Regional Convergence Within and Across Countries:
Evidence from Germany and the European Union
Miriam Wagner*
June 2016
Supervisor: Dr. P. Miliones
Abstract
Using NUTS2 and NUTS3 data from Eurostat, regional convergence is investigated using a panel data approach based on the neoclassical growth model. Regions of Germany and of the current 28 member countries of the EU are inspected. More specifically, convergence within Germany, across the EU members, within East and West Germany each separately and convergence along former and current national borders are examined. This exercise is performed along two dimensions, namely income and human development. The time span ranges from 2000 until 2015, which allows to inspect the impact of the European debt crisis on convergence. For all cases and along both considered dimensions convergence can be unambiguously concluded to have taken place in that period. However, the strength of convergence is not equal: Most convergence is found across regions located along border and within East and West Germany separately. Comparing Germany with the EU, convergence is concluded to be stronger within the the former. Further, there seems to be stronger convergence in terms of the HDI that in terms of GDP per capita.
1. Introduction
Germany as well as the European Union (EU) are appealing settings to study convergence: Both, the reunification of Germany and the establishment of the EU can be seen as external shocks leading to similar politics and goals. Further, there were sizeable differences across East and West Germany as well as across the EU member countries before their respective 'fusions'. How did those external shocks affect convergence? Was it beneficial for or obstructive to it? For both cases, Germany and the EU, a great amount of convergence has already been achieved in the past. This is not surprising, as similar frameworks are assumed to be beneficial for convergence. However, some differences between East and West Germany as well as between EU countries still persist. These remaining differences give rise to the analysis of this thesis: Does the process of convergence still prevail in Germany after the year 2000 or has the process of convergence ended without closing the gap? How does the Eastern European enlargement in 2000 affect convergence within the EU?
Why are economists and politicians concerned about convergence? One major reason is that convergence represents or follows from integration, harmonisation and unity (Halmai and Vásáry 2012, Otoiu 2015). For the case of Germany it means the integration of the East German regions, i.e. the 'new federal states'. The regional transfer system and the special investment subsidies for the East demonstrate the importance the German government places on integrating the new federal states. They are supposed to help East Germany to catch up with West Germany in several aspects. For the case of the EU it means to integrate new member states. The treaty of the European Union claims 'the promotion of economic, social and territorial cohesion, and solidarity among Member States' to be one of its major goals. Convergence is a way of achieving this goal. Further, the European cohesion policy aims at 'reducing disparities between the various regions and the backwardness of the least-favoured regions'. It redistributes between regions. As is further elaborated in the following, there are still sizeable discrepancies within reunified Germany as well as within the EU.
those countries divergence is observed. Convergence within particular parts or federal states of Germany is also not necessarily observed. This has to be borne in mind when giving policy recommendations. Greater inequality is also associated with more fragile growth (Ostry et al. 2014). As economic growth is desired, national and supranational policies should aim at reducing inequality. Further does inequality undermine growth via other channels, such as health and education (human development) or via political and economic instability.
A special feature of this thesis is the nature of the data: All countries are subdivided into several regions according to the NUTS (Nomenclature of Territorial Units for Statistics) system of Eurostat. More precisely, the NUTS2 and NUTS3 level are consulted, i.e. there are two levels of aggregation considered in this thesis. This subdivision gives rise to study regional convergence.
These data allow to study convergence on a national and on an international level, as well as to study convergence across regions within specific parts of a country. In this particular case it means that convergence within Germany and across the EU countries is inspected whilst accounting for regional or national heterogeneity. Further, convergence within East and West Germany each separately is investigated. It also gives rise to compare regions located along specific borders. More specifically, convergence along three borders is inspected: Firstly, the former German border between East and West Germany, secondly, the external West German border and thirdly the external East German border. The results of those four cases are compared.
Further, subdividing a country into regions multiplies the number of observations and thus leads to a more precise analysis. Another feature of the thesis is that convergence is conditionally tested. Furthermore, it is investigated along two dimensions, namely income and human development. Considering two dimensions provides a more profound understanding of conditional convergence. It is conceivable that convergence only occurs along one of the dimensions, or that convergence is not equally strong along the dimensions.
The combination of applying a dynamic panel data approach on this regional Eurostat data and testing for convergence in terms of income and human development can be seen as innovative.
created a more similar setting. However, the difference is that Germany is a single country whereas the EU consists of 28 independent countries. How does this differentiate convergence between these two cases?
The remainder of this thesis is as follows: Section 2 gives a literature review, split up for Germany and the EU. In section 3 the neoclassical growth model implying convergence is briefly presented, including justifications why it is well-applicable to East and West Germany and to the EU members. Next, the methodology and data are described in section 4 and 5. Section 6 presents and discusses the results. The last section concludes and gives some policy implications.
2. Literature Review
The literature on convergence within Germany and within the EU is extensive, stressing further the relevance of this topic. Therefore, a selective representative overview is given for both. As this thesis focusses on convergence in terms of income and human development, the presented literature also relates to these two subjects. However, none of the presented literature has considered income and human development, measured by means of the Human Development Index (HDI), at the same time. Further, the literature review considers the concepts of cross-country and cross-regional convergence, whereas this thesis focusses only on the latter. The difference between the two is whether countries as a whole or regions within one country or within several countries are converging. None of the presented literature has considered convergence along the borders studied in this thesis, nor did any of the presented literature investigate convergence for both, Germany and the EU. Investigating these two cases simultaneously can thus also be considered as a novelty.
2.1 Literature Review for East and West Germany
pronounced. Afterwards it levelled off. This is in line with the neoclassical growth model according to which most of the catching-up occurs initially. Strong income convergence in the early 1990's is also confirmed by Brück and Peters (2009), Goebel et al. (2009) and Brenke and Zimmermann (2009).
Kosfeld et al. (2006) investigate, besides others, conditional convergence in terms of income within Germany from 1992 until 2000 applying a spatial econometric approach. They confirm convergence within Germany for that period. They also investigate East and West Germany separately. They find convergence within both parts. However, convergence within East Germany is more than four times stronger in East than in West Germany according to them. Contrary to this, Brück and Peters (2009) find slight divergence from the mid 2000's. Yet, they find convergence in terms of living standards after 2000. The living standard is also partly captured by the HDI, which is investigated by this thesis. A great difference is their data, which is microeconomic and originating from the GSOEP (German Socio-Economic Panel). Their finding of divergence in terms of income could be traced back to their different data as well as their different investigated time period.
Goebel et al. (2009) investigate the catching-up of income and the living standards from 1990 until 2008 using the GSOEP data. As opposed to Brück and Peters (2009), they find overall convergence from 2000 to 2008 in terms of income. Regarding the livings standards, they find slight divergence just after the reunification. Starting in the late 1990's, living standards started approaching each other again. In 2008, the difference in subjective living standard had narrowed significantly, but still remained.
Finally, convergence in terms of GDP per capita after 2000 is also confirmed by Brenke and Zimmermann (2009). They confirm the pattern of the process of convergence eventually slowing down. In 2008, East Germany still had capacity to catch up.
Zimmermann 2009). The former describes the mechanism of redistribution of financial means. Thus, the donor federal states co-finance the borrowing federal states. West German federal states are more likely to be donor federal states which accords with the intuition that West Germany is more wealthy or developed and that East German regions are the ones ought to catch up.
Overall, income convergence and convergence in terms of human development within Germany has been found to have taken place ever since the reunification. It is presumed to have been the strongest in the early 1990's. But also, after the year 2000, income convergence was still found. However, the gap between East and West Germany still persists. The transfers to and subsidies for East Germany aim at closing this gap.
This thesis and the literature presented above share the above-presented literature have in common the finding of convergence. A contrast is the approach: the considered literature focuses on a cross-country convergence approach, whereas this thesis follows a cross-regional approach. Thus, the data and its level of aggregation are dissimilar as well. Lastly, most of the presented literature does not use a panel data approach, unlike this thesis.
2.2 Literature Review for the EU Countries
Convergence in terms of income within the EU has been subject to several studies. Kaitila (2014) finds convergence, with some minor pauses, in terms of GDP per capita in the EU from 1960 up until 2012. However, a difference to this paper is the definition of the EU: throughout this paper, the EU refers to the 28 current member countries whereas Kaitila (2014) investigates the EU15 and the then-current members. The finding of convergence within the EU before the turn of the millennium (1960-1995) is also confirmed by Yin et al. (2003). They investigated convergence in terms of per capita GDP.
those two groups are then compared: They find considerably different but significant convergence rates for the two groups. This implies heterogeneity between the two groups, i.e. countries within the groups converge, but they do not find convergence between the two groups.
Income convergence, within the back-then EU25, has also been found by Paas and Schlitte (2006). They apply a spatial econometric model to test for conditional convergence from the period 1995 to 2003 and use the same data source as this thesis. Siljak (2015) finds convergence in terms of GDP per capita within the EU also in a more recent time period, from 1995 until 2013.
For the case of the EU, cross-regional convergence has already been inspected. Marques and Soukiazis (1998) apply a similar approach as this thesis: They follow the neoclassical growth model and apply the Eurostat NUTS data to test for conditional convergence in terms of income in the EU. However, they inspect a longer time period: 1975- 1995. In general, the process of convergence is found to be quite low in the EU. Yet, over those 20 years the countries were converging.
Montfort (2008) extends the analysis of regional convergence across the EU countries to the time period of 1995 until 2005. He investigates whether there is convergence in terms of GDP per capita on the NUTS2 level. He observed that, indeed, EU countries were converging over that time period. Yet the convergence speed was low and unsteady.
Otoiu and Titan (2014) use disposable income per capita, unemployment rates and incidence of low work intensity to represent the socio-economic well-being of the EU population as they are presumed to reflect the purchasing power, health and poverty. With respect to those measures they find regional convergence within the EU after 2000.
3. Conceptual Framework
The conceptual framework follows the neoclassical growth model of Solow (1956) and Swan (1956). It pioneered exogenous growth models implying convergence. Besides that, also relevant extensions of it, such as models including human capital or R&D, yield similar convergence properties: They all predict that poorer economies in terms of capital per capita grow faster than richer economies and will thus catch up. This is because the returns to inputs, including capital per capita, are diminishing according to the model. Thus, economics with an initially lower capital per capita are expected to grow faster. Catching-up of poorer economies is also getting boosted by the common awareness that imitation is easier and cheaper than innovation. Therefore, developing economies have an advantage in the sense that they can grow by imitating whereas more developed economies have to make more difficult and expensive inventions.
Another implication of the neoclassical growth model is that the convergence speed is higher, the further away an economy is from its steady state, i.e. the more an economy is lagging behind. Therefore, the speed of convergence is expected to be initially higher and to eventually decelerate. It also implies that there is a negative correlation between the initial level of the capital per capita and its subsequent growth rate.
More precisely, the neoclassical growth model implies conditional convergence, i.e. convergence only across economies which are similar (Barro and Sala-i-Martin 1995). Also, empirically conditional convergence has predominantly proven to hold, although other factors need to be controlled for in order to be able to refer to similar countries.
In addition, Abramovitz (1986) stresses the importance of the so-called social capabilities. These account for education, political, and financial institutions and thus capture social advancement. To give rise to convergence across economies, social capability has to be similar across them. More specifically, social capabilities are necessary in order to grow or converge by means of technological improvements: Imitations are easier and cheaper than innovations, but in order to implement them some similarities in a socio-economic sense must be given. They are needed to absorb technologies from more developed economies and to implement them in less developed ones. If social capabilities are not similar, or not adequately established in an behind-lagging economy, convergence may not take place.
economies will eventually catch up with richer economies because they exhibit higher growth rates. The closer an economy gets to its steady state, the slower it grows. This suggests that richer economies are closer to their steady state than are poorer ones. It also implies that there is a negative correlation between the growth rate of capital per capita and its level. Secondly, to give rise to convergence economies have to be similar, in an economic and socio-economic sense (social capabilities). If the condition of similarity does not hold, economies might strive for different steady states, i.e. might not converge.
3.1 Applicability to the Case of East and West Germany
After having been two separate countries for 41 years, the former Federal Republic of Germany (FRG) in West Germany and the former German Democratic Republic (GDR) in East Germany were reunified in 1990. The former had a free market economy and capitalism, whereas the latter had a centrally planned economy and a socialist system. In the course of the reunification in 1990, the West German economic and political system was imposed upon East Germany. Thus, the former GDR underwent drastic changes which had a huge impact on its evolution. West Germany was considerably less affected by the changes due to the reunification. The reunification led to a sudden politically and economically similar framework which was expected to lead to convergence.
Data Source: Eurostat. Own calculations.
Most importantly, two conclusions can be drawn from graph 1. Firstly, West Germany was and still is clearly ahead of East Germany in terms of GDP per capita: East Germany is indeed lagging behind. For both parts of Germany, the paths of the NUTS2 and NUTS3 level are almost identical.
Secondly, a small narrowing between the gap of East and West Germany's GDP per capita can be concluded: On the NUTS2 level, the difference declined by 23.4% from 2000 until 2014; on the NUTS3 level, the difference declined by 19.62% over 13 years. Thus, East Germany is catching up with West Germany. However, the narrowing of the gap is hardly visible. Also, the closing of the income gap by roughly 20% over 13 years implies that the process of catching-up is very slow. It could also suggest that the gap is rather persistent or that East and West German regions are not converging to the same steady state.
For both, East and West Germany, and for both NUTS level, a decline is observable in 2009, caused by the European debt crisis. However, West Germany was stronger affected than East Germany: The GDP per capita in West Germany declined by 0.57% on the NUTS2 and by 0.53% on the NUTS3 level. For East Germany, it only declined by 0.39% on both NUTS levels.
only reached its performance from 2007 on the NUTS2 level and only surpassed it by 0.004 points on the NUTS3 level, East Germany recovered much better: On the NUTS2 level it increased its performance by 0.1, comparing 2007 and 2010. The respective number for the NUTS3 level is 0.12.
Data Source: Eurostat and UIS. Own calculations.
Regarding the HDI, the same two conclusions can be drawn. Graph 2 undoubtedly demonstrates that East Germany was and still is lagging behind West Germany.
The closing of the gap is not as apparent from the graph as for the GDP per capita, but taking a closer look at the numbers, catching-up can be concluded as well: The difference between the HDI in West and East Germany decreased by 19.2% between 2000 and 2012. The finding of convergence along both dimensions, GDP per capita and the HDI, is in line with other empirical evidence presented in the literature review. However, as was also found for the development of the GDP per capita, the gap is quite persistent and narrows only slowly. It is also noteworthy that the gap between East and West Germany is much more pronounced for the GDP per capita than for the HDI. This is presumably because human development never exhibited great differences. Also, a large amount of differences among human development might have been already overcome before 2000.
The HDI declined from 0.899 to 0.896, or by 0.33%. In 2010 it had further declined to 0.877. In 2011 it recovered to 0.907. The HDI covers three components: Income, education and health. The drop in 2009 is mostly caused by the income component, as education and health did not change as much during the crisis. For East Germany there is a decline in HDI between 2008 and 2009 from 0.884 to 0.882. By 2010, East Germany had already recovered and surpassed its performance from 2008. Overall, West Germany was more negatively affected by the crisis than East Germany; also, it struggled more with recovering.
Even though the institutional and political framework is equal on a national level, there are differences arising from distinct federal policies. Education, science, cultural and economic policies are especially affected: The educational system may differ across federal states, R&D investments by the federal state government may follow different strategies, the infrastructure of cultural offerings is not equal across federal states and federal state governments may support or subsidize different industries. Nevertheless, the federal states have to obey national law and thus have a limited scope of action.
Conditional convergence requires economies or regions to be similar in an economic sense. They should also exhibit similar social capabilities. East and West Germany have a very similar legal framework, the same constitutions and are similar in a socio-economic sense. After having been reunified for ten years, the condition of similarity is assumed to be given. The two German parts thus constitute a favourable setting to test for conditional convergence. It is particularly interesting to find out whether convergence is still taking place after ten years of reunification: Is the gap narrowing further or is it persistent?
3.2 Applicability to the Case of the EU
The EU has its historical roots in 1950, when the European Coal and Steel Community began to interconnect European countries economically and politically. Over the past 66 years, the EU expanded to 28 country members1. The last country to join was Croatia in 2013. Ever
since 1950 several steps have been taken to achieve more integration, harmonisation and 1 The 28 countries are: Austria, Belgium, Bulgaria, Croatia, Cyprus, Czech Republic, Denmark, Estonia,
unity: The creation of Common and Single Market, the implementation of free movement of goods, services, people and money, the Schengen Agreement and the implementation of the Euro as common currency for most of the member states.
To show that there is room for convergence within the EU, the countries are grouped, following previous literature: One group consists of the EU15, the other group contains the remaining 13 member countries2. The rationale behind this division is that the EU15
constitutes the founders and countries which joined before April 2004. None of them is located in Eastern Europe. Afterwards, the first Eastern European countries became EU members. The division into the EU15 and the remaining 13 countries thus represents, firstly, this historic event of the Eastern European enlargement and, secondly, a cut in time. Graph 3 illustrates the average development in terms of the logarithm of the GDP per capita of those two groups.
Data Source: Eurostat. Own calculations.
First of all, for both groups of countries the graphs for the NUTS2 and NUTS3 level correspond. For the 13 more recent EU members the graphs of the NUTS2 and NUTS3 level exhibit a constant gap over the 14 years which might be caused by commuters. Besides that, 2 The EU15 contains Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy,
the EU15, on average, have a higher GDP per capita than the other remaining 13 countries on average. Also, it is apparent from the graph that the gap between the two groups narrowed over the 14 years, especially due to the 13 other countries growing faster than the EU15. Lastly, the gap between these two groups of countries is larger than the gap between East and West Germany.
The negative effect of the crisis is also reflected in the lines, as they all show a decline in 2009. The decline was slightly more pronounced for the EU15 countries: On the NUTS2 level, it decreased by 0.021 and by 0.022 on the NUTS3 level compared to 0.021 for both NUTS levels for the other 13 countries. By 2010, both groups were approximately back on their 2008 levels.
Graph 4 deals with the average development of the EU15 and the remaining countries in terms of the HDI.
Data Source: Eurostat and UIS. Own calculations.
larger when considering the GDP per capita than the HDI. The same is true for East and West Germany. However, in terms of GDP per capita as well as the HDI, the two EU country groups exhibit larger differences than East and West Germany.
The crisis had no negative impact on the HDI development of the other 13 countries. For the EU15, the HDI decreased on average by 0.001 from 2008 to 2009, which is a negligible effect. The HDI of Germany, more precisely both parts of it, was much stronger affected by the crisis than the EU country groups on average.
The EU started off as an economic union. Nowadays, it is also a political union: There are EU laws as well as common institutions, such as the European Parliament and the Court of Justice of the European Union. EU policies aim at, among others, cooperation at justice and home affairs. There are supranational laws and a strive for harmonisation in several respects such as equal standards and similar economic policies. Thus, the EU countries exhibit similarities in terms of politics and monetary policies. They are also approaching each other in terms of social capability due to, for instance, educational and vocational mobility as well as free movement of goods, capital and people. However, the countries are more autonomous than Germany's federal states and thus also their contained regions exhibit more differences. Nevertheless, unity, harmonisation and integration are major objectives of the EU which is why convergence is a desirable outcome. These objectives are manifested in the treaty of the European Union and its Cohesion Policy.
Besides that, it is interesting to study how the Eastern European enlargement affected convergence. Are the new EU member countries adequately similar to exhibit convergence with the EU15? The effect of the crisis is especially appealing as Portugal, Ireland, Spain and Greece are in the group of the EU15. Those countries were strongly affected by the crisis. How did this affect the convergence between the EU15 and the other 13 EU countries?
4. Methodology
prediction of the above-explained neoclassical growth model which constitutes the conceptual framework. Secondly, as elaborated in the literature review, conditional convergence has been empirically observed across countries and within countries. Thirdly, studying conditional convergence allows to discover whether poorer countries grow faster than richer ones and whether there is convergence across countries or regions with different structures or dissimilarities.
In order to analyse whether convergence holds for the case of East and West Germany and for the case of the EU countries, two aspects are considered: Convergence in terms of income and in terms of human development. GDP per capita serves as a proxy for the former, whereas the Human Development Index (HDI) serves as a proxy for the latter.
The neoclassical growth model implies convergence in terms of capital per capita, conditional on certain assumptions. This idea is captured by testing for convergence in terms of GDP per capita. The reasons for checking for convergence also in terms of the HDI are manifold: Firstly, economic growth or development of a country are more often investigated not only by looking at the GDP growth rates, but also by looking at alternative measures, such as the HDI, which consider additional aspects. This increase of use of alternative aspects is motivated by the second argument: Economic growth in terms of GDP growth is closely related to the education and health of a country's population: Education and health contribute to growth; and, at the same time, education and health are consequences of growth. Also Abramovitz (1986) stresses the importance of a similar social advancement for convergence to happen. Thirdly, as Becker et al. (2005) argue, incorporating life expectancy expands the analysis from a qualitative approach (in terms of GDP per capita) by a quantitative approach. Life expectancy is one of the three components of the HDI. This allows for a broader notion of the population's well-being. At the same time, they argue that increased life expectancy contributes to improvements of overall welfare. They also stress that income and life expectancy development must not necessarily have the same implications in terms of convergence. Fourthly, considering two aspects simply allows a larger insight on whether East and West Germany as well as the EU countries are converging in an even broader sense. The overall inference drawn is thus based on two investigations instead of one.
capita. The arithmetic mean of the three constitutes the HDI3. Due to data limitations, the
calculation of the HDI was only possible for the NUTS2 level.
An issue whilst calculating the HDI was that Eurostat does not provide data on expected years of schooling. These data have thus been retrieved from the UNESCO Institute for Statistics Data Centre. Unlike the Eurostat data, there is no subdivision of the countries into regions. Therefore, values of the expected years of schooling for a given year and country have been applied to all regions of that country. Another issue regarding Germany's HDI was that the UNESCO Institute for Statistics Data Centre only includes the expected years of schooling for the year 2014. Consequently two HDIs have been calculated: The first uses the German 2014 value for the expected years of schooling for all years and all regions of Germany. The second averages all EU values for the expected years of schooling in 2014 and applies this number to all years and all EU regions. These two HDIs have been used for the analysis for Germany and their results compared. However, both HDIs do not feature variability over time and are thus limited at detecting trends. As the results did not differ significantly, only the first HDI is reported to simplify matters and preserve clarity.
This thesis focusses on cross-regional convergence. This means that the data are aggregated on a regional and not on a country level. Section five elaborates further on the NUTS2 and NUTS3 level, which describe different levels of aggregation. Looking at distinct aggregation levels involves some benefits: Firstly, it leads to more precision. Further, if the results from both levels correspond it leads to higher credibility. However, if a region is very small, which especially concerns the NUTS3 level, it is very likely that a substantial fraction of its GDP per capita is attributed to commuters. This complicates interpreting the results. Yet aligning the results from both NUTS levels contributes to more reliability and robustness.
As opposed to the most of the presented literature, this thesis is going to apply a panel data approach. Advantages of using longitudinal data will be explained in section 5.
The analysis follows Islam (1995) who studies conditional convergence using panel data. This means that convergence is tested for after controlling for differences across regions, as well as year effects and other demographic factors. Convergence in terms of HDI is tested for 3 The exact procedure to calculate the HDI can be found in the Human World Development Report 2015 by the
United Nations Development Programme at:
conditional only on year effects and region fixed effects. In general, the effects are ought to be fixed and not random because in the latter case it is assumed that they are not correlated with the exogenous variables of the model. In this analysis the attributes of the regions are not the results of random variation; the exogenous variables and fixed effects are rather correlated. The basic specification is as follows:
git=ß1y(it−1)+ß2Xit+ßi+ßt+uit . (1)
where git=yit− y(it−1) . The dependent variable is either GDP per capita or the HDI. The
variable Xit captures demographic effects, ßt captures year effects and ßi region or
country fixed effects, depending on the specification, and uit is the error term. Equation (1) can be rewritten as
yit=(1+ ß1)y(it−1)+ß2Xit+ßi+ßt+uit , (2)
with (1+ ß1) capturing the idea of convergence: If
(1+ ß1)<1 , (3)
convergence can be concluded: A higher GDP per capita (or the HDI respectively) in the previous period (t − 1) results in a lower corresponding value for the current period t . Thus, a higher value for y(it−1 ) implies a lower growth rate and vice versa. As Islam (1995)
puts it, there is convergence if there is a negative correlation between the initial level of yit
and its subsequent growth rate y(it−1 ) . This is also an implication of the neoclassical growth
model (see section 3). Equation (2) describes the econometric specification actually estimated, leading to the presented results in section 6. If equation (3) holds, there is convergence.
included, the impact of the regions is averaged. This disregards regional differences. Including year effects is supposed to capture the impact of economic cycles.
To correct for possible heteroscedasticity and autocorrelation, the standard errors have been clustered. Mostly, three distinct regressions are reported for comparability as well as for robustness motifs: One includes no control variables, one includes all variables and one includes the baseline controls.
The selected set of control variables determines to what extent regions are considered similar in the sense of Solow (1956), Swan (1956) and Abramovitz (1986), i.e. what dissimilarities are controlled for. Similarity of economies is considered a necessary condition for convergence. Originally, the neoclassical growth model allowed for differences only with respect to the savings rate, the population growth rate and technological progress. This has been relaxed; rather more relevant and realistic control variables are included. They are suspected to significantly influence the GDP per capita.
details about the variables can be found in Appendix A.
To shed multifaceted light on the issue of convergence, the analysis is split up: Firstly, convergence within Germany and within the EU is tested for in terms of GDP per capita as well as in terms of the HDI. As argued above, both constitute favourable settings to test for convergence. The results of those regressions also allow to answer the questions whether Germany still exhibits convergence after being reunified for ten years and whether the EU15 and the other 13 countries are capable to converge. The impact of the crisis in 2008 on convergence within Germany and within the EU is also being reported for both, the GDP per capita and for the HDI. Further, convergence within East and West Germany, each separately, is examined to detect which part of Germany exhibits stronger convergence. This has also been conducted in terms of GDP per capita and HDI. This is appealing since they have been distinct countries for 41 years. It is conceivable that convergence within these two parts differ, as dissimilarities might still have remained after the year 2000. Thirdly, convergence along borders has been inspected. For this purpose, three borders are investigated: The first takes regions close to the former German border between the FRG and the GDR, i.e. the former German border, into account. The second considers nearby regions along the West German external border, including the respective German regions4. The third captures regions nearby
the East German external border5. Clearly, the number of regions and observations is thus
strongly unbalanced, i.e. there are more regions and observations for the external West German border than for the East German border. The rationale for inspecting borders is that they might exhibit more similarities and thus more convergence due to following the same trajectory. Also are spillovers more likely at close proximity. Furthermore, nearby regions along a border are more integrated. This is especially pronounced for the former German border.
4 The countries neighbouring Germany in the West are Austria, Belgium, Denmark, France, Luxembourg and The Netherlands. Austria is, despite its geographical location, considered to be rather Western than Eastern due to economic, political and historical aspects. This division also corresponds to the territories during the Cold War.
5. Data
The main source of the data used for this analysis is the Eurostat Database6. Only the data for
the expected years of schooling are retrieved from the UNESCO Institute for Statistics (UIS) Data Centre7. The Eurostat data is subdivided into the NUTS (Nomenclature of Territorial
Units for Statistics) system, where NUTS2 includes basic regions for the application of regional policies and NUTS3 contains small regions for specific diagnoses. Thus, NUTS3 is the finer subdivision. Considering Germany, the NUTS2 level includes 39 regions and 429 regions on the NUTS3 level. Those regions which are located in the former GDR are pooled and referred to as 'East Germany' in the following analysis; regions which are located in the former FRG are pooled and referred to as 'West Germany'8. Thus, each part of Germany
consists of several regions. The European level contains 28 countries, with 273 regions on the NUTS2 level and 1324 regions on the NUTS3 level. The 28 countries constitute the EU, thus the EU15 as well as the other 13 EU countries.
Most of the data starts in 2000 and ranges until 2015; the average observation period is 14 years. The beginning of the data collection in 2000 coincidences almost with the introduction of the Euro as the common currency of the EU. This constituted a major achievement of the EU. As for the case of East and West Germany, it can be argued that a finding of convergence even ten years after the German reunification shows that the process of catching-up is still not completed. Additionally, after having been a reunified country for ten years, it is reasonable to assume that East and West Germany have become even more similar in various aspects which favours convergence to take place.
The constructed data sets are unbalanced panel data sets. The number of observations differs depending on the NUTS level and on whether the German or the European data set is considered. The complete German dataset has around 700 observations on the NUTS2 and about 6600 observations on the NUTS3 level. The corresponding numbers for the complete European dataset are roughly 5100 and 23200. Due to data availability, the number of 6 Regional Statistics by NUTS Classification. Available at http://ec.europa.eu/eurostat/data/database.
7 Theme: Education, School Life Expectancy. Available at http://data.uis.unesco.org/.
observations and regions might differ between regressions.
In order to investigate convergence, longitudinal data is used. Clearly, the usage of panel data has some advantages over applying a cross-sectional approach: Firstly, the increased precision as the data is more informative and shows more variability. Secondly, the increased efficiency due the exploitation of the variation across individuals and over time: individual and time heterogeneity can be controlled for. Thirdly, panel data allows to study dynamic relationships and dynamics of adjustment, for example developments and trends. This is a relevant feature for the analysis of convergence since variables can be lagged. Fourthly, it enables one to take a potential omitted variable bias into account. Fifthly, panel data allows the inclusion of year as well as country or region fixed effects.
Correlation tables can be found in Appendix B.
6. Results
This section is subdivided into the four cases which are investigated: Convergence within Germany, convergence within the EU, convergence within East and West Germany each separately and convergence along current and former national borders. The cases are also compared. Convergence is present if the equation (3) holds, thus if the coefficient on the lagged GDP per capita (or HDI) is smaller than one. The smaller the coefficient, the stronger is convergence.
As mentioned earlier, a detailed description of the variables can be found in Appendix A. Correlation tables are presented in Appendix B.
6.1 Convergence within Germany
The coefficient on the lag of GDP per capita is smaller than one for all of the five regressions. Thus, convergence is present in each of the cases. However, the magnitude of convergence varies depending on the chosen set of controls. It is the largest for the second regression where all controls are included. Nevertheless, the number of observations drops sizably when the investment share and the agricultural employment share are controlled for, compared to regression (3) where they are excluded. Variables which are significant have the expected sign, apart from the population growth which has a positive impact. The effect of the crisis in 2008 can be evaluated by looking at the regressions (4) and (5). Since the coefficient on the lag of GDP per capita is higher after the crisis than before the crisis, it should be concluded that the crisis slowed convergence.
Table 2 shows the corresponding results of the NUTS3 level: They are in line with those of the NUTS2 level: Convergence is present in each of the five regressions. Also, the strongest convergence can be found in the second regression where all controls are included. For the NUTS3 level, the loss of observations due to the inclusion of the agricultural employment
Table 1: Convergence within Germany in terms of GDP per Capita on the NUTS2 Level No Controls All Controls Baseline Controls Before Crisis After Crisis
(1) (2) (3) (4) (5)
Dependent Variable Logarithm of the GDP per Capita
Log of GDP per Capita, Lagged 0.983*** 0.476*** 0.741*** 0.479*** 0.533*** (0.00252) (0.131) (0.0532) (0.0948) (0.0706) Mean Years of Schooling -0.00244 0.00942 -0.00687 0.0261
(0.0621) (0.0108) (0.0131) (0.0235) Life Expectancy 0.0214* -0.00339 -0.00560 -0.00326 (0.0121) (0.00676) (0.00569) (0.00727) Unemployment Rate -0.00200 -0.00181** -0.00334*** -0.00358 (0.00489) (0.000766) (0.000942) (0.00232) Population Growth 0.0240** -0.00227 -0.00709 0.000380 (0.0106) (0.00170) (0.00698) (0.00139) Investment Share 0.289 (0.181) Agricultural Employment Share 0.0202 (0.0299)
Year Effects No Yes Yes Yes Yes
Region Fixed Effects No Yes Yes Yes Yes
Observations 532 69 418 210 208
R-squared 0.804 0.902 0.945 0.924 0.869
Number of Regions 38 35 35 35 35
share is even more pronounced than on the NUTS2 level: It declines from 532 to 69. Also does the R squared suffer from its inclusion (0.684 compared to 0.453).
The greatest difference from the results on the NUTS2 level can be found when looking at the effect of the crisis: Whereas it was found that convergence decreased after the crisis on the NUTS2 level, the opposite holds for the NUTS3 level: Convergence increased after the crisis. Another difference is the negative impact of the population growth on the logarithm of GDP per capita; the impact was positive on the NUTS2 level. This can partly be explained by the stronger positive correlation of population growth and GDP per capita on the NUTS2 level compared to the NUTS3 level (see Appendix B). Further, the impact is only slightly positive on the NUTS2 level and solely significant at the 5% level. When only the baseline controls, and not the full set of controls, are included its impact is positive again. Probably, there is some bias due to correlation of the exogenous variables.
Next, convergence within Germany in terms of the HDI is looked at (see table 3). Due to data limitations, this is only possible on the NUTS2 level.
Table 2: Convergence within Germany in terms of GDP per Capita on the NUTS3 Level No Controls All Controls Baseline Controls Before Crisis After Crisis
(1) (2) (3) (4) (5)
Dependent Variable Logarithm of the GDP per Capita
Log of GDP per Capita, Lagged 0.989*** 0.107** 0.707*** 0.518*** 0.428*** (0.00187) (0.0493) (0.0230) (0.0374) (0.0422) Population Growth -0.0110*** -0.00595*** -0.000737 -0.0110***
(0.00373) (0.00202) (0.00189) (0.00336) Agricultural Employment Share -0.00697
(0.0110)
Year Effects No Yes Yes Yes Yes
Region Fixed Effects No Yes Yes Yes Yes
Observations 5,226 1,166 4,955 2,625 2,330
R-squared 0.684 0.453 0.790 0.618 0.582
Number of Regions 402 396 396 375 396
As can be seen from the coefficients on the lagged HDI, there is convergence in Germany with respect to the HDI; the coefficients are smaller than one in all four regressions. The R squared is very high throughout. This is partly caused by the fact that a part of the HDI for Germany does not vary over time because, when constructing it, data for expected years of schooling in Germany are available only for 2014 (see section 4). As it was the case with convergence in terms of GDP per capita on the NUTS3 level, the crisis in 2008 led to a raise in convergence: The coefficient on the lagged HDI after the crisis is smaller than the coefficient on the lagged HDI before the crisis.
Next, the effect of the crisis is discussed, as the results regarding the GDP per capita were conflicting when comparing the NUTS2 and NUTS3 level. In terms of the HDI, more convergence was found after the crisis. Thus, a tendency for an increased convergence after the crisis can be presumed. As is shown in section 3.1, the descriptive evidence suggests convergence over the whole of the time span. Further, a decline in the HDI after 2009 was observed as well as a drop in the GDP per capita in 2009 for both German parts, founded in the effects of the crisis. For both dimensions, the decline was more pronounced for West than for East Germany. Thus, the former was more negatively affected by the crisis. Also, graph 1 and 2 suggested that, when comparing the respective numbers from 2007 with those from 2010, that East Germany recovered better, i.e. outperformed itself whereas West Germany only slightly surpassed its performance from 2007. East Germany took the crisis as a chance to catch up. Combining the fairly straightforward descriptive evidence and the results for the
Table 3: Convergence within Germany in terms of the HDI on the NUTS2 Level No Controls Baseline Controls Before Crisis After Crisis
(1) (2) (3) (4)
Dependent Variable HDI
HDI, Lagged 0.974*** 0.735*** 0.659*** 0.447*** (0.00433) (0.0353) (0.0460) (0.119)
Year Effects No Yes Yes Yes
Region Fixed Effects No Yes Yes Yes
Observations 772 772 393 379
R-squared 0.921 0.954 0.917 0.855
Number of Regions 74 74 67 74
HDI, which both indicate towards more convergence after the crisis, with the ambiguous results for the GDP per capita, it is concluded that the crisis amplified convergence within Germany. This is to a great extent due to East Germany's superior performance.
Brück and Peters (2009) find slight divergence after mid 2000 in terms of income but convergence in terms of living standards. As to the extent that the HDI represents living standards, these results correspond. To reconcile their finding of divergence in terms of income with the finding of convergence in this thesis, it has to be borne in mind that the time spans differ and that this thesis averages convergence over 15 years. They are also consulting a different data source. As is discussed in further detail in the next paragraph, the crisis strengthened convergence within Germany on average from 2000 until 2015, although Brück and Peters (2006) found some divergence within a part of that period.
Overall it can be concluded that Germany is converging, even after ten years of reunification, in terms of GDP per capita as well as in terms of the HDI. With regards to this, the results from both NUTS levels correspond. The gap between East and West is closing, yet only very slowly. On the NUTS2 level, the crisis seems to have slowed convergence in terms of GDP per capita. However, for the NUTS3 level and with respect to the HDI the crisis is found to have amplified convergence.
6.2 Convergence within the EU
This section evaluates convergence within the EU following the process of section 6.1: Convergence in terms of GDP per capita on the NUTS2 and NUTS3 level and convergence in terms of the HDI on the NUTS2 level.
To begin with, all of the six regressions exhibit convergence. The overall convergence coefficients are between 0.788 and 0.991. These numbers are slightly higher than for the case of East and West Germany. Thus, according to the results from the NUTS2 level, there is somewhat less convergence within the EU than within Germany.
The effect of the crisis on convergence is sizeable: The coefficient on the lagged GDP per capita dropped from 0.836 to 0.560 which implies a stronger convergence within the EU after the crisis. As expected, the inclusion of the investment share and the agricultural employment share significantly lowers the number of observations and regions.
Comparing regression (3) and (4) reveals the difference between including region or country fixed effects. In both cases convergence can be found. However, more convergence can be found when region fixed effects are controlled for. The R squared is very similar. The interpretation differs in a sense that the inclusion of country fixed effects only captures differences across countries, but not across regions. Country fixed effects thus neglect the fact
Table 4: Convergence within the EU in terms of GDP per Capita on the NUTS2 Level
No Controls All Controls Baseline Controls Baseline Controls Before Crisis After Crisis
(1) (2) (3) (4) (5) (6)
Dependent Variable Logarithm of the GDP per Capita
Log of GDP per Capita, Lagged 0.984*** 0.788*** 0.817*** 0.991*** 0.836*** 0.560*** (0.00129) (0.0185) (0.0137) (0.00393) (0.0284) (0.0343) Mean Years of Schooling -0.0124* -0.00323 0.00543*** -0.0189*** 0.00493 (0.00652) (0.00322) (0.00153) (0.00578) (0.00647) Life Expectancy 0.00149 0.00478** -0.000750 -0.00132 0.0121*** (0.00393) (0.00187) (0.000536) (0.00281) (0.00290) Unemployment Rate -0.00490*** -0.00357*** -0.00104*** -0.00297*** -0.00569*** (0.000634) (0.000298) (0.000144) (0.000523) (0.000511) Population Growth -0.00291* -0.00437*** -0.00319*** -0.00736*** -0.000597 (0.00154) (0.000864) (0.000742) (0.00196) (0.00126) Investment Share 0.00804 (0.0422) Agricultural Employment Share -0.00151 (0.00103)
Year Effects No Yes Yes Yes Yes Yes
Region Fixed Effects No Yes Yes No Yes Yes
Country Fixed Effects No No No Yes No No
Observations 3,845 1,518 3,106 3,106 1,464 1,642
R-squared 0.862 0.918 0.912 0.902 0.904 0.706
Number of Regions 277 201 249 249 234 249
that regions within a country are not identical. Thus, the approach of including region fixed effects is more appropriate, also bearing in mind that the data is cross-regional.
The significant variables have corresponding and expected signs. An exception are the mean years of schooling which do not have consistent signs. Its impact on GDP per capita is supposed to be positive.
Next, the results of the analysis on the NUTS2 level are matched with those of the NUTS3 level in table 5.
Similar as on the NUTS2 level, convergence within the EU is overall slightly weaker than within Germany: Regressions (1) to (4) show convergence coefficients ranging from 0.779 to 0.994. The number of observations declines drastically when the agricultural employment share is included. Population growth and the agricultural employment share both have an expected negative and significant impact on the GDP per capita.
The effect of the crisis is the same as on the NUTS2 level: More convergence can be found afterwards as the coefficient declined from 0.715 to 0.573. Furthermore, the inference about the differences of either including region or country fixed effects are the same: Convergence is lower when controlling for country fixed effects than when controlling for region fixed
Table 5: Convergence within the EU in terms of GDP per Capita on the NUTS3 Level
No Controls All Controls Baseline Controls Baseline Controls Before Crisis After Crisis
(1) (2) (3) (4) (5) (6)
Dependent Variable Logarithm of the GDP per Capita
Log of GDP per Capita, Lagged 0.986*** 0.779*** 0.830*** 0.994*** 0.715*** 0.573*** (0.000684) (0.0137) (0.00709) (0.00128) (0.0189) (0.0169) Population Growth -0.00352*** -0.00341*** -0.00368*** -0.00124 -0.00130 (0.000989) (0.000752) (0.000526) (0.00113) (0.00112) Agricultural Employment Share -0.00183***
(0.000457)
Year Effects No Yes Yes Yes Yes Yes
Region Fixed Effects No Yes Yes No Yes Yes
Country Fixed Effects No No No Yes No No
Observations 17,410 6,633 14,899 14,899 7,494 7,405
R-squared 0.764 0.820 0.804 0.798 0.745 0.468
Number of Regions 1,342 1,064 1,242 1,242 1,161 1,242
effects. As explained earlier, including region fixed effects is more appropriate in order to control for regional heterogeneity. Overall, the results of the NUTS3 level tell the same story as those from the NUTS2 level.
Convergence within the EU in terms of the HDI is estimated and table 6 shows the results of estimating equation (2). Independent of the regression specification, convergence can be found within the EU. As with the results of regressing on the logarithm of the GDP per capita, convergence is stronger when including region fixed effects than when including country fixed effects. Similarly, convergence was weaker before than after the crisis.
When comparing these results to those from section 4.1, it becomes apparent that overall there is more convergence within Germany than within the EU. This might trace back to the fact that Germany has been reunited in 1990 and is thus 'older' than the complete set of the 28 EU countries: After the year 2000, still another 13 countries joined the EU. Besides that, the EU consists of 28 independent countries, whereas Germany is a single country. Naturally, the political and institutional framework is more consistent across regions and thus more favourable for convergence within Germany than within the EU. This indicates that regional convergence within a country is stronger than across countries, which is intuitive.
Various explanations for the slightly stronger convergence within Germany are conceivable: Firstly, linking back to the need of similar social capabilities (Abramovitz 1986), German regions might exhibit more similarities in this sense than the whole set of the EU regions and
Table 6: Convergence within the EU in terms of the HDI on the NUTS2 Level
No Controls Baseline Controls Baseline Controls Before Crisis After Crisis
(1) (2) (3) (4) (5)
Dependent Variables HDI
HDI, Lagged 0.977*** 0.702*** 0.967*** 0.627*** 0.497*** (0.00234) (0.0236) (0.00633) (0.0500) (0.0597)
Year Effects No Yes Yes Yes Yes
Region Fixed Effects No Yes No Yes Yes
Country Fixed Effects No No Yes No No
Observations 2,129 2,129 2,129 1,077 1,052
R-squared 0.862 0.938 0.902 0.882 0.808
Number of Regions 182 182 182 175 180
are thus more equipped to converge. Secondly, also the neoclassical growth model which implies convergence requires countries (regions) to be similar, but rather in a sense of economic measures. Clearly, the EU countries did exhibit and still do exhibit diversity as they are autonomous countries. This might hinder convergence. Thirdly, the neoclassical growth model predicts that regions (countries) catch up faster the further they are away from their respective steady state, i.e. the further they are away from catching-up with their peer regions (countries). Therefore it might also be the case that some regions in Germany were significantly lagging behind and are thus catching up faster, which in turn implies higher convergence. Differences across EU countries might be relatively major, causing them to exhibit less convergence. It also has to be borne in mind that the EU (already) showed great convergence before 2000 (Kaitila 2014, Yin et al. 2003). Afterwards, mainly Central European and Eastern European countries joined the EU. Traditionally, they exhibit less similarities with the West European countries which hinders convergence. It is also possible that the 13 new EU countries are converging to a different steady state than the EU15, as claimed by Caveneile and Dubois (2011).
The finding of convergence along both dimensions, GDP per capita and HDI, is in line with the implications from the descriptive evidence in section 3.2. Also, the finding of stronger convergence after the crisis corresponds with the descriptive evidence. It has been shown in section 3.2, that the EU15 countries were more negatively affected by the crisis and also struggled more to reach their pre-crisis levels than the other 13 EU countries. Just as East Germany used the crisis to catch up with West Germany, the 13 other countries caught up with the EU15 because they recovered faster from the crisis and were not suffering as much from the crisis. Thus, the crisis unambiguously strengthened convergence within the EU. This holds for both, the GDP per capita as well as for the HDI and for both NUTS levels.
strong convergence after 2012 might have amplified the finding of more convergence after the crisis. Also, as mentioned above, convergence already slowed from 2000 onwards according to her, which might be another reconciliation with the finding of less convergence before the crisis in this thesis.
Why could the crisis have increased convergence across the European countries? Several reasons are possible: In section 3.2 it has been shown that the EU15 countries suffered more from the crisis than the remaining 13 countries. The economies of the EU15 were affected more negatively and their growth slowed more, giving more room for other countries to catch up. Stronger convergence after the crisis has also been found for East and West Germany. However, the increase in convergence is more pronounced for the EU countries than for Germany. Comparing the recovery of the GDP per capita of West Germany, East Germany, the EU15 and the remaining 13 countries, it becomes apparent that East Germany outperformed itself the most (see section 3.1). West Germany and the other 13 EU countries achieved roughly their post-crisis level. The EU15 was not even able to reach its 2007 level of GDP per capita by 2014, most likely due to Southern European countries being affected the most by the crisis (think of Greece, Portugal, Italy). This is a possible reason for why convergence increased more after the crisis across the EU countries than within Germany: The EU15 grew comparably much slower, giving rise to the other 13 EU countries' catching-up. Generally, it seemed that there is less convergence happening in the whole EU than in Germany, probably due to less similar initial conditions or probably due to more convergence having already taken place. However, the crisis boosted convergence across the EU.
6.3 Convergence within East and West Germany separately
Since East and West Germany have been separate countries for 41 years, it is also interesting to examine whether there is convergence within both parts separately and not just within the whole of Germany. These results are compared. This analysis has been conducted for the GDP per capita on the NUTS2 and on the NUTS3 and for the HDI on the NUTS2 level.
Starting with the first, table 7 shows ambiguous results: On the NUTS2 level, there is more convergence within East Germany than within West Germany. Contrary, more convergence can be found in West Germany on the NUTS3 level. However, on the NUTS2 level the convergence coefficient for East Germany is significant only at the 10% level whereas it is significant on the 1% level on the NUTS3 level. There are only 46 observations for East Germany on the NUTS2 level. This number might be too small to capture the real effect, i.e. to reveal a significant, proper effect. Also, there are only four NUTS regions located in East Germany on the NUTS2 level, which might further hamper drawing decent inferences. Therefore, the results from the NUTS3 level should be preferred, although the number of
Table 7: Convergence within East and West Germany in terms of GDP per Capita
NUTS2 NUTS3
West Germany East Germany West Germany East Germany
(1) (2) (3) (4)
Dependent Variable Logarithm of GDP per Capita Logarithm of GDP per Capita Log of GDP per Capita, Lagged 0.766*** 0.362* 0.679*** 0.771***
(0.0532) (0.122) (0.0227) (0.0482) Mean Years of Schooling 0.00646 0.0678
(0.0114) (0.0343) Life Expectancy -0.00312 0.0135* (0.00685) (0.00528) Unemployment Rate -0.00311 -0.00284 (0.00186) (0.00168) Population Growth -0.00130 -0.0322 -0.0112*** -0.000119 (0.00173) (0.0164) (0.00218) (0.00244)
Year Effects Yes Yes Yes Yes
Region Fixed Effects Yes Yes Yes Yes
Observations 372 46 4,227 728
R-squared 0.939 0.989 0.784 0.823
Number of Regions 31 4 326 70
observations and regions between East and West Germany is still highly unbalanced. They indicate higher convergence in West than in East Germany. Life expectancy is the sole significant control variable on the NUTS2 level with an expected positive impact. On the NUTS3 level, the population growth has a significantly negative impact. Overall it remains unclear if convergence is stronger within East or West Germany in terms of GDP per capita, but it is concluded that convergence is stronger in West Germany.
Besides that, it is also interesting to look at the different impacts of the population growth on the GDP per capita in West and in East Germany. As regression (2) only has 46 observations, the NUTS3 level is considered: Population growth is less obstructive to increasing the GDP per capita in East Germany. Generally, the population density is much lower in East Germany. It is considered to be so low that it is actually hindering economic growth (Brenke and Zimmermann 2009): The labour supply is too small and particularly too little educated to contribute to economic growth.
Lastly, convergence within East and West Germany separately has been investigated in terms of the HDI.
Contrary to the results for the GDP per capita, table 8 suggests more convergence within East than in West Germany; the convergence coefficient on East Germany is less significant than is the convergence coefficient for West Germany. The R squared is higher than 0.9 in both cases. As explained in section 4, the part of the HDI that concerns the expected years of schooling is
Table 8: Convergence within East and West Germany in terms of the HDI
West Germany East Germany
(1) (2)
Dependent Variable HDI
HDI, Lagged 0.493*** 0.482** (0.0738) (0.146)
Year Effects Yes Yes
Region Fixed Effects Yes Yes
Observations 310 49
R-squared 0.983 0.991
Number of Regions 31 7
not varying over time for the analysis of Germany. This contributes to a very high R squared, especially when no other control variables are included in the regression. Obstructive to a proper inference is also the low number of observations and regions for East Germany, as already elaborated further above (see section 6.3).
Whether there is more convergence within East or within West Germany is ambiguous. Most likely, the difference is not sizeable. It is also possible that in terms of GDP per capita there is more convergence within West Germany and more convergence in East Germany considering the HDI.
Kosfeld et al. (2006) find more convergence within East Germany in terms of productivity and income for the period from 1992 to 2000. Possibly lots of convergence within East Germany took place during that period, leaving less room for convergence after 2000 in terms of income. This might be due to the collapse of the socialist system, giving more room for development and catching-up. Once again, most of the catching-up happens initially and slows down eventually.
However, it is noticeable that within both parts of Germany there is more convergence compared to within the whole of Germany. It can be argued that this is due to the required social capabilities (Abramovitz 1986) and required similarity of economic factors (neoclassical growth model): As the whole Germany does not have a very long joint history (1990 until today) compared to East and West Germany each separately (1949 until 1990), there might still be greater differences across all of the German regions than across East German and West German regions independently. It is also conceivable that West Germany is closer to its steady state than East Germany.
6.4 Convergence along Borders
Finally, regions which are located along a common border are inspected. The reasons are possible spillover effects as well as proximity in various terms such as geography, culture, politics and attitudes towards economic actions. Regions might also be more integrated and follow a more similar trajectory.
As mentioned earlier, three borders are considered: The former German border between the FRG and the GDR, the external West German border and the external East German border. The analysis is the same as in the preceding subsections: Firstly, the GDP per capita, and secondly, the HDI are inspected. The HDI analysis is only possible on the NUTS2 level.
Table 9 shows the results for the former. The first thing to notice is the similarity between the convergence coefficients from the NUTS2 and the NUTS3 level for the respective regressions: Along the former German border convergence is most pronounced and least pronounced for the external East German border. The magnitudes of convergence correspond as well.
Table 9: Convergence along Borders in terms of GDP per capita
NUTS2 NUTS3
Former German Border Western Border Eastern Border Former German Border Western Border Eastern Border
(1) (2) (3) (4) (5) (6)
Dependent Variable Logarithm of the GDP per Capita Logarithm of the GDP per Capita
Log of GDP per Capita, Lagged 0.609*** 0.723*** 0.814*** 0.703*** 0.719*** 0.910*** (0.0669) (0.0540) (0.0471) (0.0393) (0.0306) (0.0419) Mean Years of Schooling 0.0394*** 0.00411 0.0437***
(0.0107) (0.00937) (0.0105) Life Expectancy -0.00153 -0.00812 -0.000864 (0.0124) (0.00800) (0.00453) Unemployment Rate -0.00281** -0.00443*** -0.00259*** (0.00104) (0.000771) (0.000562) Population Growth 0.00102 -0.00337 -0.00879* 0.000999 -0.00437** -0.00598 (0.00338) (0.00284) (0.00476) (0.00207) (0.00209) (0.00830)
Year Effects Yes Yes Yes Yes Yes Yes
Region Fixed Effects Yes Yes Yes Yes Yes Yes
Observations 119 341 126 1,600 3,374 214
R-squared 0.955 0.887 0.981 0.775 0.773 0.949
Number of Regions 10 27 10 132 269 24
The result of strong convergence along the former German border is not surprising since those regions are within the same country, whereas regions along the external West German border as well as regions along the East German border belong to distinct countries (see section 4). Thus, regions along the former German border were presumably more similar initially and were thus expected to show more convergence. Regions within one country are more integrated. The convergence coefficients for the external East German border are higher than most of the convergence coefficients obtained in subsection 4.1 which analysed convergence in the whole Germany; the coefficients on the lagged GDP per capita for the external West German border are slightly lower on average. The same holds for the convergence coefficients along the former German border. The R squared is high in all cases and the significant control variables (mean years of schooling, unemployment rate, population growth) exhibit the expected sign.
Table 10 examines convergence along borders in terms of the HDI on the NUTS2 level. Likewise, convergence is very strong along the former German border and the weakest along the external East German border. Convergence is most pronounced along the external West German border. Also are the R squareds high. Again, the HDI for Germany is not perfectly varying over time, leading to an upward biased R squared, most likely in all of the three regressions. This is expected to be most pronounced for the first regression as it contains only German regions. The HDI, and especially the part considering the expected years of schooling, of the other countries and their respective regions is varying over time.
Table 10: Convergence along Borders in terms of the HDI
Former German Border Western Border Eastern Border
(1) (2) (3)
Dependent Variable HDI
HDI, Lagged 0.482*** 0.383** 0.847*** (0.106) (0.173) (0.143)
Year Effects Yes Yes Yes
Region Fixed Effects Yes Yes Yes
Observations 110 156 59
R-squared 0.983 0.952 0.990
Number of Regions 12 14 5
As expected, there is more convergence along borders than overall (compare to sections 4.1 and 4.2). This is especially pronounced for the former German border and the external West German border. It holds for the GDP per capita as well as for the HDI. High convergence along the former German border reflects the similarity across those regions which strongly favours convergence. As mentioned earlier, this is largely because those regions are situated in the same country. The finding of the least convergence along the external East German border might be explained by the remainder of historical differences: Poland, the Czech Republic and the former GDR had a close proximity with the Soviet Union in a political sense, they shared borders and were even partially part of it. Before the turn of the millennium, all three of them had once possessed a communist system. By means of the German reunification, the former GDR was suddenly no longer similar to Poland and the Czech Republic in these senses. Therefore, it developed differently which created dissimilar settings in the sense of Abramovitz (1986) and the neoclassical growth model. Dissimilarities are not favourable for convergence.
The reverse argument holds for the external West German border: Historically, the regions resembled and still exhibit similar politics and economies. Further, regions along the external West German border did not experience such a shock as did the regions along the external East German border due to the German reunification. Therefore, regions along the external West German border evolved increasing similarities which favours convergence, unlike regions along the external East German border.
Overall, most convergence is found along the former German border. Least convergence can be concluded for the external East German border. Regions along the former German border are more similar, giving more rise to convergence.
