the intercepts, short-run coefficients, and error variances to differ freely across groups, but the long-run coefficients are constructed to be the same.The PGME was introduced by Hashem Pesaran, Yongcheol Shin, and Ronald Smidth (1998), who proposed estimating the following autoregressive distributed lag (ARDL) model of orders 1 l and 2 l : The dependent variable in the first differences is regressed on the lagged values of dependent and independent variables in the levels and first differences.The long-run coefficients, , are defined to be the same across countries (the MGE method allows differences in long-run coefficients across individuals).
The error correction term obtained from the pooled mean group estimator is used as a test of cointegration.A negative and statistically significant error correction term,  , confirms the presence of a long-run relationship between it y and it x .The equation is estimated using the maximum likelihood procedure.
The conclusion relating to the empirical testing of the BB and the BS effect has been based mostly on FMOLS and DOLS method estimates, while PMGE and MGE techniques have been used for robustness checking.

Review of the Empirical Studies
The BS effect in CEECs has been empirically tested in numerous works.Even though authors use various econometric methods, as well as different indicators of relative prices and relative productivity, and distinguish the tradable and non-tradable sectors in different ways, their results most often confirm the presence of the BS effect (see Josip Funda, Gorana Lukinic, and Igor Ljubaj 2007, p. 326).
It is impossible to address all the papers on the subject in this article considering their large number.Therefore, we will summarize the results and the techniques applied in the most cited selected recent empirical studies of the BS effect in CEECs (for further review, see Fritz Breuss 2003;Virginie Coudert 2004;Jose Garcia-Solanes 2008;etc.).
The first round of studies, conducted from the late 1990s to the early 2000s, suggested that the BS effect is one of the major determinants of the high inflation in CEECs (see Andras Simon and Mihaly Andras Kovacs 1998;Philipp Rother 2000;Halpern and Wyplosz 2001;Adriana Lojschova 2003;etc.).Here the contribution of the BS effect to the real appreciation in a number of CEECs constitutes up to 3 percent points per annum.
For instance, Halpern and Wyplosz (2001) estimated the BS effect in CEECs using fixed and random effects, and ordinary least square and generalized least square methods for unbalanced panels.Their results confirmed the presence of the BS effect of a magnitude of 3 percent points per annum.
Lojschova (2003) tested the BS effect using time series and panel estimation techniques for Slovakia, Czech Republic, Poland, and Hungary during the period between 1995 and 2002.According to these results, the individual country estimates of the BS effect are approximately 2.5% per annum, while when using panel models, this effect is smaller and ranges from 0.8% in Hungary to 2% in Poland.Furthermore, Lojschova (2003) tested an extended version of the BB effect with wages as an PANOECONOMICUS, 2012, 4, pp.475-499 additional explanatory variable, but the estimated coefficients of productivity were not changed significantly compared to the model that did not take wages into account.
In contrast to previously mentioned studies that estimated a very strong contribution of the BS effect in CEECs, a large group of studies (for instance, Cipriani 2001;Mark De Broeck and Torsten Slok 2001;Monika Blaszkiewicz et al. 2004;Mihaljek and Klau 2004;Martin Wagner and Jaroslava Hlauskova 2004;Egert 2005;etc.)found that the BS effect is rather modest and that it is not sufficient to explain the observable real exchange appreciation in CEECs.The contribution of the BS effect to real appreciation in those studies ranged from 0% to 1.5%.
For example, De Broeck and Slok (2001) regressed the real exchange rate on the productivity differential between sectors for a range of CEECs; depending on the specifications used, the elasticities obtained varied between 0.2 and 0.6.Using a similar regression based on panel data, Corricelli and Jazbec ( 2001) obtained an elasticity of roughly 0.5.Cipriani (2001) focused on structural inflation rather than real exchange rate in his study on the BS effect.He used a panel of ten CEECs between 1995 and 1999.According to him, the BS effect is relatively weak due to the relatively small share of non-tradables in the consumer price index and the pronounced growth of productivity in the tradable and non-tradable sectors.Cipriani (2001) argued that around 0.7 percentage point of observed inflation could be explained by the productivity growth differentials between the tradable and non-tradable sectors.
Egert, Imed Drine, and Christophe Rault ( 2002) estimated the magnitude of the BS effect for nine CEECs using panel cointegration techniques.According to these authors, even in countries where the increase in relative productivity in the tradable sector is very sharp, such as Poland, the impact on relative prices compared to Germany remains moderate, between 1.2% and 2.4% per annum.Even though they detected that the productivity growth in the tradable sector is likely to bring about a non-tradable inflation, they argued that this depends on the composition of the consumer price index basket, as well as on the share of regulated prices.
Using time series models, Mihaljek and Klau (2004) found that the BS effect in six CEECs explained, on average, only between 0.2 and 2.0 percentage points of annual inflation differentials vis-a-vis the euro area.They also argued that, as the pace of catching up decelerates, these effects are likely to decrease and hence should not become a determining factor of the ability of these countries to satisfy the Maastricht inflation criterion.An updated analysis of Mihaljek and Klau (2008) confirmed the presence of the BS effects in CEECs countries in the period since the mid-1990s through the first quarter of 2008.On average, BS effects explain around 24% of inflation differentials vis-a-vis the euro area (about 1.2 percentage points on average) and 84% of the domestic relative price differentials between non-tradables and tradables, or about 16% of overall domestic CPI inflation (about 1.1 percentage points on average).Blaszkiewicz et al. (2004) conducted an empirical analysis for an unbalanced panel of nine CEECs in the period between 1995 and 2003.FMOLS and PMGE panel estimations pointed to a well-behaved and statistically significant BB effect, while PANOECONOMICUS, 2012, 4, pp.475-499 there was less evidence in favor of the BS effect.They argued that their estimates of the contribution of the BS effect to inflation (below 2 percent points per annum) and to real appreciation (below 3 percent points) must be interpreted with caution because productivity indicators are subject to short-run fluctuations, which may lead to a negligible or even negative BS effect.Wagner and Hlauskova (2004) conducted an assessment of the BS effect in eight CEECs in the period between 1993 and 2000.They investigated a variety of specifications of extended models (non-homogeneity of wages, deviations from PPP in tradables, and including demand-side variables to explain inflation differentials).Evidence of the BS effect was found, but it was relatively small (around half a percent per annum) and not sufficient to explain the inflation differentials between CEECs and euro area countries.
Following the work of Wagner and Hlauskova (2004), Machova (2008) quantified the BB and the BS effect for 24 European countries divided into Western nations, CEECs, and Delta (Romania and Bulgaria) countries.Using a panel data method in the period between 1996 and 2005, Machova (2008) obtained the BB effect quantified as percent of inflation per annum and the BS effect as less than half a percent per annum.
Paun (2010) tested the BS effect for the period between 1999 and 2007 based on Euclidian distances between real and nominal convergence, which were previously estimated for each of the CEECs.The author found unconvincing evidence of the BS effect in most CEECs; clear evidence of the BS effect was found only in Lithuania.
The literature has also identified some puzzles related to the operation of the Balassa-Samuelson effect (see Egert and Jiri Podpiera 2008;Kosta Josifidis, Emilija Beker, and Novica Supić 2008;Egert 2010).Some of those studies, which found the BS effect to be modest, provide an explanation for the tendency toward real exchange rate appreciation in CEECs using complementary determinants such as demand-side factors, as well as structural transformation and institutional factors (see Garcia-Solanes 2008, pp. 29-30).Demand-side factors have been proxied by different types of variables, such as real interest rate differential (for instance, Lojschova 2003; Christoph Fischer 2004), real income per capita and government spending as a percentage of GDP (for instance, Rother 2000; Olga Arratibel, Diego Rodriguez-Palenzuela, and Christian Thimann 2002), the budgetary deficit (Arratibel, Rodríguez-Palenzuela, and Thimann 2002), the share of private consumption over GDP (Coricelli and Jazbec 2004;Fischer 2004), and total consumption (Wagner and Hlouskova 2004), among others.
For instance, Arratibel, Rodriguez-Palenzuela, and Thimann (2002) highlighted the importance of nominal wage growth and fiscal policy, as well as the impact of liberalization-oriented reforms on lowering inflation in the non-tradable sector.According to these empirical findings, the productivity growth in the tradable sector does not seem to have been a significant variable explaining the inflation dynamics in the non-tradable goods sector in CEECs.Fischer (2004) developed and tested a model that included an investment demand channel.The BS effect still exists and, additionally, the domestic demand affects the price of non-tradables and thereby the real exchange rate.The model pre-PANOECONOMICUS, 2012, 4, pp.475-499 dicts that capital demand is negatively dependent on the price of non-tradables and on the interest rates.
Garcia-Solanes, Francisco Sancho-Portero, and Fernando Torrejon-Flores (2007) found that increases in the demand for differentiated domestic tradables, which are steered by higher economic growth and improvements in the quality of these goods, introduce an upward bias in the prices of tradables in these countries, leading to trend appreciations in the real exchange rate of tradable goods.
Michael Brandmeier (2006) found that the impact of productivity on inflation differences is probably superimposed by the growth rate of the monetary stock or by influences from the demand side of the economy.
Different explanations of the PPP puzzle are addressed in the literature; some of the most important are problems with precise sector division, incomplete substitutability of tradables in the euro area and CEECs, the existence of the non-tradables processing component introducing frictions to the international arbitrage, and imperfect competition (see Blaszkiewich et al. 2004;Garcia-Solanes and Torrejon-Flores 2009).The results of most empirical studies (see Matthew Canzoneri, Robert Cumby, and Behzad Diba 1999;Egert, Drine, and Rault 2002;Blaszkiewicz et al. 2004;Lukasz Rawdanowicz 2004;Mirjana Palic 2009;etc.)show that the stronger version of PPP is not satisfied in CEECs.

Data, Sources and Definitions
The data set used in this study consists of average labor productivity data, real wages, relative prices, and real exchange rates.The panel data cover 9 CEECs (Bulgaria, Croatia, Czech Republic, Hungary, Macedonia, Poland, Romania, Slovenia, and Slovakia).All series are transformed into natural logarithms.The empirical testing of the BB hypothesis is conducted using quarterly data covering the period from 1995 to 2007; for the BS hypothesis, the data cover the period from 1997 to 2007.The first years of transition are eliminated from this analysis due to the fact that during this period, price and productivity developments were driven more by initial reforms than by the BS effect.The BB effect and the BS effect are tested to extend the analysis to the period of the Global Economic Crisis (covering the period from 1995/1997 to the third quarter of 2010).This panel is unbalanced due to the fact that, in the interim, Slovenia and Slovakia became EMU members.
The national account data, wages, and employment data are obtained from Eurostat and national statistical offices, while nominal and real exchange rate data, as well as CPI and PPI indices, are taken from International Financial Statistics.
One crucial issue in constructing productivity and relative price variables is how to define the tradable and non-tradable sectors.No consensus has been reached in the literature on this issue, but most empirical studies refer to industry, or industry and agriculture, as the tradable goods sector.Egert, Drine, and Rault (2002) highlighted that their results are affected by the way sectors are classified, particularly whether or not agriculture is considered as part of the tradable sector.
Average labor productivity, which is a proxy for average total productivity, is computed using the above classification as a basis and by dividing the sectoral value added to the corresponding number of employees.Considering only the number of full-time employees instead of the total number could change the BS and the BB effect on inflation and real appreciation, especially in countries with a higher share of PANOECONOMICUS, 2012, 4, pp.475-499 agriculture in total gross value.However, those series were not been publicly available for all analyzed countries.
Sectoral productivities serve as a basis for calculating relative productivities, PRI (if only industry is considered as tradable) and PRIA (if both industry and agriculture are considered as tradable).All other sectors, excluding administration, are considered as non-tradable.Relative productivity differential is defined as the ratio of relative productivities in CEECs and Euro Area 12.
Following the empirical work of Blaszkiewicz et al. (2004), and Egert, Drine, and Rault (2002), we use two types of proxies as a measure of relative prices.Firstly, relative price (RP1) is defined as the ratio of the corresponding sectoral GDP deflators.The prices of non-tradable goods are given by the GDP deflator for services, while the prices of tradable goods are given by the GDP deflators for industry.Secondly, the ratio of services in consumer price index and producer price index is used as a proxy for relative prices (RP2).
The real exchange rate is calculated using the GDP deflator in the services sector (RERG).Other measures, such as prices in the tradable sector as well as CPI, could be used as the deflator for the real exchange rate.

Empirical Results
Prior to testing the long-run relationship between different macroeconomic variables, we applied panel unit root tests by In Choi (2001) and Kyung So Im, Hashem Pesaran, and Yongcheol Shin (2003) to test for non-stationarity, a necessary condition for cointegration.The tests were performed on levels and first differences, but for practical reasons, only the results of those carried out on levels are presented (Table 1).The presence of a unit root for all indicators in the specification with constant was confirmed by both tests, while for the specification with trend, the applied tests for some indicators pointed to the rejection of the null hypothesis of non-stationarity.The applied tests are inconclusive, but the presence of a unit root could not be strongly rejected in any case and leads us to conclude that the variables are non-stationary in levels.

Testing the Baumol-Bowen Effect
Empirical testing of the BB effect was conducted in three steps: between productivity and real wages in the tradable sector, between real wages in the tradable and the nontradable sector, and between relative productivity and prices.
The long-run relationship between productivity (

CEEC T PR
) and real wages

RWAG
) in the tradable sector (industry) is a necessary condition to confirm that the BS effect is one of the most important factors of real appreciation.The cointegration relationship is confirmed by most Pedroni tests for the model with constant and trend, and partly for a model with individual intercept only (Table 2).Source: Author's estimations.Source: Author's estimations.
The FMOLS and PMGE results are broadly in line with the findings from the cointegration analysis, but the estimated coefficient of real wages elasticity is far from that found in the analysis.This suggests that a significant influence of the BS effect on inflation and real appreciation is unlikely.
These findings are in line with the results of Egert's (2005) analysis for the panel composed of Bulgaria, Croatia, Turkey, Romania, Russia, and Ukraine, which revealed a real wage elasticity ranging from 0.5 to 0.8.
In the next step, the cointegration between the wages in the traded goods sector and in the non-traded goods sector was tested (Table 4).The wages in construc-PANOECONOMICUS, 2012, 4, pp.475-499 tion, transport, trade, education, and health weighted by the number of employees in those sectors were used as a proxy for non-tradable sector wages.
In contrast to the long-run relationship between relative productivity and wages, the long-run relationship between wages in the tradable and in the nontradable goods sector is less obvious and, more importantly, not confirmed by the tests that we consider as the most appropriate (group ADF, panel ADF, and panel  ).This result is of great importance for the empirical testing of the BB effect: if the assumption of wage equalization is not satisfied, the difference in wages between the tradable and the non-tradable sector should be included as a factor in domestic inflation, in addition to relative productivity.The cointegration between relative prices and relative productivity is strongly confirmed by group ADF and panel ADF statistics for relative prices and relative productivity if GDP deflators are used as a proxy for relative prices (Table 5).The estimation of price elasticity from the relative productivity growth is done by the FMOLS method.The results of the FMOLS method (Table 6) suggest that a faster productivity growth by 1% in the tradable goods sector causes increased prices in non-tradable goods, measured by GDP deflators as 0.4 percentage points.The estimation of this effect for separate countries ranges from 0.1 for Bulgaria to 0.65 for Slovakia.All estimated coefficients are positive and statistically significant.
Including the period of the Global Economic Crisis in the analysis does not significantly change the estimation of the BB effect, although the estimated the contribution of productivity differential to relative prices is lower.The Crisis significantly changed only the contribution of productivity differential to the relative prices of non-tradable goods in Bulgaria, whose coefficient became negative.
For the sake of robustness checking, we estimated the same relationship by MGE, PMGE, and DOLS methods (Table 7).In line with the FMOLS results, MGE, PMGE and DOLS pointed to a well-behaved and statistically significant BB effect.In all three models, the panel long-run coefficients are positive and statistically significant.The elasticity obtained by MGE is similar to that obtained by FMOLS, while the elasticity obtained from the PMGE and DOLS method is slightly higher.This confirms the robustness of the obtained results related to the BB effect.The estimates are not significantly changed when both industry and agriculture are included in the tradable sector (the results of this analysis have not been presented separately).Presuming that the share of agriculture in total gross value is low, except in Slovakia, Romania, and Bulgaria, the inclusion of agriculture does not change the estimation of the impact of productivity on non-tradable prices.The coefficient is negative only in Bulgaria, which has very low productivity in agriculture.The estimates obtained by MGE, PMGE, and DFE methods are, as in a previous case, very close to the FMOLS estimates.
The results of the FMOLS method suggest that 1% faster productivity in the tradable sector compared to the non-tradable sector causes an approximately 0.5% PANOECONOMICUS, 2012, 4, pp.475-499 increase in relative prices measured by the ratio of services in CPI and PPI (Table 8).This effect is smaller than most previous studies suggest (for instance, Egert, Drine, and Rault 2002;Mihaljek and Klau 2008).
When the ratio of services in the CPI and PPI indices is used as a proxy for relative prices, the estimated coefficient for each particular country except Croatia is very close to that obtained using GDP deflators.The average cointegration coefficient is higher because Bulgaria, Romania, and Macedonia are excluded from the analysis.One possible explanation for this difference in Croatia is inappropriate classification of the tradable and non-tradable sectors due to the high share of tourism in overall exports.
As in the previous case, the average cointegration coefficient is slightly lower when the period of the Global Economic Crisis is included in the analysis.Bulgaria, and Macedonia).Two lags for kernel estimators are used, but the different specification does not alter the estimation.The t statistics are for the null hypothesis that the estimated coefficient is equal to one.Therefore, for coefficients that are positive but lower than one, the test statistics are negative.
To find out the contribution of the BB effect to domestic inflation according to equation ( 5), we multiplied the share of non-tradable goods and services in the CPI by the estimated coefficient of elasticity, which was previously multiplied by the historical average of the domestic productivity differential.These findings suggest that approximately one percentage point of inflation (for RP2 measure) and 0.6 p.p. (for RP1 measure) could be explained by the BB ef-PANOECONOMICUS, 2012, 4, pp.475-499 fect.These results related to the average BB effect are consistent with the findings of Machova (2008) and Mihaljek and Klau (2008) while the individual country estimates are different.The results show that the BB effect has the highest contribution in Slovenia, Slovakia, and Croatia (measured by RP2), and the lowest contribution in Bulgaria.For instance, according to Machova (2008), the BB effect ranges from 0.15% for Bulgaria to 2.5% for Slovakia.The contribution of the BB effect in Mihaljek and Klau ( 2008) is negative for Hungary and positive but very low for Bulgaria; for the Czech Republic and Slovenia, it is much higher than in our analysis.
The main policy implication of these results is that the BB effect is not a determining factor of the ability of CEECs to satisfy the Maastricht inflation criteria.Taking into account that Slovenia and Slovakia have already satisfied those criteria and become members of the EMU, it is unlikely that other countries would not be able to do the same.
In line with equation ( 5), the standard version of the BB effect could be extended to an alternative specification that captures the fact that the assumption of wage equalization is not necessarily satisfied due to differences in skills and human capital.Therefore, we could relax the assumption of wage equalization and test the extended version of the BB effect, including wage differential as an additional factor in the increase in relative prices (as was done by Lojschova 2003; Wagner and Hlouskova 2004;Machova 2008).As the previous analysis has not provided clear evidence that wage equalization happens in CEECs, the extended version of the BB effect is tested.

Note:
The cointegration vector is estimated under the assumption that the number of lags is 2. A panel of six countries is used (excluding Romania, Bulgaria, and Macedonia).
The Johansen panel cointegration test proposed by Maddala and Wu (1999) commits that relative prices, wages, and productivity differential are cointegrated, but not their regressors (Table 10).With reference to this result, the FMOLS method could be used.
As the results in Table 11 show, the coefficient of wage differential has the opposite sign for some countries, while the coefficient of relative productivity is almost unchanged.In line with the results of Lojschova ( 2003), these findings lead to the conclusion that including wage differential as an additional factor in inflation does not contribute much to the prices of non-traded goods.Contrary to these findings, Machova (2008) found evidence of the importance of adding relative wages in the estimation of the BB effect.Source: Author's estimations.

Testing the Balassa-Samuelson Effect
Testing the BS effect consists of several steps.In the first step, the cointegration between the real exchange rate and the difference in relative productivity in CEECs relative to the euro area (as the main trading partner for most of these countries) is tested.
Most of the implemented panel cointegration tests suggest a long-run relationship between the real exchange rate and the difference in relative productivity in CEECs and the euro area (Table 12).Taking into account that all the necessary conditions are fulfilled, the next step of the analysis is to estimate the average contribution of the productivity differential to the inflation differential.Therefore, we estimate the standard version of the BS effect.The standard version of the BS effect fully corresponds to testing the cointegration between the real exchange rate and productivity differential.On average, the contribution of the BS effect to the inflation differential relative to the euro area ranges from 0.7% to 1.5% (depending on the measure of relative productivity used).This is a relatively small fraction of the rate of overall inflation.Taking into account that the average inflation differential in CEECs relative to the euro area in the analyzed period is 5%, it could be concluded that the BS effect is not the most important cause of real currency appreciation.For instance, these results are similar to those of Mihaljek and Klau (2008), while Machova (2008) reached a lower estimate of the BS effect.These findings are in line with those of a number of studies reported in Section 2.1, which also showed that more than the BS effect, other factors, such as the process of liberalization and opening of these economies to foreign direct investments, as well as the growth of public expenditure through the aggregate demand channel, contribute to real appreciation in these countries.
The results of this analysis seem to suggest that the BS effect is stronger in economies with a stronger convergence process.One of the explanations why the BS effect is less evident in Bulgaria, Macedonia, and Romania, which could not be directly explained by productivity growth and the BS effect, is the process of price liberalization, which affects the prices of both tradable and non-tradable goods and services.The low share of non-traded goods in the CPI index could be also one explanation.Moreover, what seems to characterize these countries is their fastgrowing productivity not only in the tradable goods sector but also in the non-traded goods sector.

Concluding Remarks
In this paper we investigated the importance of the BB and the BS effects in nine CEECs in the recent period (from the mid-1990s to the third quarter of 2010).Using panel estimation techniques, we found evidence in favor of the BB and the BS effects.Panel cointegration tests detected a relationship between productivity and relative prices (the BB effect), and between relative productivity differential and real exchange rate vis-à-vis the euro area (the BS effect).Similar evidence were found using the FMOLS, DOLS, PMGE, and MGE methods.
The econometric analysis shows that the BB and the BS effects are less evident, particularly in Romania, Bulgaria, and Macedonia.One possible explanation for the limited role of the BS effect in overall inflation and real exchange rate appreciation is the failure to achieve some of the most important preconditions (a stronger version of PPP holds, real wages are proportionately linked to productivity in the tradable sector, wage equalization).
The findings suggest that approximately one percentage point of inflation could be explained by the BB effect.The results also show that the BB effect has the highest contribution in Slovenia and Slovakia, and the lowest contribution in Bulgaria, where the difference in productivity between the tradable and the non-tradable sector is also the lowest.On average, the BS effect's contribution to the inflation differential and real appreciation is less than 1.5 percentage points.Based on this result, it could be concluded that the BS effect is not the most important cause of the real currency appreciation in CEECs.
The CEECs' target of adopting the euro, already reached by Slovenia in 2007 and by Slovakia in 2009, has sparked debates about right timing and the pertinence of the Maastricht criteria due to crisis implications.This in turn roused the interest of researchers and policy makers in the dynamics of the BS effect.Answering the question whether the BS effect has become stronger or weaker during the catching up process as well as during the crisis, this analysis is conducted for the subsample period from 1995 to 2007 and from 1995 to 2010.
Most of the results point out that the BS effect was not noticeably changed during the crisis, even though it was lower compared to that in the earlier stage of transition (for the period up to 2004).Thus, it can be anticipated that the slowdown of the convergence process due to a crisis will not be an obstacle for investors and that these countries will again begin the race to catch up.
The analysis suggests that the BB and the BS effects still play a role in inflation and the real exchange rate, and are likely to remain on the agenda of both policy and research for a while but not as significantly as in the earlier stage of transition.In principle, we should expect the BS effect to persist for the entirety of the catching-up period, which should continue for at least the ensuing decade, before and after the adoption of the euro; however, the size of the effect will depend on the speed of the catching-up process.

Table 1
Results of Panel Unit Root Tests In all tables, * refers to statistical significance at 10%, ** refers to statistical significance at 5%, and *** refers to statistical significance at 1%.The data source for all tables is the author's calculations.The number of lags included in the model is chosen according to the Schwarz information criterion.P values are given in parentheses.

Table 2
Pedroni Panel Cointegration Test for RWAG T CEEC and PR T Note: In all tables, * refers to statistical significance at 10%, ** refers to statistical significance at 5%, and *** refers to statistical significance at 1%.The data source for all tables is the author's calculations.The number of lags included in the model is chosen according to the Schwarz information criterion.

Table 3
Estimation of Cointegration Relationship between Real Wages and Productivity Note: Test statistics are given in parentheses.

Table 4
Panel Cointegration Test for RWAG N CEEC and RWAG T Source: Author's estimations.

Table 5
Panel Cointegration Test for RP1 and PRI

Table 7
MGE, PMGE, and DOLS Estimators of the Cointegration Vector between Relative Prices (RP1) and Relative Productivity (PRI)

Table 8
FMOLS Estimation of Cointegation between RP2 and PRIA

Table 11 FMOLS
Estimates of the Cointegration Vector between Relative Prices, Productivities, and Wage DifferentialsNote: A panel of six countries is used (excluding Romania, Bulgaria, and Macedonia).Two lags for kernel estimators are used, but the different specification does not alter the estimated coefficient.The t statistics are for the null hypothesis that the estimated coefficient is equal to one.Therefore, for coefficients that are positive but lower than one, the test statistics are negative.

Table 12
Panel Cointegration Test of Real Exchange Rate (RERG) and Productivity Differential (RPR1)

to 2007:4 Model with individual intercepts Model with individual intercepts and individual trends
The number of lags differs across the individual members.RPR1 corresponds to productivity differential PRI.