The country fixed effects estimates from the previous section suggest that the increase in economic openness in Europe has amplified the structural differences among European economies due to the heterogeneous effect of openness on different countries. We now aim for gaining a clearer understanding of this observed heterogeneity. To this end, we start with an inductive approach and analyze the country fixed effect estimates obtained in the previous section by using hierarchical cluster analysis. In doing so, we try to identify suitable subgroups of the European countries in our data set and complement this inductive approach with theoretical considerations. Then we use sectoral export data to study the reasons underlying this structural change in European countries.
Hierarchical clustering of country fixed effects
In order to identify potential clusters of countries that show similarity in their unobserved country characteristics in response to European economic integration, we analyze the country fixed effects obtained in the previous section by using hierarchical cluster analysis (HCA, Tan et al. 2005, p. 515ff). The general idea behind HCA is to separate a set of objects into disjunctive groups, called clusters, where members of the same cluster are similar to each other, but distinct to members of other clusters. In contrast to partitional clustering, hierarchical clustering produces a set of nested clusters that are organized as a tree, usually represented as a dendogram or a factor map (see Fg. 3 below), which also allow for tracking the relation between clusters (see also Tan et al. 2005, p. 526)Footnote 4.
The results are presented in Fig. 3. Obviously, Luxembourg is quite distinct from the rest, which can be seen as a first indication that the intuition of separating countries in which the financial sector plays an outsized role into a proper sub-group might be a fruitful approach. The countries can be separated into four further groups. The cluster on the bottom consists of Austria, Denmark, Sweden, the Netherlands, Finland, and Germany. These are the typical ‘core countries’. The cluster on top, consisting of Spain, Cyprus, Portugal, Greece, Italy, France and Belgium corresponds – with the exception of Belgium and (maybe) France – to the classic conception of a European periphery. The remaining two clusters include the Eastern European catch-up countries, Malta and Ireland. Interestingly, these countries are separated into two clusters, of which the smaller one consists of Slovakia, Lithuania, Latvia and Bulgaria, while the other one comprises all other Eastern European countries as well as Malta and Ireland. This result is consistent with recent findings that highlight the presence of different sub-groups in the Eastern European countries (see e.g. Bohle 2018), which exhibit different degrees and intensities in the overall catch-up process observable in Eastern Europe.
All our clustering results are robust, not only with regard to different cluster algorithms, but also regarding the exclusion of smaller economies, such as Malta, Luxembourg, and Cyprus. An extensive robustness analysis exploring all these avenues is presented in the appendix.
In summary, although hierarchical clustering is a purely inductive way of analyzing data that does not exploit theoretical insights other than that involved in variable selection, the results are largely consistent with classifications used in the previous literature.
A country taxonomy for the EU: delineating clusters with theory and descriptive statistics
Previous taxonomies usually focused on particular subsets of the EU’s member countries. The most common distinction is that of a Eurozone core and a Eurozone periphery (e.g. Simonazzi et al. 2013; Iversen et al. 2016). Since the Eastern European countries are difficult to accommodate in this dichotomous classification, they are – if considered at all – usually treated as a third category (Bohle 2018).
Table 1 summarizes our country groups, which departs slightly from the results of our clustering analysis: although the overall clustering results are intuitive, the focus on the country fixed effects estimates as inputs for the clustering may still understate important differences with regard to some of the EU countries’ specificities in terms of their national regulations and institutions. As can be seen, we go beyond previous classifications and suggest categorizing the European Union’s members into four categories: core, periphery, catching-up countries in Eastern Europe, and financial hubs. While the classification of core – as those countries chracterized by high standards of living, a modern and highly competitive production sector and low unemployment (see Fig. 4a) – and periphery countries – as those countries with less competitive firms, higher unemployment rates and especially burdensome levels of debt (see Fig. 4b) – is rather standard, the group of financial hubs and catch-up countries, as well as the classification of France deserves further explication:
Table 1 Country taxonomy for 26 EU countries. Own illustration First, we add a proper group for financial hubs in the EU because the financial sectors in Luxembourg, the Netherlands, Malta and Ireland are outsized compared to other European countries (e.g. Karwowski et al. 2017; European Central Bank 2016; Schwan 2017, note that the UK is not part of our EU country sample)Footnote 5. In our data, this is reflected by a disproportionate amount of foreign direct investments as well as high levels of private sector debt, an exceptional share of the finance sector in gross output and relatively large incomes derived from the taxation of wealth (see Fig. 4c). On top of that, these four countries also feature an exceptionally large ‘shadow banking sector’ (Beyer and Bräutigam 2016, and chart 2 in European Central Bank 2016), where ‘shadow banking’ is understood as the non-banking part of the financial system, characterized by looser regulations and thinner public safety nets for financial institutions (Ban and Gabor 2017). Moreover, Luxembourg, Ireland, the Netherlands and Malta have followed particularly liberal and finance-friendly policies geared towards attracting foreign capital and the associated rents and profits from other (European) countries. The Netherlands has a prominent role as a hub in the ‘shadow banking system’ (Bakk-Simon et al. 2012; Broos et al. 2012; Beyer and Bräutigam 2016). Ireland has been using a low-tax and low-financial-regulation regime to attract multinational companies as well as leading global financial services firms. These low-regulation policies have played an essential part in the Irish export-led growth model (e.g. Barry and Bergin 2012; Zucman2014). Malta implemented finance-friendly policies that have led to an exceptional growth of its banking sector over the last two decades. Notably, a majority of the banking-sector’s total assets in Malta are foreign-owned (e.g. European Central Bank 2016). Finally, Luxembourg is a financial center with favorable tax policies for high-net worth individuals and institutional investors, leading to an outsized role of finance in the overall economy (e.g. Johannesen and Zucman 2014; Zucman 2015). These considerations lead us to classify these countries as financial hubs, rather than as core or periphery countries.
Second, the Eastern European countries are often termed catching-up countries; they consist of Bulgaria, Romania, Czech Republic, Estonia, Latvia, Lithuania, Hungary, Poland, Slovenia and Slovakia. These countries still display relatively low levels of income, low levels of wages and employment standards and large capital inflows. Moreover, the data indicate a weak foreign ownership position of the Eastern countries (captured in a negative difference between foreign assets and foreign liabilities of more than 75%). At the same time, their share of the industry sector in terms of employment is large in comparison to the other countries in our data set (see Fig. 4d). But along with these similarities, there are also important differences among Eastern countries. Most notably, while we can observe a certain catch-up process in terms of technological capabilities, particular for the Visegrad countries, no such process can be observed in the Baltics (see also below, as well as figure 6 in the Supplementary Material; for more details, see Bohle 2018). Nevertheless, we decided to treat these countries as a single cluster and leave a more detailed classification for future research.
Finally, while the clustering approach suggests that France is currently part of the periphery, classifying this country is difficult and one might also consider it as part of the core (such as, e.g., Artis and Zhang 2001; Campos and Macchiarelli2018). The country can be seen as an intermediate case between core and periphery and its location in the core-periphery nexus is not necessarily in line with its important political role in the EU, which is also determined by its size and its historically close relation to Germany (Gräbner et al. 2017). Nevertheless, we argue that if one focuses on the economic factors, France is closer to the periphery than to the core – especially if we take into account its development in terms of technological capabilities (see Section 4.3 and figure 6 in the Supplementary Material).
Structural change and the sectoral development of nations: assessing the directedness of technological change
While the previous sections focused primarily on the effects of European economic integration on macroeconomic indicators, we now turn to the mechanisms underlying macroeconomic convergence and divergence between countries. As suggested by the structuralist literature surveyed in Section 2, we focus on analyzing the dynamic distribution of technological capabilities. To this end, we use data on trade and economic complexity (Hidalgo and Hausmann 2009) to construct a measure for the direction of technological change relative to the rest of the world.
In particular, we compare trade volumes of all countries on the SITC-V2 4-digit product level over the two time periods 1995-1999 (pre-Eurozone and pre-crisis) and 2010-2014 (post-Eurozone and post -crisis) to assess the changes in a country’s export basket. For each country, we regress the log of the positive and negative difference in the value of exports on the average product complexity (PCI, see Hidalgo and Hausmann 2009) and weight the observations according to the share of the product in the country’s export-basket in 2012-2014. This allows us to understand, for a given country, whether export values change more drastically for more or less complex products. The weights ensure that we pay more attention to products that have recently played an important role in the country’s export-basket.
Define \(P_{c}^{+}\) as the set of products for which country c has increased its exports in 2010-2014 as compared to 1995-1999 and ϕc, i = 1 if \(i\in P_{c}^{+}\) and zero otherwise. We then estimate the following two equations for each country:
$$ \log\left( \sum\limits_{t=2010}^{2014}\phi_{c,i}\pi_{c,i,t}-\sum\limits_{t=1995}^{1999}\phi_{c,i}\pi_{c,i,t}\right)=\beta_{c}^{+}\overline{\text{PCI}}_{c,i}+\epsilon_{c,i}\quad\forall i\in P_{c}^{+} $$
(2)
and
$$ \log\left( \sum\limits_{t=1995}^{1999}(1-\phi_{c,i})\pi_{c,i,t}-\sum\limits_{t=2010}^{2014}(1-\phi_{c,i})\pi_{c,i,t}\right)=\beta_{c}^{-}\overline{\text{PCI}}_{c,i}+\epsilon_{c,i}\quad\forall i\notin P_{c}^{+} $$
(3)
In both equations, πc, i, t is the total export of product i by country c in period t ∈ ({1995,...,1999},{2010,...,2014}), and \(\overline {\text {PCI}}_{c,i}={\sum }_{t}\left [\frac {\pi _{c,i,t}}{{\sum }_{t}\pi _{c,i,t}}\text {PCI}_{i,t}\right ]\), where PCIi, t is the product complexity of product i in year t as defined in Hidalgo and Hausmann (2009). The weights ωc, i for the WLS estimation are given by \(\omega _{c,i}=\frac {{\sum }_{t}\pi _{c,i,t}}{{\sum }_{i}{\sum }_{t}\pi _{c,i,t}}\), i.e. the share of product i in the country’s export basket in 2012-2014. This way, we obtain two estimates for each country, \(\hat {\beta }_{c}^{+}\) and \(\hat {\beta }_{c}^{-}\), the first for the products for which the country has increased its export value, and the second for the remaining products.
By calculating a weighted average of these two coefficients, one arrives at a final estimate for the direction of technological change in the countries under investigation. To this end, define
$$ \gamma_{c}^{+}=\sum\limits_{t=2010}^{2014}\phi_{c,i}\pi_{c,i,t}-\sum\limits_{t=1995}^{1999}\phi_{c,i}\pi_{c,i,t} $$
(4)
as the sum of increases in exports of country c and
$$ \gamma_{c}^{-}=\sum\limits_{t=1995}^{1999}(1-\phi_{c,i})\pi_{c,i,t}-\sum\limits_{t=2010}^{2014}(1-\phi_{c,i})\pi_{c,i,t} $$
(5)
as the sum of all the absolute values of the losses in exports of country c. Then the final estimate for the direction of technological change in country c is defined as follows:
$$ \theta_{c}=\frac{\gamma_{c}^{+}}{\gamma_{c}^{+}+\gamma_{c}^{-}}\hat{\beta}_{c}^{+}-\frac{\gamma_{c}^{-}}{\gamma_{c}^{+}+\gamma_{c}^{-}}\hat{\beta}_{c}^{-} $$
(6)
If 𝜃c > 0, this indicates a relative increase in exports of more complex products for this country. In other words, if 𝜃c > 0, more complex products become relatively more important for this country’s export-basket (vice versa for 𝜃c < 0). Figure 5 provides an illustration of the results. It shows the respective regression lines as well as the composition of the underlying data for the cases of Greece and Germany with regard to expanding products (i.e. \(i\in P_{c}^{+}\)). It indicates that greater expansion of exports in Germany (right panel) is associated with higher product complexity, while greater expansion of exports in Greece (left panel) is associated with a lower technological complexity, partially driven by a reversal towards being a producer of primary inputs (such as refined oil).
Although the country-specific results do not always show such clear trends as in the examples given in Fig. 5 (for details on the other EU countries see the appendix), in sum they point to a clear pattern of the sectoral developments across Europe from the perspective of international competitiveness: we find that higher levels of overall complexity before the onset of the Eurozone (in 1999) are, on average, associated with stronger gains of complexity measured in terms of the expansion and decline of individual sectors for the larger part of the observed countries (Fig. 6, upper panel). While this result is broadly consistent with the Kaldorian prediction that “success breeds success” (Kaldor 1980), a more nuanced interpretation of this overall quadratic relationship is given in the lower panel of Fig. 6: although the catching-up of Eastern Europe has an imprint on overall developments, patterns consistent with Kaldorian effects can be identified within the Eastern European countries, where they are rather pronounced, as well as (with a weaker intensity) among all the remaining EU countries. Thereby, large parts of the variety in the results for the Eastern European catch-up economies seem to be moderated by their geographical proximity to Europe’s industrial core (Stöllinger 2016).
The patterns of technological change as depicted in Fig. 6 also allow us to emphasize four further observations. First, there is still considerable heterogeneity within the typically proposed country-groups: core countries differ in their development, mirroring the fact that some of these countries struggle to hold on to their position, while others, mostly Germany, have managed to expand their technological dominance (e.g. Storm and Naastepad 2015a). In fact, Germany is the only example of the core countries that finds itself above the value predicted by a quadratic model fitted to the data. Second, the upper panel of Fig. 6 shows that we cannot find a single periphery country with a decidedly positive technological development: Portugal is the only periphery country that manages to surpass the predicted value, albeit this country has started from a relatively low level of complexity. Third, we find that while most Eastern catch-up countries perform better than the prediction, two exceptions are actually located markedly below the regression line. This indicates that the economic catch-up process of Eastern European countries is not necessarily tied to a technological catch-up process, as evidenced most forcefully by the outliers Bulgaria and Lithuania. Fourth, the heterogeneity among financial hub countries is particularly large, but can be explained by their different financialization strategies: Ireland’s role as a corporate tax haven manifests itself in a massive technological upgrading (e.g. Regan and Brazys 2018), while strategies of the Netherlands and Malta are associated with more pronounced deindustrialization (e.g. Visser et al.2016).
As international competitiveness and technological capabilities are of prime importance for assessing the future developmental trajectories within given political and institutional constraints (Hidalgo and Hausmann 2009; Cristelli et al. 2015), it is important to note that we cannot observe convergence in terms of technological capabilities in the current European framework. Quite on the contrary, our results point to the possibility that some countries in Eastern Europe might indeed manage eventually to catch-up to the core (Czech Republic, Poland, Hungary and Slovakia), while others (such as Bulgaria or the Baltic countries) are much more likely to join the European periphery (Stöllinger 2016).