Regional resilience to the Covid-19 shock in Polish regions: how is it different from resilience to the 2008 Global Financial Crisis?

ABSTRACT The world economy has experienced two major economic shocks over the past 15 years: the Global Financial Crisis of 2008–09 and the Covid-19 pandemic. The purpose of this study is to link regional economies’ resilience to these extreme events while investigating the regional determinants of resistance to the Covid-19 shock and accounting for spatial spillovers. A total of 380 Polish NUTS-4 regions are analysed using spatial modelling techniques. Our primary finding is that the regional economic resistance to the 2008–09 global crisis and that of the Covid-19 pandemic are positively related. Specifically, the regions that were less affected by the first wave of the global crisis in 2009 were also more resistant to the Covid-19 shock in 2020, despite the fundamentally different anatomies of these shocks. Moreover, we find that the regions with higher production per capita were less resistant to the Covid-19 shock. This result could be attributed to the fact that industrial clusters are often integrated into global value chains, which were severely affected by the 2020 pandemic. Finally, we show that regions with a higher share of agricultural employment were generally more resistant to the Covid-19 shock, although the specifics of this sector in the eastern parts of Poland reduced the resistance to some extent. Generally, our results support the rationale behind the explicit modelling of spatial spillovers in the context of investigating regional resilience. We find that spatial spillovers in the resistance to the Covid-19 shock are significant and positive.


INTRODUCTION
The world economy has experienced two major economic shocks over the past 15 years. The first was the Global Financial Crisis in 2008-09 that spilled over to public finance and sparked a worldwide recession with a magnitude not seen since the Second World War. In 2020, the Covid-19 pandemic prompted governments around the globe to introduce response measures that were unprecedented in terms of scale and rapid execution, including effectively shutting down entire industries and locking citizens within their borders. These measures almost immediately shattered global value chains and redefined the rules of international trade, thereby posing challenges to the corporate sector.
The global crisis 1 of 2008-09 gave birth to keen academic interest in the concept of economic resilience, especially in regional and urban economics. Since then, a substantial body of theoretical and empirical research has accumulated, shedding light on the mechanisms of resilience and its different aspects and determinants. Even though the anatomy of Covid-19 was unique and strikingly different from that of the global crisis, its devastating effects on the economy were again substantially spatially diversified, thus reinforcing the interest in the resilience of local and regional economic systems.
In terms of determining economic outcomes, however, some country-wide features might have been similar: the industry mix, macroeconomic vulnerabilities and available fiscal space, quality of governance, or complex linkages within the international trade network. The most precise way to control these specific conditions is to perform studies on the country itself. Such an approach additionally fulfils the merits of investigating resilience at the regional level, as summarized by Sutton and Arku (2022). They emphasize that the effects of all events are ultimately resolved at the regional level, resulting in direct, long-lasting effects on regional labour markets and affecting patterns of economic growth and divergence.
In the current study we examine regions of Poland, one of the Europe's most spatially diversified country in terms of economic and business sector structure, accessibility, and income (Gajewski & Tchorek, 2017). This diversity was reflected in one of the strongest cross-regional heterogeneity of labour market resilience to the global crisis, as found by Giannakis and Bruggeman (2017), despite overall good performance of the country during the crisis episode. A similar conclusion was reached by Bristow et al. (2014). This combination of good macro-level performance and substantial cross-regional heterogeneity heats up academic interest in mechanisms and determinants driving regional economic resilience in Poland (Sensier et al., 2016).
Using Polish regions as a case study, the purpose of this study is to link regional economies' resistance to two extreme events: the 2008 and 2009 global crisis and the Covid-19 shock, while investigating regional determinants of resistance to the latter event and accounting for spatial spillovers. To the best of our knowledge, this is the first study to adopt such an approach. In contrast to other studies that investigate resilience through the lens of regional employment, our focus is on industrial output because employment during the Covid-19 shock was artificially supported by national response measures.

LITERATURE REVIEW
The spatially heterogeneous economic outcomes of the global economic crisis induced an inflow of empirical studies, which attempted to examine this heterogeneity and gave rise to the immense interest in economic resilience. However, it soon became clear that the rapidly expanding definitions, measurement methods and empirical approaches used in research threatened the feasibility of advancing academic discussion (Christopherson et al., 2010). The foundations of the consensus were established in the 2010 special issue of the Cambridge Journal of Regions, Economy and Society (vol. 3, issue 1), which offered several important contributions, including proposing, defining, framing and operationalizing the concepts of regional economic resilience.
We mainly draw from the above contributions in defining resilience within the engineering frame of reference and in the way it is operationalized (Christopherson et al., 2010;Pendall et al., 2010). Most importantly, we acknowledge that 'engineering' resilience, derived from engineering sciences and the seminal work of Holling (1973), is understood as the ability to withstand shock and recover to the initial state. Therefore, it is usually bifurcated into the initial resistance to a shock and to the subsequent recoverability, which is the ability to return to the pre-shock state (Brakman et al., 2015;Giannakis & Bruggeman, 2017;Martin, 2012;Martin et al., 2016).
Regardless of the approach, methodology, sample and variables of interest (usually output or employment), most studies investigating economic resilience to the global crisis conclude that substantial differences in resistance and recoverability were experienced at the regional level. Giannakis and Bruggeman (2017) explicitly show that Poland, France and Italy have demonstrated the most heterogeneous resilience patterns in the European Union.
Profound differences emerge from the way the forces behind these differences are conceptualized and identified. The complexity of the determinants themselves is explained by Sutton and Arku (2022), who illustrated their multi-scalar nature with a flow diagram. Simplifying this complex picture, we can conclude that they are usually sought in so-called initial conditions (Webber et al., 2018). Some of them are exogenous and broadly fixed (e.g., access to markets, extent of urbanization, administrative role, cultural factors) while others are structural (human capital, sectoral structure and specialization, demography, innovation) and more cyclical (e.g., prior economic performance, investment activity).
A particularly interesting discussion on resilience has developed in the urban/rural divide. This topic is also critical from the point of view of countries such as Poland (but also Italy, Spain, Greece, Romania, etc.) with stark regional differences in terms of the importance of agriculture or urbanization. In the case of Poland, the agricultural sector in numerous studies has been found to affect economic dynamics. In eastern parts of the country, agriculture, which is often fragmented and still partly based on subsistence farming, absorbs redundant labour forces during slowdowns at the expense of productivity. 2 Capello et al. (2015) find that regional economic resilience increases with the size of the cities. Similarly, Holl (2018) reports that Spanish urban core municipalities performed better than rural municipalities during the recession years. By contrast, Giannakis and Bruggeman (2017) and Palaskas et al. (2015) find that rural municipalities and the NUTS-2 regions in Greece were more resilient than their urban counterparts. Dijkstra et al. (2015) identify European metropolitan regions as more vulnerable to economic crises than rural regions. Giannakis and Bruggeman (2017) further confirm that a high share of agriculture positively contributes to the economic resilience of intermediate and rural regions. Finally, Ženka et al. (2017) do not find large differences in the abilities of the different types of Czech regions (metropolitan cores and hinterlands, urban regions and rural regions) to withstand the recessionary impact.
Although attractive and stimulating, the urban/rural divide is not the only axis of dispute in the regional resilience literature. Marelli et al. (2012) and Palaskas et al. (2015), for example, indicate that regions with the best-performing labour markets before the crisis subsequently became more vulnerable than the lagging regions. Martin (2012) arrives at a broadly opposite conclusion. Some studies have explored the features of the business sector, especially the size of firms. For example, Tsiapa and Batsiolas (2019) show that small firms lend resilience to their regions because of their flexibility in adapting to changes. Theoretically, larger firms could also have an advantage over small ones as they typically have easier access to financial resources during adverse conditions and can therefore cushion and manage their employment size and structure more effectively. Peric and Vitezic (2016) show that firm performance was positively related to firm size during the global crisis and Asian financial crisis, respectively. Finally, some studies find no empirical evidence of a meaningful relationship between firm size and resilience (Cainelli et al., 2019).
Some consensus is perhaps visible with regard to the role of accessibility. Östh et al. (2015) and Giannakis and Bruggeman (2017) reveal the positive role of accessibility for resilience. Meanwhile, little evidence for either supporting or impeding resilience has been found in the area of demography (Giannakis & Bruggeman, 2017).
All the studies discussed are focused on earlier crisis episodes. Regarding the Covid-19 shock, empirical evidence is still scarce as economic data (especially at the regional level) are published with a delay. Hence, the initial focus has been directed toward concerns that could have been addressed with high-frequency, instantly available data obtained from healthcare systems.
So far, a few new subsets of studies dealing with the regional economic outcomes of the Covid-19 shock have emerged. First, a robust strand of research is developing on urban resilience (e.g., Hu et al., 2022;Sharifi, 2021). Second, some far-reaching postulates suggest the 'rethinking' of regional science in the post-Covid-19 world (Martin, 2021). Third, resilience to the Covid-19 shock for regional economies can be examined with existing tools, which are now efficient and flexible owing to the substantial theoretical and methodological advancement that took place after 2008. Importantly, researchers have already turned their attention to the comparison of the Covid-19 shock with earlier crises.
The preliminary results from Chinese regions where the Covid-19 shock first emerged suggest that in addition to the severity of the pandemic, high income and a high reliance on foreign trade were associated with the strongest decline in regional gross domestic product (GDP) (Gong et al., 2020). Partridge et al. (2022) examine the resilience of counties in the United States to Covid-19-induced shock and find that larger leisure service sectors contributed to better resistance and that counties with a younger and less educated workforce were more resistant. The role of small businesses was found to be ambiguous, which contradicts their role in earlier crisis episodes where they appeared to increase regional resistance. In another attempt to connect the Covid-19 shock with previous crises, Brada et al. (2021) simulate the effects of the Covid-19 pandemic based on experiences in the 2008-09 global crisis. The closest study to ours is, however, the recently published work by Kim et al. (2022), who examine the relationship between the resilience of US states across three crisis episodes: the 2001 recession, the 2008 global crisis and the Covid-19 shock. They found that states that quickly recovered from the 2001 recession were more likely to resist the pandemic shock during the initial phase of the Covid-19 crisis, but they do not observe any link between the 2008 crisis and the Covid-19 shock. However, Kim et al. do not explicitly account for spatial dependence or spillovers.
What might be surprising is that although the spatial context is typically discussed in resilience studies, only a few explicitly model spatial spillovers. However, those that do find that the spillovers are significant in terms of either the resilience measure itself or its determinants (Annoni et al., 2019;Cainelli et al., 2019;Giannakis & Mamuneas, 2022). To the best of our knowledge, no study examining resilience to the Covid-19 shock has used spatial modelling techniques.

THE GLOBAL FINANCIAL CRISIS AND THE COVID-19 SHOCK AT A GLANCE
As in other European countries, the first wave of the global crisis in Poland was mainly a demand shock (Benguria & Taylor, 2020). Hence, it had an adverse impact on GDP growth and industrial production, as well as on the unemployment rate, while slightly decreasing consumer price index (CPI) inflation (Table 1). Capital outflows from emerging economies to safe havens led to massive depreciation in the nominal exchange rate, thus preventing GDP from plummeting. Owing to the relatively robust growth in the pre-crisis period, the deceleration did not push the country's GDP dynamics into negative territories, leaving Poland as a 'green island' among the European countries immersed in recession.
By contrast, the first wave of the 2020 Covid-19 pandemic had supply-side roots. The lockdowns and widespread restrictions that disabled operations in some industries, as well as the disruptions in global supply chains, had the same impact on Poland's industrial production as the 2008 crisis. However, the effect during this period was accompanied by accelerating inflation, and the exchange rate did not offer significant protection to GDP. Hence, the output was eventually severely affected. In addition to these developments, public support programmes required firms to shelter employment, thus causing the unemployment rate to remain broadly at the pre-Covid-19 pandemic level.
Although public aid policies to protect employment can be judged as commendable, they pose another challenge for researchers. In studies on resilience to previous crises, employment is a common choice for constructing a dependent variable. As it is artificially supported in some countries, other response variables must be sought. This study focuses on industrial production, which has not been directly targeted by public policies and has recorded a similar decline in both crisis episodes (Table 1).
Poland entered a turbulent period preceding the global crisis with an unravelling (if not overheated) economy. Industrial production grew at an annual rate exceeding 10% until 2007. During 2007 and 2008, these dynamics gradually weakened until the first quarter of 2009 when the economy took the most severe blow, as reflected in industrial production falling by 10% in annual terms. A rebound followed and continued for the remainder of 2009. Similar developments were observed in response to the Covid-19 shock, except that they were not anticipated. Hence, it was not before the first quarter of 2020 (at the end of which the first strict response measures were implemented) when industrial production was affected and its year-on-year dynamics fell almost to zero. In this case, the trough occurred in the subsequent quarter. However, the rebound was completed by the end of that year (Figure 1).
Despite the divergent mechanisms driving the two crises, the patterns of decline and rebound in industrial production have not been radically different. Moreover, extensive empirical evidence on the global crisis demonstrates substantial cross-regional heterogeneity in economic resilience. This begs the question of whether spatial differences in resistance to the Covid-19 shock were as strong as those during the global crisis and if some association exists between the regional resistance to these two types of shocks.

METHODOLOGY AND DATA
4.1. Measuring regional resistance Given that the Covid-19 pandemic is a recent event, we take a short-run perspective (i.e., we do not account for structural reorientation, etc.) and conceptualize regional resistance in line with the work of Martin et al. (2016) and Hu et al. (2022), who first construct counterfactual series of expected regional performance variables based on national development and then define resistance in terms of the deviation of the actual variables from this counterfactual. Adopting this measure is especially relevant for our study because both shocks developed in the country-wide context. Since individual regions were not explicitly targeted, they could have been expected to move in concert with the national economy. Let (DIP i ) expected be the expected annual growth rate of sold industrial production during the crisis period of 2020 or 2009, referred to as the national growth rate (DIP PL ).
(DIP i ) expected = DIP PL We define:

Data
Our dataset comprises two measures of resistance to the global crisis (resistance1) and the Covid-19 shock (resistance2) that are calculated for 380 NUTS-4 regions (poviats). Additionally, we control for the initial economic and structural characteristics of the regions, which are often discussed in the context of regional resilience. The control variables are the share of agricultural employment in total employment (share_agr), output per person (prodpop), share of firms with 10 or more employees (firms10plus), average salary, relative to Poland (salary), share of the population living in cities with population exceeding 400,000 (bigcity), 3 and share of the working-age population (pop1564). The variables capturing the initial conditions, that is, before the Covid-19 pandemic, refer to the year 2019. All data are obtained from the official local database (Bank Danych Lokalnych) maintained by the Polish Statistical Office. Table 2 presents the summary statistics of the variables. It appears that resistance to the Covid-19 shock is slightly less diversified across regions than the resistance to the global crisis. Nevertheless, the data illustrate substantial cross-regional differences in initial conditions, especially in the importance of agriculture, output per person, salaries and the share of population in 'big cities.'

The spatial model
To model spatial spillovers, we consider the spatial Durbin model (SDM), where a spatial lag of the dependent variable and a spatial lag of the independent variable matrix is added to the set of explanatory variables (Anselin, 1988): where y is the n × 1 dependent variable (resistance to Covid-19 shock) vector; a denotes the intercept coefficient; X is the n × p matrix of p independent variables, including economic characteristics and resistance to the 2009 crisis; and i is the n × 1 vector of ones. ρ denotes the spatial autoregressive (SAR) term, b is a p × 1 vector of coefficients capturing the impact of the independent variables on the dependent variable, and g is a p × 1 vector of coefficients for the spatially lagged independent variables. W is the spatial (row-standardized, contiguity) weighting matrix, and 1 is the i.i.d. error term.
To summarize, resistance to the Covid-19 shock in a given region is assumed to depend on: (1) a region's resistance to the 2009 crisis, (2) a region's economic characteristics, (3) resistance to the Covid-19 shock in neighbouring regions, and (4) resistance to the 2009 crisis in neighbouring regions and their economic characteristics. Such extensive cross-regional spillovers are often undetected, and the specification given by the SDM can be compressed to a nested model, such as the SAR model (when g = 0 and ρ ≠ 0) or the spatial error model (SEM) (when g − br = 0).
In this study we follow the well-established procedure outlined by Lesage andPace (2009) andElhorst (2010), who recommend that the SAR model should be first tested against its non-spatial counterpart and estimated by means of ordinary least squares (OLS). If the ρ = 0 hypothesis is rejected, then the SDM should be estimated first because it is the only model that yields unbiased coefficient estimates even if the true data generation process is of a different form (e.g., SAR or SEM). However, although the SDM yields unbiased parameter estimates because it is a generalization of the SAR model and SEM, testing for the above parameter restrictions and selecting the optimal model can improve the efficiency of the estimates.
In our case, the ρ = 0 hypothesis is comfortably rejected by means of the Anselin (1988) Lagrange multiplier (LM) test (Table 2). Thus, we subsequently test the SDM against its spatial Note: prodpop is expressed in logarithms; salary: Poland = 100. The data on industrial production of 28 and 33 regions (7.4% and 8.7%, respectively) in the first and second crisis periods, respectively, are unavailable.
alternatives. Ultimately, the LM test suggests that the most unbiased and efficient estimates can be obtained using the SAR model. 4 This transformation enables us to calculate the partial derivatives of y with respect to each of the k-th explanatory variable as follows: Finally, we compute the direct and indirect effects (LeSage & Pace 2009). Direct effects capture the impact of explanatory variables in region i on the dependent variable in region i. Indirect effect represents spillovers from related regions (Golgher & Voss, 2016).
The indirect effects in our model comprise two elements: the local effects due to the g 1 coefficient (impacting only immediate neighbours) and the global effects arising from the (I − rW ) −1 matrix. Global spillover effects impact all regions because they contain a series of feedback effects that arise as a result of impacts passing through neighbouring regions and returning to the region from which the impulse originated. Table 3 presents the estimation results. Our primary finding is that the regional economic resistance to the global crisis and that to the Covid-19 shock are positively related. The regions that were less affected by the first wave of the global crisis in 2009 were also more resistant to the Covid-19 shock in 2020, despite the fundamentally different anatomies of these two crises.

RESULTS
Interesting results are obtained regarding the role of the agricultural sector in resistance. We capture it with two variables: share_agr and an interactive term of this variable with a dummy 'east', which takes the value of 1 for regions located in eastern Poland (i.e., voivodships: Lubelskie, Podlaskie, Świętokrzyskie, Podkarpackie, Warmińsko-Mazurskie). Agriculture in this part of the country has historically been distinct and less productive because of land fragmentation, undercapitalization and subsistence farming. According to Gajewski and Kutan (2018), the value added per worker in agriculture in eastern voivodships is five to seven times lower than that in the western parts of the country. Our results generally find strong evidence of the protective role of agriculture against the Covid-19 shock, but this protection is partly neutralized in the east possibly because of the idiosyncrasy outlined above.
As the focus of this study is on resistance measured by industrial output, a natural question arises about the role of industrial clustering in this context. Undoubtedly, clustering brings certain benefits: it helps enterprises, especially smaller ones, to overcome growth constraints and compete in distant markets (Schmitz & Nadvi, 1999). However, whether clustering also brings insurance against external shocks or, conversely, makes a host region more vulnerable to them is a separate issue. Our results indicate that regions with higher production per capita were less resistant to the Covid-19 shock. A possible explanation is that industrial clusters are often more integrated into global value chains (which were severely affected by the 2020 pandemic) than 'ordinary' regions, where production is more locally embedded in terms of supply and demand. Other variables associated with the business sector (including firm size), presence of large cities, salary levels, and demography are not found to be statistically significant.
However, similar to the case of the 2009 crisis (e.g., Brada et al., 2021), we find positive spatial spillovers of resistance to the Covid-19. shock. Moreover, when the alternative SEM is considered instead of the SAR model, the global diffusion of shocks across regions is visible and significant. Table 4 reports the direct and indirect effects obtained from the SAR model. The direct effects appear to dominate over the indirect effects, although their inclusion raises the total effects substantially and exceeds the OLS coefficient estimates.

CONCLUSIONS
Using data on Polish NUTS-4 regions, this study aims to shed some light on the association between regional economic resistance to two severe shocks that occurred within a 15-year time span while investigating whether spatial spillovers of resistance to the recent Covid-19 shock can be detected. The results confirm that the regions that performed better during the 2009 global crisis were also more resistant to the Covid-19 shock in 2020.
Given some earlier results indicating the greater resilience of rural regions compared with urban centres, an impression might arise that the economic backwardness of a region, manifested by its outdated employment structure and low openness and accessibility, can be protective against external shocks. Admittedly, backward regions are less prone to financial bubbles and other imbalances, which tend to accumulate in 'innovative' areas that are dominated by services. Indeed, the transition crisis in Central and Eastern Europe after 1989 was least Note: t-statistics are shown in brackets; *, **, ***significance at the 0.1, 0.05 and 0.01 levels, respectively. Pseudo-R 2 is calculated as [corr(y,ŷ)] 2 , whereŷ is the reduced-form prediction of y. Table 4. Direct indirect effects.

Direct
Indirect Total devastating in backward, rural areas dominated by subsistence farming. However, our results provide a more nuanced picture.
In line with the work of Giannakis and Bruggeman (2017) and Palaskas et al. (2015), we show that agricultural regions were generally more resistant to the Covid-19 shock. An important caveat here is that although agriculture offered substantial protection, the specifics of this sector in eastern Poland (broadly characterized by relative backwardness) neutralized this effect to some extent. It was therefore the highly efficient and modernized agricultural sector that was most efficient in protecting regional economies.
Our results support the rationale behind the explicit modelling of spatial spillovers in the context of investigating regional resilience. Spatial spillovers are significant, and spatial models outperform their nonspatial counterparts.

DISCLOSURE STATEMENT
No potential conflict of interest was reported by the author. NOTES 1 We use the labels 'global crisis' and 'Covid-19' for brevity and clarity, but acknowledge that Covid-19 was by no means less global than the crisis in 2008-09. 2 Even at the NUTS-3 level in Poland, the maximum/minimum ratio of labour productivity in agriculture was 3.7, as of 2019 (source: Polish Central Statistical Office (Główny Urząd Statystyczny)). 3 As of 2019, there were seven such cities: Warsaw, Kraków, Łódź, Wrocław, Poznań, Gdańsk and Szczecin; they are usually regarded as growth engines for the regions that host them. 4 For a robustness check, we also report the results from the SEM, which is only marginally rejected in favour of the SAR model.