Policy Research Working Paper 9494 Economic Growth in European Union NUTS-3 Regions Austin Kilroy Roberto Ganau Finance, Competitiveness and Innovation Global Practice December 2020 Policy Research Working Paper 9494 Abstract This paper analyzes the growth pathways of 1,321 regions contribution of industry to growth but there is a strong in the European Union from 2003 to 2017. The aim is to association between foreign direct investment and growth. inform integrated territorial investments and other eco- Among low-growth lagging regions—the 171 small regions nomic development initiatives in lagging regions. Using in the European Union with gross domestic product per the definition of lagging regions from the European Com- capita less than 90 percent of the European Union average, mission’s Catching Up Initiative, more than two-thirds of and stagnant or negative growth performance—growth is the European Union member states have lagging regions correlated with construction and innovation. There are also when defined at the Nomenclature of Territorial Units differences in the growth pathways of rural and non-ru- for Statistics-3 scale. These small lagging regions are often ral regions: growth is associated with moving away from hidden within larger and more prosperous regions. The agriculture in rural regions, and it is associated with con- paper considers the roles of industrial structure, innovation, struction and innovation in non-rural regions. The results and inward foreign direct investment as growth-enhancing imply that a finer geographic scale can be important in factors. The findings indicate that the growth dynamics policy making and programming of Cohesion Policy Funds, in low-income regions are different from those in regions to cater to different needs and opportunities at the scale of in other income groups: there is no overall pattern in the Nomenclature of Territorial Units for Statistics-3 regions. This paper is a product of the Finance, Competitiveness and Innovation Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at akilroy@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Economic Growth in European Union NUTS-3 Regions * Austin Kilroy** World Bank Group 1818 H St NW, Washington, DC 20433, USA E-mail: akilroy@worldbank.org Roberto Ganau Department of Economics and Management “Marco Fanno”, University of Padova Via del Santo 33, 35123 Padova, Italy E-mail: roberto.ganau@unipd.it Department of Geography and Environment, London School of Economics and Political Science Houghton Street, London WC2A 2AE, United Kingdom E-mail: r.ganau1@lse.ac.uk KEYWORDS Economic Growth; Development Policy; NUTS-3 Regions; European Union. JEL CODES R11; R58. * We are grateful to Chiara Burlina for exhaustive work to assemble data from Regio (Eurostat) and fDi Markets (Financial Times) databases, and to Lewis Dijkstra for having shared long time series of data from the Regio database. We thank Anwar Aridi and Todor Milchevski, who worked on a previous version of the paper. The paper has benefitted from guidance by Natasha Kapil and Ilias Skamnelos, and from comments on an earlier version of the paper by Paulo Correa, Thomas Farole, Soraya Goga, John Nasir, and Paula Restrepo Cadavid as well as participants to the Regional Studies Association Conference in Cluj-Napoca (September 2017). The paper was commissioned as part of the World Bank Group’s work on lagging regions in the European Union. The content of this paper does not reflect the official opinion of the World Bank Group. Responsibility for the information and views expressed therein lies entirely with the authors. ** Corresponding Author. 1. INTRODUCTION Cohesion Policy in the European Union (EU) aims at reducing disparities among countries and regions. The objective of cohesion was defined in the 1986 Single European Act, for “reducing disparities between the various regions and the backwardness of the least-favored regions” (Article 130a). During the EU’s current 2014-2020 budget period, about €469 billion of EU and national budgets have been allocated under Cohesion Policy funds through the European Regional Development Fund (ERDF), the European Social Fund (ESF), and the Cohesion Fund (CF). 1 In the previous 2000-2006 and 2007-2013 budget periods, €198 billion and €347 billion, respectively, was committed under Cohesion Policy initiatives. 2 The EU has successfully reduced inequalities between member states in average incomes, but inequalities between regions within countries have increased. Since the last big wave of new Member States joining the EU in 2004, the average gross domestic product (GDP) per capita of the three poorest EU countries – namely, Bulgaria, Latvia, and Romania – has risen from 15.7% of the EU average in 2004 to 29.3% in 2018. 3 In this sense, the EU has been a ‘convergence machine’ for its Member States. However, most of that convergence has been driven by economic growth in leading regions within each country – usually, the capital cities. For example, Sofia achieved a 9.2% average yearly GDP per capita growth rate over the period 2004-2017, Riga an 8.3%, and Bucharest a 12.5%. 4 Overall, despite the positive results in terms of economic convergence at the country level, regional inequality has started to rise again since the early 2000s both across and within countries, after a promising reduction observed during the 1990s with respect to the previous decade (Storper, 2018; Iammarino et al., 2019). Regional economic performance has been analyzed almost always at the NUTS-2 level. Given the large share of EU and national budgets allocated to reduce regional inequality and promote harmonized growth, economists, economic geographers and regional scientists have investigated with great scrutiny the economic impact of the EU Cohesion Policy (e.g. Cappelen et al., 2003; Beugelsdijk and Eijffinger, 2005; Bussoletti and Esposti, 2008; Dall’erba and Le Gallo, 2008; Ramajo et al., 2008; Pellegrini et al., 2013; Rodríguez-Pose and Garcilazo, 2015; Di Cataldo, 2017; Bachtrögler et al., 2020; Crescenzi and Giua, 2020), and the factors underlying regional economic growth and convergence (e.g. Quah, 1996; Fingleton, 1997; Magrini, 1999; Rodríguez-Pose, 1999; Canova, 2004; Gardiner et al., 2004; Basile, 2008, 2009; Tselios, 2009; Arbia et al., 2010; Crescenzi and Rodríguez-Pose, 2012; Rodríguez-Pose and Ketterer, 2020). A commonality characterizing the empirical works analyzing the effectiveness of cohesion investments in particular, and the determinants of economic growth and convergence in general, is that they consider the NUTS-2 level geography as unit of analysis, coherently with the EU policy design. Indeed, policy and programming have conventionally focused on NUTS-2 regions, and eligibility for Cohesion Policy funds is determined according to GDP per capita measured by NUTS-2 regional boundaries. However, NUTS-2 regions are large entities, with populations of 800,000 to 3 million people, which hides substantial heterogeneity inside the regions. The pitfall of the NUTS-2 level as spatial unit for policy design and programming is that this large size often hides substantial variations in economic performance and economic potential. For example, the Emilia-Romagna region (Italy) sees the coexistence of large and wealthy NUTS-3 regions, such as Bologna and Modena, with lagging ones, such as Ferrara and Rimini. In Croatia, a single NUTS-2 region includes the richest city of the country – i.e. Zagreb – and five of the EU’s poorest counties. In Bulgaria, the NUTS-2 region of Yugozapaden includes Sofia Province and Sofia City – that recorded a 10.6% and 1 Authors’ elaboration on data available at “https://cohesiondata.ec.europa.eu/overview”. 2 Various figures appear in different sources (e.g. European Commission, 2015). The breakdown by year and by fund in each country available under ‘Data on budget commitments’ at https://ec.europa.eu/regional_policy/en/policy/evaluations/data-for-research/ indicates the totals cited here. 3 Authors’ elaboration of Eurostat data. 4 Authors’ elaboration on Eurostat data. The spatial unit of reference corresponds to the level 3 of the Nomenclature des Unités Territoriales Statistiques (NUTS) adopted by the EU. 2 a 9.5% average yearly GDP per capita growth rate over the period 2003-2017, respectively – and also includes the provinces of Pernik, Blagoevgrad, and Kyustendil – that recorded growth rates of 7%, 6.7%, and 4.7%, respectively. Such a pattern does not characterize only Mediterranean and Eastern countries – i.e. those generally lagging behind in the EU – but also ‘top’ performing Member States, such as Germany. As an example, the German region of Oberfranken shows a 4.9% difference in the average yearly GDP per capita growth rate over the period 2003-2017 between its NUTS-3 region that has grown the most, namely Coburg (6.7%), and the one that has grown the least, namely Hof (1.8%). 5 Furthermore, in five EU Member States, namely, Cyprus, Estonia, Luxembourg, Latvia, and Malta, one NUTS-2 region covers the whole country. Overall, this substantial variation in economic performance within NUTS-2 regions makes it hard to configure sub-national policies and investments. Accordingly, a growing emphasis is being put on development initiatives at a smaller scale, at NUTS-3 or similar levels. NUTS-3 regions usually have populations of 100,000 to 800,000. In the 2014-2020 and 2021-2027 programming periods, there is a growing use of Integrated Territorial Investments (ITIs) and other instruments of ‘sustainable urban development’ (SUD). Of the ERDF, 5% should be allocated to SUD in the 2014-2020 programming period, and this will rise to 6% for the 2021-2027 programming period. 6 ITIs combine resources for an integrated program of initiatives at the scale of a group of towns, a county, or one medium-sized city and its hinterland. 7 Currently there are fully 329 ITIs in 17 EU Member States, and a further 55 SUD initiatives at the city level include economic development aspects. 8 Despite increased attention to the NUTS-3 level, there is scarce empirical evidence to inform the choice of investments and actions by local leaders, especially in lagging regions. There is a growing body of evidence on sub-national competitiveness at NUTS-2 and similar levels globally (e.g. OECD, 2012; World Bank, 2015). Yet these studies have not usually examined heterogeneity of growth dynamics in different types of regions. What works in leading and fast- growing regions may not work in lagging and slow-growing ones. The need for guidance may be strongest among lagging NUTS-3 regions, which are usually suffering from economic underperformance, unemployment, and emigration, and cannot find many examples of success among their peers. There are some notable examples of economic analysis at the NUTS-3 level (e.g. Geppert and Stephan, 2008; Becker et al., 2012; Panzera and Postiglione, 2014; Percoco, 2017; Butkus et al., 2018), to which this paper aims to contribute. Research and evaluations must become more oriented towards the pragmatic dilemmas with which local leaders struggle, if they are to be useful. The European Commission’s evaluations of EU Cohesion Policy funds have generally focused on whether funding per se has had an impact, rather than on which interventions have been effective (Davies, 2017). A European Commission (2017) review of competitiveness in low-income and low-growth regions is one exception to this trend; it provides guidance on potential solutions to the typical problems of lagging regions, namely access to finance; low productivity and educational attainment; weak innovation systems and institutional quality; high emigration; and low public and private investment. 9 However, it does not yet bring evidence on whether lagging regions have realistically addressed these constraints, and if doing so has led to growth. Evaluations of typical interventions (e.g. access to finance, 5 Author’s elaboration on Eurostat data. 6 The total budget for ERDF during the period 2021-2027 is estimated at €201 billion, and Cohesion Policy funds will be €331 billion – a smaller total budget owing to the departure of the United Kingdom. For details, see the report “Regional Development and Cohesion Policy beyond 2020: The new framework at a glance”, available at “https://ec.europa.eu/regional_policy/en/2021_2027/”. 7 Other instruments such as Community-Led Local Development (CLLD) may also be used. CLLDs tend to focus on a smaller scale than ITIs, and are implemented by a Local Action Group (LAG) which may be constituted by several smaller municipalities. For a review of ITIs and CLLDs, see van der Zwet et al. (2014). 8 Data from Joint Research Council STRAT-Board, accessed on May 26, 2020 and available at “https://urban.jrc.ec.europa.eu/strat- board/”. Some examples are the integrated plans for Larnaca, Limassol, Paphos, and other cities in Cyprus; city strategies for Aalborg, Odense, and other Danish cities; and city contracts in France for Coeur de France town, Vierzon, Amboise, and others; and integrated city programs in Italy for Aversa, Caserta, Salerno, and several others. 9 Similar conclusions are reached also by Farole et al. (2018). 3 apprenticeships, business advisory, innovation, investment promotion, etc.) have been undertaken, but the results have differed greatly across evaluations, probably indicating that the quality of design and implementation can be as important as the category of intervention. 10 So local leaders in lagging regions cannot access a good quality reference base on ‘what works’ in regions such as theirs. We contribute to this debate by analyzing the growth pathways of EU regions at the NUTS-3 level in the period 2003-2017. We focus on correlates of economic growth in lagging regions, such as industrial structure, innovation, and inward foreign direct investments (FDI). We rely on a regional taxonomy based on income level and long-run growth criteria, which combines the standard classification adopted under Cohesion Policy with that proposed in 2015 by the European Commission under the ‘Catching Up Initiative’. Our analysis is descriptive in its nature, but contributes to the understanding of the determinants of economic growth in lagging regions, as a prelude to investigating the causal relationships from policy and investments to growth. The aim of the paper is to provide evidence on what are realistic aspirations for economic growth in different types of regions. The paper aims to give an answer to questions from national and sub-national policy makers such as: ‘What has worked in regions like mine?’; ‘How have other regions grown?’; and ‘What is realistic for my region?’ The choice of the NUTS-3 level as unit of analysis is made on the basis that a finer spatial scale can provide a more detailed picture of growth pathways than the NUTS-2 level. There are 1,348 NUTS-3 regions in the EU, compared to 281 NUTS-2 regions, thus giving a higher number of observations for study, and more variation in their economic structure and performance. Some regions have been more successful than others, and we can we learn from the growth patterns that have characterized different types of regions. We focus especially on low-income and low-growth regions, as the subset of regions which are the highest priorities for Cohesion Policy. We structure the analysis according to three research questions: on location, economic dynamism, and correlates of growth. Our research questions are the following: (i) where are NUTS- 3 level lagging regions located, and how similar are these locations to NUTS-2 lagging regions? (ii) have NUTS-3 lagging regions been growing and converging with non-lagging regions? (iii) what are the correlates of growth among NUTS-3 regions, and do these correlates differ for lagging regions? The rest of the paper is organized as follows. Section 2 provides an overview of the empirical framework by describing the data set employed, discussing the taxonomy adopted to classify regions, and presenting the empirical modeling. Section 3 presents some stylized facts on the economic geography and dynamism of regions in the EU. Section 4 presents the empirical results. Section 5 concludes the work by discussing the main results and drawing some policy implications. 2. EMPIRICAL FRAMEWORK 2.1. Data Set We use data on NUTS-3 regions, with special focus on industrial structure, innovation, and inward FDI. According to the 2016 NUTS classification, there are 1,348 NUTS-3 regions in the EU. 11 Specifically, we aim to analyze the drivers of regional economic growth – defined as yearly growth of GDP per capita – by focusing on three key economic dimensions: industrial structure, innovation, and inward FDI, while controlling also for population dynamics and agglomeration forces. The choice of these three economic dimensions is for three main reasons. First, the empirical literature has identified industrial structure, innovation, and inward FDI as key growth-engine factors (e.g. Crescenzi, 2005; Menghinello et al., 2010; Crescenzi et al., 2016). Second, these dimensions are influenced by policy makers, who could have the capacity to intervene with ad hoc measures in order to stimulate a particular industrial sector, promote the innovation capability of firms, or attract 10 One of the best attempts at a review of robust analytic evidence has been done by the What Works Centre for Local Economic Growth (“https://whatworksgrowth.org/”). 11 This number includes the five French Overseas Departments, and the Spanish extra-territorial autonomous cities of Ceuta and Melilla. 4 investments by foreign companies. Third, from a practical viewpoint, data are available on these three dimensions at the NUTS-3 level, but data are missing on some other dimensions. European official statistical sources provide information on a large set of variables for NUTS-1 and NUTS-2 regions, but only a reduced amount of information is available at the NUTS-3 level. Unfortunately, it was not possible to include data series on firm size. We would like to estimate the role of micro (fewer than 10 employees), small (10 to 49 employees), medium (50 to 249 employees), and large (250 employees or more) firms in regional economic growth. Such analysis could help reveal the relative importance of initiatives on SMEs in lagging regions, versus initiatives to boost competitiveness of large firms as drivers of growth. Unfortunately, these data series – available from the Regio database provided by Eurostat – present two key shortcomings. First, regional data are based on a too broad classification of firms with respect to their size, as data are available for zero-employment firms, firms with 1 to 9 employees, and firms with 10 or more employees. This represents a key problem, as small, medium, and large firms are grouped together in a unique size class, such that it is impossible to capture the role of SMEs versus large firms. Second, business demography data that are available only from the year 2008 – while our analysis starts in the year 2003 – do not cover nine EU Member States, and present severe missing values or inconsistent estimates over time for the remaining countries. 12 Since the lack of availability is not systematic, we would risk bias in the analysis if we included only the available data. We use data for the 2003 to 2017 period, from Eurostat, the Organisation for Economic Co-operation and Development (OECD), and fDi Markets (Financial Times). Given our empirical goal, and considering data availability constraints, we have collected the largest possible set of data at the NUTS-3 level from three main sources to cover the longest possible time period, namely from 2003 to 2017. On economic growth and industrial structure, we use Eurostat data from the Regio database on GDP, population, employment, land area, and sectoral Gross Value Added (GVA) – for agriculture, industry (mining; electricity; manufacturing), construction, market services, 13 and non-market services. 14 Missing values in the NUTS-3 level regional series have been filled in by linearly interpolating NUTS-2 level data. On innovation, we use microdata on patents filed under the Patent Co-operation Treaty (PCT) from the REGPAT database provided by the OECD, aggregated at the NUTS-3 level by priority year and inventor’s residence using the fractional count criterion. On FDI, we use data on inward ‘greenfield’ FDI from the fDi Markets database provided by the Financial Times. Of particular interest for our purposes, the fDi Markets database collects information on individual investment projects in terms of year, destination region at the NUTS-3 (or city) level, and business activity run in the host economy. 15 The merging and cleaning procedure of the above-mentioned data series left us with a sample of 1,321 NUTS-3 regions, which represents 98.5% of the EU-28 territory. We also excluded a priori the French Overseas Departments, as well as Ceuta and Melilla. As shown in Table 12 First, there are no data at all for Belgium, Cyprus, Germany, Greece, Ireland, Luxembourg, Malta, Sweden, and the United Kingdom. Second, a large number of countries present severe missing values or inconsistent estimates. A first group of countries – namely, France, Hungary, Italy, and Portugal – are covered with inconsistent regional data according to the changes occurred over time in the NUTS classification. A second group of countries is covered only for some years: Spain is covered entirely, but for 2015; Croatia and Latvia are covered only from 2011; Denmark is covered only up to 2013; Slovenia and the Netherlands are covered only up to 2010; Czech Republic has data for the year 2010, and then from 2013 onwards; Estonia has data for the period 2008-2010, and then from 2015 onwards; Poland has data for the period 2008-2010, and then for the year 2016. Therefore, a complete a reliable data series from 2008 to 2016 is available only for Austria, Bulgaria, Finland, Romania, and Slovak Republic. This lack of data would greatly reduce the number of regions and countries on which estimates for all variables could be made – thereby diminishing the accuracy of all results. 13 Wholesale and retail trade; transport; accommodation and food service activities; information and communication; financial and insurance activities; real estate activities; professional, scientific and technical activities; administrative and support service activities. 14 Public administration and defence; compulsory social security; education; human health and social work activities; arts, entertainment and recreation; other service activities; activities of household and extra-territorial organisations and bodies. 15 It is worth noting that the fDi Markets database represents the best available source to analyze FDI-related phenomena at the sub- national level over a long time period. It covers ‘greenfield’ FDI, that includes new investments and enlargements of existing investments, but excludes ‘brownfield’ investments such as mergers and acquisitions. Despite the limitation of not covering FDI in the form of mergers and acquisitions, its overall validity and reliability is confirmed by the many empirical studies that have used it (e.g. Crescenzi et al., 2014; Dogaru et al., 2015; Castellani and Pieri, 2016). 5 1, the sample covers entirely all EU Member States but Poland, for which data were partially unavailable in 20 of 73 regions. 16 Table 1: Structure of the sample and geographic coverage. Regions Country Population Sample Percentage Covered Austria 35 35 100.00 Belgium 44 44 100.00 Bulgaria 28 28 100.00 Cyprus 1 1 100.00 Czech Republic 14 14 100.00 Germany 401 401 100.00 Denmark 11 11 100.00 Estonia 5 5 100.00 Greece 52 52 100.00 Spain 57 57 100.00 Finland 19 19 100.00 France 96 96 100.00 Croatia 21 21 100.00 Hungary 20 20 100.00 Ireland 8 8 100.00 Italy 110 110 100.00 Lithuania 10 10 100.00 Luxembourg 1 1 100.00 Latvia 6 6 100.00 Malta 2 2 100.00 Netherlands 40 40 100.00 Poland 73 53 72.60 Portugal 25 25 100.00 Romania 42 42 100.00 Sweden 21 21 100.00 Slovenia 12 12 100.00 Slovak Republic 8 8 100.00 United Kingdom 179 179 100.00 Total 1,341 1,321 98.51 Notes: Percentages are defined on row values. The five French Overseas Departments, and the Spanish extra-territorial autonomous cities of Ceuta and Melilla have been excluded from the sample à priori. 20 Polish regions are out of the sample due to data availability issues. 2.2. Defining Lagging Regions The NUTS-2 typology of regions provides the basis for our analysis at NUTS-3 level. Under Cohesion Policy, NUTS-2 level regions are defined according to their GDP per capita level as being in one of three regional types: (i) ‘more developed’, with a GDP per capita over 90% of the EU average; (ii) ‘transition’, with a GDP per capita between 75% and 90% of the EU average; and (iii) ‘less developed’, with a GDP per capita less than 75% of the EU average. 17 Under the European Commission’s ‘Catching Up Initiative’, 18 two further subsets of lagging regions were defined: (i) ‘low-income regions’, with a GDP per capita under 50% of the EU average in 2013; and (ii) ‘low- 16 Specifically, the sample does not include the Polish NUTS-3 regions of Nowosądecki, Nowotarski, Oświęcimski, Koszaliński, Szczecinecko-pyrzycki, Szczeciński, Nyski, Opolski, Grudziądzki, Inowrocławski, Świecki, Włocławski, Słupski, Chojnicki, Starogardzki, Ciechanowski, Płocki, Ostrołęcki, Siedlecki, and Żyrardowski. 17 Definition taken from “https://ec.europa.eu/eurostat/web/regions/background”. 18 The initiative was launched in 2015 with the aim of providing more targeted technical assistance to support regional development in a subset of regions facing the greatest challenges in unlocking growth potential. 6 growth regions’, with a GDP per capita under 90% of the EU average in 2013, and which have not converged towards the EU average between 2000 and 2013. 19 We transpose these two types of lagging regions to the NUTS-3 level, to identify ‘low income’ regions and ‘low growth’ regions. Combining the standard taxonomy from Cohesion Policy with the 2015 ‘catching up’ taxonomy, we classify ‘lagging’ NUTS-3 regions among our sample of 1,321 regions. 20 Income level is defined by average yearly GDP per capita over the period ������2003−2017 ). Growth rate is defined by the long-run GDP per capita growth rate 2003-2017 ( 2003−2017 21 between the years 2003 and 2017 (∆ ). Regions are classified by income as: (a) ‘high- ������ income’, if 2003−2017 ≥ 90% of the sample average; (b) ‘transition’, if 75% ≤ ������ 2003−2017 < 90% of the sample average; (c) ‘less developed’, if 50% ≤ ������ 2003−2017 < 75% of the sample ������2003−2017 < 50% of the sample average. Regions are average; and (d) ‘low-income’, if 2003−2017 classified by growth performance as: (a) ‘high-growth’, if ∆ ≥ 50% of the sample 2003−2017 average; and (b) ‘low-growth’, if ∆ < 50% of the sample average. According to this classification, 16.1% of NUTS-3 regions are ‘low income’ lagging regions, and 12.9% of NUTS-3 regions are ‘low growth’ lagging regions. Table 2 reports the distribution of NUTS-3 regions according to the two-sided classification adopted. ‘Low-income’ regions represent the 16.1% of the sample, while ‘low-growth’ regions represent the 33.4% of the sample regions (including also ‘high income’ regions). Among low-growth regions, 61.2% are ‘high- income’, 18.4% are ‘transition’ regions, 17.7% are ‘less developed’, and only 2.7% are ‘low-income’. ‘Low growth’ lagging regions are those in the transition, less developed, and low-income groups. Of the sample NUTS-3 regions, 12.9% are described by this category. Table 2: Regions’ distribution by income level and long-run growth rate. Lon-Run Growth Rate Income Level High Low Total No. % No. % No. % High-Income 480 36.34 270 20.44 750 56.78 Transition 113 8.55 81 6.13 194 14.69 Less Developed 87 6.59 78 5.90 165 12.49 Low-Income 200 15.14 12 0.91 212 16.05 Total 880 66.62 441 33.38 1,321 100.00 Notes: Authors’ elaboration on Eurostat data. Percentages are defined on the total number of regions in the sample. High-income regions are those with an average yearly GDP per capita over the period 2003-2017 equal to or greater than the 90% of the sample average. Transition regions are those with an average yearly GDP per capita over the period 2003-2017 equal to or greater than the 75% of the sample average, but lower than the 90% of the sample average. Less developed regions are those with an average yearly GDP per capita over the period 2003-2017 equal to or greater than the 50% of the sample average, but lower than the 75% of the sample average. Low-income regions are those with an average yearly GDP per capita over the period 2003-2017 lower than the 50% of the sample average. High- (low-)growth regions are those recording a long-run growth rate of GDP per capita between the years 2003 and 2017 equal to or greater than (lower than) the 50% of the sample average. 19 Definition taken from “https://ec.europa.eu/regional_policy/en/policy/how/improving-investment/lagging_regions/”. 20 It is worth noting that Rodriguez-Pose and Ketterer (2020) have employed the official ‘catching up’ taxonomy proposed by the European Commission to analyse the relationship between institutional quality and economic growth at the NUTS-2 level. 21 With respect to the income level dimension, the choice of considering the average yearly GDP per capita evaluated over the period 2003-2017 rather than the 2013 figure, or that referring to the last available year of observation, is aimed at relaxing, first, potential biases due to extreme or outlying values ascribable to abnormal economic performances recorded by a region in a particular year – in the case, for example, of a natural disaster, or the closure of a large plant –, and, second, issues related to linearly interpolation of missing data. With respect to the growth dimension, the choice of considering the long-run growth rate between the years 2003 and 2017 rather than the relative convergence criterion proposed by the European Commission is aimed at capturing the absolute growth capacity of a region rather than its relative performance. 7 Table 3: Regions’ distribution by income level and country. Income Level High-Income Transition Less Developed Low-Income Total Country % % % % No. No. No. No. No. Category Country Category Country Category Country Category Country Austria 31 4.13 88.57 3 1.55 8.57 1 0.61 2.86 0 0.00 0.00 35 Belgium 31 4.13 70.45 7 3.61 15.91 6 3.64 13.64 0 0.00 0.00 44 Bulgaria 0 0.00 0.00 0 0.00 0.00 0 0.00 0.00 28 13.21 100.00 28 Cyprus 0 0.00 0.00 1 0.52 100.00 0 0.00 0.00 0 0.00 0.00 1 Czech Republic 1 0.13 7.14 0 0.00 0.00 3 1.82 21.43 10 4.72 71.43 14 Germany 297 39.60 74.06 80 41.24 19.95 24 14.55 5.99 0 0.00 0.00 401 Denmark 11 1.47 100.00 0 0.00 0.00 0 0.00 0.00 0 0.00 0.00 11 Estonia 0 0.00 0.00 0 0.00 0.00 1 0.61 20.00 4 1.89 80.00 5 Greece 5 0.67 9.62 4 2.06 7.69 32 19.39 61.54 11 5.19 21.15 52 Spain 17 2.27 29.82 17 8.76 29.82 23 13.94 40.35 0 0.00 0.00 57 Finland 19 2.53 100.00 0 0.00 0.00 0 0.00 0.00 0 0.00 0.00 19 France 57 7.60 59.38 37 19.07 38.54 2 1.21 2.08 0 0.00 0.00 96 Croatia 0 0.00 0.00 0 0.00 0.00 2 1.21 9.52 19 8.96 90.48 21 Hungary 0 0.00 0.00 1 0.52 5.00 0 0.00 0.00 19 8.96 95.00 20 Ireland 8 1.07 100.00 0 0.00 0.00 0 0.00 0.00 0 0.00 0.00 8 Italy 69 9.20 62.73 10 5.15 9.09 31 18.79 28.18 0 0.00 0.00 110 Lithuania 0 0.00 0.00 0 0.00 0.00 1 0.61 10.00 9 4.25 90.00 10 Luxembourg 1 0.13 100.00 0 0.00 0.00 0 0.00 0.00 0 0.00 0.00 1 Latvia 0 0.00 0.00 0 0.00 0.00 1 0.61 16.67 5 2.36 83.33 6 Malta 0 0.00 0.00 0 0.00 0.00 1 0.61 50.00 1 0.47 50.00 2 Netherlands 38 5.07 95.00 1 0.52 2.50 1 0.61 2.50 0 0.00 0.00 40 Poland 1 0.13 1.89 0 0.00 0.00 4 2.42 7.55 48 22.64 90.57 53 Portugal 1 0.13 4.00 1 0.52 4.00 15 9.09 60.00 8 3.77 32.00 25 Romania 0 0.00 0.00 0 0.00 0.00 1 0.61 2.38 41 19.34 97.62 42 Sweden 21 2.80 100.00 0 0.00 0.00 0 0.00 0.00 0 0.00 0.00 21 Slovenia 1 0.13 8.33 0 0.00 0.00 8 4.85 66.67 3 1.42 25.00 12 Slovak Republic 1 0.13 12.50 0 0.00 0.00 1 0.61 12.50 6 2.83 75.00 8 United Kingdom 140 18.67 78.21 32 16.49 17.88 7 4.24 3.91 0 0.00 0.00 179 Total 750 100.00 56.78 194 100.00 14.69 165 100.00 12.49 212 100.00 16.05 1,321 Notes: Authors’ elaboration on Eurostat data. By-category percentages are defined on column totals. By-country percentages are defined on row totals. High-income regions are those with an average yearly GDP per capita over the period 2003-2017 equal to or greater than the 90% of the sample average. Transition regions are those with an average yearly GDP per capita over the period 2003-2017 equal to or greater than the 75% of the sample average, but lower than the 90% of the sample average. Less developed regions are those with an average yearly GDP per capita over the period 2003-2017 equal to or greater than the 50% of the sample average, but lower than the 75% of the sample average. Low-income regions are those with an average yearly GDP per capita over the period 2003-2017 lower than the 50% of the sample average. 8 Table 4: Regions’ distribution by high- vs. not high-income, long-run growth rate and country. Long-Run Growth Rate High-Income, High-Growth High-Income, Low-Growth Not High-Income, High-Growth Not High-Income, Low-Growth Total Country % % % % No. No. No. No. No. Category Country Category Country Category Country Category Country Austria 31 6.46 88.57 0 0.00 0.00 4 1.00 11.43 0 0.00 0.00 35 Belgium 30 6.25 68.18 1 0.37 2.27 11 2.75 25.00 2 1.17 4.55 44 Bulgaria 0 0.00 0.00 0 0.00 0.00 28 7.00 100.00 0 0.00 0.00 28 Cyprus 0 0.00 0.00 0 0.00 0.00 1 0.25 100.00 0 0.00 0.00 1 Czech Republic 1 0.21 7.14 0 0.00 0.00 13 3.25 92.86 0 0.00 0.00 14 Germany 273 56.88 68.08 24 8.89 5.99 103 25.75 25.69 1 0.58 0.25 401 Denmark 11 2.29 100.00 0 0.00 0.00 0 0.00 0.00 0 0.00 0.00 11 Estonia 0 0.00 0.00 0 0.00 0.00 5 1.25 100.00 0 0.00 0.00 5 Greece 1 0.21 1.92 4 1.48 7.69 0 0.00 0.00 47 27.49 90.38 52 Spain 13 2.71 22.81 4 1.48 7.02 26 6.50 45.61 14 8.19 24.56 57 Finland 18 3.75 94.74 1 0.37 5.26 0 0.00 0.00 0 0.00 0.00 19 France 18 3.75 18.75 39 14.44 40.63 1 0.25 1.04 38 22.22 39.58 96 Croatia 0 0.00 0.00 0 0.00 0.00 20 5.00 95.24 1 0.58 4.76 21 Hungary 0 0.00 0.00 0 0.00 0.00 20 5.00 100.00 0 0.00 0.00 20 Ireland 3 0.63 37.50 5 1.85 62.50 0 0.00 0.00 0 0.00 0.00 8 Italy 7 1.46 6.36 62 22.96 56.36 3 0.75 2.73 38 22.22 34.55 110 Lithuania 0 0.00 0.00 0 0.00 0.00 10 2.50 100.00 0 0.00 0.00 10 Luxembourg 1 0.21 100.00 0 0.00 0.00 0 0.00 0.00 0 0.00 0.00 1 Latvia 0 0.00 0.00 0 0.00 0.00 6 1.50 100.00 0 0.00 0.00 6 Malta 0 0.00 0.00 0 0.00 0.00 2 0.50 100.00 0 0.00 0.00 2 Netherlands 28 5.83 70.00 10 3.70 25.00 2 0.50 5.00 0 0.00 0.00 40 Poland 1 0.21 1.89 0 0.00 0.00 52 13.00 98.11 0 0.00 0.00 53 Portugal 0 0.00 0.00 1 0.37 4.00 23 5.75 92.00 1 0.58 4.00 25 Romania 0 0.00 0.00 0 0.00 0.00 42 10.50 100.00 0 0.00 0.00 42 Sweden 20 4.17 95.24 1 0.37 4.76 0 0.00 0.00 0 0.00 0.00 21 Slovenia 1 0.21 8.33 0 0.00 0.00 11 2.75 91.67 0 0.00 0.00 12 Slovak Republic 1 0.21 12.50 0 0.00 0.00 7 1.75 87.50 0 0.00 0.00 8 United Kingdom 22 4.58 12.29 118 43.70 65.92 10 2.50 5.59 29 16.96 16.20 179 Total 480 100.00 36.34 270 100.00 20.44 400 100.00 30.28 171 100.00 12.94 1,321 Notes: Authors’ elaboration on Eurostat data. By-category percentages are defined on column totals. By-country percentages are defined on row totals. High- (not high-)income regions are those with an average yearly GDP per capita over the period 2003-2017 equal to or greater than (lower than) the 90% of the sample average. High- (low-)growth regions are those recording a long-run growth rate of GDP per capita between the years 2003 and 2017 equal to or greater than (lower than) the 50% of the sample average. 9 ‘Catching up’ regions are found in 19 of 28 EU Member States. 22 ‘Low-income’ regions are found in Bulgaria, Czech Republic, Estonia, Greece, Croatia, Hungary, Lithuania, Malta, Poland, Portugal, Romania, Slovenia, and the Slovak Republic – see Table 3. ‘Low-growth’ lagging regions are found in Belgium, Germany, Greece, Spain, France, Croatia, Italy, Portugal, and the United Kingdom – see Table 4. Almost all EU Member States show a relatively high within-country variability in growth rates. Figure 1 plots average yearly GDP per capita growth rate recorded over the period 2003-2017. Large variations are found both in larger countries among the EU-15 (Germany, the United Kingdom, Spain, Greece), smaller countries in the EU-15 (Ireland, Belgium, and others), and post-2004 enlargement countries. In other words, lagging regions are a widespread problem. Figure 1: Within-country variability in average yearly GDP per capita growth. Notes: Authors’ elaboration on Eurostat data. Time average of the regional yearly growth rate of GDP per capita over the period 2003- 2017. Percentage values. The dashed line refers to the sample average, while the dots refer to country-level average values. The spatial distribution of NUTS-3 regions shows a geographic divide between the types of lagging regions in newer and older Member States, and between north and south. Figure 2 complements Table 2 by mapping the spatial distribution of NUTS-3 regions by category. Several observations can be made: (i) ‘low-income, high-growth’ regions seem to make up the bulk of post- 2004 enlargement countries; (ii) Mediterranean countries are relatively heterogeneous, and include regions belonging to almost all categories; (iii) Northern countries and Germany are much less heterogeneous. In particular, almost all Northern countries’ regions tend to be ‘high-income, high- growth’, except Germany which is characterized by a clear West-East divide, with ‘high-income, 22 EU-28 includes the United Kingdom until the end of the transition period. 10 high-growth’ regions making up the bulk of the Western part of the country, and ‘transition, high- growth’ or ‘less developed, high-growth’ regions being the majority in East Germany. 23 Figure 2: Spatial distribution of regions by income level and long-run growth rate. Notes: Authors’ elaboration on Eurostat data. High-income regions are those with an average yearly GDP per capita over the period 2003-2017 equal to or greater than the 90% of the sample average. Transition regions are those with an average yearly GDP per capita over the period 2003-2017 equal to or greater than the 75% of the sample average, but lower than the 90% of the sample average. Less developed regions are those with an average yearly GDP per capita over the period 2003-2017 equal to or greater than the 50% of the sample average, but lower than the 75% of the sample average. Low-income regions are those with an average yearly GDP per capita over the period 2003-2017 lower than the 50% of the sample average. High- (low-)growth regions are those recording a long-run growth rate of GDP per capita between the years 2003 and 2017 equal to or greater than (lower than) the 50% of the sample average. 2.3. Empirical Model We use a simple growth model to estimate the role of industrial structure, innovation, and inward FDI on economic growth. The growth model is of the form: ∆ = 0 + 1 log(−1 ) + 2 ℎ + +3 log( −1 ) + 4 log(ℎ −1 ) + +5 log(ℎ −1 ) + 6 log(ℎ −1 ) + +7 log(ℎ −1 ) + +8 log(ℎ −1 ) + 9 log(−1 ) + +10 log( −1 ) + + + (1) 23 In an Appendix, available on request, Figures A1 and A2 map the spatial distribution of NUTS-3 regions by income level and long- run growth rate, respectively. Table A1 reports the distribution of NUTS-2 regions by income level and long-run growth rate, while Figures A3 and A4 map the spatial distribution of NUTS-2 regions by income level and long-run growth rate, respectively. Tables A2 and A3 report the distribution of NUTS-2 regions by country, and by income level and long-run growth rate, respectively. 11 where the dependent variable for region = 1, . . . , 1321 at time = 2003, . . . , 2017 is defined in terms of yearly growth of GDP per capita as follows: ∆ = log( ) − log(−1 ) (2) The right-hand side of Equation (1) includes three baseline controls. These controls are: GDP per capita at the beginning-of-the growth period (−1 ) to capture relative - convergence; population change ( ℎ ) as a dummy variable taking value one if a region has recorded a null or strictly positive change in population between times and − 1, and zero otherwise, to ensure that ‘growth’ in GDP per capita does not merely reflect a declining population; and employment density ( −1), defined as number of employed persons per square kilometer, to reflect the influence of agglomeration-related forces. The explanatory variables of interest are aimed at capturing industrial structure, innovation, and inward FDI. First, industrial structure is proxied by a set of variables defined in terms of sectoral share of GVA: ℎ −1, denoting the share of regional GVA ascribable to the agriculture sector; ℎ −1, denoting the share of GVA ascribable to the industrial sector; ℎ −1, denoting the share of GVA ascribable to the construction sector; ℎ −1 , denoting the share of GVA ascribable to the market services sector; and ℎ −1 , denoting the share of GVA ascribable to the non-market services sector. Second, regions’ innovativeness is captured by the variable −1 , defined as the fractional number of PCT patents per 100,000 inhabitants. Third, the role of inward FDI is captured by the variable −1 , defined as the number of ‘greenfield’ investments set up in a region per 100,000 inhabitants. Finally, and denote vectors of region- and year-specific fixed effects (FE), respectively, while denotes the error term. 24 In particular, the two-way FE estimation approach allows us to relax unobserved heterogeneity and omitted variable biases. Indeed, region-specific FEs allow us to control for time-invariant unobservable factors affecting each specific region, while year-specific FEs allow us to control for unobserved factors affecting all regions in a given year. The role of inward FDI is further dissected by examining the business activity of the investing company. In some places in our analysis, we modify Equation (1) by replacing the variable capturing the log-number of inward FDI per 100,000 inhabitants with a categorical variable ( −1) aimed at capturing whether a region has received ‘greenfield’ inward FDI in a certain year. If it has done so, the variable records the business activity with the highest number of investment projects: 25 (i) headquarter (strategic activities, legal, finance, public affairs, government relations, accounting); (ii) innovation (consisting of research and development, design, testing, education and training); (iii) production (extraction, manufacturing, construction); (iv) logistics, distribution, and transportation; (v) marketing and sales (consisting of activities to inform buyers, support services to customers, sales and after sale services). The aim of this exercise is to evaluate whether regions that have received inward FDI have registered a ‘growth premium’ with respect to non-receiving ones, and also whether this premium is associated with a particular business activity run by the multinational company in the host regional economy. Table 5 reports some descriptive statistics of the dependent and the explanatory variables; Table 6 reports the correlation matrix of explanatory variables. Given our focus on the ‘lagging’ typology of regions, the two-way FE estimation of Equation (1) and its modified version has been performed on the regional categories defined according to income level and long-run growth rate criteria, as well as the whole sample of regions. 2 24 An Inverse Hyperbolic Sine transformation of the form = �0.5 � + �1 + �� has been applied to handle variables including ‘zero’ entries, namely those capturing inward FDI and patents. 25 The categories of business activities are defined according to the fDi Markets taxonomy as Crescenzi et al. (2014). 12 Table 5: Time average of yearly GDP per capita growth over the period 2003-2017. Variable Mean Std. Dev. Min. Max. ∆GDPpcrt 0.03 0.06 -0.57 0.55 log(GDPpcrt−1 ) 9.94 0.66 7.17 13.07 NonNegative Population Changert 0.61 0.49 0 1 log(Employment Densityrt−1 ) -9.45 1.51 -14.11 -3.35 log(Share GVA Agriculturert−1 ) 0.03 0.04 0.00 0.36 log(Share GVA Industryrt−1 ) -1.59 0.54 -4.87 -0.16 log(Share GVA Constructionrt−1 ) -2.83 0.44 -6.62 -0.83 log(Share GVA Market Servicesrt−1 ) -0.87 0.54 -9.41 -0.13 log(Share GVA NonMarket Servicesrt−1 ) -1.55 0.59 -10.18 -0.66 log(Patentsrt−1 ) 2.00 1.55 -0.69 6.05 log(Inward FDIrt−1 ) -0.33 0.54 -0.69 5.43 Max Inward FDIrt−1 None 0.53 0.50 0 1 Headquarter 0.03 0.18 0 1 Innovation 0.02 0.13 0 1 Production 0.17 0.37 0 1 Logistics, Distribution, Transportation 0.04 0.19 0 1 Marketing and Sales 0.22 0.41 0 1 Table 6: Time average of yearly GDP per capita growth over the period 2003-2017. Variable [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] log(GDPpcrt−1 ) [1] 1 NonNegative Population Changert [2] 0.41 1 log(Employment Densityrt−1 ) [3] 0.48 0.25 1 log(Share GVA Agriculturert−1 ) [4] -0.69 -0.31 -0.57 1 log(Share GVA Industryrt−1 ) [5] -0.19 -0.23 -0.21 0.07 1 log(Share GVA Constructionrt−1 ) [6] -0.31 0.06 -0.40 0.22 0.00 1 log(Share GVA Market Servicesrt−1 ) [7] 0.30 0.18 0.20 -0.35 -0.44 -0.24 1 log(Share GVA NonMarket Servicesrt−1 ) [8] 0.20 0.06 -0.01 -0.22 -0.38 -0.20 0.81 1 log(Patentsrt−1 ) [9] 0.72 0.30 0.40 -0.58 0.07 -0.31 0.20 0.10 1 log(Inward FDIrt−1 ) [10] 0.12 0.13 0.30 -0.14 -0.05 -0.14 0.07 -0.06 0.10 1 Max Inward FDIrt−1 None [11] 0.00 -0.15 -0.20 0.09 -0.05 0.01 -0.02 0.09 -0.04 -0.70 1 Headquarter [12] 0.08 0.07 0.10 -0.06 -0.06 -0.06 0.05 0.01 0.06 0.14 -0.19 1 Innovation [13] 0.06 0.05 0.02 -0.04 0.00 -0.02 0.02 0.01 0.07 0.06 -0.14 -0.02 1 Production [14] -0.22 -0.04 -0.09 0.10 0.17 0.10 -0.12 -0.10 -0.16 0.27 -0.48 -0.08 -0.06 1 Logistics, Distribution, Transportation [15] 0.03 0.04 0.05 -0.04 0.03 0.01 0.02 -0.01 0.03 0.10 -0.21 -0.04 -0.03 -0.09 1 Marketing and Sales [16] 0.13 0.15 0.25 -0.14 -0.09 -0.08 0.10 -0.02 0.13 0.48 -0.55 -0.10 -0.07 -0.24 -0.10 1 13 3. ECONOMIC GEOGRAPHY OF NUTS-3 INCOME LEVELS AND GROWTH NUTS-3 level data confirm that inequalities among regions have been rising since 2009 and have risen much faster than inequalities between countries. This is clearly shown in Figure 3, which plots yearly variations in GDP per capita across countries, across NUTS-3 regions, and across NUTS-3 regions within countries over the period 2003-2017. Figure 3: Yearly coefficient of variation in GDP per capita. Notes: Authors’ elaboration on Eurostat data. The plot reports the coefficient of variation. The year 2003 is set equal to 1 as reference value for the time series. Data cover the 28 EU Member States for a total of 1,321 NUTS-3 level regions. The five French Overseas Departments, and the Spanish extra-territorial autonomous cities of Ceuta and Melilla have been excluded à priori. The sample does not include 20 Polish regions for which data are not available. The within-country cross-region coefficient of variation is defined by, first, calculating the cross-regional coefficient of variation by country, and, second, by averaging the country-level coefficient of variation by year. The within-country cross-region coefficient of variation excludes Cyprus and Luxembourg due to the presence of a unique region at the NUTS-3 level. The performance of NUTS-3 regions appears to vary substantially even within the same NUTS-2 regions. Figure 4 maps the spatial distribution of the average yearly GDP per capita growth rate at the NUTS-3 level and shows how ‘good’ and ‘bad’ performing NUTS-3 regions coexist within the same NUTS-2 region. This seems to be the case particularly in Eastern EU countries, i.e. those that benefitted the most from cohesion funds since the 2004 enlargement. 26 26 Appendix Table A6 reports the distribution of NUTS-3 regions and corresponding NUTS-2 regions by income level and long-run growth rate. Appendix Figure A5 replicates Figure 4 considering country rather than NUTS-2 level borders. 14 Figure 4: Spatial distribution of regional average yearly GDP per capita growth. Notes: Authors’ elaboration on Eurostat data. Data cover the 28 EU Member States for a total of 1,321 NUTS-3 level regions. The five French Overseas Departments, and the Spanish extra-territorial autonomous cities of Ceuta and Melilla have been excluded à priori. The sample does not include 20 Polish regions for which data are not available. Time average of the regional yearly growth rate of GDP per capita over the period 2003-2017. Percentage values. The darker the shade, the higher the growth rate. This finding implies there is an urgent need to design and program Cohesion Policy initiatives at the NUTS-3 level. Intervention strategies designed for NUTS-2 regions are less likely to have addressed the heterogeneous experience of NUTS-3 regions within their territories. In principle, investments and actions at NUTS-2 level would specifically address leading and lagging regions within their territory – for example in ‘linking’ the various types of regions for mutual benefit. 27 Unfortunately, spatially integrative programs are rarely proposed by NUTS-2 regions, owing to coordination failures between and within local governments. 28 So programming at the NUTS-3 level may be more realistic to address these challenges. NUTS-3 level analysis can inform NUTS-3 programming decisions – such as on ITIs and other SUD instruments. Turning to data on economic growth, we can observe immediately that the performance of NUTS-3 regions has been strongly heterogeneous in the 2003 to 2017 period. Table 7 shows the average yearly growth rates of GDP per capita for all income and growth categories. Growth rates are quite similar among the income groups (around 2% per year), except for low-income regions which have grown much more (around 6% per year). Low-growth lagging regions have grown at an average of only 0.72% per year during the 2003-2017 period. This group of 171 regions – about 12.9% of the sample – may be a special cause for concern. They have incomes of less than 90% of 27 NUTS-3 level analysis would also be essential in these cases, to give a more precise picture of economic performance and potential within these large and heterogeneous regions. 28 Some exceptions include, for example, the Pomorskie NUTS-2 region in Poland – which has considered the different and complementary roles of its major cities in linking to Gdansk’s economy, and Emilia-Romagna NUTS-2 region in Italy – which has an ‘inner areas’ strategy for its lagging NUTS-3 regions. 15 average GDP per capita in the EU, but have grown slower than low growth regions that are already high income. Indeed, 17% of this group of 171 regions recorded a negative average yearly GDP per capita growth rate over the period 2003-2017. Examples include the Greek region of Thesprotia which recorded an average growth rate of minus 2.4%, and the Greek region of Ioannina which recorded an annual growth rate of minus 0.02%. Figure 5 plots the growth rates against income level. Clearly the most variation in growth rate is among the lowest income regions: these regions have grown faster on average, but with very substantial heterogeneity. Table 7: Time average of yearly GDP per capita growth over the period 2003-2017. Taxonomy Average Yearly GDP Per Capita Growth Income Level High-Income 2.21 Transition 2.19 Less Developed 2.11 Low-Income 6.12 Long-Run Growth Rate High-Income, High-Growth 2.83 High-Income, Low-Growth 1.11 Not High-Income, High-Growth 4.87 Not High-Income, Low-Growth 0.72 Decline in Average GDP per capita around the Great Recession Not Declined 3.31 Declined 0.82 Notes: Authors’ elaboration on Eurostat data. Time average of the yearly GDP per capita growth rate in percentage terms over the period 2003-2017. High-income regions are those with an average yearly GDP per capita over the period 2003-2017 equal to or greater than the 90% of the sample average. Transition regions are those with an average yearly GDP per capita over the period 2003-2017 equal to or greater than the 75% of the sample average, but lower than the 90% of the sample average. Less developed regions are those with an average yearly GDP per capita over the period 2003-2017 equal to or greater than the 50% of the sample average, but lower than the 75% of the sample average. Low-income regions are those with an average yearly GDP per capita over the period 2003-2017 lower than the 50% of the sample average. High- (not high-)income regions are those with an average yearly GDP per capita over the period 2003-2017 equal to or greater than (lower than) the 90% of the sample average. High- (low-)growth regions are those recording a long-run growth rate of GDP per capita between the years 2003 and 2017 equal to or greater than (lower than) the 50% of the sample average. Decline of GDP per capita around the Great Recession is defined by comparing the average yearly GDP per capita over the pre-crisis period 2003-2007 with the average yearly GDP per capita over the subsequent period 2008-2017. The varied performance of NUTS-3 regions is shown particularly in their trajectories after the Great Recession. We examine the performance of NUTS-3 regions in the pre-crisis period 2003-2007 and the subsequent period 2008-2017, to find the proportion of regions that experienced a decline in the average yearly GDP per capita between the two periods. According to data summarized in Table 8, the share of ‘declining’ NUTS-3 regions is 20.4% among ‘high-income’ regions, 28.9% among ‘transition’ regions, 27.3% among ‘less developed’ regions, and only 2.8% among ‘low-income’ regions. 29 Such a pattern emerges clearly from Figure 6, which plots the average GDP per capita among each category of NUTS-3 region. On average, ‘low-income’ regions have recorded a better trend than regions in the other three categories. This finding suggests that ‘low- income’ regions may not be as much of a problem for Cohesion Policy than declining and ‘low- growth’ regions in the less developed and transition categories. In subsequent analysis, we group the transition, less developed, and low-income regions together as a ‘not high-income’ category. 29 Appendix Table A7 (available on request) reports the distribution of regions by their declining vs. non-declining status in terms of average yearly GDP per capita around the Great Recession and by country. Figure A6 (in appendix, available on request) spatially identifies those regions that have recorded a decline in average yearly GDP per capita between the pre-crisis period 2003-2007 and the subsequent period 2008-2017. 16 Figure 5: Regions’ distribution by income level and long-run growth rate. Notes: Authors’ elaboration on Eurostat data. The dashed lines refer to mean values. Average yearly GDP per capita is defined over the period 2003-2017. Long-run GDP per capita growth rate is defined between the years 2003 and 2017. The plot represents the 99.85% of the sample, as the UK regions of Camden and City of London (UKI31) and Westminster (UKI32) have been excluded due to extremely high values of the average yearly GDP per capita over the period 2003-2017. Specifically, Camden and City of London (Westminster) recorded an average yearly GDP per capita over the period 2003-2017 equal to about €382 billion (€284 billion), and a long-run GDP per capita growth rate equal to about 67% (49%). Table 8: Regions’ distribution by income level and decline in average yearly GDP per capita around the Great Recession. Average Yearly GDP Per Capita Declined After the Great Recession Income Level No Yes Total No. % No. % No. % High-Income 597 45.19 153 11.58 750 56.78 Transition 138 10.45 56 4.24 194 14.69 Less Developed 120 9.08 45 3.41 165 12.49 Low-Income 206 15.59 6 0.45 212 16.05 Total 1,061 80.32 260 19.68 1,321 100.00 Notes: Authors’ elaboration on Eurostat data. Percentages are defined on the total number of regions in the sample. High-income regions are those with an average yearly GDP per capita over the period 2003-2017 equal to or greater than the 90% of the sample average. Transition regions are those with an average yearly GDP per capita over the period 2003-2017 equal to or greater than the 75% of the sample average, but lower than the 90% of the sample average. Less developed regions are those with an average yearly GDP per capita over the period 2003-2017 equal to or greater than the 50% of the sample average, but lower than the 75% of the sample average. Low-income regions are those with an average yearly GDP per capita over the period 2003-2017 lower than the 50% of the sample average. Decline of GDP per capita around the Great Recession is defined by comparing the average yearly GDP per capita over the pre-crisis period 2003-2007 with the average yearly GDP per capita over the subsequent period 2008-2017. 17 Figure 6: Temporal dynamics of GDP per capita by income level category. Notes: Authors’ elaboration on Eurostat data. Yearly log-GDP per capita is averaged by income level regional category. High-income regions are those with an average yearly GDP per capita over the period 2003-2017 equal to or greater than the 90% of the sample average. Transition regions are those with an average yearly GDP per capita over the period 2003-2017 equal to or greater than the 75% of the sample average, but lower than the 90% of the sample average. Less developed regions are those with an average yearly GDP per capita over the period 2003-2017 equal to or greater than the 50% of the sample average, but lower than the 75% of the sample average. Low-income regions are those with an average yearly GDP per capita over the period 2003-2017 lower than the 50% of the sample average. The growth performance of NUTS-3 regions indicates a special concern for ‘low-growth’ lagging regions. Figure 7 plots the temporal dynamics of the average yearly GDP per capita by ‘high- income’ vs. ‘not high-income’ (i.e. regions with an average yearly GDP per capita over the period 2003-2017 lower than 90% of the sample average) categories, and shows their long-run growth pattern between the years 2003 and 2017. Among the high-growth regions (solid lines on the graph), the ‘not high-income’ group of regions has grown faster than the ‘high-income’ group. However, among the low-growth regions (dashed lines on the graph), the ‘not high-income’ group has grown slower than the ‘high-income’ group. This again indicates a special cause for concern about ‘low- growth’ lagging regions: these regions are relatively poor but still are not growing. Figure 8 plots the average GDP per capita among the bottom and top deciles of regions in these groups. The graphs show a much larger heterogeneity in economic performance among the ‘not high-income’ regions (red lines) than among the high-income regions (black lines). Particularly from 2009 onwards, there has been an absolute and sustained decline in income among the low-growth regions (dashed red line on the graph). 30 30 This heterogeneity in growth among not high-income regions explains the apparently contradictory findings from Figure 3 (that inequality among regions has been rising) and Figure 7 (that the average incomes of the four categories of regions have been converging). The inequality among regions is driven by strong within-group variation in growth rates, displayed in Figure 5 and Figure 8. 18 Figure 7: Temporal dynamics of GDP per Figure 8: Temporal dynamics of GDP per capita by income level and long-run growth capita for top and bottom 10% regions by rate category. income level and long-run growth rate. Notes: Authors’ elaboration on Eurostat data. Yearly log-GDP Notes: Authors’ elaboration on Eurostat data. Yearly GDP per per capita is averaged by income level and long-run growth rate capita is averaged for top and bottom 10% regions defined with regional category. High- (not high-)income regions are those respect to the distribution of log-GDP per capita, and then by with an average yearly GDP per capita over the period 2003- income level and long-run growth rate regional category. High- 2017 equal to or greater than (lower than) the 90% of the sample (not high-)income regions are those with an average yearly GDP average. High- (low-)growth regions are those recording a long- per capita over the period 2003-2017 equal to or greater than run growth rate of GDP per capita between the years 2003 and (lower than) the 90% of the sample average. High- (low-)growth 2017 equal to or greater than (lower than) the 50% of the sample regions are those recording a long-run growth rate of GDP per average. capita between the years 2003 and 2017 equal to or greater than (lower than) the 50% of the sample average. Figure 9: Temporal dynamics of GDP per capita for top and bottom 10% regions by income level. Notes: Authors’ elaboration on Eurostat data. Yearly GDP per capita is averaged for top and bottom 10% regions defined with respect to the distribution of log-GDP per capita, and then by income level regional category. High-income regions are those with an average yearly GDP per capita over the period 2003-2017 equal to or greater than the 90% of the sample average. Transition regions are those with an average yearly GDP per capita over the period 2003-2017 equal to or greater than the 75% of the sample average, but lower than the 90% of the sample average. Less developed regions are those with an average yearly GDP per capita over the period 2003- 2017 equal to or greater than the 50% of the sample average, but lower than the 75% of the sample average. Low-income regions are those with an average yearly GDP per capita over the period 2003-2017 lower than the 50% of the sample average. 19 The growth performance of less developed regions has been worse than low-income regions. Figures 9 plots the average GDP per capita among the bottom decile and top decile of regions within each income group. The red lines (less developed and low-income regions) show that growth performance of low-income regions has generally been stronger than less developed regions – especially among those in the bottom decile of their group. In other words, low-income regions have exhibited a strong performance in catching up, even after the Great Recession, while the less developed regions have been more likely to exhibit low-growth and declining incomes. So the less developed regions may require special attention by policy makers. Low-income regions have been able to improve their relative position more often than low-growth lagging regions. Figure 10 plots income levels of all low-income regions in 2003 and 2017. The regions in the top-left quadrant of the graph are those that had incomes below the low- income group average in 2003, and above average for the group in 2017 – i.e. indicating unusually fast growth. Examples include the Croatian region of Koprivnica-Križevci which recorded an average growth rate of 1.7%. Meanwhile the regions in the bottom-right were above average in 2003 but below average in 2017, indicating unusually slow growth. Examples include the Greek region of Rhodope, which recorded annual average growth of the average yearly GDP per capita growth rate of minus 1.8%. Regions above a 45-degree line gained in relative position in the group, while regions below a 45-degree line lost in relative position. When comparing this pattern to Figure 11, which plots income levels of all low-growth lagging regions in the same years, we see a much tighter distribution around the 45-degree line. This indicates less change in the ranking of regions among low-growth regions. Figure 10: GDP per capita distribution for Figure 11: GDP per capita distribution for ‘not ‘low-income’ regions in 2003 and 2017. high-income, low-growth’ regions in 2003 and 2017. Notes: Authors’ elaboration on Eurostat data. The dashed lines Notes: Authors’ elaboration on Eurostat data. The dashed lines refer to mean values. Low-income regions are those with an refer to mean values. Not high-income and low-growth regions average yearly GDP per capita over the period 2003-2017 lower are those with an average yearly GDP per capita over the period than the 50% of the sample average. 2003-2017 lower than the 90% of the sample average, and recording a long-run growth rate of GDP per capita between the years 2003 and 2017 lower than the 50% of the sample average. These results are not strongly influenced by emigration or population change: they reflect the economic performance of the regions. One of the checks we made while conducting the analysis was on the relationship between GDP performance and GDP per capita performance. We wanted to ensure that the high growth rates in GDP per capita experienced by low income regions were not an artefact of population shrinkage. For example, if low income regions have high rates of emigration, their GDP per capita will rise, even if GDP is flat or shrinking. However, this appears to be the case 20 only in very few regions. Figure 12 shows the very close relationship between GDP and GDP per capita growth rates. The very few regions in the top-left quadrant are those that have negative GDP growth and positive per capita growth (owing to population shrinkage). 31 Overall, the close relationship between GDP and GDP per capita indicates we can focus on GDP per capita as our outcome variable of interest for measuring economic performance, without serious caveats about population change. Overall, this analysis has shown that NUTS-3 regions are heterogeneous, with a large group of slow and declining lagging regions, but that there are success cases from which we can learn. First, as is already well known, the EU is characterized by the coexistence of ‘rich’ and ‘poor’ territories, as well as of more and less dynamic ones. Second, the persistence of a gap in income level among 171 NUTS-3 regions, which are low income but not growing, shows that EU Cohesion Policy has not yet succeeded in addressing challenges in a substantial proportion of lagging regions. The NUTS-3 lens on regional economic performance can help identify places that need additional support. Third, our evidence has highlighted some successful cases, i.e. lagging regions that over a quite long time period (from 2003 to 2017) have been able to improve their income level. Thus, it is worth looking at these exceptions, and assessing whether we can learn from them. To this aim, we must understand the economic forces driving growth in the various types of regions. In the remainder of the paper, we use quantitative analysis to understand the correlates of growth. Figure 12: Correlation between yearly GDP per capita growth and yearly GDP growth. Notes: Authors’ elaboration on Eurostat data. Dashed lines refer to sample mean values. The correlation coefficient of the two variable is equal to 0.987, with p-value equal to 0.000. 31 There is a greater number of regions in the bottom-right quadrant which have positive GDP growth but negative GDP per capita growth (owing to population growth). 21 4. EMPIRICAL RESULTS 4.1. Main Results First, looking at all regions together, we see that growth is correlated with an increased share of industry, construction, and market services, plus innovation and FDI. Table 9 reports the results of the two-way FE estimation of various versions of the baseline Equation (1), and its modified version accounting for business activities of inward FDI, for the whole sample of regions. The results show a general convergence process, as denoted by the negative coefficient on the variable capturing GDP per capita at the beginning of the period. Agglomeration forces, as captured by the employment density variable, are positively associated with GDP per capita growth. The results indicate that a 10 percent change in employment density is associated with a 0.4 to 0.6 percent change in GDP per capita. The correlation with population change is negligible. Looking at specifications (4), (5), and (6), we see that GDP per capita growth is positively associated with increases in the industry, construction, and market services shares of GVA, while negatively associated with increases in the non-market services share of GVA. The agriculture sector, on the contrary, does not show a significant correlation with economic growth. Furthermore, both innovation and inward FDI emerge as growth-enhancing factors, and the coefficients on the industry variables are relatively consistent with the addition of these variables. Inward FDI is most beneficial to incomes growth when it is in production activities, logistics, distribution and transportation. These results emerge looking at specification (7), which replaces the continuous variable for inward FDI with the categorical one ( −1), showing the business activity most commonly observed in the host region among positive FDI. The evidence implies that regions have recorded a ‘growth premium’ when they attract inward FDI in production activities and logistics, distribution and transportation. Disaggregating the regions by income category, we see that correlates of growth differ for low-income regions – in which none of the industries is significantly correlated with growth, but the role of FDI is strongest. Table 10 reports the results of two-way FE estimation of Equation (1) and its modified version obtained by accounting for regional heterogeneity in terms of income level. Drawing on the income-based taxonomy previously presented, regions have been split into ‘high- income’, ‘transition’, ‘less developed’, and ‘low-income’. Comparison of the results reported in specifications (1) to (4) indicates the existence of different growth pathways for different types of regions. First, looking at the industrial structure of regions, it emerges that increases in the industry, construction, and market services shares of GVA are positively associated with economic growth in all regions, except for ‘low-income’ regions where there is no pattern in the industries contributing to growth. For agriculture, there is a positive association with growth; and for non-market services there is a negative association with growth – but both of those relationships seem to affect only ‘transition’ and ‘less developed’ (‘high-income’) regions. Second, innovation (proxied by the number of patents) seems to be a growth-enhancing factor in all regions, except for ‘transition’ regions. This may be because transition regions rely more on technology adoption, rather than pioneering innovation, but this requires further investigation. Third, inward FDI seems to be associated with growth only in ‘low-income’ regions. In those regions, the relationship is very robust (significant at the 0.1 percent level), but it is relatively weak in impact: a doubling of the number of foreign investments per 100,000 inhabitants is associated with a 1.1 percent increase in GDP per capita growth. For example, a region with an average per capita income of €4,000 and 2 foreign investments per 100,000 inhabitants, would find on average that doubling the number of investments to 4 per 100,000 inhabitants is associated with an increase in per capita income of €44. This should not discourage low-income regions, since the relationship with FDI is still robust and positive, but indicates that increases in GDP per capita for the whole region are associated only with a dramatic increase in the number of foreign investments. 32 32 Some readers will question why the number of investments is used, rather than the value of investments or the jobs created in the host region through inward FDI. Both values are available from the same fDi Markets database. However, they are generally regarded 22 Table 9: Baseline estimates – Whole sample. Dependent Variable ∆GDPpcrt (1) (2) (3) (4) (5) (6) (7) log(GDPpcrt−1 ) -0.168**** -0.170**** -0.174**** -0.203**** -0.204**** -0.204**** -0.204**** (0.005) (0.005) (0.005) (0.006) (0.006) (0.006) (0.006) Population Changert <0 … Ref. Ref. Ref. Ref. Ref. Ref. ≥0 … 0.004**** 0.002 0.000 0.000 0.000 0.000 (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) log(Employment Densityrt−1 ) … … 0.062**** 0.039**** 0.038**** 0.036**** 0.038**** (0.009) (0.009) (0.009) (0.010) (0.009) log(Share GVA Agriculturert−1 ) … … … -0.056 -0.059 -0.058 -0.059 (0.055) (0.055) (0.055) (0.055) log(Share GVA Industryrt−1 ) … … … 0.032**** 0.032**** 0.032**** 0.032**** (0.005) (0.005) (0.005) (0.005) log(Share GVA Constructionrt−1 ) … … … 0.028**** 0.028**** 0.028**** 0.028**** (0.003) (0.003) (0.003) (0.003) log(Share GVA Market Servicesrt−1 ) … … … 0.051**** 0.051**** 0.050**** 0.051**** (0.007) (0.007) (0.007) (0.007) log(Share GVA NonMarket Servicesrt−1 ) … … … -0.033**** -0.033**** -0.032**** -0.033**** (0.007) (0.007) (0.007) (0.007) log(Patentsrt−1 ) … … … … 0.003**** 0.003**** 0.003**** (0.001) (0.001) (0.001) log(Inward FDIrt−1 ) … … … … … 0.004*** … (0.001) Max Inward FDIrt−1 None … … … … … … Ref. Headquarter … … … … … … 0.001 (0.003) Innovation … … … … … … 0.004 (0.003) Production … … … … … … 0.002* (0.001) Logistics, Distribution, Transportation … … … … … … 0.005** (0.002) Marketing and Sales … … … … … … 0.002 (0.001) Region FE Yes Yes Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Yes Yes No. Observations 18,494 18,494 18,494 18,494 18,494 18,494 18,494 No. Regions 1,321 1,321 1,321 1,321 1,321 1,321 1,321 R2 0.33 0.33 0.34 0.35 0.35 0.35 0.35 Adjusted R2 0.28 0.28 0.28 0.30 0.30 0.30 0.30 406.74 380.82 358.97 281.76 269.55 258.36 220.55 Model F Statistics [p-value] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] Notes: * < 0.1; ** < 0.05; *** < 0.01; **** < 0.001. Robust standard errors are reported in parentheses. The variables capturing patents and inward FDI are defined per 100,000 inhabitants. All variables are defined in logarithm terms, except for the dummy variable capturing population change between times and − 1, and the categorical variable included in specification (7) that captures the most relevant business activity of inward FDI that interested a region, for which the category used as reference is no inward FDI received. as less precise, since they reflect announced investment values (i.e. as publicized in the media) rather than actual investment values. The actual value of greenfield investments often differs sharply from the announced values. There is no overall consensus among researchers on using number of investments or monetary (or job) value of investments. However, we performed a check on these results to determine if they are substantially different if using the monetary and job value of investments. Specifically, we have tested the robustness of the inward FDI-growth relationship for low-income regions by replicating specification (4) in Table 8 using both a monetary value-based variable and a job-based variable for inward FDI, rather than the number of investments set up in a region. The results confirm the positive association between inward FDI and GDP per capita growth in low-income regions, although the estimated elasticities are equal to 0.001 (rather than 0.011) and are statistically significant at 10 (rather than 0.1) percent level. However, incomplete information on some investments concerning the monetary value or the number of jobs associated with inward FDI could help explaining this reduction in magnitude. Furthermore, it is worth noting that the sign, magnitude, and significance level of all the other variables are confirmed when considering the two alternative variables for inward FDI. 23 Table 10: Estimates by income level category. Dependent Variable ∆GDPpcrt Regional Taxonomy High-Income Transition Less Developed Low-Income (1) (2) (3) (4) (5) log(GDPpcrt−1 ) -0.265**** -0.222**** -0.144**** -0.206**** -0.206**** (0.010) (0.016) (0.016) (0.011) (0.011) Population Changert <0 Ref. Ref. Ref. Ref. Ref. ≥0 0.000 -0.005* -0.003 0.012*** 0.012** (0.001) (0.003) (0.003) (0.005) (0.005) log(Employment Densityrt−1 ) 0.039** 0.073*** 0.040* 0.058**** 0.061**** (0.016) (0.027) (0.023) (0.016) (0.016) log(Share GVA Agriculturert−1 ) 0.152 0.406** 0.212* -0.091 -0.105 (0.132) (0.177) (0.122) (0.071) (0.072) log(Share GVA Industryrt−1 ) 0.047**** 0.109**** 0.021* 0.016 0.016 (0.007) (0.020) (0.012) (0.011) (0.010) log(Share GVA Constructionrt−1 ) 0.037**** 0.059**** 0.039**** 0.001 0.001 (0.004) (0.008) (0.005) (0.006) (0.006) log(Share GVA Market Servicesrt−1 ) 0.110**** 0.143*** 0.066** 0.001 0.001 (0.011) (0.047) (0.030) (0.013) (0.013) log(Share GVA NonMarket Servicesrt−1 ) -0.070**** 0.002 -0.027 0.007 0.008 (0.009) (0.030) (0.021) (0.013) (0.013) log(Patentsrt−1 ) 0.002* -0.001 0.005*** 0.004** 0.004** (0.001) (0.001) (0.002) (0.002) (0.002) log(Inward FDIrt−1 ) 0.002 0.003 -0.001 0.011**** … (0.001) (0.003) (0.005) (0.003) Max Inward FDIrt−1 None … … … … Ref. Headquarter … … … … 0.004 (0.010) Innovation … … … … -0.011 (0.011) Production … … … … 0.007** (0.003) Logistics, Distribution, Transportation … … … … 0.014* (0.008) Marketing and Sales … … … … 0.005 (0.004) Region FE Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes No. Observations 10,500 2,716 2,310 2,968 2,968 No. Regions 750 194 165 212 212 R2 0.39 0.37 0.36 0.48 0.48 Adjusted R2 0.34 0.31 0.31 0.44 0.43 Model F Statistics [p-value] 177.10 [0.000] 42.93 [0.000] 47.22 [0.000] 83.12 [0.000] 71.49 [0.000] Notes: * < 0.1; ** < 0.05; *** < 0.01; **** < 0.001. Robust standard errors are reported in parentheses. High-income regions are those with an average yearly GDP per capita over the period 2003-2017 equal to or greater than the 90% of the sample average. Transition regions are those with an average yearly GDP per capita over the period 2003-2017 equal to or greater than the 75% of the sample average, but lower than the 90% of the sample average. Less developed regions are those with an average yearly GDP per capita over the period 2003-2017 equal to or greater than the 50% of the sample average, but lower than the 75% of the sample average. Low-income regions are those with an average yearly GDP per capita over the period 2003-2017 lower than the 50% of the sample average. The variables capturing patents and inward FDI are defined per 100,000 inhabitants. All variables are defined in logarithm terms, except for the dummy variable capturing population change between times and − 1, and the categorical variable included in specification (5) that captures the most relevant business activity of inward FDI that interested a region, for which the category used as reference is no inward FDI received. Looking more closely at the role of FDI in low-income regions, we see growth is generated most by production, logistics, distribution and transportation activities. Specification (5) uses the categorical variable for inward FDI in the sub-sample of ‘low-income’ regions. The results indicate that ‘low-income’ regions exhibited a ‘growth premium’ when they attract inward FDI in production activities and logistics, distribution and transportation. These are usually not high value 24 activities, and may not be adding much in terms of the sophistication of the local economy. What they do add is jobs: if low-income regions often lack the firms and investment that can generate jobs, this is what FDI can provide – and thus increases GDP per capita. We may surmise that FDI is providing the capital and entrepreneurship that is ‘missing’ in low-income regions. This is not the case in the other income categories of regions. 33 Meanwhile in low-growth lagging regions, growth is correlated systematically only with construction and patents, and is in a negative relationship with FDI. Table 11 considers NUTS- 3 regions using the long-run growth taxonomy. In ‘high-income’ regions – see specifications (1) and (2) – economic growth is correlated positively with industry, construction, and market services; negatively with non-market services; and shows no correlation with agriculture. The role of patenting activity is positive but small in high-growth regions, and the role of inward FDI is positive but small in low-growth regions (a 100 percent increase in FDI is associated with a 0.5 increase in GDP per capita growth). In ‘not high-income’ regions – see specifications (3) and (4) –, the findings are quite different: economic growth is associated with a move away from agriculture among high-growth regions, and is associated with an increased share of the construction industry among low-growth regions. 34 FDI is strongly and positively associated with growth among high-growth regions (a 100 percent increase in FDI generates more than twice the increase in GDP per capita as in high-income low-growth regions), but is negatively associated with growth among low-growth regions. This is a puzzling result. 35 In specification (5), we disaggregate FDI into the various business activities of investors, and find that there is ‘a growth premium’ for FDI in high-valued headquarter activities, while all the other business activities of inward FDI are in a negative and negligible association with economic growth. Among low-growth lagging regions that declined in the post-2008 period, their decline is correlated with shrinkage in industry and construction sectors, population decline, and not with FDI. In view of the role of the Great Recession as a threshold event for the economic performance of regions, we have further focused on ‘low-growth’ lagging regions during the post-2008 period. We are most interested in differences between those regions that did and did not register a decline in income level during the post-2008 period. Table 12 reports the two-way FE estimation results. Specification (1) refers to all regions that did not decline, specification (2) refers to all regions that did decline; specifications (3) and (4) refer to the ‘low-growth’ lagging regions that did decline. The results suggest that economic growth in ‘low-growth’ lagging regions is associated with the industry and construction shares of GVA, as well as with innovation, population dynamics, and agglomeration forces. In other words, the decline in those regions has typically been associated with shrinkage in industry, construction, innovation, and population – therefore those weaknesses could be goals for local policy makers to address and mitigate such risks. Growth for this type of region is not correlated with inward FDI. 33 FDI is not systematically associated with growth in less developed, transition, or high-income regions – see specifications (1), (2), and (3) in Table 10. We may thus conclude that it is the strong role of FDI in low-income regions that is generating the correlation when all regions are considered – see Table 9, specifications (6) and (7). The same logic applies for the conclusions on business activities: the significant coefficients on production, logistics, and distribution and transportation from Table 9 can be traced to their role in generating growth specifically in low-income regions. 34 The coefficient on the agriculture share of GVA among high-growth regions is negative and relatively large (a 10 percent decrease in the share of agriculture is associated with a 1.35 percent increase in GDP per capita growth). 35 It is puzzling especially because we found FDI to be strongly correlated with growth among low-income regions. However, we cannot investigate more closely the subset of low-growth low-income regions, since there are only 12 such regions – too few to permit robust statistical analysis. 25 Table 11: Estimates by long-run growth rate category. Dependent Variable ∆GDPpcrt High- vs. Not High-Income High-Income Not High-Income Long-Run Growth Rate High-Growth Low-Growth High-Growth Low-Growth (1) (2) (3) (4) (5) log(GDPpcrt−1 ) -0.342**** -0.374**** -0.233**** -0.254**** -0.255**** (0.017) (0.017) (0.009) (0.017) (0.018) Population Changert <0 Ref. Ref. Ref. Ref. Ref. ≥0 0.002 0.002 0.005** 0.003 0.003 (0.002) (0.003) (0.002) (0.003) (0.003) log(Employment Densityrt−1 ) 0.087**** 0.015 0.054**** 0.062*** 0.063*** (0.019) (0.020) (0.013) (0.021) (0.021) log(Share GVA Agriculturert−1 ) 0.251 0.094 -0.135** 0.036 0.030 (0.164) (0.210) (0.067) (0.124) (0.124) log(Share GVA Industryrt−1 ) 0.040**** 0.035* 0.010 0.018 0.019 (0.008) (0.018) (0.008) (0.013) (0.013) log(Share GVA Constructionrt−1 ) 0.026**** 0.042**** -0.000 0.045**** 0.045**** (0.005) (0.008) (0.005) (0.006) (0.006) log(Share GVA Market Servicesrt−1 ) 0.095**** 0.046 0.014 0.028 0.031 (0.012) (0.042) (0.011) (0.038) (0.038) log(Share GVA NonMarket Servicesrt−1 ) -0.062**** -0.049** -0.007 -0.037 -0.035 (0.010) (0.023) (0.010) (0.026) (0.026) log(Patentsrt−1 ) 0.002* 0.002 0.001 0.004** 0.003** (0.001) (0.002) (0.001) (0.002) (0.002) log(Inward FDIrt−1 ) -0.001 0.005* 0.012**** -0.009* … (0.001) (0.003) (0.002) (0.005) Max Inward FDIrt−1 None … … … … Ref. Headquarter … … … … 0.013* (0.008) Innovation … … … … -0.003 (0.012) Production … … … … -0.001 (0.003) Logistics, Distribution, Transportation … … … … -0.002 (0.007) Marketing and Sales … … … … -0.000 (0.004) Region FE Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes No. Observations 6,720 3,780 5,600 2,394 2,394 No. Regions 480 270 400 171 171 R2 0.39 0.56 0.44 0.41 0.41 Adjusted R2 0.34 0.52 0.39 0.36 0.35 Model F Statistics [p-value] 92.39 [0.000] 141.55 [0.000] 103.48 [0.000] 62.72 [0.000] 53.10 [0.000] Notes: * < 0.1; ** < 0.05; *** < 0.01; **** < 0.001. Robust standard errors are reported in parentheses. High- (not high-)income regions are those with an average yearly GDP per capita over the period 2003-2017 equal to or greater than (lower than) the 90% of the sample average. High- (low-)growth regions are those recording a long-run growth rate of GDP per capita between the years 2003 and 2017 equal to or greater than (lower than) the 50% of the sample average. The variables capturing patents and inward FDI are defined per 100,000 inhabitants. All variables are defined in logarithm terms, except for the dummy variable capturing population change between times and − 1, and the categorical variable included in specification (5) that captures the most relevant business activity of inward FDI that interested a region, for which the category used as reference is no inward FDI received. Finally, examining differences between rural and non-rural regions, we find that economic growth in rural ‘low-income’ regions is associated with patenting and a move away from agriculture; and growth in non-rural ‘low-income’ regions is associated with a strong role for FDI and of agglomeration. As a final exercise, we have focused on lagging regions characteristics in terms of urbanization level. All NUTS-3 regions are classified as urban, 26 intermediate, or rural. 36 Prior evidence indicates that convergence processes among newer EU Member States have been driven by capital cities, but also that small- and medium-sized cities can generate growth (e.g. Frick and Rodríguez-Pose, 2016, 2018). Table 13 reports the results of the two- way FE estimation of Equation (1) by rural vs. non-rural regions. Looking at ‘low-income’ regions – see specifications (3) and (4) –, we see similar results to those in Table 10 in both rural and non-rural regions, except that economic growth in rural regions is positively associated with decreases in the agriculture share of GVA and positively associated with innovation. Meanwhile in non-rural ‘low- income’ regions, yearly GDP per capita growth seems to be positively affected by agglomeration forces. Among low-growth lagging regions, Box 1: Correlation, not causation growth in rural regions is associated with an Readers may be tempted to attribute some kind of increased role for agriculture and construction, causation to the results of this analysis. This should and negatively associated with FDI; growth in be avoided. For example: “We need to boost the non-rural regions is associated with construction industry in our region to get more growth”. construction and innovation. The results The statistical techniques utilized in this paper do not reported in specification (7) indicates that rural permit causation or determinism to be attributed. They allow only the attribution of correlation between regions’ economic growth is positively associated industry structure, innovation, FDI, and growth. with increases in the agriculture and construction Causation would require additional techniques, such as shares of GVA, and negatively associated with the use of instrumental variables. inward FDI. Meanwhile in non-rural ‘low- The results in the paper should be interpreted as growth’ regions – see specification (8) –, showing trends and patterns in economic growth of economic growth is positively associated with regions. The results can be interpreted in the form: increases in the construction share of GVA, “Economic growth in low-income regions is correlated with additional inward FDI”; or “Economic growth in innovation, as well as agglomeration forces. rural regions is correlated with a decreased share of Finally, interesting results emerge when agriculture in GVA”. These interpretations allow considering the subset of ‘low-growth’ lagging regions to know more about the growth pathways of regions that declined in income during the post- regions ‘like theirs’, on average. The results should not 2008 period. Growth in declining rural regions – preclude regions from adopting radically different growth strategies, based on local endowments and see specification (9) – was linked to market opportunities. The results simply allow us to know what services and was not linked to agriculture (all has typically characterized regional economic growth in other correlates are the same as in non-declining each type of region. rural regions). Growth in declining non-rural regions – see specification (10) – was linked to population change and to industry (all other correlates are the same as in non-declining non-rural regions). 36The European Commission classifies NUTS-3 regions as ‘predominantly urban’, ‘intermediate’, and ‘predominantly rural’. Drawing on this taxonomy, we have considered two regional typologies due to the limited number of regions resulting in the different categories when considering also the lagging status of regions. Specifically, we have considered rural (including ‘predominantly rural’) vs. non- rural (including ‘intermediate’ and ‘predominantly urban’) regions. The official taxonomy is available at “https://ec.europa.eu/eurostat/web/rural-development/methodology”. 27 Table 12: Estimates for ‘not high-income, low-growth’ regions that declined around the Great Recession. Dependent Variable ∆GDPpcrt Decline in Average Yearly GDP Per Capita No Yes Regions All All Not High-Income and Low-Growth (1) (2) (3) (4) log(GDPpcrt−1 ) -0.219**** -0.305**** -0.253**** -0.254**** (0.007) (0.016) (0.021) (0.021) Population Changert <0 Ref. Ref. Ref. Ref. ≥0 0.002 0.018**** 0.012*** 0.012*** (0.001) (0.004) (0.005) (0.005) log(Employment Densityrt−1 ) 0.040**** 0.035* 0.066*** 0.068*** (0.010) (0.018) (0.025) (0.025) log(Share GVA Agriculturert−1 ) -0.105* -0.036 -0.052 -0.060 (0.059) (0.150) (0.166) (0.165) log(Share GVA Industryrt−1 ) 0.018**** 0.048*** 0.033* 0.034* (0.005) (0.016) (0.018) (0.018) log(Share GVA Constructionrt−1 ) 0.014**** 0.057**** 0.047**** 0.046**** (0.003) (0.007) (0.007) (0.007) log(Share GVA Market Servicesrt−1 ) 0.028**** 0.100** 0.060 0.060 (0.008) (0.041) (0.051) (0.051) log(Share GVA NonMarket Servicesrt−1 ) -0.015** 0.011 -0.026 -0.023 (0.007) (0.025) (0.035) (0.035) log(Patentsrt−1 ) 0.001* 0.004* 0.005** 0.005** (0.001) (0.002) (0.002) (0.002) log(Inward FDIrt−1 ) 0.004*** 0.001 -0.009 … (0.001) (0.003) (0.006) Max Inward FDIrt−1 None … … … Ref. Headquarter … … … 0.014 (0.009) Innovation … … … -0.010 (0.017) Production … … … -0.001 (0.005) Logistics, Distribution, Transportation … … … -0.006 (0.010) Marketing and Sales … … … 0.002 (0.005) Region FE Yes Yes Yes Yes Year FE Yes Yes Yes Yes No. Observations 14,854 3,640 1,428 1,428 No. Regions 1,061 260 102 102 R2 0.36 0.56 0.45 0.45 Adjusted R2 0.31 0.52 0.39 0.39 Model F Statistics [p-value] 193.73 [0.000] 147.55 [0.000] 46.47 [0.000] 39.44 [0.000] Notes: * < 0.1; ** < 0.05; *** < 0.01; **** < 0.001. Robust standard errors are reported in parentheses. (Not) Declining regions are those with an average yearly GDP per capita over the pre-crisis period 2003-2007 (equal to or lower than) greater than the corresponding value defined over the subsequent period 2008-2017. Not high-income regions are those with an average yearly GDP per capita over the period 2003-2017 lower than the 90% of the sample average. Low- growth regions are those recording a long-run growth rate of GDP per capita between the years 2003 and 2017 lower than the 50% of the sample average. The variables capturing patents and inward FDI are defined per 100,000 inhabitants. All variables are defined in logarithm terms, except for the dummy variable capturing population change between times and − 1, and the categorical variable included in specification (4) that captures the most relevant business activity of inward FDI that interested a region, for which the category used as reference is no inward FDI received. 28 Table 13: Estimates for ‘catching up’ and declining regions by rural vs. non-rural regional typology. Dependent Variable ∆GDPpcrt Income Level Low-Income Not High-Income Long-Run Growth Whole Sample High-Growth Low-Growth … Decline in Average GDP Per Capita … … Declining Region Rural Region Yes No Yes No Yes No Yes No Yes No (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) log(GDPpcrt−1 ) -0.189**** -0.212**** -0.201**** -0.218**** -0.243**** -0.225**** -0.201**** -0.303**** -0.200**** -0.290**** (0.010) (0.007) (0.015) (0.017) (0.014) (0.012) (0.026) (0.024) (0.033) (0.027) Population Changert <0 Ref. Ref. Ref. Ref. Ref. Ref. Ref. Ref. Ref. Ref. ≥0 -0.003 0.002 0.018*** 0.012** 0.001 0.007** -0.003 0.005 0.000 0.018*** (0.002) (0.001) (0.007) (0.006) (0.003) (0.003) (0.004) (0.004) (0.006) (0.007) log(Employment Densityrt−1 ) 0.024 0.050**** 0.021 0.066**** 0.049** 0.048*** 0.025 0.150**** 0.021 0.174**** (0.015) (0.012) (0.027) (0.020) (0.022) (0.017) (0.030) (0.031) (0.035) (0.037) log(Share GVA Agriculturert−1 ) 0.009 -0.134 -0.211** 0.014 -0.186** -0.079 0.280* -0.309 0.150 -0.222 (0.072) (0.091) (0.090) (0.131) (0.086) (0.117) (0.154) (0.221) (0.197) (0.286) log(Share GVA Industryrt−1 ) 0.009 0.044**** 0.017 0.023 0.004 0.020* 0.009 0.022 0.014 0.054* (0.008) (0.006) (0.014) (0.016) (0.011) (0.011) (0.019) (0.017) (0.023) (0.028) log(Share GVA Constructionrt−1 ) 0.026**** 0.027**** -0.000 0.001 -0.009 0.006 0.045**** 0.044**** 0.045**** 0.044**** (0.005) (0.003) (0.008) (0.008) (0.008) (0.006) (0.008) (0.008) (0.010) (0.011) log(Share GVA Market Servicesrt−1 ) 0.029*** 0.065**** -0.011 0.015 0.006 0.025 0.081 -0.030 0.124* -0.016 (0.011) (0.009) (0.016) (0.023) (0.014) (0.017) (0.052) (0.050) (0.064) (0.071) log(Share GVA NonMarket Servicesrt−1 ) -0.015 -0.044**** 0.020 -0.006 -0.002 -0.015 -0.046 -0.032 -0.056 0.011 (0.010) (0.009) (0.015) (0.022) (0.013) (0.016) (0.034) (0.038) (0.043) (0.052) log(Patentsrt−1 ) 0.002 0.004**** 0.007** 0.001 0.000 0.001 0.003 0.005* 0.004 0.006* (0.001) (0.001) (0.003) (0.003) (0.002) (0.002) (0.002) (0.003) (0.003) (0.004) log(Inward FDIrt−1 ) -0.001 0.005**** 0.014** 0.011*** 0.015**** 0.011**** -0.014** -0.002 -0.016* -0.008 (0.003) (0.001) (0.006) (0.004) (0.004) (0.003) (0.006) (0.007) (0.009) (0.007) Region FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes No. Observations 5,824 12,670 1,400 1,568 2,380 3,220 1,134 1,260 672 756 No. Regions 416 905 100 112 170 230 81 90 48 54 R2 0.34 0.36 0.51 0.47 0.47 0.42 0.40 0.56 0.42 0.53 Adjusted R2 0.28 0.31 0.47 0.42 0.42 0.37 0.34 0.40 0.35 0.47 Model F Statistics [p-value] 73.01 [0.000] 195.13 [0.000] 44.31 [0.000] 42.89 [0.000] 50.79 [0.000] 55.88 [0.000] 27.76 [0.000] 39.43 [0.000] 20.78 [0.000] 31.21 [0.000] Notes: * < 0.1; ** < 0.05; *** < 0.01; **** < 0.001. Robust standard errors are reported in parentheses. Low-income regions are those with an average yearly GDP per capita over the period 2003-2017 lower than the 50% of the sample average. Not high-income regions are those with an average yearly GDP per capita over the period 2003-2017 lower than the 90% of the sample average. High- (low-)growth regions are those recording a long-run growth rate of GDP per capita between the years 2003 and 2017 equal to or greater than (lower than) the 50% of the sample average. Declining regions are those with an average yearly GDP per capita over the pre- crisis period 2003-2007 greater than the corresponding value defined over the subsequent period 2008-2017. The variables capturing patents and inward FDI are defined per 100,000 inhabitants. All variables are defined in logarithm terms, except for the dummy variable capturing population change between times and − 1. 29 5. DISCUSSION AND CONCLUSIONS Cohesion Policy has channeled very large investments from EU and national budgets to regional economic development. Unfortunately, regional economic inequalities have sharply risen during the 2003-2017 period, especially within countries. This paper contributes to the debate on regional economic growth in the EU by analyzing the growth performance of NUTS-3 regions, with the aim of identifying growth pathways that have characterized different types of regions. We have placed a special focus on lagging regions, namely ‘low-income’ regions and ‘low-growth’ lagging regions, by combining the NUTS-2 taxonomy adopted under Cohesion Policy with the taxonomy proposed by the European Commission under the ‘Catching Up Initiative’. Our focus on NUTS-3 regions is precisely because of the heterogeneity of performance among NUTS-3 regions. In the introductory sections of this paper, we suggested that the strong heterogeneity in economic performance within NUTS-2 regions will handicap the ability of Cohesion Policy to adequately target and address the challenges of regional development at that level. NUTS- 2 regions include large and rich NUTS-3 regions (besides capital cities) and poor rural ones located within the same NUTS-2 boundaries. Inequality among NUTS-3 regions has been increasing over time. Therefore, the paper focuses on analyzing the economic performance of NUTS-3 regions in order to provide insights for policy makers and local officials. Due to data constraints, our analysis considers only a subset of growth determinants – namely, industrial structure, innovation, and inward FDI. Overall, our empirical results suggest that different types of regions are characterized by different economic growth pathways, such that a ‘one-size-fits-all’ approach to policy design could not be effective. Within those constraints, we have attempted to provide sub-national policy makers with novel evidence useful for the definition of development interventions that are coherent with their specific type of region. Referring back to our three research questions, the following conclusions can be made: (i) Where are NUTS-3 level lagging regions located, and how similar are these locations to NUTS-2 lagging regions? Transposing the definitions of lagging regions from the European Commission’s ‘Catching Up Regions’ initiative, we see that 16% of NUTS-3 regions are low-income, and 13% of NUTS-3 regions are low-growth. These lagging regions are found in 19 of 28 EU Member States. Low-income lagging regions are found mainly in post-2004 enlargement countries; low-growth lagging regions are found in a variety of EU Member States, including the richest ones. Often the lagging regions are hidden within large NUTS-2 regions, which implies that Cohesion Policy initiatives should have an increased focus on NUTS-3 level programming. (ii) Have NUTS-3 lagging regions been growing and converging with non-lagging regions? Low-income regions have grown relatively fast: an average of 6% per year between 2003 and 2017, compared to 2% per year for other income categories. As such, low-income regions have been converging with other regions. However, low-growth regions require more attention: there has been an absolute and sustained decline in income among low-growth regions. Even within the low-growth category, there is less dynamism among regions: they rarely move up and down the rankings within their group. (iii) What are the correlates of growth among NUTS-3 regions, and do these correlates differ for lagging regions? Our analysis shows that correlates of growth differ among NUTS-3 regions, by income level and by long-run growth category. Low-income regions have no uniform pattern in the role of economic sectors in contributing to growth, but do show a strong relationship between FDI and growth – especially production, logistics, distribution and transportation activities. Low-growth lagging regions show a correlation between growth and construction, patenting, and puzzlingly show a negative relationship with FDI and growth. This is explained by disaggregating low-growth regions into income categories, since FDI loses its importance to growth among transition and less developed 30 regions. Among rural low-income regions, growth is associated with a move away from agriculture; and among non-rural low-growth regions, growth is associated with construction and innovation, as well as agglomeration. Further work in this area could focus on ‘how’ lagging regions have turned around their economies. Research could focus especially on those that had experienced slow or negative growth in one period and positive growth in another period. Studying this special category of successful ‘low- growth’ lagging regions would provide the most value for policy makers and local officials. Low- income regions have already tended to grow fast, while low-growth lagging regions are the main barrier to convergence in the EU’s regions. On the basis of evidence in this paper that growth pathways differ substantially between different income groups, growth profiles, and rural/non-rural regions, such studies should be made only of similar regions rather than lagging regions ‘in general’. What policy actions could be taken on the basis of these results? • First, the results can help local leaders to take actions for growth in different types of regions. The results clearly show that growth pathways differ for different types of regions – by income group, growth profile, and rural vs. non-rural regions. Regions designing their growth strategies must look at their own profile, their endowments, and their realistic opportunities. The analytic results for each type of region show the factors that are typically correlated with growth in that type of region, and can thus help local leaders narrow down their realistic options. Local governments can influence several of the variables in this analysis, through targeting industries for support, through fostering innovation, and through promoting and facilitating inward FDI. • Second, the results should alert regional administrations (at the NUTS-2 level) and national governments to the heterogeneity within NUTS-2 regions, and the need for deliberate actions to link leading NUTS-3 areas with lagging NUTS-3 areas. The analytic results in this paper show that rich areas within NUTS-2 regions will not necessarily ‘pull’ up growth in nearby poorer areas: the rich and poor areas can diverge in growth over time. Regional economic strategies will need to be designed differently if better outcomes for lagging NUTS-3 regions are desired. Richer NUTS-3 areas can provide opportunities for poorer NUTS-3 areas, but this will require an explicit spatial strategy to link the areas, since it does not appear to happen naturally. These links could be in terms of, for example, transport infrastructure and services to improve physical mobility between neighboring areas; labor market information to improve the efficiency of job searches; firm-level matchmaking to increase access to neighboring markets and participate in supply chains; or in other areas. In each region, a diagnostic could be made of barriers to economic integration of neighboring NUTS-3 areas, and actions could be designed to address these. • Third, the results imply that policy makers at the European and national levels should address inequalities at the NUTS-3 level as a source of discontent. Policy makers will usually be torn between the two objectives of efficiency and equality. Efficiency (i.e. the shortest route to growth at a national level) may imply that resources should be focused on locations that are already leading, since those locations have already demonstrated their potential for innovation, competitiveness, and investment. Equality will, by contrast, require a redistribution of resources to poorer and lagging regions, to try to resuscitate the economies in those places. Recent evidence has demonstrated a robust link between regional inequalities and discontent in the EU (e.g. Dijkstra et al., 2018; Rodríguez-Pose, 2020), and thus implies that the balance of policy actions needs to be oriented further towards equality than is currently the case. The evidence presented in this paper indicates that inequalities are increasing at the NUTS-3 level, thus the situation is likely to get worse. Cohesion in the EU will require renewed and concerted action at the NUTS-3 level. 31 REFERENCES Arbia G., Battisti M. and Di Vaio G. (2010) Institutions and geography: Empirical test of spatial growth models for European regions. Economic Modelling 27(1): 12–21. Bachtrögler J., Fratesi U. and Perucca G. (2020) The influence of the local context on the implementation and impact of EU Cohesion Policy. Regional Studies 54(1): 21–34. Basile R. (2008) Regional economic growth in Europe: A semiparametric spatial dependence approach. Papers in Regional Science 87(4): 527–544. Basile R. (2009) Productivity polarization across regions in Europe: The role of nonlinearities and spatial dependence. International Regional Science Review 31(1): 92–115. Becker S. O., Egger P. H. and Von Ehrlich M. (2012) Too much of a good thing? On the growth effects of the EU’s regional policy. European Economic Review 56(4): 648–668. Beugelsdijk M. and Eijffinger S. C. W. (2005) The effectiveness of structural policy in the European Union: An empirical analysis for the EU-15 in 1995–2001. Journal of Common Market Studies 43(1): 37–51. Bussoletti S. and Esposti R. (2008) Impact of objective 1 funds on regional growth convergence in the European Union: A panel-data approach. Regional Studies 42(2): 159–173. Butkus M., Cibulskiene D., Maciulyte-Sniukiene A. and Matuzeviciute K. (2018) What is the evolution of convergence in the EU? Decomposing EU disparities up to NUTS 3 level. Sustainability 10(5): 1–37. Canova F. (2004) Testing for convergence clubs in income per capita: A predictive density approach. International Economic Review 45(1): 49–77. Cappelen A., Castellaci F., Fagerberg J. and Verspagen B. (2003) The impact of EU regional support on growth and convergence in the European Union. Journal of Common Market Studies 41(4): 621–644. Castellani D. and Pieri F. (2016) Outward investments and productivity: Evidence from European regions. Regional Studies 50(12): 1945–1964. Crescenzi R. (2005) Innovation and regional growth in the enlarged Europe: The role of local innovative capabilities, peripherality, and education. Growth and Change 36(4): 471–507. Crescenzi R. and Giua M (2020) One or many Cohesion Policies of the European Union? On the differential economic impacts of Cohesion Policy across member states. Regional Studies 54(1): 10–20. Crescenzi R. and Rodríguez-Pose A. (2012) Infrastructure and regional growth in the European Union. Papers in Regional Science 91(3): 487–513. Crescenzi R., Luca D. and Milio S. (2016) The geography of the economic crisis in Europe: National macroeconomic conditions, regional structural factors and short-term economic performance. Cambridge Journal of Regions, Economy and Society 9(1): 13−32. Crescenzi R., Pietrobelli C. and Rabellotti R. (2014) Innovation drivers, value chains and the geography of multinational corporations in Europe. Journal of Economic Geography 14(6): 1053−1086. Dall’erba S. and Le Gallo J. (2008) Regional convergence and the impact of European structural funds over 1989–1999: A spatial econometric analysis. Paper in Regional Science 87(2): 219–244. Davies S. (2017) Does EU Cohesion Policy work? Meta-review of research in Cohesion Policy effectiveness. European Policy Research Paper No. 99, European Policies Research Centre, University of Strathclyde, Glasgow. Di Cataldo M. (2017) The impact of EU Objective 1 funds on regional development: Evidence from the U.K. and the prospect of Brexit. Journal of Regional Science 57(5): 814–839. Dijkstra L., Poelman H. and A. Rodríguez-Pose (2018) The Geography of EU Discontent, Working Paper 12/2018, European Commission DG REGIO. Dogaru T., Burger M., Karreman B. and van Oort F. (2015) Functional and sectoral division of labour within Central and Eastern European countries: Evidence from greenfield FDI. Tijdschrift voor Economische en Sociale Geografie 106(1): 120–129. European Commission (2015) Ex post evaluation of Cohesion Policy programmes 2007-2013, focusing on the European Regional Development Fund (ERDF) and Cohesion Fund (CF) – Work Package Zero: Data collection and quality assessment. Directorate-General for Regional and Urban Policy, Brussels. European Commission (2017) Competitiveness in low-income and low-growth regions: The lagging regions report. Staff Working Document No. 132, European Commission, Brussels. Farole T., Goga S., Ionescu-Heroiu M. (2018) Rethinking lagging regions: Using Cohesion Policy to deliver on the potential of Europe’s regions. World Bank Report on the European Union, World Bank Group, Washington DC. Fingleton B. (1997) Specification and testing of Markov chain models: An application to convergence in the European Union. Oxford Bulletin of Economics and Statistics 59(3): 385–403. Frick S. A. and Rodríguez-Pose A. (2016) Average city size and economic growth. Cambridge Journal of Regions, Economy and Society 9(2): 301–318. Frick S. A. and Rodríguez‐Pose A. (2018) Big or small cities? On city size and economic growth. Growth and Change 49(1): 4–32. Gardiner B., Martin R. and Tyler P. (2004) Competitiveness, productivity and economic growth across the European regions. Regional Studies 38(9): 1045–1067. 32 Geppert K., and Stephan A. (2008) Regional disparities in the European Union: Convergence and agglomeration. Papers in Regional Science 87(2): 193–217. Iammarino S., Rodríguez-Pose A. and Storper M. (2019) Regional inequality in Europe: Evidence, theory and policy implications. Journal of Economic Geography 19(2): 273–298. Magrini S. (1999) The evolution of income dispersion among the regions of the European Union. Regional Science and Urban Economics 29(2): 257–281. Menghinello S., De Propris L. and Driffield N. (2010) Industrial districts, inward foreign investment and regional development. Journal of Economic Geography 10(4): 539–558. OECD (2012) Promoting Growth in All Regions, OECD Publishing, Paris, ISBN: 9789264174634. Panzera D. and Postiglione P. (2014) Economic growth in Italian NUTS 3 provinces. The Annals of Regional Science 53: 273–293. Pellegrini G., Terribile F., Tarola O., Muccigrosso T. and Busillo F. (2013) Measuring the effects of European regional policy on economic growth: A regression discontinuity approach. Papers in Regional Science 92(1): 217–233. Percoco M. (2017) Impact of European Cohesion Policy on regional growth: Does local economic structure matter? Regional Studies 51(6): 833–843. Petrakos G, Kallioras D. and Anagnostou A. (2011) Regional convergence and growth in Europe: Understanding patterns and determinants. European Urban and Regional Studies 18(4): 375–391. Quah D. (1996) Regional convergence clusters across Europe. European Economic Review 40(3–5): 951–958. Ramajo J., Màrquez M. A., Hewings G. J. D. and Salinas M. M. (2008) Spatial heterogeneity and interregional spillovers in the European Union: Do cohesion policies encourage convergence across regions? European Economic Review 52(39): 551–567. Rodríguez-Pose A. (1999) Convergence or divergence? Types of regional responses to socio-economic change in Europe. Tijdschrift voor Economische en Sociale Geografie 90(4): 363–378. Rodríguez-Pose, A. (2020) The Rise of Populism and the Revenge of the Places That Don’t Matter. LSE Public Policy Review, 1(1), p.4. Rodríguez-Pose A. and Garcilazo E. (2015) Quality of government and the returns of investment: Examining the impact of cohesion expenditure in European regions. Regional Studies 49(8): 1274–1290. Rodríguez-Pose A. and Ketterer T. (2020) Institutional change and the development of lagging regions in Europe. Regional Studies 54(7): 974–986. Storper M. (2018) Separate Worlds? Explaining the current wave of regional economic polarization. Journal of Economic Geography 18(2): 247–270. Tselios V. (2009) Growth and convergence in income per capita and income inequality in the regions of the EU. Spatial Economic Analysis 4(3): 347–370. van der Zwet A., Miller S. and Gross F. (2014) A first stock take: Integrated territorial approaches in Cohesion Policy 2014-20. IQ-Net Thematic Paper No. 35(2), European Policies research Centre, University of Strathclyde, Glasgow. World Bank (2015) Competitive Cities for Jobs and Growth: What, Who, and How, Washington DC: World Bank. 33