University of Groningen
Faculty of Economics and Business
Master Thesis International Economics and Business
The European Union and the Convergence Criteria: Are the Central and
Eastern European Countries and the EU15 Converging?
Name Student: Elroy Tibbe Student ID Number: S2784432
Student Email: e.tibbe.1@student.rug.nl Submission date: 02-02-2017
2 Abstract
This research has focused on the convergence within the European Union between ten Central and Eastern European countries (CEE10) and the high-income West-European countries (EU15). The convergence hypothesis implies that initially poorer countries will catchup to the initially richer countries. In these tests for convergence I control for the effects of the convergence criteria which set a limit to the government deficit and government debt and were imposed by the European Union to stimulate convergence between member states. I have found that convergence has taken place within the European Union between the CEE10 and the EU15 between 1995 and 2015. This convergence has been conditional on several other determinants of economic growth and the convergence criteria indeed seem to have had an impact on the economic growth of all countries and on the convergence between both country groups.
3 1. Introduction
The European Union (EU) faced some severe setbacks in 2016 with the United Kingdom leaving the union, Italy getting into a serious banking and public debt crisis and Greece still being in crisis after several years of measures. And with elections coming in countries that provide a home to seemingly constantly growing EU critical parties and some parties calling for a same EU referendum as took place in the United Kingdom the future of the union is at stake. The Euro critical book from American Nobel Prize winner in Economics Joseph Stiglitz has contributed further to the EU and Euro skepticism. Stiglitz (2016) argues in his book why the common currency is threatening the future of Europe as a whole. He states that the convergence criteria that have been adopted by all EU countries, result in divergence instead and are to blame for the poor growth performance of European countries during the last years.
The aim of this research is to find out whether or not the Central and Eastern European countries are converging in GDP per capita terms towards to high-income countries of the EU and what the influence has been of the convergence criteria. Neoclassical economic growth theory predict economic convergence between countries on the basis of diminishing returns to capital. However, so far little evidence has been found for convergence between countries worldwide. So, despite the economic growth models and the convergence that gets hypothesized within some of these models, it is still of great interest for growth economist to see whether or not convergence is actually taking place and what is driving this possible faster economic growth of countries that try to catchup. Due to the fact that so far little evidence is found for convergence worldwide, that the effects of the convergence criteria on economic growth are under discussion and the economic performance of all EU member states is of great interest for the EU and its policy makers the findings of this research can be of great interest.
4 Main finding of this research is that convergence is taking place between the CEE10 and EU15 between 1995 and 2015. The results show that, both, β-convergence and σ-convergence are statistically significant over the full timeframe. The σ-σ-convergence further showed that the standard deviation of the log GDP per capita of all countries decreased over the full time frame. The σ-convergence occurred at the fastest pace between 2000 and 2008 and stagnated heavily after the global financial crisis. The impact of the convergence criteria on economic growth and on convergence between the CEE10 and EU15 show that the government balance criterion is working towards convergence, whereas the government debt criterion seems to have a positive influence on the economic growth rates for both country groups.
The remainder of this paper is structured as follows. Section 2 presents a literature review of the convergence and economic growth literature. Section 3 describes the methodology and the data used in the research. Followed by, section 4 which presents the empirical results of this paper. Lastly, section 5 that presents the conclusions of the research.
2. Literature review
The EU faced some severe setbacks in 2016, the union that has grown ever since its establishment, now has lost a member state for the first time. The United Kingdom voted in a referendum against further EU participation and are now preparing to leave the union. Furthermore, Italy is getting into a more and more serious banking and public debt crisis, whereas Greece is still in crisis after several years of measures. Because of the ongoing financial crises during recent years and the current migration crisis Eurosceptic political parties are rising in several member states and with elections coming up in several of these countries the future of the European Union is at stake. The Euro critical book from American Nobel Prize winner in Economics Joseph Stiglitz has contributed further to the EU- and Euro skepticism. Stiglitz (2016) states that the common currency is what is threatening the future of Europe as a whole and argues that the so-called convergence criteria, that have been adopted by all EU member states, have so far only led to divergence instead. This paper will research if convergence has taken place between the CEE10 and EU15 between 1995 and 2015 and whether or not the convergence criteria had indeed a converging or a diverging effect.
5 research. Finally, I define other possible important drivers of economic growth which will act as control variables in the remainder of this research.
Convergence literature
The convergence hypothesis implies that richer economies grow slower than poorer economies such that eventually the poorer economies are able to catch-up to the richer economies. The convergence literature describes two main concepts of convergence: β-convergence and σ-convergence (Sala-i-Martin, 1990, 1996). If initially poorer economies grow faster than richer economies β-convergence occurs. If the dispersion of GDP per capita between economies tend to decrease over time σ-convergence occurs. These two types of convergence are intuitively related to one another, since the existence of β-convergence is a necessary condition for the existence of σ-convergence and the existence of β-convergence will generate σ-convergence. However, theoretically they do not always have to occur together, in a situation in which the poorer economy becomes the richer economy and vice versa β-convergence occurs whereas σ-convergence does not. Figure 1 gives a graphical presentation of these types of σ-convergence in a two country example. In panel a the growth rate of country A is smaller than the growth rate of country B, so we find that β-convergence is taking place. Panel a also shows σ-convergence because the dispersion between the two country’s log(GDP) at time t+T is smaller than it was at time t. Panel b shows an example in which neither one of the convergence types is present. The richer country A is growing faster than the poorer country B, and therefore, the dispersion between the two countries log(GDP) levels increases. So, the countries are not converging but instead diverging. Panel c gives an example in which β-convergence is present, however, σ-convergence is not. Since the initially poorer country B is growing faster than the initially richer country A β-convergence is present. However, country B became the richer country at time t+T and the dispersion between the two country’s log(GDP) levels did not shrink, which implies that σ-convergence did not occur.
Figure 1: The relationship between β and σ-convergence
6 The classical convergence hypothesis that is based on the neoclassical economic growth model by Solow (1956) and Swan (1956) hypothesizes that convergence takes place because countries face diminishing returns to capital. This prediction that countries with an initial lower GDP per capita achieve higher growth rates depends on the assumption that the only aspect in which countries differ is their initial capital to labor ratio. This implies that all countries have an equal savings rate, which means that they can invest an equal amount into new capital. Since the returns on capital are diminishing, eventually all countries end up with the same capital to labor ratio. In the neoclassical model, having no other differences than the initial capital to labor ratio implies that all countries should have an equal savings rate, depreciation rate, population growth rate and access to the same level of technology. If this assumption holds the model predicts β-convergence in the absolute sense and all countries would converge to the same steady state level with the same capital to labor ratio and an equal GDP per capita. However, in reality, countries differ in lots of things such as their level of technology, savings rate, and or population growth rates. If countries differ in these and other characteristics the convergence forces apply in a conditional sense which, for instance, implies that if an initial poorer country saves less it will not entirely catch up to the richer countries with a higher savings rate. The different steady state levels arise because of the positive relationship the steady state level has with the savings rate and the negative relationship with the population growth rate. If conditional convergence is the case each country converges to its own steady state which can vary in levels of capital to labor and GDP per capita.
7 exhibit a different savings rate which results in different investment curves (all else held equal). In this case both countries converge to different steady states (k*A and k*B) with different capital-labor ratios, a different output per worker but an equalized growth rate.
It was not until the late 1980s that convergence caught the eye of mainstream economist, the classical approach used for convergence analysis is set apart in the research by Sala-i-Martin (1996). Empirical research on economic convergence during that time was still scarce since the only available data was on rich countries. In a research by Baumol (1986) strong cross-country convergence was found, especially, after World War II. However, these results were criticized because of ex-post sample selection bias. After expansion of the Madison dataset that Baumol initially used the positive results for convergence disappeared (Baumol and Wolff, 1988). Whereas the Madison dataset mostly contained data from richer countries, the Summers-Heston dataset that Sala-i-Martin (1996) used in his research contained data on more than one hundred countries. However, also within the new dataset no convergence between countries occurred.
The lack of convergence across countries worldwide has led economic growth theorist to come up with new models of endogenous growth (Romer, 1986, Rebelo, 1990). These first models of endogenous growth did not include diminishing returns to capital which means they do not predict convergence in the way the neoclassical growth model did. The models rely on the existence of externalities, increasing returns to scale and the believe that inputs can be accumulated through policy decisions (for instance human capital, through higher investments in education).
8 Convergence criteria
Stiglitz (2016) argues that the monetary union within Europe, the Eurozone, has been growth diminishing instead of growth enhancing since its introduction in 2002. According to Stiglitz the rules, regulations and institutions that govern the Euro are to blame for the poor performance of the whole region after the global financial and the Euro crisis. Moreover, he states that the so-called convergence criteria (Maastricht criteria), that were imposed to ensure the stability of the European Monetary Union and got adopted by all EU member states in the Stability and Growth Pact, are actually resulting in divergence. The two most salient convergence criteria are the ones that limit government deficit to 3 percent of total GDP and total government debt to 60 percent of total GDP. Main thought behind these criteria is that low deficits and low debts would ensure convergence within the European Union and would prevent crises. However, the global financial crisis and the Euro crisis showed that also countries with low deficits and low debts faced crises. For these countries the debts and deficits did not cause their crisis, but the crisis created their debts and deficits.
These convergence criteria lead to severe cuts on government investments, which according to Stiglitz is fatal for countries that are already in an economic crisis and are trying to get out of it. Since the countries that were part of the Eurozone could not use currency devaluation to restore the balance of payment they used measures of internal devaluation to lower their export prices (cutting production costs, especially lowering wages) and thereby increasing their international competitiveness. However, due to the free mobility of labor it was possible for (especially high-skilled) workers to move out of the countries with high debts. The same principle holds for the free mobility of capital and goods, which made it possible for money to flow out of the banking systems in countries affected by the crises towards the countries that have been less affected by it.
Conclusively, we can state that the goal of the European Union behind these convergence criteria contradicts the findings by Stiglitz. In this research I will control for the effects of the government deficit and government debt measures on the convergence between the CEE10 and EU15.
Hypotheses
9 convergence. The β-convergence hypothesis implies that countries with an initially lower GDP per capita level achieve higher growth rates than countries with an initially higher GDP per capita, thus that GDP per capita is negatively related to the growth rate of GDP per capita. The σ-convergence hypothesis implies that the standard deviation of GDP per capita of all the countries decreases over time.
Next to the overall convergence question I will research what has been the influence of the convergence criteria that sets a limit to government deficit and government debt to the convergence process. I control for these effects by adding government balance and government debt to the model which tests for β-convergence. The literature review already showed that there are conflicting arguments on which effect these variables have on economic growth. If I follow the reasoning from the EU behind the convergence criteria the government balance should have a positive relationship with economic growth and the government debt should have a negative relationship with economic growth. If the government balance shows a positive relationship with economic growth rates and the government debt a negative relationship with economic growth the reasoning from the EU is confirmed. If I find that these relationships are vice versa the reasoning from Stiglitz (2016) holds. To also be able to make statements about the impact the convergence criteria have on convergence I will add two interaction variables to the model which are multiplications of the government balance and the government debt variables with a dummy variable that takes the value of one if the country is from the CEE10 group. By adding these interaction variables to the model I can control whether or not the convergence criteria have a different effect on economic growth if the country is initially poor or initially rich.
Control variables
10 depend on the dynamics of technological catchup and in the long run growth rates equalize across economies and convergence has taken place (Howitt, 2000). However, Gerschenkron already argued that the catchup prophecy is hard to fulfill and far from an automatic process, instead it requires large amounts of efforts and institution building. Romer (1986, 1990) and Agion and Howitt (1998) add to this that country specific factors such as R&D policies and market structure are important determinants for the long run technology growth and thereby for closing the technology gap.
One of the ways in which a country can close the technology gap is by obtaining FDI and benefit from the positive externalities that come with it, such as technology and knowledge spillovers. Romer (1993) found that the gaps in technology and knowledge between the poor and the rich countries can diminish through FDI. Furthermore, it is expected that the transfer of technology and business know-how boosts the productivity of the economy as a whole and not just that of the firm attracting the FDI. Borensztein, De Gregorio and Lee (1998) add that the effects of FDI on economic growth is dependent on the absorptive capability of advanced technology within the host economy. This means that without an adequate level of human capital within the host economy the possible positive externalities of FDI will not be transferred to its full potential.
11 Through this higher productivity level openness to trade can have a positive influence on the economic growth of a country.
The EU (at the time the European Economic Community) was founded in the aftermath of the Second World War to foster economic cooperation between members and create economical interdependent countries. The EU brings down barriers to trade and investments between member countries to foster this economical interdependency. Overall, economic integration is believed to increase trade and investment, mostly, between members of the agreement. Myrdal (1956) describes economic integration as the social and economic process that brings down social and economic barriers between participants of economic activity. Some definitions go a little bit further and mention that is all about optimizing the international economy by removing barriers and introducing elements for unification (Tinbergen, 1954). Balassa (1961) took multiple definitions within the literature into account and described economic integration as a process and as a state of affairs. Viewed as a process it encompasses measures aimed to abolish discrimination between economic units with different nationalities and as a state of affairs it is the absence of discrimination between national economies. Trade creation and trade diversion were found by Viner (1950) as effects of international integration. Trade creation refers to the idea that if an union is created all tariffs between members disappear and that the consumers’ demand for the now lower priced goods increases which creates new trade. Trade diversion refers to the idea that trade might divert away from a more efficient global producer towards a less efficient member of the union, less trade now exists between members of the union and their old trading partners. Viner concluded that trade creation benefits the member countries, whereas trade diversion has a negative influence. Viner’s view became criticized by others who find that his view neglects other favorable effects of economic integration and that trade expansion might overcome the negative impact of trade diversion (Balassa, 1961; Meade, 1955). Roughly the same holds for (foreign direct) investments and economic integration, through the abolishment of investment barriers investment creation and investment diversion can arise (Lloyd & MacLaren, 2004). According to a World Bank (2000) research also an political aspect to trade blocs is of great importance, through increased security and greater international bargaining power intentions are offered for greater economic integration between nations.
12 higher level of human capital in a country results in a higher rate of domestically produced technological innovation. This finding is in line with earlier research done by Romer (1990). Second, human capital levels affect the speed of adoption of new innovative technologies from abroad. This finding is in line with earlier research by Nelson and Phelps (1996) and also corresponds with the finding by Borensztein, de Gregorio and Lee (1998) on human capital and FDI as earlier mentioned in this literature review. Mankiw, Romer and Weil (1992) showed the importance of human capital by adding it to the neoclassical Solow-Swan model. They found that whereas the basic Solow-Swan model could explain half of the variation in wealth between countries in their dataset, their extended growth model accounted for nearly 80 percent of the cross-country variation in income.
Overall, we can conclude that the neoclassical model and the classical convergence literature provides a good starting point to research convergence between the CEE10 and EU15. However, we must also conclude that the neoclassical Solow-Swan model of economic growth omits some very important determinants of growth. The endogenous technology models provide several other determinants of growth and eventually convergence between poor countries that lay behind the technological frontier and rich countries that are on the technological frontier. In particular, the positive externalities that are obtained from inward FDI flows and international trade are of interest and therefore also the effectiveness with which these externalities can be absorbed. The absorbability power relies for instance on the available human capital within the economy, on own R&D investments, but also on the institutions that govern and protect international trade and investments.
3. Methodology
This section provides the methodology of the research. First it describes the scope of the research, followed by the model that I have used and a specification of all the variables and finally a first glance at the GDP per capita data.
Scope of the research
13 group Bulgaria and Romania joined the EU in 2007 whereas all the others joined the EU in 2004. The EU15 group consists of Austria, Belgium, Denmark, Finland, France, Germany, Greece, Luxembourg, Ireland, Italy, the Netherlands, Portugal, Spain, Sweden and the United Kingdom.
The timeframe of the research will be from 1995 to 2015. The former USSR countries gained or gained back their independence in the early 1990s. So, for most of the CEE10 countries data on GDP is only available since 1995 in the World Bank Databank. To check for convergence I construct a sample with the CEE10 countries and all the countries of the EU15. The EU15 makes a good reference group for checking for convergence with the CEE10 since 14 countries of the EU15 were ranked as a high income country in 1995 according to the World Bank classification (Greece was ranked upper middle), whereas all CEE10 countries were ranked lower- or upper- middle income countries in 1995.
Model and hypotheses
Based on the literature review I construct a model with which I can control for the different types of convergence between the CEE10 and EU15 and for the effects of the convergence criteria. Dependent variable throughout the research will be the GDP per capita growth which is computed according to equation 1:
ℎ = ln , (1)
where ℎ is the annual GDP per capita growth of country i at time t and and are the GDP per capita levels of country i at time t and time t-1, respectively. To check for convergence the initial GDP per capita levels are vital, that is why the natural logarithm of the GDP per capita with a one-year lag is added as core explanatory variable in the model. I have added the variable in the form of a natural logarithm to reduce the large positive skewness within the original data. If the sample exhibits β-convergence the coefficient for the lagged GDP per capita variable has to be negative.
14 relationship with respect to the GDP per capita growth. The second variable, government debt, controls for the effects of the measure that sets a limit of 60 percent to total government debt. Again the goal of the measure is to achieve a more stable and higher economic growth, this time by lowering the government debt. Thus, I expect the government debt variable to have a negative impact on the economic growth rates. Next to the effects on economic growth these convergence criteria are also believed to result in convergence, to be able to find the effect of these criteria on the actual convergence between the CEE10 and EU15 I add two interaction variables. The two interaction variables are interactions between a dummy variable that takes the value of one if a country belongs to the CEE10 group with, respectively, the government balance variable and the government debt variable. If indeed these criteria result in convergence I expect to see impacts of the interaction variables on economic growth in such a way that they result in a higher growth rate for the initial poorer CEE10 countries.
15 such as property rights, corruption, business- and labor freedom and several more (all ten indicators are mentioned in appendix A). Since open and outward oriented countries are expected to outperform more closed economies and good institutions are required for countries to close the technology gap I expect a positive coefficient for the economic freedom variable with respect to economic growth. I further added dummy variables to control for the effects of EU membership and being a member of the Eurozone on economic growth. The EU and Euro dummy variables, respectively, take the value of one for a specific year if a country was a member state of the EU and had the Euro as its national currency in that same year. Next to these dummy variables I add two dummy variables that control for different period effects, I will elaborate on the different time periods further below in the data section.
I will use a panel econometric approach to take both the cross-sectional and the time-series components of the data into account. Furthermore, I will make use of a fixed effects model so that the model controls for country specific effects. To test whether pooled ordinary least squares, fixed effects or random effects where appropriate for the model I tested the full model with each three estimation methods and ran Hausman tests. The first Hausman test on the fixed effects and the pooled OLS models rejected the null hypothesis that both models are appropriate estimation methods and accepted the alternative hypothesis that the fixed effects method is the best estimation method. The second Hausman test on the fixed effects and random effects model again rejected the null hypothesis, which indicates that the fixed model is the model that fits the data best. The results from the second Hausman test is confirmed by the Breusch-Pagan Lagrangian multiplier test for random effects, which rejected the null hypothesis that a random effects model is the appropriate estimation method. The results of both Hausman tests and the Breusch-Pagan Lagrangian multiplier test for random effects are presented in appendix B.
First, I will check whether or not absolute β-convergence took place between the CEE10 and EU15 countries. Because absolute β-convergence implies that initially poorer countries should always grow faster than initially richer countries I only add the lagged variable for GDP per capita as explanatory variable to the model with as dependent variable the growth rate of GDP per capita. This results in the regression equation 2:
16 Absolute β-convergence is present within the data if the coefficient for is negative in regression equation 2. This results in the following testable hypotheses for absolute β-convergence:
H : ≥ 0 H : < 0
Thereafter, I will test for conditional β-convergence within the data and the effects of the convergence criteria by adding all the variables of interest and control variables to the right-hand side of the regression equation. This results in the equation 3:
ℎ = + + ℎ + +
+ + + + + 2
+ 3 + +
+ 10 ∗ + 10 ∗ + + ,
(3)
where is the fixed-error component and is the overall error component. All the other variables have already been described and are summarized in table 1. We find that conditional β-convergence is present if the coefficient for is negative in regression equation 3 and thus we have the same testable hypotheses for conditional β-convergence:
H : ≥ 0 H : < 0
The effect of the convergence criteria are measured in equation 3 through the government balance and government debt variables. If we accept the reasoning behind the convergence criteria the government balance has a positive relationship with economic growth and the government debt has a negative relationship with economic growth. This results in the following testable hypotheses:
H : ≤ 0 and ≥ 0 H : > 0 and/or < 0
17 before mentioned relationships of government balance and government debt on economic growth. So, we end up with the same testable hypotheses for the two interaction variables as for the regular variables of government balance and government debt:
H : ≤ 0 and ≥ 0 H : > 0 or < 0
Lastly, I test for σ-convergence between the CEE10 and EU15 countries. If the standard deviation of the GDP per capita for the full sample decreases over time then we find that
σ-Table 1: Variable specification and data collection
Variable Description of variable Source Expected sign growth Natural logarithm of the annualized growth rate of GDP per capita (in current US$) World Bank Dependent variable gdppct-1 A one-year lagged value of the natural logarithm of the initial GDP per capita (current US$) World Bank -
hc Percentage of labor force with tertiary education (% of total) World Bank + grcapf Natural logarithm of gross capital formation (% of GDP) World Bank + fdi Natural logarithm of FDI inward stock (% of GDP) UNCTAD + trade Natural logarithm of exports plus imports of goods and services (% of GDP) World Bank + free Index of economic freedom Foundation Heritage + EU &
Euro Dummy variables for being a EU member state and part of the Eurozone + & + Per2 &
Per3 Dummy variables for respectively the second and third time period (Per1 is reference group) + & -/+
govbal Governance balance (% of GDP) Eurostat +
govdebt Governance debt (% of GDP) Eurostat -
CEE10 * govbal
Interaction variable of the dummy variable CEE10 (takes the value of 1 if the country is part of the
CEE10 group) and the variable govbal +
CEE10 * govdebt
Interaction variable of the dummy variable CEE10 (takes the value of 1 if the country is part of the
18 convergence is present. The trend line of dispersion in GDP per capita between countries can be estimated with the following ordinary least squares regression equation (equation 4):
sd = + , (4)
where the dependent variable is the standard deviation of the natural logarithm of GDP per capita levels of all countries, the coefficient for shows whether or not the data exhibits σ-convergence and t is a time variable (t =1, …, 21 for the time period 1995 to 2015). If is negative in regression equation 4 is confirmed within the data. This results in the following testable hypotheses for σ-convergence:
H : ≥ 0 H : < 0 Data
The primary data that I use to research whether or not convergence is taking place is GDP per capita (in current US$), which is derived from the World Bank database. This data is available for all twenty-five countries for the full time period of 1995 to 2015. This implies that 25 countries deliver 20 observations each for the dependent variable which results in 500 observations. All the datasets for the other explanatory variables are all almost completely filled for the time period of 1995 to 2015. By adding all the variables to the model the dataset loses 54 observations due to missing data, thus we end with still 446 observations.
19 4 supports the indication of convergence from figure 3. It shows on the y-axis the average annualized growth rates per country over the period of 1995 to 2015 and on the x-axis the log GDP per capita level in 1995. Clearly visible is a pattern of higher growth rates for countries with an initially lower level of GDP per capita. Based on these findings I will from now on
Figure 3: Growth of GDP per capita for the CEEC and EU15
Figure 4: Average annualized growth rates of GDP per capita over the period 1995-2015
10 0 20 0 30 0 40 0 In de x 1995 2000 2005 2010 2015 Year CEE10 EU15
Growth of GDP per Capita (Index: 1995 = 100)
AUT BEL BGR CZE DNK EST FIN FRADEU GRC HUN IRL ITA LVA LTU LUX NLD POL PRT ROU SVK SVN ESP SWE GBR .0 2 .0 4 .0 6 .0 8 .1 Lo g G ro w th R at e 7 8 9 10 11
Log GDP per Capita 1995
20 occasionally talk about three different sub-periods, namely period 1 from 1995 to 2000, period 2 from 2000 to 2008 and period 3 from 2008 to 2015. I will also add dummy variables to the model to control for the different period effects.
Table 2 gives a summary of the GDP per capita levels per country and the average level of each group. The second and the third column give the GDP per capita of 1995 and 2015 respectively. If we take a look at the second column we see that the average GDP per capita level of the EU15 was more than 5.5 times as high as the average of the CEE10, with the largest
Table 2: GDP per capita (in current US$) and growth per country and groups
(2) (3) (4) (5) (6) (7)
Country / Group GDP per capita 1995 GDP per capita 2015 Growth '95-'15
21 gap between Luxembourg and Bulgaria (33 times as high) and the smallest gap between Portugal and Slovenia (only 10 percent). So, obviously large differences between countries in each group already existed in 1995. In the fourth column the percentage growth of GDP per capita between 1995 and 2015 is presented. Over this full time period we find that the average GDP per capita of the EU15 increased with 63 percent whereas the average GDP per capita of the CEE10 increased with 248 percent. Which results in the facts that the gap in average GDP per capita decreased from 5.5 times as high in 1995 to 2.3 times as high in 2015 and that the largest gap between countries, which is still between Luxembourg and Bulgaria, also decreased significantly (from 33 to 14 times as high). Moreover, if we look at columns 5 to 7, which present the GDP per capita growth rates per period, we see that on average the CEE10 growth rates are higher (or less negative) than the EU15 growth rates for each period, which suggests that convergence is taking place throughout the whole time period. The average growth rates per group per period, logically, correspond with the graph in figure 3, that convergence started of slowly during the first period, went more rapidly in the second period and stagnated in the third period. By comparing individual countries of both groups we can see that Slovenia has already surpassed the GDP per capita levels of Greece and Portugal and that, in particular, the Czech Republic and Estonia are also already at the same level. However, the Czech Republic and Slovenia were already above the average GDP per capita of the CEEC in 1995 and their achieved growth over the time period 1995-2015 is much lower than, for instance, the growth achieved by the three Baltic States.
22 model on the y-axis and the fitted values on the x-axis indicate that the residuals are normally distributed. I then used a Kernel density estimate to plot the distribution of the residuals. This graph confirms the previous finding that the residuals are normally distributed. All the complete results of the tests referred to above can be found in appendix C.
4. Results
The results section starts with the models controlling for absolute- and conditional β-convergence between the CEE10 and EU15. After the tests for β-β-convergence the results for σ-convergence will be presented. Followed by several robustness tests for both kind of convergence tests.
Beta convergence and convergence criteria
The β-convergence regression results are presented in table 3. In line with most of the existing convergence literature I start with the results of equation 2 that controls for the presence of absolute β-convergence. These results for absolute β-convergence between the CEE10 and EU15 over the timeframe 1995-2015 are presented in the first column of table 4. It presents solely the relationship between the one year lagged value of the log GDP per capita and dependent variable is the achieved. The GDP per capita variable shows a negative and statistically significant coefficient at the 1 percent level. This result implies that over the full time period absolute β-convergence is present, hence countries with initially lower GDP per capita levels have been unconditionally converging towards countries with a initially higher GDP per capita level. Because we know that in 1995 the country with the highest GDP per capita from the CEE10 was still lower than the country with the lowest GDP per capita from the EU15, we can state that absolute β-convergence has occurred between the CEE10 and EU15 between 1995 and 2015. However, the explanatory power of this model in the first column is rather low. The adjusted r-squared shows that only a small 6 percent of the variation within economic growth rates could be explained by the initial GDP per capita level. The following results on conditional β-convergence will show whether or not the faster growth from the CEE10 as compared to the EU15 within the second period has been conditional on the convergence criteria and/or on other determinants of economic growth.
23 Table 3: Regressions results for β-convergence
(1) (2) (3) (4)
VARIABLES growth growth growth growth
GDP per Capita -0.0574*** -0.112*** -0.114*** -0.113*** (0.00428) (0.0187) (0.0180) (0.0164) Human capital 0.000914 -0.000048 0.000184 (0.000839) (0.000797) (0.000875) Capital formation 0.103*** 0.217*** 0.203*** (0.0229) (0.0440) (0.0445) FDI 0.00378 0.00652 0.00794 (0.0162) (0.0165) (0.0177) Trade 0.136*** 0.101** 0.0896* (0.0403) (0.0437) (0.0485) Economic freedom 0.00130 0.00207 0.00190 (0.00137) (0.00128) (0.00127) EU 0.0455** 0.0440* 0.0495** (0.0205) (0.0213) (0.0196) Euro 0.0473*** 0.0539*** 0.0543*** (0.0111) (0.0122) (0.0109) Period 2 0.0995*** 0.0982*** 0.0955*** (0.0105) (0.0130) (0.0134) Period 3 -0.00207 0.00696 0.00267 (0.0158) (0.0202) (0.0212) Gov. balance 0.00247* 0.000487 (0.00128) (0.00206) Gov. debt 0.00135*** 0.00127*** (0.000276) (0.000295) CEE10*Gov. balance 0.00748* (0.00404) CEE10*Gov. debt -0.000368 (0.000354) Constant 0.606*** -0.0113 -0.290 -0.199 (0.0420) (0.275) (0.276) (0.312) Observations 500 451 446 446 Number of id 25 25 25 25 R-squared 0.059 0.525 0.536 0.547 Adj. R-squared 0.057 0.515 0.523 0.533 F-test 180.5 197.8 132.4 231.3 Prob > F 0.000 0.000 0.000 0.000
24 r-squared of 0.52 shows that the explanatory power of the model increases drastically as compared to the first model. The GDP per capita variable is again negative and highly statistically significant, moreover the coefficient becomes bigger which confirms that the β-convergence has been conditional on several economic growth determinants. All of the added control variables behave as expected or are statistically insignificant. The capital formation and FDI variables are both statistically significant at a 1 percent level and show a positive relationship with respect to economic growth, which supports the capital accumulation of the neoclassical growth models and the externalities of the endogenous growth models, respectively. The dummy variables that were added to control for the effects of EU membership and being part of the Eurozone on economic growth also both show positive coefficient and are, respectively, statistically significant at a 5 percent and a 1 percent level. The period dummy variables behave as expected and in accordance with figure 3. The second period variable returns a highly statistical significant positive coefficient which indicates that overall economic growth rates have been higher in the second period as compared to the first period. The insignificant third period dummy variable is insignificant which is in line with the earlier findings.
25 rather than a linear relationship (Reinhart & Rogoff, 2010; Cherita & Rother, 2012). These researches show that the turning point of the inverted-U is at roughly a government debt of 90 to 100 percent as percentage of GDP. However, if I add a quadratic variable for government debt to the model the results show an insignificant coefficient for this variable (p-value of 0.318). This theory, however, could still explain the positive coefficient since the mean government debt over the full timeframe for the CEE10 and EU15 group are respectively, 33 and 68 percent of GDP. Also must be taken into account that, both, the coefficients of government balance and government debt are rather small as compared to the coefficients of, for instance, capital formation and trade.
26 Sigma convergence
If the standard deviation of the log GDP per capita of all countries decreases over time σ-convergence is taking place within the sample. Figure 5 plots the standard deviation of the log GDP per capita over the time period of 1995 to 2015. Three linear fits are added to the figure for each of the sub-periods. Overall, the figure clearly shows a decreasing standard deviation which confirms that σ-convergence is taking place between the CEE10 and EU15 countries. The standard deviation of log GDP per capita of the full sample starts with the value of 1.098 in 1995 and reaches its lowest value of 0.660 in 2015. By taking a look at the different time periods it becomes clear that during each of the three sub-periods σ-convergence is taking place between the CEE10 and EU15. However, the linear fits show that the pace at which this took place varies greatly. Within the first period and second period the standard deviation of log GDP per capita decreased at a relatively constant pace. The standard deviation of log GDP per capita decreases a bit faster in the second period and therefore we can conclude that the σ-convergence was the fastest during the time period of 1995-2008. The third period still shows a decreasing standard deviation, however the speed of the σ-convergence decreases drastically as compared to the first and second period.
Table 4 shows the so-called trend line of dispersion in log GDP per capita for the full sample. It shows that for each time period the coefficient of the time variable is negative and
Figure 5: Sigma convergence
.6 .8 1 1. 2 St . D ev . 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 Year
27 statistically significant. The regressions give numerical support to figure 5 and confirm that the fastest σ-convergence took place between 2000 and 2008 and that the speed of σ-convergence decreased drastically in the last time period of 2008-2015.
The literature already pointed out that β-convergence and σ-convergence are related to one another. It showed that β-convergence is a necessary condition for σ-convergence, but vice versa σ-convergence is not a necessary condition for β-convergence (which was shown in panel C of figure 1). However, table 1 already showed that the CEE10 countries on average did not overtake the EU15 in terms of GDP per capita, so there is no situation as sketched in panel C of figure 1. I have found that, both, β-convergence and σ-convergence are taking place between the CEE10 and EU15 between 1995 and 2015. So, I can conclude that β-convergence and σ-convergence are related within the used dataset.
Robustness tests
To check if the convergence tests have been influenced by the some large growth differences across countries I conduct several robustness tests in which I drop certain countries from the sample. Also, the annual setup of the sample might be influenced by business cycle fluctuations. By constructing a dataset with three year averages I overcome this possible bias and this transformation also makes the sample less likely to be serial correlated.
The summary of the data in table 2 already showed that initial GDP per capita and the growth rates over time do not only differ between groups but also within both groups. For instance, the GDP per capita growth between 1995 and 2015 ranges from 94 to 553 percent
Table 4: Regressions results for σ-convergence
Full timeframe 1st period 2nd period 3rd period
(1) (2) (3) (4)
VARIABLES stdev stdev stdev stdev
t -0.0253*** -0.0242*** -0.0388*** -0.00379* (0.00144) (0.00523) (0.00230) (0.00159) Constant 1.117*** 1.128*** 1.234*** 0.740*** (0.0180) (0.0204) (0.0237) (0.0281) Observations 21 6 9 8 R-squared 0.942 0.842 0.976 0.486
28 within the CEE10 group and from 30 to 167 percent within the EU15 group. To test whether this affects the results I construct two samples, one in which I drop the five countries with the lowest growth rates from the CEE10 and the five countries with the highest growth rates from the EU15 and a sample in which I drop the five countries with the highest growth rates from the CEE10 and the five countries with the lowest growth rates from the EU15. Based on the data in table 2, the countries that are dropped from the first sample are Czech Republic, Hungary, Poland, Slovak Republic, Slovenia, Ireland, Luxembourg, Spain, Sweden and the United Kingdom. Intuitionally, we would expect convergence taking place at a faster pace since the average growth of these five CEE10 countries is higher than the average of all CEE10 countries and the average growth of the ten EU15 countries is lower than the average of all EU15 countries. In the second sample I drop the following countries: Bulgaria, Estonia, Latvia, Lithuania, Romania, Austria, Belgium, France, Germany and Greece. In this case the previous reasoning is vice versa and we would expect convergence taking place at a slower pace. For the sake of convenience I will refer to the first sample as the “fast sample” and to the latter sample as the “slow sample”.
29 variables. The only statistically significant variable is the government debt variable which again is positive and statistically significant.
Figure 6 present the sigma convergence results for the fast and slow sample. The figure shows the standard deviation of the natural logarithm of GDP per capita of the full sample, the fast sample and the slow sample. Noticeably, the fast sample containing the fast growing CEE10 countries and the slow growing EU15 countries had a standard deviation of 1.34 in 1995 whereas the slow sample containing the slow growing CEE10 countries and the fast growing EU15 countries had a much smaller standard deviation of only 0.91. The overall conclusion from the figure is that sigma convergence has taken place in both samples. However, remarkable is the fact that in 2013 the standard deviation of the fast sample becomes smaller than the standard deviation of the slow sample. This occurs because of the fact that since 2008 the ten countries within the slow sample have been diverging in GDP per capita terms. Because this sample groups the countries with the lowest growth from the CEE10 with the highest growth countries from the EU15 it is not surprising that the first sign of divergence in this arises in this group right after the global financial crisis.
To test whether business cycle fluctuations have influenced the overall results I test the model again with three year averages of the dataset. This means that for each variables three year averages are constructed and that the dataset now contains seven observations in time per country for each variable, whereas before the dataset contained twenty-one observations in time
Figure 6: Robustness test sigma convergence
.6 .8 1 1. 2 St . D ev . 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 Year
Full sample Fast Sample Slow Sample
31 Table 5: Robustness checks regression results
Fast sample Slow sample 3 year averages
(1) (2) (3) (4)
VARIABLES growth growth growth growth
GDP per Capita -0.113*** -0.115*** -0.133*** -0.101*** (0.0164) (0.0259) (0.0232) (0.0201) Human capital 0.000184 0.00131 0.00138 -0.00109 (0.000875) (0.00138) (0.00154) (0.000672) Capital formation 0.203*** 0.172** 0.126** 0.105*** (0.0445) (0.0643) (0.0583) (0.0161) FDI 0.00794 -0.00298 -0.00634 0.0157 (0.0177) (0.0226) (0.0260) (0.0134) Trade 0.0896* 0.160** 0.0326 0.0315 (0.0485) (0.0591) (0.0514) (0.0325) Economic freedom 0.00190 0.00473** 0.00137 -0.00253** (0.00127) (0.00202) (0.00151) (0.00101) EU 0.0495** 0.0695** 0.0535* 0.0336 (0.0196) (0.0276) (0.0253) (0.0227) Euro 0.0543*** 0.0708*** 0.0404*** 0.00172 (0.0109) (0.0208) (0.0131) (0.0108) Period 2 0.0955*** 0.0677*** 0.117*** 0.122*** (0.0134) (0.0158) (0.0199) (0.00930) Period 3 0.00267 -0.0496** 0.0339 0.0738*** (0.0212) (0.0225) (0.0263) (0.0177) Government balance 0.000487 -0.00563** 0.00193 0.000821 (0.00206) (0.00248) (0.00157) (0.00126) Government debt 0.00127*** 0.00105** 0.00100** -0.000041 (0.000295) (0.000421) (0.000381) (0.000277) CEE10*Gov. balance 0.00748* 0.0163** 0.00512 0.000486 (0.00404) (0.00697) (0.00431) (0.00290) CEE10*Gov. debt -0.000368 -0.000154 -0.000414 0.000034 (0.000354) (0.000483) (0.000710) (0.000342) Constant -0.199 -0.574 0.571 0.606** (0.312) (0.362) (0.342) (0.252) Observations 446 267 270 169 Number of id 25 15 15 25 R-squared 0.547 0.613 0.489 0.834 Adj. R-squared 0.533 0.591 0.461 0.819 F-test 231.3 8003 122767 385.6 Prob > F 0.000 0.000 0.000 0.000
32 5. Conclusion
The aim of this research has been to find out whether or not the CEE10 countries have been converging towards the EU15 countries and what has been the influence of the convergence criteria. The research focusses on convergence in GDP per capita terms between the initially poorer CEE10 countries towards the initially richer EU15 countries and spans the timeframe of 1995 to 2015. I have conducted tests for the appearance of β-convergence, (higher growth rates for initially lower GDP per capita countries) and σ-convergence (a decreasing standard deviation of log GDP per capita) between both groups. I have controlled for the effects of the two of the best-known convergence criteria on economic growth and convergence between both groups. The first convergence criterion that I control for is the one that sets a limit of 3 percent of GDP to the government deficit and the second convergence criterion which limits the government debt to 60 percent of GDP. If the relationship between these convergence criteria and the economic growth rates differs between the CEE10 and EU15 group we indeed find an impact on the convergence between both country groups.
A first glance at the GDP per capita data has learned that the CEE10 as a group achieved a higher growth rate between 1995 and 2015 than the EU15. Furthermore, a graph of the GDP per capita growth with 1995 as index year allowed me to specify three sub-periods in which the growth rate between the CEE10 and EU15 grew apart or stayed constant. The first period from 1995 to 2000 is characterized by small differences in growth in favor of the CEE10. The second period starts in 2000 and ends in 2008, the year that the global financial crisis hit. This period is characterized by an increasing gap in GDP per capita growth in favor of the CEE10. The period after the global financial crisis and the ongoing Euro crisis from 2008 to 2015 is characterized by economic decline. All twenty-five countries in the sample saw their GDP per capita decline during this period.
33 findings corresponds with the predictions from the neoclassical growth model and the predictions from the endogenous technology models. The first results on the convergence criteria showed that for the full sample, both, the government balance and government debt have a positive effect on the economic growth rates. After adding two interaction variables that interact the government balance and the government debt variables with a CEE10 dummy variable the results for the government deficit variable change. The results show that a higher government balance has a positive impact on the economic growth of the CEE10 countries, whereas it becomes insignificant for the EU15 countries. This implies that this convergence criteria indeed is leading towards convergence between both country groups. The interaction variable for government debt is insignificant and thus does not seem to result in convergence. The robustness checks have shown that if the five CEE10 countries with the lowest growth rates and the five EU15 countries with the highest growth rates are dropped from the sample the convergence force of the government balance variable even becomes bigger and returns a higher statistical significance level. The effects of the convergence criteria variables in the so-called slow convergence sample and in the sample containing three year averages are hardly statistically significant.
The tests for σ-convergence have shown that the standard deviation of the log GDP per capita levels of the full sample has decreased between 1995 and 2015. The standard deviation of the full sample also declined in every sub-period, however the speed at which the standard deviation declined varied strongly per period. This decline mainly took place in the first and second period. In the third period, after the global financial crisis hit the standard deviation of GDP per capita still kept declining, only at a much slower pace.
Overall, I can conclude that convergence in terms of GDP per capita has taken place between the CEE10 and EU15 between 1995 and 2015. The higher growth rates of GDP per capita for the CEE10 has not only been because of the initial lower GDP per capita levels, but also depends on the formation of capital, trade flows and the level of economic integration. Despite the mixed results from the robustness checks, the main results of the convergence criteria show that the government balance criterion is working towards convergence, whereas the government debt criterion seems to have a positive influence on economic growth for both country groups. It is also clear from the σ-convergence results that the convergence stagnated heavily after 2008, the year in which the global financial crisis hit.
34 boundary to the timeframe. Moreover, the CEE10 countries have been in transition, mainly in the first period, from a planned economy towards an open market economy which most likely resulted in more opportunities to achieve economic growth. The economic freedom variable, however, has not been significant in the main results and show mixed results within the robustness checks. Another limitation is the fact that the data is heteroscedastic and auto-correlated for which I needed to report the results with cluster robust standard errors.
Whereas most of the research on convergence between countries did not find any evidence of convergence taking place this research did find evidence of convergence. This research found that CEE10 countries with an initial lower GDP per capita have been converging towards the EU15 countries that had an initial higher GDP per capita. I showed that this convergence has not been unconditional, however conditional on capital accumulation trade flows and economic integration. The effects of the convergence criteria that have been imposed by the EU on economic growth show that the government balance criterion indeed has been resulting in higher growth rates for the CEE10 as compared to the EU15, which has contributed to the convergence between both groups. And that the effect of government debt on economic growth is equal across both country groups, thus does not specifically result in convergence.
35 References
Abramovitz, M. (1986). ‘Catchin up, forging ahead, and falling behind’. Journal of Economic History, XLVI, 385-406.
Aghion, P. and Howitt, P. (1998). ‘Endogenous growth theory’, MIT press, Cambridge, MA. Balassa, B. (1961). ‘Towards a theory of economic integration’. Kyklos, 14(1), 1-17.
Baumol, W. J. (1986). 'Productivity growth, convergence, and welfare: what the long-run data show.', American Economic Review, vol. 76, no. 5 (December), pp. 1072-85.
Baumol, W. J. and Wolff, E. (1986). 'Productivity growth, convergence, and welfare: reply.', American Economic Review, vol. 78, no. 5 (December), pp. 1195-9.
Benhabib, J., & Spiegel, M. M. (1994). ‘The role of human capital in economic development evidence from aggregate cross-country data’, Journal of Monetary economics, 34(2), 143-173.
Bernard, A. B. and Jensen, J. B. (1995). ‘Exporters, jobs, and wages in US manufacturing: 1976-1987.’, Brookings papers on economic activity. Microeconomics, 1995, 67-119. Borensztein, E., De Gregorio, J., and Lee, J. W. (1998). ‘How does foreign direct investment
affect economic growth?’, Journal of international Economics, 45(1), 115-135.
Checherita-Westphal, C., & Rother, P. (2012). ‘The impact of high government debt on economic growth and its channels: An empirical investigation for the euro area.’, European Economic Review, 56(7), 1392-1405.
Dowdrick, S. and Nguyen, D. (1989). ‘OECD comparative economic growth 1950-85: catch up and convergence’. American Economic Review, 79, 1010-30.
Gerschenkron, A. (1962). ‘Economic backwardness in historical perspective: a book of essays (No. HC335 G386)’. Cambridge, MA: Belknap Press of Harvard University Press. Grossman, G.M. and Helpman, E. (1991). ‘Innovation and Growth in Global Economy’, MIT
Press, Cambridge
Howitt, P. (2000). ‘Endogenous growth and cross-country income differences’. American Economic Review, 90, 829-46.
IMF. (1997). ‘World Economic Outlook’, Washington.
Krueger, A. O. (1998). ‘Why trade liberalization is good for growth.’, The Economic Journal 108 (September):1513-1522.
Lloyd, P. J., and MacLaren, D. (2004). ‘Gains and Losses from Regional Trading Agreements: A Survey’, Economic Record, Vol 80 No.251, pp:445-467.
36 Mankiw, N.G., Romer, D., Weil, D. (1992). ‘A contribution to the empirics of economic
growth.’, Quarterly Journal of Economics. Vol 107 (2), pp. 407-37.
Myrdal, G. (1956). ‘An international economy; problems and prospects’, New York, Harper, p. 11.
Nelson, R., Phelps, E. (1966). ‘Investment in humans, technological diffusion, and economic growth’, American Economic Review: Papers and Proceedings 61, 69-75.
OECD. (1998). ‘Open Markets Matter: The Benefits of Trade and Investment Liberalization.’, Paris: OECD. 1998.
Reinhart, C. M., & Rogoff, K. S. (2010). ‘Growth in a time of debt (digest summary).’, American Economic Review, 100(2), 573-578.
Romer, P. (1986). ‘Increasing Returns and Long Run Growth’, Journal of Political Economy, Vol. 94, pp.1002–37.
Romer, P. (1990). ‘Endogenous technological change’, Journal of Political Economy 98, S71-Sl02.
Romer, P. (1993). ‘Idea gaps and object gaps in economic development’, Journal of Monetary Economics 32, No.3. December.
Sala-i-Martin, X. (1990). ‘On growth and states.’, Ph.D. Dissertation, Harvard University. Sala-i-Martin, X. (1996). ‘The classical approach to convergence.’, Economic Journal, 106,
1019-36.
Solow, R. M. (1956). ‘A contribution to the theory of economic growth.’, Quarterly Journal of Economics, vol. 70, pp. 65-94.
Stiglitz, J. E. (2016). ‘De euro: Hoe de gemeenschappelijke munt de toekomst van Europe bedreigt’, Amsterdam: Athenaeum – Polak & Van Gennep.
Swan, T. (1956). ‘Economic growth and capital accumulation. Economic Record’, vol. 32, no. 63 (November), pp. 334-61.
Tinbergen, J. (1954). ‘International economic integration.’, Amsterdam, Elsevier, p. 95. Viner, J. (1950). ‘The Custom Union Issue’, Carnegie Endowment for International Peace, New
York.
37 Appendices
Appendix A – Heritage Foundation’s Index of Economic Freedom
The Index of Economic Freedom grades countries on a scale from 0 (no economic freedom) to 100 (absolute economic freedom). A countries overall score is derived by averaging the following indicators for economic freedom (which all are based on a scale from 0 to 100): property rights, freedom from corruption, fiscal freedom, government spending, business freedom, labor freedom, monetary freedom, trade freedom, investment freedom and financial freedom.
Appendix B – Determining right panel estimation method
Hausman test on the fixed effects model and the pooled OLS model
Hausman test on the fixed effects model and the random effects model
38 Appendix C – Multiple regression diagnostics
Modified Wald test for group wise heteroscedasticity in fixed effect regression model
Wooldridge test for autocorrelation in panel data
Correlation table of all variables gro wth gd pp cL
1 hc grcapf fdi trade free eu euro per2 per3
go
vb
al
go
vd
ebt vbal cee10go vdebt cee10go
Fisher-type unit root test for growth (based on augmented Dickey-Fuller tests)
Residuals plot and normal distribution of the residuals
-.2 -.1 0 .1 .2 R es id ua ls -.2 -.1 0 .1 .2 .3 Fitted values Residuals scatter plot
0 2 4 6 D en sit y -.2 -.1 0 .1 .2 Residuals
Kernel density estimate Normal density
kernel = epanechnikov, bandwidth = 0.0175