Portal de Periódicos da CAPES

Economic resilience and regionally differentiated cycles: Evidence from a turning point approach in Italy

2023; Elsevier BV; Volume: 102; Issue: 2 Linguagem: Inglês

10.1111/pirs.12725

1435-5957

Hasan Engìn Duran, Ugo Fratesi,

Regional Economics and Spatial Analysis

The literature on regional resilience often neglects the timing of recessions and simply uses national cycles. Region-specific cycles and turning points might bias the results, however, and affect the choice of regions to target with policies. This paper investigates the geography and determinants of regional resilience with a regional turning point approach, using data for Italy, a country with a well-known and sizeable regional divide. The results show that the timing of regional cycles varies substantially and that the detected resilience determinants are different across the two approaches, implying that the policy levers may be wrongly estimated with national turning points. La bibliografía sobre la resiliencia regional suele pasar por alto el momento de las recesiones y se limita a utilizar los ciclos nacionales. Sin embargo, los ciclos y los puntos de inflexión específicos de las regiones podrían sesgar los resultados y afectar a la elección de las regiones a las que dirigir las políticas. Este artículo investiga la geografía y los determinantes de la resiliencia regional con un enfoque de punto de inflexión regional, para lo cuál emplea datos de Italia, un país con una división regional bien conocida y considerable. Los resultados muestran que el momento de los ciclos regionales varía sustancialmente y que los determinantes de la resiliencia detectados se pueden distinguir en los dos enfoques, lo que implica que las palancas políticas pueden estimarse erróneamente si se emplean puntos de inflexión nacionales. 地域のレジリエンスに関する研究は、経済不況のタイミングを無視し、単純に国の景気サイクルを使用していることが多い。ただし、地域特有のサイクルや転換点が偏った結果を生み、対象とする地域の選択に影響を与える可能性があります。この論文では、大きな地域格差があることでよく知られているイタリアのデータを使用し、地域の転換点に対するアプローチを用いて地域の回復力の地理と決定要因を調査した。この結果は、地域のサイクルのタイミングが大幅に異なり、検出された回復力の決定要因が2つのアプローチで区別されることを示しており、国の転換点により政策レバーが誤って推定される可能性があることを示唆している。 Resilience is one of the most desirable phenomena in engineering, environmental studies, and social sciences, including economics and regional science. From an engineering standpoint, resilience refers to the resistance of an economic unit to unanticipated negative movements and the ability to achieve the pre-shock situation. The ecological perspective defines resilience rather as the long-term ability to stand robustly against the shocks and to absorb them adequately while not moving to a new equilibrium. Alternatively, an evolutionary or adaptive approach conceives of resilience as the capacity of the systems to reorient, reconfigure, and transform their structure to reach a superior and sustainable long-run equilibrium path (Boschma, 2015; Di Caro, 2017; Di Caro & Fratesi, 2018; Faggian et al., 2018; Fingleton et al., 2012; Folke et al., 2004; Giannakis & Bruggeman, 2015, 2017; Han & Goetz, 2015; Martin, 2012; Martin & Gardiner, 2019; Martin & Sunley, 2015; Modica & Reggiani, 2015; Pike et al., 2010; Simmie & Martin, 2010). These definitions introduce three components of resilience: first, resistance of economies to shocks; second, speed of recovery from recessions; and, third, adaptability of economies by reorienting their economic and industrial structure to maintain a new long-run growth path (Bristow & Healy, 2018; Di Caro, 2017; Faggian et al., 2018; Fingleton et al., 2012; Folke, 2006; Folke et al., 2004; Foster, 2007; Fratesi & Perucca, 2018; Giannakis & Bruggeman, 2015, 2017; Han & Goetz, 2015; Hill et al., 2008, 2011; Martin, 2012; Pike et al., 2010). The analyses of regional resilience may be biased, however, because almost all the literature so far analyses the different behaviours of regions by assuming that the regions are affected by the same shocks and follow the same national economic cycles, which implies that resilience is measured using the statistical national peaks (Faggian et al., 2018; Giannakis & Bruggeman, 2015, 2017; Martin, 2012; Sedita et al., 2017). The purpose of this study is to address an important shortcoming of the literature. Regions may have different cycles, and the fact that they belong to the same country and share the same currency and macroeconomic policy is not a sufficient condition for full cycle synchronization, as the literature on the syncing of national cycles following the start of the Euro shows (Camacho et al., 2006). In the academic literature, the timing of the crisis for regions is, however, mostly defined on the basis of the national business cycle. The possibility of significant region-specific cycles and differential turning points is largely ignored, because these are difficult to detect with annual data, which is what is normally available at the regional level. However, failing to relax the assumption of synced regional cycles might seriously bias the results, as many regions might exhibit quite different timings from the national cycle. Indeed, a stream of scholars has explored large asymmetries across business cycles. Focusing mostly on the Eurozone, they have detected relatively heterogeneous business cycles (Montoya & De Haan, 2008). This paper is intended to show the importance of pursuing resilience analyses on the basis of turning points specific to regions, but not to the national turning points, which may often be biased. The paper seeks to show explicitly the extent of the bias that occurs under national turning points and to show that this bias has important policy implications: (i) using national turning points might lead to the selection of some regions as being most affected by a crisis while others might be in a more severe recession; and (ii) using national instead of regional turning points can lead to wrongly estimating the determinants of regional resilience, resulting in the choice of less effective regional policy levers. To the best of our knowledge, the implications of this shortcoming have never been addressed in the literature, and only a few papers have provided evidence of it. Indeed, only a small number of papers have focused on regional turning points, most of which focus on concordance, diffusion, and synchronization of the regional cycles with each other and with the national business cycle. Owyang et al. (2005), Hall and McDermott (2004), Duran (2014), and Magrini et al. (2013) represent some examples of this stream of scholars who analyse states in the US (except Hall & McDermott, 2004, who focus on regions in New Zealand). Only Sensier et al. (2016) have conducted an analysis for Europe, finding that regional cycles in Europe are inhomogeneous, although their analysis is constrained by annual data and does not look to region-specific determinants, nor specific policy implications. Some of this stream also study the underlying economic or spatial determinants of the bilateral synchronization of fluctuations (Magrini et al., 2013). With regard to the structure of the paper, Section 2 presents an account of the existing studies. In Section 3, the aggregate and region-specific turning points are estimated using an adaption of Bry and Boschan's (1971) algorithm. The differences in regional and national turning points are presented in detail. In Section 4, the extent of the asymmetries and asynchronization across regional business cycles are shown with the help of bilateral cycle correlations, diffusion, and concordance indexes. In Section 5, regional resilience scores are computed using both national and regional turning points. The results are presented comparatively with illustrative graphs. To complement these comparisons, some statistical evidence of differences is provided. Section 6 investigates the economic and demographic determinants of regional resilience patterns using three-stage least squares (3SLS) regressions, and they are shown to differ under different turning points. Finally, Section 7 concludes the study by illustrating the relevance of the work for the choice of the regions to be assisted and the levers that must be selected for policies. The issue of resilience has been thoroughly and heatedly debated in the economic literature. It is considered politically crucial, as systems lacking resilience suffer seriously high unemployment, welfare losses, poverty, in-utilization of the productive labour base and other related socioeconomic inadequacies (Fingleton et al., 2012). Resilience is also important within countries, as different parts of the same country can show different rates of resilience. This was particularly evident after the financial crisis of 2007–2008, after which the term became widespread in the analysis of regional economies. Empirically, the literature has often focused on estimating the resilience levels of regions and exploring the geographical patterns. Some examples of these studies are Fingleton et al. (2012), who analysed the UK regions; Giannakis and Bruggeman (2017), who focused on 268 EU NUTS 2 regions over the period 2008–2013; Cellini and Torrisi (2014), who analysed Italian regions; Faggian et al. (2018), who focused on Italian local labor system over the period 2007–2011; Han and Goetz (2015), who studied the resilience across 3,138 US counties over the 2003–2014 period; Di Caro (2017), who analysed 20 NUTS 2 Italian regions over the period 1992–2012; Fratesi and Perucca (2018), who analysed the different patterns of resilience of NUTS 3 regions in Europe depending on territorial capital endowment; and Eraydin (2016)), who focused on the NUTS 2 Turkish regions during the 1987–2001 period. A general finding of this stream of research is the evidence of heterogeneity across regional resilience depending on their industrial structure, sectoral specialization, diversification, productive capacity, innovativeness, level of human and social capital, economic openness, and urban or rural setting, among other factors (Capello et al., 2015; Crescenzi & Rodríguez-Pose, 2011; Di Caro, 2017; Di Caro & Fratesi, 2018; Eraydin, 2013, 2016; Groot et al., 2011; Lagravinese, 2015; Ubago Martínez et al., 2019; van Bergeijk et al., 2017; van den Berg & Jaarsma, 2017). Moreover, there is new evidence that regional resilience is often different in different crises (Di Caro & Fratesi, 2022). It is important to analyse the regional economic resilience patterns (see Sutton & Arku, 2022a, 2022b; Sutton et al., 2022). From a policy standpoint, it is essential to identify which regions are more resistant or characterized with high adaptive capacity and recovery. This information will be critical to determine the place-specific policies and related necessities (Fratesi & Perucca, 2019). The problem of regional resilience has become topical again after the start of the global COVID-19 pandemic. This has been shown to have different effects on regions within the same country, with processes of spatial diffusion between one region and the other (Bloise & Tancioni, 2021; Bourdin et al., 2021; Paez et al., 2021). COVID-19 has caused not only a shock to human-health, but also it generated a socio-economic crisis which, even if international by definition, has had different effects on different countries and the different regions inside each country, in terms of magnitude, social groups affected and, especially relevant for this paper, timing (Crossley et al., 2021; Gong et al., 2020). In the context of resilience, however, far fewer studies have used regional turning points. One such study was conducted by Sensier et al. (2016), who analysed the European regions. They estimated the turning points so as to determine the resilience patterns; however, they did not show explicitly how seriously the results may change under national and regional turning points, as the current paper proposes to do. Moreover, they used annual data, which masks many fluctuations. The current paper uses quarterly data, which is known to provide more accurate turning points. Another study that uses the local turning points was carried out by Ringwood et al. (2019). They estimated the turning points (peaks) of US counties to calculate the degree of resilience to and recovery from the 2007–2009 global financial crisis (GFC). Again, while their approach is comprehensive, they do not compare the outcomes under national and local turning points, and they are uninterested in the consequences these outcomes might have for regional policy-making. Regional business cycles are important and may exhibit far different evolutions from the national one. A number of researchers have found significant differences across the turning points of US regional economies (Duran, 2014; Magrini et al., 2013; Owyang et al., 2005). The differences across regional cycles as well as between the regional and national one (a-synchronicities) are often attributed to large dissimilarities in industrial structure (Krugman, 1991), inadequate financial and trade linkages (Frankel & Rose, 1998; Imbs, 2004; Kalemli-Ozcan et al., 2001), and the differences in labour market characteristics (Duran & Ferreira-Lopes, 2017). Thus, to avoid biased or misleading results in resilience analysis, one should refer to the regional pattern of business cycles rather than the national business cycle. The case study chosen for this paper is Italy. It is a country of wide and long-standing regional disparities (Capello, 2016). Important policy efforts have been made to reduce the divide, with limited results (Dunford, 2002). Italy has also been investigated in a number of regional resilience analyses (Cainelli et al., 2019; Cellini & Torrisi, 2014; Di Caro, 2015, 2017; Terzo, 2021). Moreover, the country was seriously affected both by the public finance crisis that followed the GFC of 2007–2008 (Moro & Beker, 2016) and by the first wave of the pandemic in 2020 (Ascani et al., 2021). The following sections will show that using regional turning points instead of national ones produces two important differences. The first is that the ordering of the regions most affected by an economic crisis is changed, meaning that policy initiatives to counteract crises will need to target different regions. The second is that the determinant of regional resilience may also be different, and so the policy targets may need to be changed. To illustrate the conceptual need for a turning point approach, it is helpful to illustrate it with an example. In Figure 1, the cycles are represented for Italy and three of its regions. The three regions are large and of similar size but show different patterns. To perform the analysis, quarterly employment data were used. The data were collected over the period 1999:1–2020:2. 1 Two variables are generally used in the literature to measure resilience, namely GDP and employment. In this case, employment is focused on for two reasons. Conceptually, losses of employment are more linked with the actual suffering of regional inhabitants during a crisis (Fratesi & Rodríguez-Pose, 2016). Moreover, regional employment data are available quarterly, while GDP data are only available annually, which makes it hard to use GDP time-series techniques to detect cycles. As Figure 1 shows, the national pattern of employment growth was positive until 2008-Q3, after which the financial crisis took its toll on the national economy and employment declined sharply. Between 2012 and 2014, another important loss of employment took place after the economic crisis involved the sovereign debt of the country, which required heavy restructuring of the national financial system. From the first quarter of 2014, however, a new period of employment creation took place, which led to a peak just before the start of the pandemic. The national pattern was not followed by all regions, however. As Figure 1 shows, Veneto, a manufacturing region in the north of the country, experienced a similar pattern to the nation as a whole. The second region, Campania, is a populous region in the South. The cycle of this region is different, as the peak in total employment was reached in 2004, and this region was already in a sharp decline pattern when the national crisis of 2008 started. After 2014, a growth pattern began, but it halted well before the pandemic in 2018. The third regional case is Lazio, a populous region in the centre of the country, which hosts the national capital, Rome. In this case, there has not been any real employment crisis, only periods of faster and slower growth; the wider economic crisis did not bring any decrease but just a temporary slowing down, consistent with the expectations for a region with large shares of public employment (Rodriguez-Pose & Fratesi, 2007). The issue cannot be treated anecdotally. An initial fundamental analysis is needed to estimate the regional and aggregate level turning points. A number of studies have tried to identify the phases of business cycles and turning points in national economies or common currency zones (Artis et al., 2004; Harding & Pagan, 2003; Magrini et al., 2013). However, far fewer studies have estimated the turning points for regional economies (the notable exceptions being Hall & McDermott, 2004; Owyang et al., 2005; Sensier et al., 2016). This lack of research was a motivation for the present study, even before coming to the analysis of resilience of the following sections. There are two main, widely accepted methods for turning point detection. The first is a monthly algorithm proposed by Bry and Boschan (1971). It is designed to replicate the NBER's official monthly turning points for the US economy. Harding and Pagan (2003) converted this procedure to quarterly data (Harding & Pagan, 2002). The second method is Hamilton's Markovian regime-switching MSVAR model (Hamilton, 1989; Krolzig, 2001). It is assumed that growth varies across two different phases of the business cycle. The calculated probabilities of regimes release the turning points. Although the more recent method is more in accordance with the data-generating process, the earlier one has been shown to provide accurate results (Bry & Boschan, 1971; Hamilton, 1989; Harding & Pagan, 2003; Krolzig, 2001; Owyang et al., 2005). The former method is thus employed in this study due its simplicity and intuitiveness. In principle, the Bry–Boschan routine searches for a local set of minima and maxima in the business cycle series while imposing certain limitations on the phase and cycle length (Duran, 2011). There are many intermediate steps, but the main procedure is as follows. The algorithm searches for sets of local minima and maxima in every five quarters (window length; Bry & Boschan, 1971; Duran, 2011, 2014). It imposes the restrictions that the minimum phase duration is at least two quarters and the minimum length of the cycle is at least five quarters (Bry & Boschan, 1971; Duran, 2011, 2014; Duran & Ferreira-Lopes, 2017; Harding & Pagan, 2002, 2003). At the final stage, the points identified in the first and last two observations are discarded and the procedure is completed. The turning points of the Italian national economy and 20 NUTS 2 regions are estimated using quarterly employment data (seasonally adjusted and in natural logarithms). The results for the national economy are presented in Table 1. As a benchmark, the estimated dates are compared to those declared by the OECD's report. 2 The OECD uses GDP as a reference series and the Bry and Boschan (1971) method. The results indicate that the two chronologies are mostly consistent with each other. The BBQ algorithm matches all turning points estimated by the OECD while detecting five more additional turning points. There are moderate disparities between the dates. These differences seem plausible, since the OECD uses GDP as a reference variable, which might have different cyclical properties than employment. The economic crises during 2008–2010 and 2011–2013 are detected. GFC was triggered by the failure of mortgage markets in the US (McKibbin & Stoeckel, 2010). The overvaluation of mortgage-backed assets created an artificial bubble. The downturn in expectations led to a collapse of the financial system, and so a serious drop was observed in GDP and employment in many countries. Sovereign debt crises were observed, particularly in Southern European economies, and Italy is among the countries heavily hit by both crises. A similar exercise in turning point estimation is performed for 20 NUTS 2 regions using the same parameters as in the national case. The results are presented in Table 2. A heterogeneous timing of recessions across the regions is observed, including diverse timing of peaks and troughs and also many different idiosyncratic recessions and expansions. Hence, these results provide a visualization of the low synchronization between the regions. The extent of differences in timing across regional business cycles is an important factor in this study (Duran, 2011). The presence of asynchronous cycles may imply the necessity of region-specific calculations of economic resilience indicators. Studies on business cycle co-movement often find an increasing but still low cycle synchronization across the Eurozone (Artis & Zhang, 1999; Montoya & De Haan, 2008). The lack of co-movement is usually attributed to industrial dissimilarity (Krugman, 1991), weak financial integration (Kalemli-Ozcan et al., 2001), inadequate trade ties (Frankel & Rose, 1998; Imbs, 2004), and differential institutional and labour market characteristics (Duran & Ferreira-Lopes, 2017). The business cycles in Italy show similarly substantial structural differences in economic oscillations across regions. Some examples of these studies are Mastromarco and Woitek (2007), focusing on the period 1951–2004, and Duran and Fratesi (2020), focusing on the period 1978–2016. This section aims to demonstrate the degree of asynchronization among regional business cycles using three types of statistical indicators. The first is the bilateral Pearson's correlation calculated using the first differences of employment (in natural logarithms). The results are documented in Table 3. The last two rows highlight the average and standard deviation of the bilateral correlations across regions. Although many high and low bilateral associations are observed, the level of correlation is low on average (0.14). This means that two regional cycles in Italy can scarcely move in a synchronous manner. This ratio is calculated for regions and presented in Table 4. The results show low concordance rates. The minimum concordant region is the Friuli-Venezia Giulia (FVG), with a score of 0.53, and the maximum concordant region is Veneto with a concordance rate of 0.78. The average rate is 0.66, which means that in 66% of the quarters, one region is in the same phase as the national economy, and in 34% of the quarters they are in different phases. Similar analyses have been made by various scholars. sectional regression model and Granville (2017) analysed European countries' business cycles over the period 1961–2013 and found that, most recently, the concordance index is 0.67 for the 1993–2002 period, 0.75 for 2003–2008, and 0.96 for 2009–2013. Similarly, focusing on concordance among US states, Duran (2014) found it to be 0.73 for 1993–2008, and Owyang et al. (2005) found it to be 0.8 for the period 1979–2002. Compared to these benchmark studies, the values in this study are lower, indicating once more the severity of heterogeneity across regional cycles and how seriously the results on economic resilience might be biased if national turning points are used instead of region-specific ones. In the regional resilience measurement, four different ratios are computed and summarized in Table 5. Resistance = 1- Sensitivity Index Sensitivity index= E peak − E through / E peak ) (Faggian et al., 2018; Fingleton et al., 2012; Giannakis & Bruggeman, 2015, 2017; Han & Goetz, 2015; Martin, 2012; Sensier & Artis, 2016; Sensier et al., 2016) E: Employment First, sensitivity measure (which represents 1-resistance) is defined by the percentage loss in regional employment at trough compared to the previous peak level (Faggian et al., 2018; Fingleton et al., 2012; Giannakis & Bruggeman, 2015, 2017; Han & Goetz, 2015; Martin, 2012; Sensier & Artis, 2016; Sensier et al., 2016). Second, the recovery index is defined by the period (quarter) length required to reach the previous peak level of regional employment. The index is defined in categories, since there are some regions which have not yet had a trough after the crisis in 2008. Third, adaptability is defined as the per quarter average growth rate of regional employment through to about 2013 until the most recent peak of the region. Finally, resistance, recovery, and adaptability scores are converted into relative values by applying (X-Min)/(Max-Min), where X is the value to be converted (Chhaochharia et al. 2020). Then the three relative indicators are chained by taking averages, and a composite resilience indicator is constructed. Details of these calculations are shown in Table 5. The two national recessions are assumed to be unique recessions. Hence, this study uses the regional peak levels corresponding to the start of the 2008–2009 crisis and the regional trough levels corresponding to the end of the 2011–2013 crisis The corresponding four resilience indicators are calculated using national and regional turning points. The results are plotted in Figures 3-5 (maps) and in Table 6 (for recovery, for which a map would be less informative). In regard to resistance (Figure 3), there is a relatively strong correspondence between the results obtained under national and regional turning points, with some exceptions, most notably Puglia and Basilicata. A North–South (Mezzogiorno) dualism is clearly observed, with regions in the North and Mezzogiorno seemingly more resistant to the crises. In other words, these regions experience a smaller decrease in employment between peak and trough levels. The most resistant regions are Lazio (resistance score: 1.00), Trentino Alto Adige (0.88), and Lombardia (0.88), which experience approximately only 0–2% loss in employment. In contrast, Southern regions and islands located in the periphery (Sicilia, Sardegna) are more vulnerable and less resistant to the crises. The three least resistant regions are Calabria (0.82), Molise (0.85), and Sicilia (0.87), with a high loss in employment (18%, 15%, and 13%, respectively). With respect to recovery (Table 6), quite different patterns are observed across the scores calculated under different turning point types. When national turning points are employed, no distinct geographical pattern is observed. However, once the region-specific turning points are used, Mezzogiorno-North–South dualism becomes more clear. Some Northern regions, especially Mezzogiorno, seem to be fast recovering zones, while most of the Southern regions are slow or not recovering. In detail, central and Northern regions such as Lazio, Liguria, Marche, Trentino-Alto Adige, and Valle D'Aosta recover fast in 0–10 quarters. In contrast, Southern regions such as Sicilia, Sardegna, Calabria, Puglia, Molise, and Abruzzo have not yet recovered. What is more striking is that there are five regions (four in the lagging Mezzogiorno) that seem to have had no recovery in terms of national turning points and that have experienced no recovery yet in terms of regional turning points. In regard to the adaptability patterns (Figure 4), the dualism is more obvious when regional turning points are used. Southern regions seem to be more adaptable to a high growth path. The faster growth pattern is observed in Molise, Sardegna, Calabria, Campania, and Lazio, whereas the slowest growth pattern is observed in Valle D'Aosta, Lombardia, Liguria, Sicilia, and Umbria. Moreover, some regions' ranking changes significantly, most notably Puglia, but also Umbria and Basilicata. Finally, composite resilience scores provide an overview (Figure 5). Sizable differences can be observed between the scores obtained under national and regional turning points. For the later ones, the regions in Mezzogiorno and the North East seem to have more resilience, whereas Southern regions are less resilient. The most resilient regions are Lazio, Trentino Alto Adige, and Marche, with resilience scores of 0.83, 0.79, and 0.64, respectively; the least resilient ones are Sicilia, Umbria, and Puglia, with resilience scores of 0.16, 0.19, and 0.26, respectively. These results deviate from those of previous studies on Italy in two important respects. First, previous empirical studies on Italian regions have found a North–South dualism. However, the analyses in this study point to some regions of Mezzogiorno as being more resilient than the average. Second, the analyses conducted under national turning points yield remarkably different results from those implemented under regional turning points, particularly in recovery but also moderately in adaptability and the composite resilience index. To support this result, several tests are applied to examine the significance of the differences: H1.The resilience score distributions computed by using national and regional turning points are identical. H2.The resilience score distributions computed by using national and regional turning points are not identical. Kolmogorov–Smirnov (KS) tests are applied along with bi-lateral correlations, and some descriptive statistics are provided (Kolmogorov, 1933; Smirnov, 1939, 1948). The results are presented in Table 7. The KS tests indicate significant differences in two resilience types (recovery and adaptability) between the results under the two different turning points. The correlation coefficients are also informative. In the resistance context, the results under the two different assumptions are almost perfectly correlated (0.9). However, this correlation is weak for recovery (0.32), low for the composite resilience index (0.45), and moderate for adaptability (0.6). This section has shown the importance of considering regional turning points, since the ranking