The role of Cohesion Policy for sustaining the resilience of European regional labour markets during different crises

ABSTRACT Whilst the analysis of the impact of Cohesion Policy funds is well documented, little is known about the short-term consequences of Cohesion Policy during economic crises and its implications for regional resilience. To study these issues, we assemble novel, panel time-series data on Cohesion Policy covering almost four programming periods and different shocks in the European Union. Using heterogeneous coefficients panel models, we find a general positive impact of Cohesion Policy on regional resilience, although with region- and crisis-specific patterns during different shocks. Our results also suggest the presence of regional variation regarding regional labour market resilience over the past three decades.


INTRODUCTION
The decades-old economic resilience approach, which received new attention during the years of the Great Recession, has been useful to study the uneven consequences of shocks such as economic crises and natural disasters across economies and actors located in different places.Since the contribution of Martin (2012), where the four main dimensions of economic resilience have been disentangled (i.e., resistance, recovery, reorientation and renewal), several empirical works have analysed the role of crises on the evolutionary dynamic process of robustness and adaptability of regional and local economies, and labour markets (Modica et al., 2019). 1 Many studies have been conducted for a single and/or group of countries such as the European Union (EU) (for a recent survey, see Pontarollo & Serpieri, 2020a).Most of the contributions focusing on the EU regions, however, have limited the attention to the years of the Great Recession, without comparing the resilience of European regional economies during different, in terms of timing, origins, and effects, crises.
Understanding the root causes of regional vulnerability during different shocks is important for strengthening economic resilience and constructing appropriate early warning resilience indicators (Caldera-Sánchez et al., 2017).Recently, Sensier and Uyarra (2021) pointed out that the resistance of a given area to a particular shock can be the consequence of complex policy actions that took place during and after previous crises.Over three decades, Cohesion Policy has been the main financial instrument available for sustaining regional economies in the EU.Although such long-lasting policy has encountered institutional (e.g., EU enlargement; Brexit, etc.) and economic (i.e., the Euro, migration flows, etc.) shocks, few papers have analysed the countercyclical role of the EU funds to date (Di Pietro et al., 2020).Therefore, it is timely to produce policy-relevant empirical research that provides new knowledge on how European regional economies and labour markets react to different shocks, and whether the role of EU Cohesion Policy can help to explain the presence of asymmetric vulnerability patterns over different crisis periods (Clark & Bailey, 2018).
This paper has two main research goals.First, to study the heterogeneous impact of the major economic crises occurred in Europe over the past three decades in terms of regional labour market resistance.To achieve this objective, we assemble a novel panel time-series dataset and we apply statistical indicators that allow us to map the resistance of the EU regional labour markets to different crises.In the years 1992-94, most of the EU regions in the new and old member states (MS) experienced relevant employment losses.In the old MS, the German unification shock and fall of the European Exchange Rate Mechanism (ERM) that mounted with the Italian lira and the UK pound currency crises of September 1992 produced significant negative economic effects in many regions (Buiter et al., 1997).In central and eastern MS, the institutional transformations following the fall of the Berlin Wall generated a sudden shock for regional labour markets (Sokol, 2001).From 2008 to 2013, the EU experienced a combination of two connected downturns: the Global Financial Crisis started in 2008 and the Euro sovereign debt crisis in , with asymmetric consequences on countries and regions (Lane, 2012).In this work, we explicitly investigate where and how EU-wide economic crises produced consequences on regional employment.
Second, we assess the ability, and degree of asymmetry, of EU Cohesion Policy to sustain the resistance of the European regional labour markets during different economic crises.Although the countercyclical role of EU funds is consistent with the idea that Cohesion Policy, which finances public and private investments and expenditures, can produce short-term demand effects both directly and indirectly through Keynesian multiplier effects, there is limited evidence to date.Canova and Pappa (2021) find that the EU Structural Funds can have average positive short-term impact on all regional macroeconomic variables in European regions, making them potentially useful for countercyclical purposes.We contribute to this research question as follows.We use novel panel time-series data on EU funds (European Commission, 2017) that allow us to disentangle the consequences of Cohesion Policy during the crises registered over the past three decades.Moreover, we apply a panel time-series empirical framework (Pesaran & Smith, 1995), which is useful to analyse the heterogeneous role of Cohesion Policy on regional employment in the EU during different shocks.Finally, we explore possible spatial patterns across regions and over different crises.
Our findings can be summarized as follows.We find that the shocks of the early 1990s and the early 2000s (for new MS only) and the Global Financial Crisis had major effects on regional employment vulnerability in the EU by producing relevant regional employment losses.We also detect regional heterogeneity during the different crises, with some regions more affected than others from a given shock.During the Global Financial Crisis, for instance, more than 50% of EU regions experienced significant employment losses with impact on the regional labour market performance.Our evidence can be read as complementary to the results obtained by other researchers who look at the effects of crises on EU regions (Canova & Pappa, 2021;Sensier et al., 2016).As for Cohesion Policy, our results suggest a positive short-term impact of Cohesion Policy funds on sustaining regional labour market resilience mainly during the Global Financial Crisis, though with regional differences.The role of Cohesion Policy differs across new and old MS and regions, with major impact registered in more assisted regions.
The remainder of the paper is structured as follows.Section 2 presents the background literature and our research aims.Section 3 describes the data, the identification of shocks and the construction of resilience indicators.Section 4 contains the econometric analysis and a discussion of the results.Section 5 concludes with general messages and policy implications.Appendix A in the supplemental data online supports the robustness of the analysis by reporting additional results.

BACKGROUND AND RESEARCH AIMS
2.1.Related literature Regional resilience works focusing on the EU have progressively shed light on the different characteristics of this multifaceted analytical concept (Alessi et al., 2020), and the need of operationalizing such a complex framework (Sensier et al., 2016).The majority of the existing studies look at the resistance and/or shock-absorption dimension of resilience, defined as the relative economic performance of a given area with respect to a benchmark (i.e., European and/or national context, different periods, etc.) during a particular crisis (Fingleton et al., 2015); some recent papers also show interest in the renewal capacity of regions (Pontarollo & Serpieri, 2020b;2021).The measurement of resilience has been mostly operated using economic variables such as regional gross domestic product (GDP) and gross value added (GVA) with limited attention paid to regional labour markets (Giannakis & Bruggeman, 2020).Most of the studies have been conducted at the NUTS-2 level of observation (Brakman et al., 2015), but some analyses have used information at a lower level of aggregation (Doran & Fingleton, 2016;Fratesi & Perucca, 2018).In this work, we add to these studies along different dimensions.Our analysis covers three decades and different economic shocks, rather than looking at the years of the Great Recession only.We compare regional labour markets in the old and new MS by providing new evidence on spatial labour market resilience across Europe.This is a relevant aspect due to differences in employment vulnerability registered in the EU over the crises of 1990s and the Great Recession (Van Rie & Marx, 2012).Moreover, we explicitly look at the relation between Cohesion Policy and regional labour market resilience in the EU over different programming periods by adding to the few works that analysed the link between resilience and EU funds during the Great Recession (Healy & Bristow, 2020). 2  We also contribute to the few studies investigating the role of Cohesion Policy during the crises. 3Although the main objective of Cohesion Policy is to promote longrun growth and regional convergence, it can also contribute to support regional economies in the short-run through different demand-side channels (Psycharis et al., 2020).Bachtrögler (2016) study the effects of EU funds on regional growth during the period 2007-13 by finding a reduction of the impact of Cohesion Policy during the Great Recession.Crescenzi and Giua (2020) show that Cohesion Policy funds had asymmetric national and The role of Cohesion Policy for sustaining the resilience of European regional labour markets during different crises 2427 regional effects on GDP and employment during the period 2007-13;Di Pietro et al. (2020) document that, in the short-run, EU funds can have a role in sustaining the resilience of EU regions, as measured in terms of regional GDP, though with spatial differences.In this paper, we make two main advancements with respect to the existing literature.We analyse the consequences of Cohesion Policy funds on resilience over three decades that encompass different economic crises.We apply a novel panel time-series approach that allows us to disentangle the heterogeneous relation between EU funds and regional employment resilience.In this direction, we add to the growing literature that looks at the region-specific effects of EU funds (Becker et al., 2018;Bourdin, 2019;Cerqua & Pellegrini, 2020;Di Caro & Fratesi, 2022;Le Gallo et al., 2011), where there is no particular focus on regional resilience during shocks, which is one of the main objectives of our work.

Research objectives
The first research goal of this work is to analyse the possible heterogeneous consequences of EU-wide economic crises on regional employment resilience over the past three decades (research question 1).The presence of spatial differences regarding the impact of shocks on regional labour markets can be motivated by different reasons.In EU regions, economic (i.e., specialization patterns, human capital levels, etc.) and institutional (i.e., quality of government, labour market rules, etc.) characteristics are unevenly distributed with implications on the place-specific ability of reacting to particular economic shocks (Annoni et al., 2019).This asymmetry is further justified by noting that the vulnerability of a given area in the short-term can be related to its long-run economic performance (Webber et al., 2018).Different shocks, moreover, such as financial crises, currency shocks and institutional transformations can have asymmetric regional effects depending on differences in the channels of transmission of the shock (Alessi et al., 2020).
Our second research objective is to investigate the short-term consequences of EU Cohesion Policy on regional employment resilience over different crises by looking for possible heterogeneous effects across regions (research question 2).This question is motivated by the fact that although the EU funds are long-term oriented, they are used for public investments and expenditures that can support countercyclical demand effects directly and indirectly by means of Keynesian multipliers (Neumark & Simpson, 2014).
The EU funds can affect regional resistance through various channels.First, they can sustain, among others, the following demand-side effects: increase in government consumption and public investment; support of private investment also through the reduction of risk premium; support of private consumption also through direct transfers to households (Monfort et al., 2021).Therefore, we can expect a positive effect of Cohesion Policy funds on regional economic performance during downturns.
Moreover, Cohesion Policy funds can help the resistance of regional labour markets during shocks through the provision of additional financial resources directly funded at the EU level, particularly when the EU cofinancing rates are increased (Camagni & Capello, 2017).This effectively happened during the Great Recession and the coronavirus crisis, when the European Commission reformed Cohesion Policy to support the resistance of MS and regions (Crescenzi et al., 2021).
Finally, the positive impacts of Cohesion Policy on regional labour markets during shocks, such as the smoothing of the consequences of aggregate shocks on earnings and employment (Hijzen et al., 2018), are also expected when the Structural Funds are recalibrated to provide funding for job-and income-insurance supporting schemes.The expected positive effects of Cohesion Policy during shocks can vary across areas because of the presence of heterogeneity in the distribution of different conditional factors, among which the quality of regional institutions plays a crucial role (Charron et al., 2014). 4

Data description
We assemble a novel dataset for 255 European regions (26 MS plus the UK) for the period 1980-2015.We exclude Croatia for which data are not available for a sufficient number of years; 5 we also exclude Romania and Bulgaria that joined the EU in 2007 when they started effectively benefiting from Cohesion Policy funds (Incaltarau et al., 2020). 6To measure labour market resilience, we use regional data on total employment that allow us to calculate the annual change of employment in a given region.Our preference for labour market resilience can be motivated as follows.Using employment data as a measure of resilience is useful to identify the impact of a recession on labour markets, which can be larger than declines in output (Fingleton et al., 2012).The focus on employment goes beyond the idea that the resilience of a region is simply a matter of production, but it also has social implications (Fratesi & Rodríguez-Pose, 2016;Sensier et al., 2016).Employment data, moreover, do not need to be deflated by ruling out the difficulties of adopting different regional price indices (Cecchetti et al., 2002).For a discussion on the pros and cons of different resilience variables, see Cellini et al. (2017).
To measure EU Cohesion Policy, we use historic data provided by the European Commission on 'regionalized' NUTS-2 (NUTS-2013 version) annual EU expenditures for the following funds: European Regional Development Fund (ERDF), Cohesion Fund, European Agricultural Fund for Rural Development (EAFRD)/European Agricultural Guidance and Guarantee Fund (EAGGF), and European Social Fund (ESF).We use 'modelled' annual data per capita that represent the actual expenditures per year by reducing the presence of outliers since they do not contain gaps in the series (Fidrmuc et al., 2020).We have observations that cover almost four programming periods: data are available from around 1990 (2000) onwards for the old (new) MS.Our focus on NUTS-2 regions is motivated by data availability at this level of aggregation and, more compellingly, since NUTS-2 regions are the relevant targeted areas of EU Cohesion Policy in terms of eligibility to the different objectives. 7 Figure 1 reports total employment in levels (a) and annual variation (b) in the EU regions: data for the EU-15 (EU-10) regions are available from 1980 (1990). 8  From 1990 onwards, the European regions registered an average employment growth of about 0.6% (EU-15) and −0.2% (EU-10), with a standard deviation (SD) of about 0.6% (EU-15) and 1% (EU-10).This suggests the presence of different growth performances both across and within old and new MS, as discussed, for instance, by Cuaresma et al. (2014).
In the old MS, two main shocks occurred over the past three decades.The first can be dated to 1992-94, when regional labour markets experienced negative growth rates following the German unification shock and the fall of the ERM (Jordà et al., 2012).The second recession covers the Great Recession and encompasses two connected shocks, namely the financial crisis (2008-09) and the European sovereign debt crisis  that mostly affected specific countries: Greece, Italy, Portugal and Spain.From 2009 to 2013, total employment fell by more than 3% across the EU-15 regions, while during the early 1990s, employment fell by about 2%.In the new MS, employment losses were −3% and −4% during the early 1990s shock and the Great Recession, respectively.Note that we prefer to keep separate the two shocks part of the Great Recession for comparative purposes with the rest of the EU (Lane, 2012).From 1999 to 2002, new MS registered an additional recessionary event, with a drop of employment of about 6%, which was primarily due to the transition of these economies towards a market economy (Shleifer & Treisman, 2014).
To save space, Appendix A in the supplemental data online provides the main statistics on the variables used in the paper.Our dataset also contains regional data on population, GDP and GVA for the main sectors from 1980 ( 1990) to 2016 for old (new) MS, which are useful to have some knowledge on other regional economic conditions.

Identification of shocks
One of the key challenges in resilience studies is how to identify when a particular adverse event has actually occurred, that is, the timing of the shock under observation, against which the impact can be measured from a comparative regional analysis perspective (Martin & Sunley, 2014).In the existing literature, two main shock-identifying approaches have been used, which follow the external and internal business cycle dating procedures discussed by Harding and Pagan (2003) (for a recent overview, see Alessi et al., 2020).According to the first approach, which is adopted by most of the resilience works focusing on the EU regions, information from raw data observation and the knowledge of the official timing of crises published by institutions such as central banks are used to construct time dummies that cover the period of a given shock.This approach is useful when analysing the heterogeneous consequences of systemic-wide shocks in cross-comparative settings (Martin, 2012).It also reduces a researcher's interferences when data are available at an annual frequency (Cainelli et al., 2019).The second approach treats each observational units as an individual entity and aims at detecting the different timing (i.e., entry and exit) and magnitude of regionspecific shocks by adopting business cycle peak-through identifying techniques (Sensier et al., 2016).The main merit of this approach is to investigate the unique evolutionary trajectory of a region facing cycles and shocks (Pontarollo & Serpieri, 2020a).
Since we are interested in the assessment of the impact of systemic, EU-wide, crises on regional labour market resilience, and we have data at annual frequency, we follow the first approach that allows us to define the shocks under observation by using information on the dating of the main EU crises provided by official institutions such as the European Central Bank (ECB) (Lo Duca et al., 2017).We complement this information with data-driven knowledge obtained from our dataset.Specifically, for the EU-15 regions, we define the following time dummies for shocks: the early 1990s crisis (1992-94); the Global Financial Crisis (2008-09); and the Euro sovereign debt crisis .For the EU-10 regions, we have the following common shocks: the early 2000s crisis (1999)(2000)(2001)(2002); the Global Financial Crisis (2008-09); and the Euro sovereign debt crisis .9 Notably, our dating of the shocks is consistent with the timing of the severe recessions that occurred in Europe, as identified by Hermansen and Röhn (2017), and with the Organisation for Economic Co-operation and Development's (OECD) dating of crises (Caldera-Sánchez et al., 2017).
Table 1 reports information on the main economic crises that occurred in the EU from 1990 to 2015.For each shock, we have calculated the regional cumulative employment variations registered during the years of the crisis for all 255 regions in the sample.More than 70% of EU regions registered employment losses during the early 1990s and early 2000s (only for new MS) crises, and the Global Financial Crisis. 10 This means that about 180 out of 255 regions experienced negative employment growth over the different crises.We have also calculated the share of regions that registered cumulative employment losses higher than 2% (in negative terms), defined in Table 1 as regions in severe recession, by finding that relevant employment losses occurred in the majority of the EU regions during the different crises.Interestingly, our findings are in line with those obtained by applying different procedures of shock-identification in the EU (Pontarollo & Serpieri, 2020a;Sensier et al., 2016).

Resilience mapping
Figure 2 maps employment resistance in the EU regional labour markets to the different shocks registered over the past three decades.We measure resistance as regional The role of Cohesion Policy for sustaining the resilience of European regional labour markets during different crises 2429

REGIONAL STUDIES
growth deviation in the period of crisis with respect to a given benchmark (Fratesi & Rodríguez-Pose, 2016), which has the advantage of being readable in terms of percentage points of growth. 11For each crisis, we calculate two resilience indicators depending on the benchmark (i.e., EU or national) against which comparing the regional growth deviation.We construct the following indicators: where E t i is the total employment in region i at time t; E t EU is the total employment at the EU level over the same period; and E t C is the total employment at the country level.The time intervals t and t -1 refer to the last and first year of a given crisis, respectively.The indicator in (1) is reported on the left-hand side, while the indicator in ( 2) is reported on the right-hand side of Figure 2. A value higher (lower) than zero represents a region where the resistance to the crisis, as measured by the percentage employment growth, has been higher (lower) than the EU or the country average, respectively.
The maps for the different crises never fully overlap by suggesting the presence of differences in the spatial deployment of each crisis, in line with the results of Sensier et al. (2016).From the resilience mapping at the EU level (left-hand side graphs), moreover, we can observe a certain degree of similarity among the different crises: some regions have suffered more in all downturns and others that show different reactions during the different shocks.These preliminary findings confirm the ideas that some areas are structurally more/less resistant to crises than others, and many regions experience different outcomes depending on the specific crisis.
We observe some national effects when looking at all the shocks here considered.Regions in Spain and Portugal have been on average less resistant during economic crises; conversely, regions in Western Germany have been more resistant during all shocks.National effects, however, hide regional consequences as emerges from the observation of the maps obtained with the national benchmark (right-hand side graphs).Indeed, regional specificities inside countries are overall less or more resistant to the different shocks.We detect 45 regions, about 17% of our sample, which have been hit less than the respective country in all the three crises.We also find that 60 regions, about 22% of our sample, have been systematically hit more than their country in all the three crises.Noteworthy, the relative resistance of a given region with respect to the benchmark country depends on the crisis: this is true for 164 regions, almost 61% of our sample.In detail, it is possible to list 26 regions that have been systematically more resistant in every crisis: they show a resilience indicator higher than 0.5% with respect to the rest of the sample. 12 Most of these regions are advanced regions and/or regions with large urban areas (Capello et al., 2015).Conversely, we can observe 29 regions that are systematically less resistant to crises than the EU regions as a whole. 13 The maps also provide information on the regional distribution of EU Structural Funds.In particular, in all the graphs in Figure 2, shaded (unshaded) areas denote the regions where the amount of Structural Funds per capita has been above (below) the average of their respective group, namely EU-15 and EU-10, over the observation period.This is useful to provide a preliminary association between regional resistance and territorial distribution of EU funds.This indicator has been preferred since the eligibility criteria for Objective 1 regions, lagging regions and less developed regions have changed over time.Moreover, we have selected the average amount of EU funds to consider the presence of overlapping among programming periods, given the possibility that funds can be spent at the end of programming periods (i.e., N + 2 and N + 3 rules).
From the left-hand side graphs, it can be seen that, with the exception of the Greek regions in the early 1990s crisis, Southern and Mediterranean EU regions have been both more assisted and less resilient than the average of the EU regions.In the rest of the EU-15, we find mixed evidence: apart from the regions in East Germany, we do not find specific relations between Structural Funds and regional labour market resistance.In the EU-10 regions, the evidence is also mixed, given that we observe both high and low degrees of relative resistance in these highly funded areas.
These patterns are mostly confirmed when looking at the resilience mapping with respect to the national benchmark (right-hand side graphs in Figure 2) Indeed, Southern and Mediterranean regions have been more assisted and less resistant than the rest of their countries, particularly in the early 1990s crisis and during the Sovereign debt crisis.This is true also for most of the regions in Eastern Germany, where a large amount of Structural Funds is related to low labour market resistance.In the EU-10 regions, we observe different patterns, with the regions with capital cities, which are richer and getting lower EU funds, are often more resilient than others, but in the case of the financial crisis.

Methodology
We empirically address our research questions by starting from the following panel time-series model specification (Pesaran & Smith, 1995): where the dependent variable is the (log of) regional employment in region i (i ¼ 1, … , I, with I ¼ 255) at time t (t ¼ 1990, … , 2015); for the EU-15 (EU-10) regions, we use data from 1990 (1998) to 2015, the last available observation.The explanatory variable EUfund it−j describes the (log of) cohesion fund payments per capita on a regional level; we add 1 before taking logs since The role of Cohesion Policy for sustaining the resilience of European regional labour markets during different crises 2431 Note: Maps report the regional differential growth (%) in the three periods of crisis: early 1990s/(early 2000s for new member states) (first from above row); Global Financial Crisis (second row); and Euro sovereign debt crisis (third row).Differential growth is calculated with respect to the European Union (EU) average (left-hand side) and country average (right-hand side).Indicators are calculated as in equations ( 1) and ( 2).Darkness denotes a region where the resistance to a given crisis is higher than the EU (left-hand side) or the country (right-hand side) benchmark.The maps also show the regions with average amount of Structural Funds per capita above (below) the group average, as described by shaded (unshaded) regional areas.The reference group is again the EU (left-hand side) or the country (right-hand side).
2432 Paolo Di Caro and Ugo Fratesi REGIONAL STUDIES some regions receive no Cohesion Policy funds in some years.We introduce the Cohesion Policy variable with a lag of one year because it is likely that projects financed by Cohesion Policy become effective for regional economies after some time lag (Mohl & Hagen, 2010). 14The vector x it−j contains additional variables (e.g., time dummies for shocks, interaction variables, etc.) and controls; we always include the (log of) regional population as standard control in dynamic regional growth models (Chodorow-Reich et al., 2012).Since we are interested in investigating the short-term consequences of shocks and Cohesion Policy expenditure on regional employment growth, it is convenient to rewrite relation (3) in the following reduced form (Pesaran et al., 1999), where, to simplify the notation, we have only reported the variables of interest: 15 where the impact of the EU-wide shocks in our sample, as identified above, is described by the dummy variable Crisis with z = 3.We also define three interaction variables EUfund it−1 * Crisis z capturing the consequences of Cohesion Policy during each specific recessionary event in our sample.Therefore, the main coefficients of interest are g iz , and d iz .
The term a i denotes regional fixed-effects and allows for the consideration of the autonomous growth rate component (Hsiao, 2014). 16The error term 1 it is adjusted for taking into account outlier-robust mean of parameter coefficients across groups and intragroup correlation (Ditzen, 2018).Relation ( 4) is estimated separately for each region and then the resulting coefficients are averaged across regions by applying the mean group (MG) estimator (Pesaran & Smith, 1995).The consideration of coefficient heterogeneity, namely the coefficients g iz and d iz , can vary across regions, is useful to generate consistent estimates that cannot be obtained when using homogenous models such as fixed-effects and random coefficient models (Pesaran et al., 1999).In our case, this model is useful to study the region-specific impact of each shock on regional employment growth and the consequences of Cohesion Policy funds during crises, which are our research questions.The preference for the MG model against alternative models is supported by Hausman-type tests (Ditzen, 2018).This approach has been recently used to study the heterogeneous impact of Cohesion Policy on regional growth (Di Caro & Fratesi, 2022) and the countercyclical role of Cohesion Policy funds with a macro-perspective (Canova & Pappa, 2021), by providing new knowledge on regional heterogeneity in Europe.

Aggregate results
Table 2 reports the MG estimates of relation ( 4) for the regions belonging to the EU-15 (specifications A-C)   The role of Cohesion Policy for sustaining the resilience of European regional labour markets during different crises 2433 and the EU-10 (specifications D-F) MS.For both groups of regions, we find that Cohesion Policy had positive effects on regional employment growth over the past 25 years.In the EU-15, a positive variation of EU funds per capita produces about 0.4% positive regional employment growth, in line with evidence on GDP using the same Cohesion Policy funds data (Fidrmuc et al., 2020).
As for the new MS, we find that a positive variation of EU funds per capita produces about 0.2% positive regional employment growth, also in line with the literature (Crescenzi et al., 2017).We find that, on average, the three main recessionary events produced different adverse impacts on employment growth.In the old MS, the regional employment losses caused by the Global Financial Crisis have been higher than those registered during the early 1990s crisis, and the Euro debt crisis of 2011-12 by confirming the severity of the Great Recession on the EU-15 regions (Fingleton et al., 2015).In the EU-10 regions, we find that regional employment losses caused by the Global Financial Crisis have been higher than those of the early 2000s crisis, and the Euro debt crisis.In the new MS, moreover, the Great Recession produced higher negative employment losses on a regional level than in the old MS.These findings confirm the presence of crisis-and region-specific patterns when analysing the resistance of the EU regions (Sensier et al., 2016).
Our results also suggest that EU Cohesion Policy played a different role during each recessionary event.In the old MS, we find that the positive consequences of Cohesion Policy on regional employment growth were reduced during the early 1990s crisis, in line with preliminary evidence discussed in the previous section.In detail, from specification (C), we note that the sum of the coefficients of interest is 0.0019 ¼ (0.0035 -0.0016).The relative low amount of EU funds committed at the beginning of Cohesion Policy implementation and the initial focus on structural, rather than countercyclical, objectives can justify this result (European Commission, 2010).
Differently, we find that during the Global Financial Crisis, Cohesion Policy sustained the resistance of regional labour markets in the EU-15, where the positive role of EU funds was amplified.In the new MS, on average, we do not find a clear different impact of Cohesion Policy on regional employment growth during all the crises, with the exception of some mitigating effects registered during the early 2000s shock when the EU-10 regions started to effectively benefit from Cohesion Policy funds.The role of Cohesion Policy for sustaining the resilience of European regional labour markets during different crises

Region-specific results
In this section, we interpret the estimated regional coefficients obtained from relation ( 4) to throw some light on the region-specific patterns observed across Europe during the different crises.For each crisis, we first focus on the number of regions that show: positive and significant effects of Cohesion Policy; negative and significant effects; and (statistically) non-significant effects.We also report the number of regions with different coefficients and significance levels regarding the impact of crisis dummies on employment growth.The results are shown in Table 3 for old and new MS separately.
Our results confirm the presence of crisis-and regionspecific patterns when looking at the resistance of the EU regional labour markets.There are several regions with negative and significant coefficients for the variable of interest; we also find a certain number of regions for which we detect coefficients that are not (statistically) significant.One of the possible explanations for the latter result can be that the impact of a given crisis was not sizable enough to produce a relevant change in regional labour market dynamic.Moreover, we find that the financial crisis produced the major adverse effects across the EU regions: this is true for old and new MS as well.Looking at the Cohesion Policy variable only, we mostly detect a positive impact on employment growth, though in some regions we do not find statistical significance.
There are several regions where the impact of Cohesion Policy during every shock has been significantly larger/smaller than in the rest of regions.This suggests that the short-run effects of each crisis can have heterogeneous implications for the effectiveness of Cohesion Policy, that is, some regions are more able to exploit their funds to react to a given shock than others.
To investigate the nexus between crisis resistance and Structural Funds effectiveness, we look at the statistical correlation of the variables of interest, without any claim of causality.As reported in Table 4, we keep separate the results for new and old MS.The development of a complete causal analysis is not possible here since data for most of the factors potentially affecting the relationship of interest are not available over the full time period considered.To describe the significance of the impact of a given variable, we use the z-tests of the estimated coefficients from relation (4).This is useful given that we obtain regional coefficients that show different levels of statistical significance.In detail, high and positive values of the ztests can be read as instances of the presence of a positive, highly significant impact of a given variable on a specific region.Conversely, very negative values of the z-statistics denote a negative, highly significant relation of the variables of interest; values around zero signal the absence of a significant relation.Moreover, Z-values are continuous, which is useful when producing a correlation matrix.For Structural Funds assistance, we construct two variables: a dummy for highly assisted regions, as described in Section 3.3; and a continuous variable for the average amount of Structural Funds per capita at a regional level.As for EU-15 regions, the various impacts of the different crises in our sample are positively correlated, that is, there are regions that are always more resilient to shocks, though with regional differences.If we look at the Structural Funds assistance, we find a positive correlation between Structural Funds and the coefficients describing the consequences of the early 1990s and financial crises.This suggests that more assisted regions registered a better market labour performance in such years, differently from other shocks, such as the Sovereign debt crisis, when negative effects were more relevant such as in regions in Greece and Spain.We do not find a strong correlation between the level of assistance and the impact of Structural Funds on employment growth by confirming the presence of heterogeneous aspects when analysing the effects of Cohesion Policy with panel time-series data over different programming periods (Di Caro & Fratesi, 2022).
We also find a negative correlation between the amount of Structural Funds received and the differential impact of Structural Funds during the financial crisis by pointing out the reduced role of Cohesion Policy in highly assisted regions during this shock.Differently, we find that a positive correlation in the case of the Sovereign debt crisis, that is, this shock had major adverse consequences on poorer EU-15 regions, where the Structural Funds contributed to smooth such negative effects.
As for the regions in the new MS, we find slightly less significant coefficients, probably because of the lower number of time observations.Yet, some heterogeneous patterns emerge also in this area.For instance, regions that suffered more in the early 2000s crisis experienced high resistance during the Euro sovereign debt crisis.In terms of Cohesion Policy assistance, region-specific results are less clear-cut in EU-10 regions.However, our findings suggest that regions with Structural Funds assistance above the average of the relative group reduced the negative drop in employment during the Sovereign debt crisis.This can be justified, among other factors, by the relatively low aggregate adverse effects of such shock in Eastern Europe and/or by the ability of these regions to absorb the Structural Funds deployed in the EU-10 area.The role of Cohesion Policy for sustaining the resilience of European regional labour markets during different crises 2437 Finally, it is worth commenting that the effectiveness of Structural Funds in the EU-10 regions has been larger for the most assisted regions during the Sovereign debt crisis by suggesting an important crisis-mitigating effect of Structural Funds.The last question that it is important to address is the relationship between the regional ability to use Structural Funds and the impact of the different crises.Although this question needs further research effort than that developed in this study, from Table 3 we can note that the regions with larger effects of Structural Funds on regional employment growth are those less affected by the Sovereign debt crisis.

Discussion, checks and future developments
We start discussing some practical issues that can occur in our empirical analysis and suggest how to address them.The Cohesion Policy literature has identified a large set of conditional factors 17 that can explain the different degree of effectiveness of Structural Funds in EU regions (Mohl, 2016).In our case, however, the inclusion of a large set of controls in the econometric model is hampered by data availability, given that our study covers 25 (15) years for the old (new) MS, and we prefer to keep the panel structure of our dataset.Based on the literature focusing on regional labour markets during shocks (Chodorow-Reich et al., 2012), and recognizing the data limitation issue, we have estimated relation ( 4) by enlarging the set of control variables, as follows.
Tables 5 and 6 report the results of some alternative specifications for the regions in the old and new MS, respectively.In detail, we add lagged employment levels to check for employment dynamics rather than population dynamics (column i).We include lagged regional GDP in levels (column ii); no significant differences are registered when using GDP growth rates.To control for pre-period regional economic trends (Bondonio & Greenbaum, 2007), we add the growth of employment registered before the observation period: from 1980 to 1990 for old MS; from 1990 to 1998 for new MS (column iii).In column iv we add the cross-sectional averages of employment growth in order to take into account the influence of

2438
Paolo Di Caro and Ugo Fratesi spatial patterns.Results (available from the authors upon request) are unchanged when adding sector-specific employment shares at a regional level.Table 5 (column v) restricts the analysis to 42 regions in the old MS belonging to the convergence and transition group, as defined in the 2007-13 EU programming period.For these regions, we find a larger impact of Cohesion Policy funds during the financial crisis than in the rest of the regions.Table 6 (column v) restricts the analysis to the regions in the old MS that have adopted the Euro, namely Cyprus, the Baltic republics, Malta, Slovenia and Slovakia.The main findings of our work remain unchanged after modifying the baseline specifications.
We cannot include additional factors in our specifications, such as the absorption of Cohesion Policy funds (Incaltarau et al., 2020), human capital (Cappelli et al., 2021) and regional institutions (Ezcurra & Rios, 2019), that can be relevant when analysing the relation between regional resilience and Cohesion Policy (Di Caro & Fratesi, 2018), mainly due to data limitations.The introduction of region-fixed effects in our model partly accounts for the lack of data of these variables, since such explanatory factors show temporal stability across regions (Vedrine & Le Gallo, 2021).
As future research developments, however, the results presented here can be extended along two main directions.First, we can analyse in more depth the interplay among selected conditional factors of Cohesion Policy and regional resilience by limiting attention to a particular recessionary event and/or a specific group of MS (Arbolino et al., 2020).Second, we can apply a two-step empirical approach to construct real-time indicators of resilience and Cohesion Policy in the short run, which is useful to forecast the resistance of regional labour markets during different shocks (Arbolino & Di Caro, 2021).

CONCLUSIONS
The set of evidence presented in this study contributes to the understanding of the resilience of regional labour markets in the EU over the past three decades, with a particular focus on regional employment resistance.Our empirical investigation shows that the different recessions that occurred in Europe from 1990 onwards produced heterogeneous consequences on regional labour markets.For each crisis, we find regional-specific patterns both within and across the European countries, with major adverse effects registered during the early 1990s and the early 2000s (for new MS only) shocks, and the Global Financial Crisis.In detail, we document that some European regions, in both the old and new MS, have been systematically more/less vulnerable to recessionary events, notwithstanding the origin, duration and type of economic crisis considered.We also find that some regions show a different degree of resistance depending on the crisis under observation, that is, the different mechanisms of the various crises have heterogeneous impact across regions.
As for the role of Cohesion Policy support during downturns, two main messages derive from our findings.First, the effects of EU funds on regional employment growth turn out to be in general positive and significant during all the recessionary events here considered, although with important regional differences.
Second, Cohesion Policy had a more relevant impact for sustaining resilience in the EU-15 during the Great Recession by confirming the positive role of Cohesion Policy funds in the short run, once the recalibration of the objectives has been explicitly operated (Di Pietro et al., 2020).
However, we document that the most assisted regions are not necessarily those more resilient to crises, given the influence of different factors that are not fully considered here mainly due to data limitation.
From a policy perspective, therefore, our analysis supports the view that Cohesion Policy funds can play a role in the post-pandemic recovery, as happened during the Global Financial Crisis, although the reactivity of regions to these stimuli is differentiated.In other words, the recalibration towards recovery of Cohesion Policy for the period 2021-27, as in the REACT-EU initiative within the Next Generation EU package, can sustain resilience in Europe if regional specificities are adequately considered (Bachtrögler et al., 2020;Di Cataldo & Monastiriotis, 2020).

Figure 1 .
Figure 1.Employment levels and growth in the European Union (EU), 1980-2015: (a) employment levels: total employment (thousands of employees) in the EU-15 (left axis) and EU-10 member states (right axis); and (b) employment growth: annual growth of total employment in the EU-15 (left axis) and EU-10 member states.Sources: Authors' elaborations; data from Cambridge Econometrics.

Figure 2 .
Figure 2. Maps of employment[AQQ19] resistance indicators in the various crises.Note: Maps report the regional differential growth (%) in the three periods of crisis: early 1990s/(early 2000s for new member states) (first from above row); Global Financial Crisis (second row); and Euro sovereign debt crisis (third row).Differential growth is calculated with respect to the European Union (EU) average (left-hand side) and country average (right-hand side).Indicators are calculated as in equations (1) and (2).Darkness denotes a region where the resistance to a given crisis is higher than the EU (left-hand side) or the country (right-hand side) benchmark.The maps also show the regions with average amount of Structural Funds per capita above (below) the group average, as described by shaded (unshaded) regional areas.The reference group is again the EU (left-hand side) or the country (right-hand side).
of significance are calculated at 5%.The exceptions (positive values) in the crises are represented by the following regions: DE30 (Berlin) in the Early 1990s crisis, CZ05 (Liberec) in the Early 2000s crisis; PL61 (Kujawsko-Pomorskie) in the Global Financial Crisis; PL63 (Pomorskie) in the Debt crisis.2434PaoloDi Caro and Ugo Fratesi

Table 1 .
Main European Union (EU) economic crises observed from 1990 to 2015.

Table 2 .
Main estimates, old and new member states.

Table 4 .
Correlations between the Z-tests of coefficients.