Review on Energy Resilience

Ann Reg Sci (2018) 60:393–410 https://doi.org/10.1007/s00168-017-0822-9

SPECIAL ISSUE PAPER

Regional economic resilience: the experience of the Italian local labor systems

Alessandra Faggian1,2 · Roberta Gemmiti3 · Timothy Jaquet1 · Isabella Santini3

Received: 16 June 2016 / Accepted: 20 February 2017 / Published online: 11 April 2017 © Springer-Verlag Berlin Heidelberg 2017

Abstract After defining the concept of resilience and its application to the regional context, the paper presents a preliminary evaluation of regional economic resilience in the case of the Italian regions. In doing so, we follow the approach by Martin (J Econ Geogr 12:1–32, 2012) and Martin and Sunley (2015) who identify three different dimensions to regional economic resilience: (a) resistance, i.e., the degree of sensitivity or depth of reaction of a regional economy to a recessionary shock; (b) recovery, i.e., the speed and magnitude of the recovery; (c) reorientation and renewal, i.e., the ability of a region to adapt in response to the shock and renew its growth path. The analysis is conducted at the local labor systems (LLS) geographical level and focuses, at this stage, only on the first two dimensions of resilience, i.e., resistance and recovery. The recessionary shock (2009–2010) is defined following the Italian National Statistical Institute approach for which a recession implies a decrease in GDP for three consecutive trimesters. The pre-recessionary period is 2007–2008 and the recovery period 2011 (as a new recession started again in Italy at the end of 2011). The results clearly point at very heterogeneous resilience for the Italian LLS.

B Alessandra Faggian [email protected]

Roberta Gemmiti [email protected]

Timothy Jaquet [email protected]

Isabella Santini [email protected]

1 AED Economics Department, The Ohio State University, Columbus, OH, USA 2 Gran Sasso Science Institute (GSSI), Social Sciences, L’Aquila, Italy 3 Department of Methods and Models for Economics, Territory and Finance,

Sapienza University of Rome, Rome, Italy

123

394 A. Faggian et al.

JEL Classification R1 · R11 · O4

1 Introduction

Following the recent economic crisis that affected first the USA and then Europe, the concept of resilience is gaining popularity in economics, especially in urban and regional economics and economic geography. It is clear that different regions show varying ability to successfully mitigate or react to economic crises, and one cru- cial question is: “Why?” What are the factors explaining a high level of regional “resilience”? Moreover, are these factors different than those driving regional growth in periods of relative stability?

After briefly defining the concept of resilience and its recent application to the regional context, this paper presents a preliminary evaluation of regional economic resilience in the case of Italian Local Labor Systems (LLS). In doing so, we follow the approach by Martin (2012) and Martin and Sunley (2015) who identify three different dimensions of regional economic resilience: (a) resistance, i.e., the degree of sensitivity or depth of reaction of a regional economy to a recessionary shock; (b) recovery, i.e., the speed of recovery or degree of recovery after a certain period of time; (c) reorientation and renewal, i.e., the ability of a region to adapt in response to the shock and return to, or even improve, its long-run growth path.

– Our initial analysis in this paper focuses on the first two aspects of resilience, resistance and recovery. In an effort to identify what factors foster higher regional resilience, we analyze the employment growth paths and their determinants in Italian local labor markets for the period 2007–2011. In our model, we employ a set of explanatory variables, which include indexes of specialization, diversification and concentration as proposed by Mameli et al. (2008).

The Italian National Statistical Institute (ISTAT) defines a recession as a decrease in GDP for at least three consecutive trimesters. In our sample, this covers 2009–2010. We are defining the pre-recessionary period as 2007–2008, and the recovery period as 2011 (as a new recession started again in Italy at the end of 2011). The results clearly point at very different levels of resilience for the Italian LLS.

The paper is organized as follows. Section 2 briefly summarizes the multitude of contributions in regional science and economic geography over the last few years to the concept of regional resilience. It also describes the scope of our paper and the empirical modeling strategy. Section 3 introduces the data used in the analysis and our case study. Section 4 presents some preliminary results and discusses them. Section 6 offers some concluding remarks and avenues for future research.

2 A brief history of regional resilience

Rarely a concept finds such an immediate popularity as resilience has in the last decade. Resilience seems the new “buzz word” of the 2000s and 2010s, and it now plays the role that sustainability did in the 1980s and 1990s.

123

Regional economic resilience: the experience of the Italian... 395

Resilience is an appealing concept because of its apparent simplicity. Generally it is defined as the ability to “bounce back” from some sort of stress. As such, it has been used in a variety of disciplines. The concept’s origins are usually attributed to ecology (Holling 1973) but has since been adapted to engineering, psychology, and more recently, social sciences. In economics, and regional economics in particular, the concept started appearing in the literature in the 2000s. Reggiani et al. (2002) offer one of the first comprehensive reviews (and empirical application) on the topic, highlighting that there are at least two different interpretations of resilience. The first, also referred to as “engineering resilience,” refers to a region’s ability to return to the pre-shock state (Hill et al. 2008). The second definition of resilience, labeled “ecological resilience,” measures the size of shock that a system can absorb without being shifted to a new equilibrium or state of being. This definition was developed with a multiple equilibrium system in mind, but is can also be viewed as minimizing the distance a shock shifts a region from its long-run growth path (e.g., Balland et al. 2015). A third definition has emerged that is centered around an area’s ability to adapt and reorganize in response to a given shock. In this context, resilience does not necessarily mean going back to the pre-existing state, but potentially shifting to a new (ideally better) one (Modica and Reggiani 2014). This “adaptive” capacity of a system has been highlighted by a series of more recent contributions such as Metcalfe et al. (2006), Hassink (2010), Martin (2012), Martin and Sunley (2015) and Boschma (2015).

However, resilience, like most “fast-rising-in-popularity” concepts, is not without its critics. Despite its appeal, a certain degree of fuzziness remains around the concept of regional resilience (Davoudi and Porter 2012; MacKinnon and Derickson 2013). If defining and interpreting resilience is difficult, measuring it, it is even more challenging (Carpenter et al. 2001). Three are the fundamental questions to answer before any empirical investigation on resilience:

1. Resilience “to what”? 2. Resilience “of what”? 3. Resilience “over what period”?

2.1 Resilience “to what”?

The concept of regional resilience found a fertile ground in disaster studies (Rose and Liao 2005; Cutter et al. 2008; Carpenter 2015, Jara and Faggian, forthcoming), but it has rapidly expanded to encompass other topics such as housing foreclosures (Swanstrom 2008), economic downturns (Hill et al. 2012), recessions (Fingleton et al. 2012; Cellini and Torrisi 2014; Di Caro 2017) and influx of immigrants (Mollenkopf and Pastor 2013). There is also a need to clearly identify a “threshold” to define what constitutes a “shock,” such as a minimum magnitude of loss to variables of interest or minimum duration of negative growth.

Obviously it would be ideal to somehow assess the “overall” resilience of an economic system to all possible external stressors, but artificially creating these cir- cumstances or finding perfect natural experiments is nigh impossible. Furthermore, regional economies are dynamic systems with a multitude of components, so we

123

396 A. Faggian et al.

lack ecology or engineering’s ability to precisely model or replicate system dynamics (Anderies et al. 2013). Hence, it is fundamental in any resilience study to clearly state the scope of the analysis. A region, which might be resilient to a certain type of shock, might not be to another type. Identifying trade-offs between the different dimensions of resilience is desirable, but extremely difficult (Anderies 2015). However, some recent contributions such as Irwin et al. (2016) seek to create a more systemic framework by framing the issue through the lens of social well-being

Similarly to the studies by Cellini and Torrisi (2014), Lagravinese (2015) and Di Caro (2017), we are interested in studying the effects of the economic recession in Italy. To identify the recessionary period, we follow the definition by the Italian National Statistical Office (ISTAT), which defines a recessionary period as a decrease in GDP for three consecutive trimesters. Based on this definition, the recessionary period in Italy covers the years 2009 and 2010.

2.2 Resilience “of what”?

This second question is even more controversial. Once we identify the external stressor thatinformsthescopeofouranalysis,howdowemeasuretheresilienceofaneconomic system? There are two compound issues here: a. Indicator: what variable or variables are most appropriate to measure regional resilience? b. Geographical areas: how do we define an economic system?

2.2.1 Resilience indicator

Most contributions in economic geography and regional economics have so far relied on traditional economic indicators such as employment (Martin 2012; Fingleton et al. 2012; Lagravinese 2015), employment growth rate (Augustine et al. 2013) or per- capita GDP (Cellini and Torrisi 2014). Balland et al. (2015) look at patent generation as a proxy for innovation, and Carpenter (2015) looks at occupation rates in housing. Foster et al. (2010), in their review, list among others: wages, unemployment and poverty rates, income inequality, out-migration, local government debt and revenues. There have been suggestions that a more complex, multi-dimensional index needs to be built; however, something along these lines has yet to emerge.

A different approach to measuring resilience is found in other fields, such as soci- ology or environmental studies where the term “regional” resilience is traditionally replaced by “community” resilience. Community resilience is defined along a series of dimensions or “capacities,” which are summarized in a resilience index. Sherrieb et al. (2010), for example, develop an index which includes not only employment lev- els but also income equity, educational level of the population and social capital. The proponents of resilience indexes point out that a single variable to measure resilience is inappropriate because the resulting outcome is largely dependent on the indicator chosen for the analysis (Irwin et al. 2016). However, an index is also subjective as there is an intrinsic level of subjectivity in the selection of the elements to be included in the index and, equally, in the weights used for their aggregation (Davoudi and Porter 2012). Moreover, two different communities, with the same exact index score,

123

Regional economic resilience: the experience of the Italian... 397

might have very different underlying values for the different components, concealing potentially important heterogeneity.

Despite the potential value of exploring the creation of such an index, we follow previous literature and look at regional resilience in Italy in terms of employment levels, similar to work done by Martin (2012) and Fingleton et al. (2012) in the UK.

2.2.2 Geographical areas

Most of the studies on resilience use administrative boundaries at different geograph- ical scales to assess the resilience of an economic system. One drawback is that some of these administrative units are quite large and do not do a good job of capturing the dynamics of local labor systems. Cellini and Torrisi (2014), for instance, look at the 20 Italian regions (NUTS1 level), which are rather large. Their rather surprising result that these regions display a high degree of homogeneity in their recovery from shocks might be partially due to their spatial unit of analysis. However, Di Caro (2017) uses the same spatial scale and finds a more heterogeneous response by the different regions. Fingleton et al. (2012) and Martin (2012) analyze the resilience of counties in the UK (NUTS2) and find that the two most densely populated regions—greater London and the South East—although more vulnerable to the recessionary shock were also more successful in recovering.

Ideally, an analysis of economic resilience measured by some kind of labor market- related variable, such as employment, should be done at a more disaggregated level. These areas should somehow be representative of the local labor market, or in other words, be “functional” rather than administrative areas, such as local labor systems (LLS) or travel-to-work areas (TTWA) (ISTAT 1997; Coombes et al. 2012). However, data at this level of analysis are more difficult to find and often require an elaborate (and time consuming) process of data cleaning, especially when a cross-temporal comparison is needed.

However, despite the challenges of working with functional areas, we believe that this is a fundamental issue in the study of economic resilience, and we present analysis of Italy at the local labor systems (LLS) level. A more detailed description of the Italian LLS and data is given in Sect. 3.

2.3 Resilience “over what period”?

Some economic systems might be able to bounce back or move toward a different locally stable equilibrium faster than others. So, the question here is: “What is the appropriate temporal framework to assess the resilience of a region?” One could also ask: is bouncing back faster a sign of being more successful even in long run? Is there a correlation between short- and long-term resilience? More complete modeling of the complementarities and trade-offs between different types of resilience represents a core knowledge gap on the frontier of resilience research, but is beyond the scope of this paper.

These questions are particularly difficult to answer in the recent case of the Euro- pean economic recession, where only few years of post-recession data are currently

123

398 A. Faggian et al.

available. The case of Italy is complicated further by subsequent waves of recessionary shocks. Following the recession of 2009 and 2010, there was a brief period of recovery (the year 2011) followed by a new recession. As such, our analysis can only present a short-run analysis of the 2011 recovery. The availability of new data (post-second recession) will allow us on the future to extend our current analysis to a longer time- frame and potentially uncover the correlation (if any) between the 2011 recovery and the longer-run growth.

3 Data and methodology

As stated in the previous section, our paper presents an initial exploratory analysis of the short-run recovery of the Italian LLSs following the recession of 2009 and 2010. As an initial step, we collected and collated data available at LLS level.

The LLSs were defined by the Italian National Statistical Institute (ISTAT) for the first time in 1981 in collaboration with Istituto Regionale per la Programmazione Economica della Toscana (IRPET) and the University of Newcastle in the UK.1 Since they are functional areas and depend on commuting flows, they are dynamic in nature, and their definition has been revised since their creation in conjunction with the release of the decennial Censuses (1991, 2001 and 2011). Not surprisingly, as commuting flows lengthened over time, the number of LLS has decreased from an initial number of 955 in 1981 down to 616 in 2011. Given the time frame of our analysis (2007–2011), we employ the 2001 definition of LLS.

Figure 1 shows a comparison between the 2001 LLS and the Italian regions, which have been used in previous analysis of resilience in Italy (Cellini and Torrisi 2014; Lagravinese 2015 and Di Caro 2017). In 2001, there were 686 LLSs, as compared to the standard 20 Italian regions.

It is worth noting that the LLSs are much smaller areas than regions and that, in some cases, an LLS is not contained just in one region but crosses over an administrative boundary. In fact, of the 686 LLSs, 49 do not belong to a unique region, but rather cross over two different regions (ISTAT 2005). This potentially adds noise to the data on labor markets when simply looking at administrative boundaries.

We use data on employment at LLS level to:

1. Run an initial exploratory analysis of employment trends in the period 2007–2011 following the framework and indicators used in Martin (2012);

2. Classify the LLS according to their behavior in resistance and recovery; 3. Link some of their structural characteristics to the likelihood of belonging to a

different class, using a multinomial logit model (MNL).

Martin (2012) provides a useful and simple framework of analysis by defining resilience as a process with different phases, namely resistance, recovery, reorien- tation and renewal. Similar to the study by Fingleton et al. (2012) in the UK, we are

1 For a more detailed definition of the methodology used, see ISTAT (1997).

123

Regional economic resilience: the experience of the Italian... 399

Fig. 1 Italian regions versus Local Labour Systems (LLSs)

primarily interested in the resistance and recovery phases. Resistance is proxied by a sensitivity index (SI) à la (Martin 2012).2

Formally, our SI can be expressed as:

SI = Er,tEr,t−1 /

En,t En,t−1

(1)

where Er represents total employment in region (r) and En represents total national employment. Period t is the recessionary period (2009–2010) and t − 1 is the pre- recessionary period (2007–2008). This index is similar to a “location quotient” for total employment and hence is centered around 1, so its interpretation is straightfor- ward. A value above unity means that the region was more resistant than the overall nation, while a value below one shows that the region was more affected by the reces-

2 The original formula in Martin (2012, p. 16) is actually (�Er/Er)/(�EN/EN). However, the formula does not match what Martin himself reports in the same paper in Table 5, p. 22. A ratio between percentage change as the one in the formula above can have both negative and positive value and it is not centered around the value of 1, while the values reported in Table 5 seem to be centered around 1 and are all positive (this is possible, but seems a bit implausible). Hence, we modified the formula to match the “sensitivity index” values reported by Martin (2012) in Table 5.

123

400 A. Faggian et al.

sion. Recovery is simply the regional percentage change in employment in 2011. As explained in Sect. 2.3, this represents a short-run recovery and a choice dictated by the fact that Italy entered a new recessionary period in 2012. A nice extension to this initial analysis would be to test the relationship between the short-run recovery and the long-run trends.

The sensitivity and recovery indexes are then used to classify the LLS into four groups:

1. High resistance/fast recovery (group I); 2. High resistance/slow recovery (group II); 3. Low resistance/slow recovery (group III); 4. Low resistance/fast recovery (group IV).

With high resistance being LLS with an SI above 1 and fast recovery being LLS with positive growth rates during the recovery post-recession period. These four groups, in turn, become the dependent variable in a multinomial logit model where the probability of belonging to each group is a function of a series of structural characteristics of each LLS, or more formally:

Pr (y = m|x) = e xβm|III∑ j e

xβj|III (2)

Equation (2) expresses the probability of an SLL of belonging to a certain group, relative to belonging to a base group, as a function of characteristics summarized by the x vector. In the above example, we use group III as the base group because they are the worst performers. At LLS level, we have information on the following explanatory variables that make up our x vector:3

(a) Population size of the SLL. We divided the SLL into 5 population classes: up to 10,000 inhabitants (n = 102); between 10,001 and 50,000 (n = 314); between 50,001 and 100,000 (n = 138); between 100,001 and 500,000 (n = 116); and above 500,000 (n = 16). Four dummy variables were created for the last 4 classes (i.e., up to 10,000 was the reference category). Size is important, as it can signal the presence of agglomeration economies.

(b) Industrial districts Industrial districts are an important reality in Italy, so a series of dummies were included to signal whether an SLL belonged to an industrial district. Twelve types of industrial districts are identified by ISTAT (in brackets we put the number of SLL belonging to each type): leather (11), shoes (22), textile (18), clothing (49), furniture (28), glasses (8), machinery (35), food (61), metals (14), transportation (16), construction (7), chemical and oil (19).

(c) Tourism and ports ISTAT also provides information on SLL that are considered to have a touristic vocation and/or that are ports. Both can be an important source of income in Italy, so dummy variables for SLLs with a touristic vocation (82) and for ports (26) were also included.

(d) Agriculture While tourism, in principle, might help during a national recession, agricultural areas, being also more peripheral, are normally disadvantaged during a crisis. To control for this, we included a dummy for SLL with an agricultural

3 Summary statistics on the explanatory variables are in Appendix, Table 3.

123

Regional economic resilience: the experience of the Italian... 401

vocation. ISTAT classifies only 24 LLSs as having a predominantly agricultural vocation.

(e) Urban areas ISTAT also classifies the SLLs has being urban or non-urban. Urban SLLs are, in turn, classified as being “urban and highly specialized” (where both specialization and urbanization economies should be present), “urban and moder- ately specialized” (where urbanization economies should dominate specialization economies), and “urban and non-specialized” (where there should be only urban- ization economies).

(f) Regional fixed effects: To control for other factors that might be region specific, we include a dummy for which region they belong to.

4 Results and discussion

4.1 Resistance and recovery of the LLSs

As a first step, we calculated resistance and resilience indexes for each LLS. As we cannot present a table with the values for all the 686 SLLs, we summarize the information in Fig. 2a, b.

Some interesting facts emerge by looking at these maps. First, there is no significant correlation between resistance and recovery. In fact, the correlation index between the two is −0.075 and insignificant even at 10% level. Second, there is a lot of heterogene- ity between LLSs within each region, which is hidden when looking at data at regional level. To further explore this point, Table 1 (panels a and b) presents a summary of

Fig. 2 a Resistance of the Italian LLSs (sensitivity index). b Recovery of the Italian LLSs. Source: Own elaboration

123

402 A. Faggian et al.

Table 1 Summary of resistance and recovery indexes by region

Region (#LLS) Resistance index (SI)

Min. Max. Spread (std. dev.)

(a) Resistance (sensitivity index)

1. Piedmont (37) 1.017 0.978 1.112 0.026

2. Val d’Aosta (3) 0.998 0.969 1.032 0.032

3. Lombardy (58) 1.015 0.964 1.072 0.021

4. Trentino Alto Adige (33) 1.036 0.995 1.105 0.022

5. Veneto (34) 1.000 0.946 1.049 0.031

6. Friuli Venezia Giulia (11) 0.985 0.947 1.012 0.019

7. Liguria (16) 0.990 0.944 1.029 0.029

8. Emilia Romagna (41) 1.003 0.965 1.073 0.023

9. Tuscany (53) 1.008 0.956 1.056 0.021

10. Umbria (17) 0.999 0.972 1.013 0.013

11. Marche (33) 1.009 0.964 1.078 0.028

12. Lazio (25) 1.031 0.976 1.078 0.024

13. Abruzzo (19) 0.972 0.885 1.017 0.035

14. Molise (9) 0.977 0.920 1.064 0.044

15. Campania (54) 0.963 0.901 1.030 0.028

16. Puglia (44) 0.976 0.922 1.045 0.030

17. Basilicata (19) 0.965 0.934 1.004 0.021

18. Calabria (58) 0.973 0.875 1.116 0.035

19. Sicilia (77) 0.995 0.924 1.151 0.040

20. Sardegna (45) 0.985 0.875 1.057 0.038

Region (#LLS) Recovery (in %) Min. Max. Spread (std. dev.)

(b) Recovery

1. Piedmont (37) 0.178 −4.486 3.643 1.935 2. Val d’Aosta (3) 0.700 −0.084 2.029 1.157 3. Lombardy (58) −1.305 −6.772 3.027 1.934 4. Trentino Alto Adige (33) 0.807 −2.792 11.145 2.408 5. Veneto (34) 1.825 −1.831 6.065 2.354 6. Friuli Venezia Giulia (11) 0.355 −1.587 3.903 2.003 7. Liguria (16) 1.917 −3.002 7.295 3.295 8. Emilia Romagna (41) 0.398 −4.435 5.044 2.260 9. Tuscany (53) −1.056 −5.596 4.320 2.348 10. Umbria (17) 0.286 −1.407 2.837 1.185 11. Marche (33) −1.264 −3.990 2.017 1.462 12. Lazio (25) 0.246 −4.329 3.364 1.836 13. Abruzzo (19) 2.801 −2.036 8.631 2.690

123

Regional economic resilience: the experience of the Italian... 403

Table 1 continued

Region (#LLS) Recovery (in %) Min. Max. Spread (std. dev.)

14. Molise (9) −1.659 −5.142 1.697 2.273 15. Campania (54) −2.068 −11.661 6.987 4.236 16. Puglia (44) −0.257 −8.546 9.060 4.095 17. Basilicata (19) −0.188 −6.694 6.073 3.351 18. Calabria (58) −1.481 −11.088 10.287 3.439 19. Sicilia (77) −0.301 −11.393 10.919 3.607 20. Sardegna (45) 2.177 −1.762 6.182 2.156

resistance and recovery indexes by region including the minimum and maximum val- ues and a measure of variability (standard deviation). As traditionally done by ISTAT, the regions are ordered from north to south.

Two trends are clear from looking at the numbers in Table 1. First—and not surprisingly—there is a marked North/South divide in Italy. All the southern regions showed a lower level of resistance to the recession, with a sensitivity index below unity. The center—sometimes referred to as “Third Italy” and well known for its productive structure organized around industrial districts and SMEs—was the most resistant, together with the largest northern regions (Lombardy, Piedmont, Veneto, Trentino and Emilia). The same result applies to the recovery phase, although with some differences. Abruzzo, for example, although vulnerable during the crisis, shows a clear positive path in the recovery phase. However, Abruzzo should be considered an outlier because the effects of the recession were compounded with the effects of an earthquake in April 2009, which destroyed most of its main city, L’Aquila and its surrounding areas. The faster recovery of the north is evident with all the regions, save Lombardy, displaying positive employment growth rates.

Second, every region displays a significant degree of heterogeneity and this het- erogeneity is more pronounced the more southern the regions are. For instance, the recovery value of Sicily is −0.301%, which is close to zero growth. However, if we look at the distribution of values for the 77 LLSs contained in Sicily, the variation is quite large, ranging from −11.393% to +10.919%. This is the type of heterogeneity that is not captured by simply looking at regional results.

4.2 Classification of the LLSs and multinomial logit model results

In the second step of our analysis, we used our measures of resistance and recovery to classify the LLSs into the four groups described in Sect. 3. Figure 3 shows a scatterplot with the results. The 686 SLLs are almost equally sub-divided in the four quadrants with no clear correlation (neither positive nor negative) between resistance and recovery. However, if we restrict to LLSs classified by ISTAT as being part of industrial districts, the picture looks a bit different, with the largest proportion of these LLSs (36.54%) belonging to Quadrant II (high resistance/slow recovery). The process of lock-in typical of traditional industrial districts might have been a factor in

123

404 A. Faggian et al.

Fig. 3 Scatterplot of resistance versus recovery—all SLLs (n = 686)

explaining why, although less vulnerable initially, they found it difficult to go back to their long-run growth path, once hit.

Although the scatterplot is useful to look at the possible correlation between recov- ery and resistance and to have an idea of how the LLSs are distributing into the four quadrants, it is not very informative on the geographical distribution of the obser- vations. As done with the measures of resistance and recovery, we mapped the four groups in Fig. 4.

Quadrant I (in white) represents the best performing LLSs (high resistance/fast recovery) as opposed to the worst performers of Quadrant III (in maroon). Quadrants II (blue) and IV (green) are the intermediate cases where LLSs were good in one dimension but not the other. The South of Italy is predominantly in Quadrants III or IV, meaning that the real issue was low resistance to the recession. The LLSs in the north and center of Italy are more likely to be in Quadrants I or II (where an initial resistance was then followed by post-recession problems), although there are also some pockets of LLSs belonging to the worst performing group (III).

Given the rather complex distribution of LLSs into the different groups, it is difficult to point at just one element responsible for their resistance and recovery. Although some North/South patterns are visible, this is not clearly the whole story. To better identify the contribution of different LLS characteristics to their probability of belong- ing to each of the four groups, we run a multinomial logit model shown in Eq. (2). To mitigate possible endogeneity problems, we lagged the explanatory variables by one year. Although we are aware that this does not completely solve the issue, it is a first step to avoid possible sources of endogeneity such as reverse causality.4 We are less concerned about omitted variable issues, such as differences in regional governance or culture, because of the inclusion of regional fixed effects.

4 While deep lags would be ideal, because of the shifting LLS boundaries, this is not possible without intensive manual aggregation of the data to reconstruct consistent areas.

123

Regional economic resilience: the experience of the Italian... 405

Fig. 4 Distribution of the Italian LLSs into the four quadrants

Table 2 presents the results of our multinomial logit model expressed in odds ratios.5

An odd ratio bigger than one means that the explanatory variable makes it more likely for an observation (LLS) to belong to a certain group (quadrant, in our case).

The first thing we notice in Table 2 is that the most urban and largely populated LLSs in Italy were not the most successful. In fact, the areas most likely to have a higher resistance and a faster recovery were LLSs with a population between 50,001 and 100,000, followed by LLSs with a population between 100,001 and 500,000. These results are consistent with other studies. Dijkstra et al. (2015) found exactly the same result for Europe, concluding that “both rural remote and urban regions were more vulnerable to the crisis than the intermediate and rural regions close to a city” (p. 935).

Aside from the size, one variable strongly correlated with the success of an area is its touristic vocation. LLSs defined by ISTAT as being “touristic” were over ten times more likely to be in the most consistently successful group (Quadrant I) than the consistently below average one (Quadrant III). This is potentially surprising, since travel represents a luxury good that should be elastic to changes in wealth during a

5 The Hausman test for IIA supports the IIA hypothesis for all categories. Although a variance inflation factor (VIF) cannot be calculated for a nonlinear model, we ran an OLS model and test for multicollinearity of the regressors. None of the regressors was multicollinear—with a maximum VIF of 2.75 (for one of the regional fixed effects) and an average of 1.69.

123

406 A. Faggian et al.

Table 2 Multinomial logit model results (odds ratios)—base category: Quadrant III

Variables Quadrant I (strong in both)

Quadrant II (strong res.)

Quadrant IV (strong recovery)

Urban areas highly specialized

1.90 (3.55) 1.31e−08 (0.00) 0.65 (1.08)

Urban areas moderately specialized

4.44 (4.30) 1.13 (1.09) 0.62 (0.65)

Urban ports 2.23 (1.84) 2.06 (1.75) 1.65 (1.16)

Touristic vocation 10.83*** (6.29) 2.47 (1.47) 1.78 (0.93)

Agricultural vocation 4.13** (2.72) 1.36 (1.06) 1.73 (1.06)

Districts

Leather 0.60 (0.68) 0.13** (0.14) 0.19 (0.24)

Shoes 1.07 (0.93) 0.70** (0.08) 1.07 (0.73)

Textile 6.98* (6.94) 3.14 (3.31) 3.02 (2.82)

Clothing 2.36 (1.69) 0.90 (0.54) 1.57 (0.83)

Furniture 3.67 (3.45) 4.56* (3.88) 2.65 (2.19)

Glasses 0.28 (0.22) 0.89 (4.01e10) 1.27e14 (6.03e17)

Machinery 1.57 (1.33) 0.48 (0.38) 0.55 (0.44)

Food 3.10* (2.06) 1.13 (0.70) 0.77 (0.47)

Metals 0.76 (0.78) 0.17* (0.16) 0.29 (0.30)

Transportation 0.16 (0.21) 0.58 (0.54) 0.75 (0.62)

Construction 0.73 (1.05) 0.87 (1.12) 0.43 (0.58)

Chemical and oil 1.89 (1.93) 1.79 (1.71) 0.67 (0.69)

Size

10,001–50,000 2.55** (1.11) 5.24*** (2.36) 3.35*** (1.26)

50,001–100,000 12.82*** (6.97) 18.29*** (10.05) 6.68*** (3.22)

100,001–500,000 5.78*** (3.30) 3.28** (1.96) 3.49** (1.77)

Above 500,000 6.41 (8.15) 3.88 (4.77) 3.42 (3.86)

Regional fixed effects Yes

*, **, *** Significant at 0.10, 0.05 and 0.01 level, respectively

Obs (n) 686

McFadden R2 0.2902

recession. However, this might represent access to foreign capital in a country with a significant portion of international visitors.

Having an agricultural designation was also a positive factor. This may follow the first half of the conventional wisdom above, with food being a very inelastic good.

Among the industrial districts, the best performing ones were in two very traditional Italian sectors, i.e., food and textile. These are the only industrial districts with signif- icant results for the best group, although there is evidence that the furniture industry performed well in resistance, if not in recovery. However, these results are only signif- icant at the 10% level. Despite the excellent performance of the textile areas, leather

123

Regional economic resilience: the experience of the Italian... 407

Fig. 5 Significant regional fixed effects for Group I (negative in red; positive in green) (color figure online)

and shoes seem particularly weak during the recession, although this effect was not significant when applied looking at recovery.

Overall the model performs well with a value of the McFadden pseudo-R2 of 0.29. Values of 0.2–0.4 for represent an excellent fit (McFadden 1978). Figure 5 shows the regions that had significant regional fixed effects.

Figure 5 shows the regions that had significant regional fixed effects. Most of the regions were not significant, meaning the model would have performed well had we not included these fixed effects. However, most of the regions in the south have a negative and significant regional fixed effect, meaning that they under-perform compared to the model predictions. Lazio, on the opposite, is the only region with a positive and significant regional fixed effect meaning it did better than what the model would predict. One possible reason for the over-performance of Lazio might be the large public and governmental sectors in Rome (its capital). However, with our current data (and the change in the definitions of LLSs) it was impossible to create a location quotient to test formally this hypothesis. The role of the different specializations of the LLSs requires a deeper analysis and a separate paper, where the issue of endogeneity needs to take central stage (e.g., by using instrumental variables or deep lags, which imply some data challenges at LLS level).

5 Concluding remarks and future challenges

Although much still needs to be done, our paper provides some initial insights into the resilience of the Italian local labor systems to the recessionary wave of 2009–2010.

123

408 A. Faggian et al.

What clearly emerges from our contribution is the heterogeneity of responses of the different LLSs even within the same regional context. Different factors are responsible for this heterogeneity, which seems to be more than just an urban versus rural story.

Not surprisingly the North/South divide in Italy played a role in the level of resis- tance and recovery of the different LLSs, but the overall picture is much more complex than expected. Among the other variables, which played a fundamental role in defining the resilience of an area were their industrial vocation and population size. Tourism was a positive factor in helping face the recession, as was belonging to an industrial district in two of the most traditional Italian sectors, food and textile. Medium-size LLSs were much more likely to be simultaneously more resistant and to recover faster than larger (or smaller) LLSs. Although these areas have enough of a critical mass to benefit from a certain degree of agglomeration economies, they are possibly more responsive than larger cities. Our results add to the recent debate in Europe about the dynamism of medium-sized cities (often referred to also as second-tier cities). Parkin- son et al. (2015) discuss the contribution that second-tier cities can and do make to the economic performance of national economies across Europe and argue that the Euro- pean Commission should do more to ensure its strategies help realize the economic potential of second-tier cities in future. Camagni and Capello (2015) point out that, especially in the recent economic downturn, second-ranked cities have outperformed large ones.

Building on this first contribution, there are a series of possible extensions and research avenues for the future. First, the analysis could be extended to a longer time series. Although it is very challenging to create long time series at LLS level because of their dynamic nature, some data are available at municipality level and could be aggregated to form “quasi-LLSs” consistent over time. This is obviously not an easy task, which requires a proper reflection on the appropriate definition for this “quasi-LLSs” across time. Second, we restricted our analysis to the two initial aspects of resilience as defined by Martin (2012), i.e., resistance and recovery. However, reorientation and renewal are also important, especially if we marry the “adaptive” view of regional resilience. However, the study of reorientation and renewal is non- trivial as it cannot be done with data at aggregate level. Data on individual firms are necessary to properly study these phenomena (Duschl 2014). Third, given some of our results such as the positive fixed effect of the Lazio region, the role of certain sectors in Italy (such as government and public sector) needs to be explored further.

Appendix

See Table 3.

123

Regional economic resilience: the experience of the Italian... 409

Table 3 Summary statistics

Variables Mean Std. dev. Min. Max.

Population 83,084.17 222,418 2956 3,374,511

Urban areas highly specialized 0.0058309 0.076193 0 1

Urban areas moderately specialized 0.0422741 0.2013605 0 1

Urban ports 0.0379009 0.1910959 0 1

Touristic vocation 0.1195335 0.324652 0 1

Agricultural vocation 0.0349854 0.1838769 0 1

Districts

Leather 0.016035 0.1257016 0 1

Shoes 0.03207 0.1763145 0 1

Textile 0.0262391 0.1599621 0 1

Clothing 0.0714286 0.2577273 0 1

Furniture 0.0408163 0.1980089 0 1

Glasses 0.0116618 0.1074367 0 1

Machinery 0.0510204 0.2201999 0 1

Food 0.0889213 0.2848378 0 1

Metals 0.0204082 0.1414951 0 1

Transportation 0.0233236 0.1510393 0 1

Construction 0.0102041 0.1005719 0 1

Chemical and oil 0.0276968 0.1642224 0 1

References

Anderies JM (2015) Managing variance: key policy challenges for the Anthropocene. Proc Natl Acad Sci USA 112(47):14402–14403

Anderies JM, Folke C, Walker B, Ostrom E (2013) Aligning key concepts for global change policy: robust- ness, resilience, and sustainability. Ecol Soc 18(2):8

Augustine N, Wolman H, Wial H et al (2013) Regional economic capacity, economic shocks and economic resilience. MacArthur foundation network on building resilient regions. Working paper

Balland PA, Rigby D, Boschma R (2015) The technological resilience of US cities. Camb J Reg Econ Soc 8(2):167–184

Boschma R (2015) Towards an evolutionary perspective on regional resilience. Reg Stud 49:733–751 Camagni R, Capello R (2015) Second-rank city dynamics: theoretical interpretations behind their growth

potentials. Eur Plan Stud 23(6):1041–1053 Carpenter A (2015) Resilience in the social and physical realms: lessons from the gulf coast. Int J Disaster

Risk Reduct Part 3(14):290–301 Carpenter S, Walker B, Anderies JM, Abel M (2001) From metaphor to measurement: resilience of what

to what? Ecosystems 4(8):765–781 Cellini R, Torrisi G (2014) Regional resilience in Italy: a very long-run analysis. Reg Stud 48(11):1779–1796 Coombes M, Casado-Dìaz JM, Martinez-Bernabeu L, Carausu F (2012) Study on comparable Labour

Market Areas, EUROSTAT Cutter SL, Barnes L, Berry M, Burton C, Evans E, Tate E, Webb J (2008) A place-based model for

understanding community resilience to natural disasters. Glob Environ Change 18(4):598–606 Davoudi S, Porter L (2012) Resilience: a bridging concept or a dead end? Plan Theory Pract 13(2):299–307 Di Caro P (2017) Testing and explaining economic resilience with an application to Italian regions. Papers

in Regional Science 96:93–113. doi:10.1111/pirs.12168

123

410 A. Faggian et al.

Dijkstra L, Garcilazo E, McCann P (2015) The effects of the global financial crisis on European regions and cities. J Econ Geogr 15:935–949

Duschl M (2014) Regional resilience and fat tails: a stochastic analysis of firm growth rate distributions of German regions. Marburg geography, working paper on innovation and space, 1.14

Fingleton B, Garretsen H, Martin R (2012) Recessionary shocks and regional employment. J Reg Sci 52:109–133

Foster K, Pendall R, Cowell M (2010) Regional resilience: building understanding of the metaphor. Camb J Reg Econ Soc 3(1):71–84

Hassink R (2010) Regional resilience: a promising concept to explain differences in regional economic adaptability? Camb J Reg Econ Soc 3(1):45–58

Hill EW, Wial H, Wolman H (2008) Exploring regional economic resilience. Working paper 2008–04, Institute of Urban and Regional Development, Berkeley

Hill E, St Clair T, Wial H, Wolman H, Atkins P, Blumenthal P, Ficenec S, Friedhoff A (2012) Economic shocks and regional economic resilience. Build Resil Reg Urban Reg Policy Eff 4:193–274

Holling CS (1973) Resilience and stability of ecological system. Annu Rev Ecol Syst 4:1–23 Irwin E, Campbell J, Wilson R, Faggian A, Moore R, Irwin N (2016) Human adaptations in food, energy,

and water systems. J Environ Stud Sci 6(1):127–139 ISTAT (1997) I sistemi locali del lavoro 1991, Roma, pp 19–26 ISTAT (2005) I sistemi locali del lavoro. Censimento 2001. Dati definitivi, Roma. http://dawinci.istat.it/

MD/download/sll_comunic_solo_testo.pdf Jara B, Faggian A (forthcoming) Labor market resilience and re-orientation in disaster scenarios In: Baycan

T, Pinto H (eds) Regional resilience. Springer Lagravinese R (2015) Economic crisis and rising gaps North–South: evidence from the Italian regions.

Camb J Reg Econ Soc 8:331–342 MacKinnon D, Derickson KD (2013) From resilience to resourcefulness: a critique of resilience policy and

activism. Prog Hum Geogr 37(2):253–270 Mameli F, Faggian A, McCann P (2008) Employment growth in italian local labour systems: issues of

model specification and sectoral aggregation. Spat Econ Anal 3(3):343–360 Martin R (2012) Regional economic resilience, hysteresis and recessionary shocks. J Econ Geogr 12:1–32 Martin R, Sunley P (2015) On the notion of regional economic resilience: conceptualization and explanation.

J Econ Geogr 15(1):1–42 McFadden D (1978) Quantitative methods for analyzing travel behaviour on individuals: some recent

developments. In: Hensher D, Stopher P (eds) Behavioural travel modelling. Croom Helm London, London, pp 279–318

Metcalfe JS, Foster J, Ramlogan R (2006) Adaptive economic growth. Camb J Econ 30(1):7–32 Modica M, Reggiani A (2014) An alternative interpretation of regional resilience: evidence from Italy. In:

ERSA conference papers (no. ersa14p369). European Regional Science Association Mollenkopf J, Pastor M (2013) Struggling over strangers or receiving with resilience? The metropolitics

of immigrant integration. Working paper by the MacArthur foundation network on building resilient regions. University of California, Berkeley

Parkinson M, Meegan R, Karecha J (2015) City size and economic performance: is bigger better, small more beautiful or middling marvellous? Eur Plan Stud 23(6):1054–1068

Reggiani A, de Graaff T, Nijkamp P (2002) Resilience: an evolutionary approach to spatial economic systems. Netw Spat Econ 2:211–229

Rose A, Liao S-Y (2005) Modeling regional economic resilience to disasters: a computable general equi- librium analysis of water service disruptions. J Reg Sci 45:75–112

Sherrieb K, Norris H, Galea S (2010) Measuring capacities for community resilience. Soc Indic Res 99(2):227–247

Swanstrom T (2008) Regional resilience: a critical examination of the ecological framework. Institute of Urban and Regional Development, Berkeley

123