Trade liberalization and economic crises : a rocky marriage

(1)

rocky marriage

Author: Marco Oglietti

Student number: 12237302

Supervisor: Sebastian Krapohl

June 2019

Master thesis Political Science, track International Relations

(2)

1

1. Introduction

Over the recent years, news outlets and academic publications have witnessed the re-emergence of talks about what supposedly had become a relic of the past: protectionism and trade wars. With the explosion of the current wave of globalization during the mid-1980s, with an unprecedented surge of market and capital liberalizations, as well as the end of the Cold War and the birth of the WTO in 1995, the threat of protectionism seemed to have become a distant memory. Both the general public and academic debate, with the emergence of a new strain of literature centered around the so-called “Open Economy Politics” (OEP) (Lake, 2009), embraced an overly optimistic view with respect to the then state of the global trade regime and considered the current wave of economic liberalization irreversible. However, the recent political developments that emerged especially in advanced western democracies, in particular the election of US president Donald Trump and the success of both left- and right-wing populist parties, have been putting the stability of the status quo into question and marked a comeback of long-gone protectionist fears. Some scholars (Thompson and Vescera, 1992, Krapohl et al., 2018) had previously noted the presence of cyclical patterns of trade cooperation, implying that the current protectionist backlash may simply be marking the end of the current upward phase of economic liberalization. In addition, other better-known cyclical processes characterize developed economies; particularly, economists have long theorized the presence of “business cycles”, alternating phases of economic expansions to those of contraction (Minsky, 1994, Kindleberger and Aliber, 2001). Because of these apparently analogous patterns and of the presence of well-known historical precedents, in particular that of the break down of international trade after the Great Depression of 1929, debate ensued on whether worsening economic conditions lead to increased domestic market protection. In light of the above recent policy developments, such debate acquires significant social and political relevance, as the global economy recovered only recently from its most serious crisis since the end of the Second World War, i.e. the Great Recession of 2008-2009. It is thus possible that the current protectionist pressures are indeed a consequence of the latter, despite of the presence of an almost decade-long gap between the two.

The aim of this research project will therefore be to both actively contribute to this debate and explain the recent political developments by investigating the relationship between economic and financial crises and the probability of countries to engage in protectionist behavior. More precisely, the analysis will focus on the various channels and mechanisms by which worsening macroeconomic conditions

(4)

3

may be able to influence the trade policy of a country and what kind of differences there are between them. In short, the aim of the project can be effectively summarized by the following research question:

How do economic crises affect a country’s probability to adopt protectionist policies?

To answer to this question, we will draw upon the notions of a variety of fields of study, including trade theory, comparative politics, political economy and even behavioral studies. Results will then show that there are two mechanisms linking these two elements: first, a short-term one operating

through industry lobbying channels, which appears in the immediate aftermath of a crisis, and second, a long-term one related to labor market distortions and shifting policy preferences, of which the effects may manifest even after several years. While the former has lost importance in the current globalized world, despite having historically been the main source of post-crisis protectionist pressures, the latter has instead become much more preponderant and could pose a serious threat to the liberal trade regime. The paper will be structured in the following way: after having briefly summarized the literature and past research on this topic, we will define the structure of the theoretical framework on which the main hypotheses will be based, followed by a brief description of the research method and of the data and sample selection. The subsequent sections will then focus on the analysis itself by which the main hypotheses will be tested. Finally, after having discussed the main limitations of the analysis and some further remarks, we will conclude by briefly recapitulating our findings and the possible implications of this research.

2. Literature review

Due to the aforementioned structural similarities, scholars started to investigate the presence of any sort of correlation between trade cooperation and macroeconomic fluctuations. For instance, McKeown (1984) focused on how negative phases of business cycles may lead import-competing industries to gain political momentum and push for protectionist measures. Eichengreen (1986) then applies this framework to the most well-known case of post-crisis protectionism, i.e. that of the 1930 Smoot-Hawley tariffs following the Great Depression, identifying the key role of a coalition of “border agriculture” and light industry, whose utility of engaging in lobbying increased in the wake of the crisis. Due to fears related to this infamous precedent, a new wave of studies on this subject emerged after the Global Financial Crisis or Great Recession of 2008-2009. For instance, Brown and Crowley (2012) investigate the relationship between several macroeconomic indicators, namely domestic and foreign GDP growth and real exchange rate, and temporary trade barriers allowed within the current

(5)

4

WTO framework. They find that while the extent of such measures remained neglectable, an important relationship between protectionist tendencies and negative domestic economic growth was still present, while behavior “switched” with respect to foreign growth, namely from targeting stagnating partners to targeting “growing” ones. Gawande et al. (2011) focus instead on the possible reasons why the world did not witness any kind of protectionist surge like that of the 1930s in the aftermath of the Great Recession. They find that, while the WTO has played a considerable role in “policing” world trade, the main factor that avoided a new trade war has been the increasing importance of trade in intermediate goods and the emergence of Global Value Chains (GVCs), thus ending on an optimistic note. Similarly, Ballard-Rosa et al. (2018) note that the political power of import-competing industries is highly limited by both the counter-lobbying action of exporters and own budget constraints, meaning that while we may witness an increase in protectionist tendencies in the initial stages of a crisis, the ability of industries to fully sustain the costs of lobbying efforts starts waning after having reached a “peak”. Additionally, a case study conducted on Belgian firms by Behrens at al. (2013) find that the fall in the number of traded goods in 2008-2009 can be almost fully attributed to a contraction in global demand, while no significant sign of crisis of the international trade system itself has been found.

Georgiadis and Gräb (2016), on the other hand, seem to carry more pessimistic views, as they find that the relationship between economic growth and protectionist policy demand has not changed in the wake of the Great Recession, meaning that a decade-long period of relatively stagnant economic conditions could potentially exert significant pressures on trade openness. Finally, in their analysis of European Union countries, Bussiére et al. (2011), although reaching the same conclusions as the other papers written in the immediate aftermath of the crisis recognizing the counter-productive nature of such a political move, curiously note the presence of long-standing “public pressures” (pp. 26-28) that, since roughly the mid 2000s, have been pushing for more protection against globalization shocks. This last point, as we will clarify later, is of vital importance in the context of this research.

Two important limitations within the literature emerge from this brief review: firstly, analyses mostly focus on industry-level effects of crises and, as a corollary, only consider the pressures for

protectionism stemming from interest group lobbying. Secondly, only short-term shocks are taken into account, thus implying that the system “reverts” back to the status quo after economic growth has recovered. The consequence is that other significant elements affected by crises which may have repercussions on trade policy in the long-term, i.e. even after apparent full recovery, are largely neglected. These include labor market and income shocks, which are known to be among trade

(6)

5

liberalization’s negative distributional effects (Rodrik, 2018, pp. 5-6). Thus, our aim will be to fill this gap in the literature and investigate both short-term and long-term effects of crises on the possible sources of protectionist pressures.

3. Theoretical framework and hypotheses

In order to avoid problems with omitted variables and overcomplicate models, we will frame our research in a probabilistic way, meaning that our dependent variable will be the probability of adopting protectionist policies. For instance, in Krapohl et al. (2018) model of evolutional game theory applied to trade cooperation there are two sources of potential instability which may lead to protectionist tendencies, termed by the authors “noise” and “mutations”. The former refers to the temporary defections caused by external pressures which trigger the necessity to protect the domestic market, while the latter refers to sudden changes in trade strategy independently from the motive. Both elements are expressed as randomly given probabilities to engage in protectionist behavior. Here we argue instead that such probabilities are not entirely randomly assigned, but rather are influenced by well-defined exogenous socks, namely economic crises. This happens via two mechanisms, the first of which is related to the patterns of lobbying described by McKeown (1984) and takes the form of immediate industry-level demand for protection.

3.1. Industry lobbying

Being it related to changes in firms’ opportunity costs, lobbying appears in the initial outbreak of a crisis and is the main source of protectionist pressure in the short term, as it had been proven by the reaction of the US government to the Great Depression. More specifically, the lobbying potential of interest groups is “activated” during negative phases of business cycles, i.e. when the general state of the economy worsens, for two main reasons: firstly, the fact that liberal trade policies are perceived as more risky not only by import competing industries but also by those somewhat “in the middle” due to the distribution of payoffs, allowing the former to gain political power, and, secondly, the lower number of new entrants in a sector during a crisis allowing firms to better deal with coordination problems. Therefore, since both of these effects are strictly related to the current state of the economy, it is expected that they will manifest in concomitance with the “peak” of the crisis. However, following Gawande et al. (2011), we will expect such dynamics to be considerably constrained in more recent times by the presence of GVCs and trade in intermediate goods, both constituting a source of counter-protectionist pressure as trade barriers would disrupt the activity of a large variety of domestic

(7)

6

producers highly dependent on them. Moreover, the policing role of the WTO, thanks to its reciprocity mechanism and most-favored-nation (MFN) principle, will limit the possibility of unilaterally imposing tariff barriers. In short, it is unlikely to witness a reaction to an economic crisis of the same extent and severity as that of the 1930s following the Great Depression.

3.2. Redistributive effects of trade liberalization

The most important effects of crises in this respect are instead the shocks affecting labor markets and income distribution, as they are often ignored or ineffectively tackled by policymakers due to the problematic nature of redistributive fiscal policies in a globalized world, therefore constituting the main source of long-term demand for protectionist policies. Such effects are the product of a complex causal chain of events; the first of them is based on the argument made by trade theory, both classic and new, that liberalization has redistributive effects which will lead to a relative decline in wealth among a certain subset of the population depending on their position within a country’s economic structure. A particular remark should be made with respect to how these redistributive processes are framed in the context of this research: even though the literature on this topic mostly focuses on redistribution between factors of production following an approach based on the Heckser-Ohlin model (and its derivative Stolper-Samuelson theorem), whereby changes in prices of goods lead to increasing or decreasing returns in the factors of production depending on how much they are used in the good favored by liberalization, we will instead use a sector- or industry-based approach. In this case, wealth will instead shift to industries with a competitive edge, which will make gains by exporting to foreign markets, from those that will suffer the competition of cheaper imports that enter the domestic markets (Helpman, 2006, p.595).

Two practical reasons lie behind this choice: first, it will allow us to relax the strong assumption of perfectly mobile labor and capital (see Lin and Chang, 2009, p. 7). For instance, the relative immobility of factors of production allows wealth to shift from import-competing sectors, usually more relyant on low-skilled workers and more labor-intensive, to exporting sectors, which are instead more capital- and high-skill-intensive, often leaving some of the most vulnerable strata of society worse-off (Milner, 1999, p. 95). This is a focal point, as it will become clear later, for the analysis of long-term effects. Secondly, focusing on industry effects rather than on factors of production allows us to better identify the actual impact of both import competition and crises. This is due to the fact that while the latter are more vaguely-defined, the various kinds of sectors and industries have a specific relationship with trade

(8)

7

patterns, as evinced by import and export flows (Kletzer, 2001), and present definite structural

characteristics. This makes it relatively simpler to both collect more specific data and to fully analyze the geographical and social extent of the impact of crises. Why is however this relative deprivation of wealth caused by trade liberalization potentially more problematic than any other development that may have a comparable effect on labor markets? As noted by Rodrik (2018, pp. 14-16), losses originating from trade competition are regarded as more “unfair” compared to those originating from other structural developments such as demand fluctuation and technological innovation due to the perception that actors enjoying the benefits of liberalization have not been “playing by the rules” and taking advantage of socially problematic situations (e.g. issues related to outsourcing labor-intensive activities to low-income countries with weaker worker’s rights and institutions). This means that trade liberalization, especially when its marginal utility is low, is more likely to be viewed as the main cause of declining relative welfare by those affected by job loss and stagnating wages, making trade policy a much more likely target of popular discontent and thus of pressures for status quo change than, for example, automation and technological change.

3.3. The role of credit markets

Crises play an important role in this respect as their impact is not proportionally equal for all the sectors, potentially exacerbating the redistributive effects explained above. In fact, Calvo et al. (2012) note that, depending on the level of inflation, the recovery phase does not necessary lead to the full absorption of labor market and income shocks. This may lead to what has been described as a “jobless” and/or “wageless” growth, i.e. a phase in the business cycle characterized by a mismatch between an increase in macroeconomic activity and slower wage and employment growth. Moreover, such phenomena are accentuated if the crisis stems from the financial market, as it was in the case of the Great Recession: due to the fact that credit institutions will be more risk averse in the aftermath of a financial crisis, they will tend to favor capital- and skill-intensive activities, being them perceived as able to provide a more tangible “intrinsic collateral”, since parts of capital “can still be recovered by the creditors”, while “funds spent hiring labor cannot be recovered from the workers” (Calvo et al., 2012, p. 20). The result of this behavior is that credit shortages effectively harm those sectors already

negatively affected by trade liberalization more than those which are more internationally competitive. Research by Fabiani et al. (2015) indeed seems to confirm such disposition; specifically, analyzing data from surveys conducted among EU firms, they find that not only the manufacturing sector, which comprises all the most import-sensitive industries within Europe, was the one which suffered the most

(9)

8

intense demand and credit shocks during the Great Recession in mid-2009 but also that “companies employing higher shares of white-collar, high-skilled, and permanent workers were less likely to experience a strong contraction of demand and credit” and even that “firms with a higher exposure to foreign markets […] were more likely to experience both strong demand and strong credit shocks” (Fabiani et al., 2015, p. 9). Sharma and Winkler (2017) furtherly corroborate such findings showing that more “financially dependent” industries, which represent a larger share of the total labor force, experienced starker declines in employment rates and wages, mostly affecting low-skilled and temporary workers within those sectors.

3.4. Voting patterns

What is however the causal link between the worsening economic conditions of those affected by the joint action of redistributive effects of trade and increased exposition to crises’ shocks and their revealed policy preferences? To answer to this question, which will constitute the final element in this chain of events, we will rely on insights originating from both traditional theories and newer

developments in political economy and comparative politics. The first of these insights, as the analysis mostly focuses on functioning democratic states, is the classic theory of “economic voting”; the main assumption on which the theory rests is that voters will tend to reject the status quo (in this specific case represented by liberal trade policies) when their overall economic condition worsens. This is due to the fact that they believe that governments (and, as a corollary, their policy decisions) are

responsible for the current economic conditions, meaning that they will try to either oust the incumbent or, in this case, push for policy change in case their evaluation is negative. In a seminal study, Lewis-Beck (1986) found that voters in four European countries, the UK, France, Italy and Germany, not only held their respective governments accountable for past economic performance (retrospective voting) but also selected their preferred policy/candidate based on future performance expectations

(prospective voting). Other findings pertinent to this research include the prevalence of “sociotropic” voting (i.e. assessing the economy at an aggregate level) over the personal financial situation

(“pocketbook” voting), and “affective” voting (ballot choice based on “anger or delight over how things have been going” (Lewis-Beck, 1986, p. 322) rather than rational calculation). The former is relevant for aggregate-level analyses, as we will expect voters to take into account the overall situation of their region rather than exclusively their personal situation, while the latter has important

implications with respect to the perceived “unfairness” of post-liberalization redistribution described by Rodrik (2018).

(10)

9

More recent developments in academia seem instead to suggest that the effect of economic conditions on policy preference has actually been declining with increased globalization and liberalization. According to Hellwig (2008), the reason behind this change in voting behavior is the increased

perception that, due to the high levels of interconnectedness of international markets, governments are no longer able to substantially influence the state of national economies, as they will be constrained by global trends and dynamics. However, this is not inherently problematic to this study for two main reasons: first, even though national economies have indeed become clearly more liberalized and

interconnected, the effect of economic voting has not completely disappeared and still exerts a relevant influence in policy decisions in many countries, such as in the case of the US (Jensen et al., 2017, p. 428); second, Kayser and Peress (2012, 2016) model of “benchmarking” partly solves this issue, as voters will distinguish between a “global” and a “local” component of economic growth, with governments being responsible only for the latter. This implies that if a country or any other sort of constituency performs poorly compared to the median level of other comparable units (i.e. the “local” component of said country or constituency compares negatively with that of the others) the

mechanisms of economic voting will still apply. Other than offering a solution for the

post-globalization voting dilemma, this model presents an interesting insight with respect to voting patterns at a sub-national level: if, in the context of this research, a particular region which is both particularly affected by import shocks and is hit more severely by financial crises may still demand for a change in trade policy after recovery has been achieved if it keeps lagging behind the rest of the country (which serves as the “benchmark” for the region), hence explaining regionally differentiated voting behavior.

3.5. Behavioral effects

One last possible element which may add on top of this sense of relative deprivation, this time at a purely individual level, is an actual change in voters’ behavior after a crisis. Precisely, Cassar et al. (2017), in an experimental study, found that individual preferences may significantly change in the aftermath of a catastrophic and traumatic event (in our specific case of purely economic and financial nature), with increased levels of risk aversion and of impatience (discount of future earnings). For those more at risk of being affected by the negative effects of trade redistribution, this may furtherly enhance they distrust in liberal trade policies. However, trying to clearly distinguish between purely structural or rational and behavioral effects of crises on trade policy preferences is beyond this research’s scope and therefore it will not furtherly elaborate on this particular aspect. On the other hand, it is still important to remark the possible presence of such mechanisms.

(11)

10

In sum, crises increase the probability of policy demand for protectionism to rise and thus of states to close markets to foreign competition via two channels, i.e. short-term industry lobbying and long-term labor and income distortions. The structure of these mechanisms will therefore resemble that of the so-called Coleman’s (1990) “bathtub”: a macro-level event (crises) will trigger a reaction (industry lobbying) directly affecting the final macro-level outcome (trade policy) and indirectly via the

micro/individual level by affecting the conditions and resources (employment and salaries) of workers, in turn causing a shift in their preferences and, consequently, influencing the final trade policy outcome via voting and public pressure.

3.6. Hypotheses

Two main hypotheses can be deduced from the above theories:

1. Economic and financial crises will cause the probability of a state to repeal its liberal trade policies to increase in the short-term, due to the presence of industry lobbying pushing for the adoption of protectionist policies.

Even though this first hypothesis has already been tested in the past (Brown and Crowley, 2012, Georgiadis and Gräb, 2016), it is still necessary to put it into perspective within the context of this research. More precisely, replicating the findings of previous studies on this topic will provide us with a clear benchmark of “standard” crisis effects, allowing us to make clearer comparisons with the newer findings. One important remark to be made in this respect is that the presence of global value chains and WTO rules will strongly constrain this initial short-term impact, implying that:

1a. For more recent episodes, the first initial short-term impact of stagnating economic growth on a country’s willingness to adopt protectionist policies will be highly constrained by

structural and institutional elements and thus of highly limited extent.

The second part of the research, which will be the true “innovation” of this research project, will instead focus on the aforementioned long-term effects, positing that:

2. Economic and financial crises will cause long-term shocks within the labor market which will not be fully absorbed even after economic growth recovers, mostly affecting those sectors already exposed to foreign competition. The prolonged effects of such distortions will lead to increasing dissatisfaction with liberal trade policies, causing bottom-up pressure for

(12)

11

One important corollary to this second hypothesis is that such pressure may fully explicitly manifest even with a several-years-long “lag” (in this case via voting behavior), thus even when the economy has apparently fully recovered from the previous crisis. It follows that if any of these distortions is still in place and/or have a long-lasting behavioral impact on the general populations when policy action does indeed materialize, it may indicate that crises indeed have a latent long-term effect on trade cooperation, even though such delay may seem to imply that there is no relationship between the two. What actually influences the length of this “lag” is beyond this project’s scope and will not be furtherly investigated.

4. Research method

In order to develop a coherent and valuable assessment of both hypotheses, we will employ a mixed approach (quantitative and qualitative) due to the very nature of the two. For short-term effects (H1 and H1a), we will employ a statistical method to evaluate the effects of macroeconomic indicators

(independent variable) on temporary trade barriers (dependent variable). To test this, we will first rely on simple ordinary least squares (OLS) regressions, directly linking variations in the macroeconomic dimension to the number of temporary protectionist measures. For the following step, in order to capture the effects of industry lobbying and to control for omitted variable bias, we will then make use of a Two-Stage Least Squares (2SLS) over the cross-sectional panel dataset. This will allow us to frame the effects of fluctuations in the national economy as a sort of “trigger” for political power of interest groups to manifest and influence the trade policies of the countries considered, as posited by H1. In short, a statistical approach, although limiting the extent of the internal validity of the analysis with respect to each particular case, will help us to investigate the general characteristics of this relationship beyond country-specific effects.

For the second part of the research, we will instead make use of a qualitative approach due to the need to get a more precise understanding of all the highly complex underlying mechanisms and processes. This will be carried out with thorough time-trend analyses and process-tracing in order to fully describe the chain of events that produce and make post-crisis long-term distortions relevant with respect to trade policy. More specifically, we will make use of descriptive data and interpret temporal variations and fluctuations in the light of the aforementioned theoretical assumptions in order to paint a coherent picture of the issues at stake. The implication here is that we will need to restrict the sample to one or two specific cases both referring to a single well-defined event. Even though this will not allow us to

(13)

12

make more generalized assumptions, the aim here will be to produce a detailed and policy relevant account and to frame the actual causal mechanisms at play. In addition, to get a clear understanding of both the geographical extent and the social dimension of all the possible long-term distortions caused by the overlapping effects of international competition and economic crises, we will divide this second part of the research in two sections. The first one will focus on aggregate-level effects, meaning that the unit of analysis will be regions or, more precisely, sub-national divisions and constituencies, as we will try to uncover the possible inter-regional discrepancies that may emerge following trade and crisis shocks. The second one instead will be centered around individual-level effects, therefore analyzing the effects of trade liberalization and economic crises on worker themselves and, more generally, on the labor markets. Structuring the analysis in this way will consequently allow us to develop a coherent understanding of how both regional and social cleavages that emerge as a consequence of a negative economic shock will translate into policy preferences and voting behavior.

5. Sample selection and data

The statistical sample for the first part of the research consists of all of the OECD countries over a period ranging from 1990 to 2017. To avoid problems with data management and definitions we will consider the European Union bloc as a single entity (of 15 countries until 2003 and of 28 since 2004), as all regulations and directives on trade matters apply equally to all member states following single market rules. Consequently, both macroeconomic and trade barrier data will refer to the bloc as a whole, as there are often considerable discrepancies in terms of the former among the various countries that may lead to a mismatch of trends. Other than the fact that more specific data are available for this set of country-years, the sample is statistically relevant as it represents more than 70% of global trade and covers a time series large enough but relatively stable with respect to the geopolitical, financial and institutional structure, with the only notable exception being the “evolution” of the GATT into the WTO in 1995. This will allow us to control for several time-invariant factors and to produce a representative enough estimate of the effects being investigated. Moreover, all of the countries in the sample are functioning democracies, a critical aspect since political accountability mechanisms play a vital role in this framework, and present a developed economic structure, thus minimizing any possible “noise” that may be produced by sample heterogeneity.

For the second part, we will instead focus on two specific countries, namely the US and the UK, over the years preceding and following the Great Recession of 2008-2009. There are two main specific

(14)

13

reasons behind this choice, once again besides simple data availability consideration: firstly, the cases of the two countries above are among the most relevant from a social and political point of view, as both of them have experienced relevant anti-globalization policy shifts in recent years. For the US, such shift is primarily embodied by the election of president Donald Trump, who not only took an explicit stance against free trade during his presidential campaign but as well, partly fulfilling his promises, imposed a series of tariffs on a variety of goods mostly from China and Europe, such as steel and aluminum, in early 2018 (Baker and Swanson, 2018), thus defying WTO rules. In the UK, the most radical departure from the status quo is instead represented by the so-called “Brexit” referendum, where voters indicated their preference for leaving the European Union. Although the anti-globalization nature of such decision is more ambiguous here, as Brexit proponents often have mentioned the possibility of getting “better” trade deals as an advantage of leaving the bloc, findings indicating the presence of a relationship between trade shocks and the leave vote (Colantone and Stanig, 2018) indeed make it a representative case of such preferences. The second of the two reason is the fact that the long-term dynamics involving labor markets (stagnating wages and employment) after the Great Recession described above are more clearly defined (Sum et al., 2014, Romei, 2018). In sum, both of these elements make these two the “most likely” cases in which we could thoroughly test the second hypothesis.

For the first half of the research, we will rely on data from the OECD’s official website and from the World Bank for the macroeconomic indicators, namely real GDP growth and government budget deficit, which will serve as the independent variables in the model. For the instrumented variable, i.e. the percentage of value added by industry, and the dependent variables, temporary trade barriers (antidumping and countervailing duties, as well as global safeguards), we will rely instead exclusively on data from the World Bank databases. Specifically, the Temporary Trade Barriers database provides information regarding the typology of trade barrier, the initiation date of the litigation process as well as the countries and goods targeted.

With respect to analysis of long-term effects, we will firstly analyze real wage fluctuations and employment levels in import-competing sectors in both the UK and the US before and after the Great Recession of 2008-2009. In addition, we will “match” clustering of import-competing industries concentration at regional and sub-national level (States for the US and local authorities for the UK) with the annual real GDP growth of said regions in order to detect any interplay between the effects of the crisis and import exposition. UK data are from the Office for National Statistics (ONS) while US

(15)

14

ones are from the Bureau for Economic Analysis (BEA) and the Bureau of Labor Statistics (BLS), respectively for macroeconomic and labor market data.

6. Short-term effects and industry lobbying

In order to put previous findings into perspective and to fully test the industry lobbying and political power hypothesis, in this section we will replicate the analysis of some of the previous studies on this topic, in particular Georgiadis (2016) and Ballard-Rosa et al. (2018). As previously anticipated, we will investigate the relationship between macroeconomic indicators, i.e. GDP growth and government budget deficit, and the number of temporary non-tariff trade barriers over time. Since the aim here is to measure a country’s willingness to engage in protectionist behavior, we will focus on the total number of litigations initiated within a year rather than exclusively on the successful ones in order to avoid problems with endogeneity that may arise due to specific characteristics of the process and of the other actors involved. For the first set of regressions a traditional OLS approach will be used, thus

momentarily ignoring the industry size effects and focusing exclusively on the relationship between the overall economic conditions of the various countries and the number of temporary protectionist

measures over the sampled years. We will also use the “Year” variable to control for any possible time trend effects. Thus, the very simple model for this first part of the analysis will be the following:

𝑦_𝑐𝑡 = 𝛼𝑥_𝑐𝑡−1+ 𝛽𝑡 + 𝑘 + 𝜖

where 𝑦 is the number of litigations for a type of temporary trade barrier initiated by country 𝑐 in year 𝑡, 𝑥 is the value of the macroeconomic indicator lagged by one year to avoid any possible issues with simultaneity, 𝑡 is the “Year” variable that measures ay possible time-related variation, 𝛼 and 𝛽 are the explanatory coefficients of the former and the latter respectively, 𝑘 is the constant and 𝜖 is the standard error. Standard errors are clustered at country level. Results of these preliminary regressions are

summarized in Table 1A. In the first three columns the independent variable is the standard annual GDP growth rate. Firstly, we find a negative relationship significant at the 5% level for anti-dumping measures: this means that on average, even though the total number of new measures has been declining (as indicated by the negative coefficient for “Year”), a weaker economic growth leads to a higher propensity of filing a complaint to the WTO to implement anti-dumping measures and, thus, an increased willingness to protect the domestic market. Moreover, the overall R-squared is of 0.125, indicating a relatively high explanatory value of these to variables alone. On the other hand, it is possible to notice that “non-benchmarked” variation in GDP growth does not have any statistically

(16)

15

significant impact on the number of countervailing duties imposed by a specific country, as well as on that of global safeguards. This is likely due to the fact that the incidence of such measure is much lower than in the case of anti-dumping duties, thus lowering the total number of observations and

consequently the precision of the estimation. Next a “benchmarked” measure of GDP growth, which has been calculated by taking the difference between a country’s own growth rate and the sample median of a given year, is used. As it clearly appears in column 5, the effects of economic growth have become even more statistically significant (at the 1% compared to the 5% of the previous measure) and their magnitude has slightly increased as well. In addition, “benchmarked” GDP growth now presents as well a negative and even significant at the 5% coefficient for countervailing duties, though its actual extent is still quite limited compared to anti-dumping measures. No relevant impact is still present for global safeguards, as statistical significance is only at the 10% and the coefficient, surprisingly with a positive sign, and is still too small to have a sizable effect on the number of measures adopted within a specific year. Results using government budget surplus as an alternative independent variable, shown in Table 1B, not only seem to reflect those obtained with the benchmarked GDP growth measure, with coefficients for both anti-dumping and countervailing duties being negative, but also show an even higher statistical significance (p-values of 0.002 and of 0.026) and even a higher R-squared. Due to the highly pro-cyclical nature of government budget deficits, it is now possible to clearly infer that non-tariff protectionist measures follow a counter-cyclical pattern in developed OECD, thus sensibly increasing in periods of negative overall economic conditions.

(17)

16

Table 1B: OLS estimates of effect of government budget surplus on the number of temporary protectionist measures

This not only confirms the findings of the previous literature, especially Georgiadis and Gräb (2016) and Brown and Crowley (2012), but possibly also the “twin deficit” hypothesis, which links

government budget deficit with trade deficit (Enders and Lee, 1990,). On the other hand, it is yet possible to observe that, although statistically and quantitatively relevant, the actual size of the effects is relatively contained. For instance, a decrease of one standard deviation (2.34%) in the

“benchmarked” growth rate on average leads to three additional instances of anti-dumping measures and 1.6 of countervailing duties applied by a specific country in the sample, while for government surplus the effect of one standard deviation decrease is of 3.6 and 2 respectively. Consider as well the fact that anti-dumping duties, which represent the most significant portion of all the temporary trade barriers, have been declining since 1990, as evinced by the negative coefficient in all of the above models. It is then evident that the actual extent of such protectionist response has become increasingly weaker and it is therefore unlikely that in the most recent times the immediate short-term effects of crises will produce a new trade war of the same proportions as that triggered by the passing of the Smoot-Hawley tariffs of 1930, even after a large traumatic event like the Great Recession of 2008-2009. Even though testing such assumption is beyond the scope of this research, it is plausible to attribute this decades-long decline to the emergence of global value chains (GVCs) and of trade in intermediate goods, as purported by Gawande et al. (2011). Furthermore, the temporary trade barriers considered in this analysis to measure the level of protectionist tendencies within the sample are a relatively weak policy tool for domestic market protection compared to traditional tariff barriers (of which the use is constrained by WTO principles) as they are able to affect only a limited set of goods and require approval by the WTO itself after due process. In sum, this OLS analysis confirms what has been posited by H1a and partly by H1, as the initial short-term effects of negative economic conditions

(18)

17

on trade policy, even though still present, are highly constrained by both structural and institutional factors, thus limiting the potentially harmful impact they might have on the global trade regime. It is now necessary to fully test the main hypothesis H1 instead, in particular concerning the role of industry lobbying in the initial response of governments to worsening conditions with respect to trade policies. To do so, we will adopt instead a two-stage least squares (2SLS) approach. The reasoning being followed here is that worsening economic condition do not directly affect the number of

temporary trade barriers as previously implied, but rather only via a change in political power balance among the various interest groups representing a specific productive sector in a given country. The model will then be represented by the following equations:

𝑦𝑐𝑡 = 𝛼𝑥𝑐𝑡+ 𝛽𝑡 + 𝑘 + 𝑣

𝑥_𝑐𝑡 = 𝜃𝑧_𝑐𝑡−1+ 𝑙 + 𝜖

where 𝑥 this time is the proxy variable for industry political power, 𝑧 is the macroeconomic indicator serving as the instrumental variable and 𝑣 is the composite standard error. Results of such calculations are shown in Table 2. In all of the models, variables are instrumented by “benchmarked” GDP growth, since it has been found to be a better predictor than standard variation and a more straightforward indicator than government budget surplus. We will also not consider global safeguards, as it has been previously shown that the limited number of observations does not allow us to make clear enough inferences. As shown in the first three columns, we first try to instrument change in industry political power, proxied by their share of total value added within a given country, considering, in order, the total aggregate share of the manufacturing sector and that of machinery and of textiles industries by themselves. Due to lack of data for certain country-years, the latter two do not yield any significant result, as the first stage of the regression is in both cases too weak for the instrumental approach to be effective. Therefore, we will drop these measures for the following analyses.

Instead, aggregate manufacturing share of total GDP, although less precise than the former two, seems to have, as expected, a significant and positive effect on the number of anti-dumping measures in the second stage of the regression. This implies that negative fluctuations of the local component of a national economy will lead exporting sectors to lose relative political weight, possibly due to falling aggregate demand and competitiveness, with respect to import-competing ones (which mostly belong to the manufacturing sector within developed economies), allowing the latter to push for protection in a more effective way as their opportunity cost of lobbying will decrease. For countervailing duties

(19)

18

instead, which showed a weaker yet significant relationship with economic growth in the OLS

estimations, no relevant effect of this change in output share is found on the number of such measures implemented by a country in a given year. This is possibly due to the simple fact that, once again, the number of observations is significantly lower compared to the case of anti-dumping duties, thus weakening the first stage of the regression. However, due to the presence of a statistically significant coefficient in the OLS regression, it might as well be possible that other factors besides industry political power that cannot be observed with the available data are actually directly linking macroeconomic fluctuations to trade policy choices.

Table 2: 2SLS estimates of the effects of variations in % total output share of each industry on the number of trade barriers

One final possibility is instead that the share of total value added is not a good enough proxy for a given industry’s political power: recall that one of McKeown’s (1984) initial arguments with respect to lobbying patterns was that, among other factors, opportunity cost for lobbying decreases because of a lower number of new firms entering the market within that specific industry, as a lower number of actors leads to less coordination problems following the Olsonian logic of collective action. In addition, Salamon and Siegfried (1977) found that while firm size is positively correlated, aggregate industry size and market concentration show instead a negative relationship with political power and successful lobbying, furtherly confirming the above implications. Thus, even though so far it has implicitly been postulated that the number of firms in a given sector is either completely stable over time or presents negligible fluctuations, it is possible instead that the share of value added is itself a function of the number of economic agents in that sector rather than simply a measure of the economic weight of a fixed set of firms. If that is the case, then the causal link between business cycles and industry lobbying powers may be compromised and the above estimates in the 2SLS models do not fully capture the

(20)

19

underlying dynamics of this process. Consequently, it is necessary to make use of an alternative proxy that measures variation in political power without being affected by the same distortions. The most obvious of such alternatives is the number of enterprises within the import-competing industries

themselves: if, controlling for long-term time trends, the total number of firms in these sector decreases during periods of economic contraction, leading to an increase in temporary protectionist measures, it is possible that the lobbying mechanisms described above indeed play a much more prominent role that previously theorized. The Structural and Demographic Business Statistics (SDBS) database of the OECD reports the number of enterprises in each sector of interest, once again at a two-digit level of aggregation (thus showing “manufacturing” as a single umbrella category). The dataset only covers a relatively limited period for most countries, starting from the year 2005, with only few exceptions (e.g. New Zealand and Norway), with a total of 124 datapoints. Surprisingly, in spite of these limitations, the first stage of the regressions shows a strong correlation (p-value below above the 1% level) between “benchmarked” GDP growth and the number of firms in the manufacturing sector, yet in the opposite (negative) direction as it would have been expected following McKeown’s (1984) reasoning.

Moreover, the final result of the 2SLS analysis is once again not statistically significant for all of the temporary trade barriers considered, possibly due to the lack of a truly comprehensive dataset and the limited time frame covered by it.

To summarize the findings of this first part of the research, protectionist tendencies, represented here by the number of procedures initiated by a country to impose temporary trade barriers, do indeed increase in the wake of a negative economic shock, even in a post-1990 highly globalized world, thus settling what has been established by the literature and partly confirming H1. On the other hand, the extent of such protectionist response is still very limited, both over time (mostly of short-term nature) and in terms of intensity, most likely due to the increasing importance of the GVCs and of trade in intermediary goods, as signaled by the decreasing number of anti-dumping processes’ initiations over the years, as well as the institutional and legislative constraints posed by the WTO, confirming the findings of Gawande et al. (2011) and Ballard-Rosa et al. (2018) and what therefore H1a. Testing industry lobbying patterns, which have been identified in the theoretical framework as the main driver of the short-term effects of crises, has instead proved to be considerably more problematic, mostly due to the lack of a global-scale proxy of industry political power and of publicly available data; therefore, it will be necessary for future research to delve deeper in this issue to develop a more coherent

(21)

20

liberal trade policies, yet not as significantly as in the pre-Bretton Woods era. It is now necessary to focus on the less evident but possibly more critical long-term effects instead.

7. Long-term effects and “lagged” demand for protectionism

As previously specified, to test the second hypothesis and thus identify the chain of events linking the impact of the Great Recession to labor market shocks and then to protectionist demand, we will analyze the overlap of crisis dynamics with negative redistributive effects of trade in the US and the UK.

7.1. Aggregate-level analysis

7.1.1. US states

We will first focus on regional-level effects and their possible implications. Starting from the US, the BLS database allows us to identify the concentration of import-competing industries identified by Kletzer (2001), namely textiles, machinery and metal industries, in order to assess a state’s exposition to import shocks. For instance, Figure 1 shows the local quotient (i.e. the ratio between local

concentration of employment in a specific sector and the national average concentration) of all states for all of the aforementioned industries.

(22)

21

Figure 1: Location quotient of import-competing industries. Source: US Bureau of Labor Statistics

For simplicity, we use the concentration of the occupation involved the most with activities related to these sectors. For instance, mechanical engineers will be the proxy for machinery manufacturing, furnace operators for the metal-refining industries and textiles workers for apparel. One striking feature that is immediately evident at first glance from this rough data is the similar relative geographical distribution of all three activities. More precisely, areas with higher concentration of these industries seem to “cluster” within specific macro-regions rather than in a more homogenously dispersed manner at national level, in particular within the so-called “rust belt” (though slightly less evident in the apparel and textiles sector). In addition, some states are among the highest ranking in all three industries’ job concentration, such as North Carolina, Indiana and Michigan. One potential implication of this significantly different within- and between-state distribution of economic activity is a rise in inter-regional inequality, as the above states are more likely to be exposed to the redistributive effects of trade and thus to bear a disproportionately heavier burden when it comes to adverse labor market and

(23)

22

income shocks. As it will be clear later, this can potentially have an effect on political agenda setting and voting behavior.

After having highlighted the general sub-national economic structure, we move to assess the impact of the Great Recession among the various states. The macroeconomic indicator that provides us with a clear preliminary overview of such effects is, expectedly, GDP growth. For simplicity, we focus on three of the states which have been previously identified as possessing a relatively high quotient in all of the above industries, namely Michigan, Indiana and Ohio. Figure 2 shows the fluctuations of quarterly figures of aggregate real GDP in millions of dollars, seasonally adjusted at annual rates, for the three states starting in 2005 and ending in 2014. It is possible to immediately notice the dramatic impact of the Great Recession on the economies of all three states: specifically, Ohio’s total aggregate output in the second quarter of 2009 (i.e. the “peak” of the crisis) fell by 7.43% compared to that of a year earlier (Q2 2008), while over the same period Indiana’s real quarterly GDP contracted by 9.38% and Michigan’s of a whopping 10.27%, making it the most adversely affected of the three considered states. Another prominent feature of these state-level trends is the relatively weak recovery, or even a lack of thereof, and a stagnant growth during the five years following the end of the Great Recession. For instance, even though Ohio’s real GDP recovered to and even surpassed pre-crisis levels in Q1 2013 and briefly experienced higher than usual quarterly growth rate of 2.05% in Q3 2014, it

immediately resumed its stagnating trend during the last quarter considered (Q4 2014) with a slightly negative quarterly growth of -0.05%.

(24)

23

Figure 2: Quarterly real GDP by state (millions of US$). Source: US Bureau of Economic Analysis

Indiana attained full recovery in Q1 2013 as well, yet its quarterly real GDP growth was only of 0.69% in Q4 2014 while its total annual output only saw an increase of only the 0.45% compared to 2013 figures due to a slight plunge in early 2014. Once again, Michigan is the most dramatic case out of the three selected: as of late 2014 the state’s real GDP had not reached pre-crisis level yet, while quarterly (Q4) and annual growth were respectively only of 0.46% and 1.88% in 2014. Thus, the picture that emerges from this brief analysis is one of states already highly exposed to import competition not only experiencing significantly larger negative shocks in the immediate aftermath of the Great Recession but also struggling to recover even three to four years after its end. Moreover, even when they did manage to finally reach the pre-crisis levels of aggregate output, they would still present a stagnant and slow-paced GDP growth rate.

This overlap between import and crisis shocks becomes even more evident if we exclusively focus on manufacturing sector data. Even though such a broad “manufacturing” category contains both the three

440000 460000 480000 500000 520000 540000 560000 2005Q 1 2005Q 4 2006Q 3 2007Q 2 2008Q 1 2008Q 4 2009Q 3 2010Q 2 20 11 Q 1 2011Q 4 2012Q 3 2013Q 2 2014Q 1 2014Q 4

Ohio Real GDP, all industries

240000 250000 260000 270000 280000 290000 300000 2005Q 1 2005Q 4 2006Q 3 20 07 Q 2 2008Q 1 2008Q 4 2009Q 3 2010Q 2 2011Q 1 2011Q 4 2012Q 3 2013Q 2 2014Q 1 2014Q 4

Indiana Real GDP, all industries

320000 340000 360000 380000 400000 420000 440000 460000 2005Q 1 2005Q 4 2006Q 3 20 07 Q 2 2008Q 1 2008Q 4 2009Q 3 2010Q 2 2011Q 1 2011Q 4 2012Q 3 2013Q 2 2014Q 1 2014Q 4

(25)

24

sectors analyzed above as well as other less import-exposed industries, the latter’s relative weight is much smaller compared to that of the former. Thus, assessing the effects of the Recession on the output of the manufacturing sector alone will to furtherly shed light on the phenomena analyzed. The story that state-level BEA data disaggregated by industry, based on the North American Industry

Classification System (NAICS), seems to tell is indeed one of an industry dramatically damaged by the impact of the crisis: in the second quarter of 2009 quarterly manufacturing output was lower (compared to Q2 2008) by a whopping 20.15%, 21.42% and 32.23% in Ohio, Indiana and Michigan respectively. Full recovery did not take place for the former two as of late 2014, while Michigan’s manufacturing sector output surprisingly reached its early 2008 values as early as Q1 2011. Moreover, in 2014 the sector’s output was growing at an annual rate of 3.97%, 2.45% and 5.88% respectively for each state. However, the reason behind relatively rapid post-crisis growth, as well as the latter state’s early recovery, may simply be some kind of structural change within the structure of the manufacturing sector itself (recall that in the BEA dataset the “manufacturing” category comprises all kinds of

manufacturing industries), such as the relocation of a particularly profitable activity within that specific state, which is beyond the scope of this research. It is therefore undeniable that the Great Recession left its negative mark on the economies of these states and, particularly, the import-competing industries already suffering from liberalization have been experiencing even more grave losses.

The analysis is however still not complete, as it is still necessary to compare these figures with both national trends and the condition of states whose economic structure is less affected by import competition. First of all, national real GDP already recovered to pre-recession levels as early as Q4 2010, i.e. a whole two years before those of Ohio and Indiana managed to achieve the same and even more than three before Michigan. Furthermore, in 2014 annual national economic output variation was of +2.6%, meaning that even Ohio, the best performing on an annual basis of the three states

considered, was lagging behind the country’s average. Moving the focus on states whose economic structure is less exposed to import competition, it is possible to notice that the Great Recession indeed had an uneven impact at sub-national level. For instance, in Q2 2009 California produced only the 5.93% less compared to Q2 2008 and, even though it recovered only one quarter earlier (Q4 2012) than Ohio and Indiana, by late 2014 it was already growing at an annual rate of 2.82%. Massachusetts, which as well presents a relatively lower location quotient in all of the three selected industries (0.25, 0.54 and 1.22 for steel, textiles and machinery respectively), only lost a total of 3.88% of its real GDP in the same four quarters and in 2014 was back at growing at an annual rate of 2.30%. Lastly, we

(26)

25

present the particular case of Idaho (location quotient of 0.64 for machinery manufacturing, other data is missing): although the state’s quarterly total output fell by 6.51% in the same period as above and subsequently experienced a period of relative stagnation, by 2014 it was growing by 2.68%, i.e. at the same rate as in 2007. All the relevant figures of this analysis are summarized in Table 3.

Table 3: US real GDP growth by state, all industries and manufacturing

Area All industries Manufacturing

Q2 2009 (comp. Q2 2008) Q4 2014 2014 (year total) Q2 2009 (comp. Q2 2008) 2014 (year total) Ohio -7.43% -0.05% 2.08% -20.15% 3.97% Indiana -9.38% 0.69% 0.45% -21.42% 2.45% Michigan -10.27% 0.46% 1.88% -32.23% 5.88% California -5.93% 0.40% 2.82% Massachusets -3.88% 0.54% 2.30% Idaho -6.51% 0.80% 2.68% National -3.96% 0.50% 2.21%

This brief descriptive analysis shows that, as previously theorized, the Great Recession’s impact was indeed uneven within the US; states already subject to import shocks and thus an already weakened economy bearing a disproportionately larger burden in the immediate aftermath of the crisis and struggling to fully absorb the shocks deriving from it, creating a long-term condition of relative stagnation compared to the rest of the country. Coupled with the fact that areas with a high

concentration of import-competing tend to cluster within the same macro-regions, thus strengthening the “geographic identity” of such phenomena, this uneven distribution of shocks will possibly produce strong and well-defined sub-national divides when it comes to policy demand and preferences. To provide some additional proof of this overlap of import competition and crisis severity, as well as of the long-term nature of this relationship, we briefly present the most recent state-level data available (Q3 2018).

(27)

26

Figure 3: percentage change in Real GDP by US state, Q3 2018

The data, as visualized in Figure 3, clearly shows the presence of a clustered distribution of growth rates within the US: specifically, states experiencing higher levels of economic growth are mostly located along the West Coast and in the Mid-West, while those in the East and those belonging to the so-called “Rust Belt” show lower overall growth. Even though there seems to be no statistically relevant difference between the latter group’s and East Coast’s rate, one should consider that, on average, Rust Belt states initial GDP value is much lower than that of Eastern ones. For example, New York’s total aggregate output grew by 2.8% in Q3 2018, i.e. at the same rate as Ohio and less than Indiana, yet its real total output was of 1675,7 billion US$ compared to the latter two’s of 672.1 and 369.2, meaning that their growth, if expressed in absolute terms, is much lower than we would expect. If we use instead Mid-Western states as benchmarks, being them similar in terms of size of the

economy, the disappointing figures of the Rust Belt become much more evident. Therefore, even the most recent data available shows that, almost a decade after the end of the Great Recession, states with more import-competing activities still seem to be lagging behind the rest of the nation, with the

“geographical overlapping” becoming evident when comparing the distribution of growth rates in

Figure 3 with that of industry concentration in Figure 1. 7.1.2. UK regions

In order to ensure that the case of the US is not just a sui generis one in this respect, we now turn to the UK, roughly following the same descriptive process as above. For convenience, we rely on the analysis

(28)

27

conducted by Colantone and Stanig (2018), which makes use of the Chinese import shock as a proxy for an area’s sensitivity to liberalization. Using the NACE (French acronym for Statistical

Classification of Economic Activities in the European Community) classification system, they firstly identify a set of import-exposed industries within the UK, all belonging to the manufacturing sector, by calculating the variation of the total quantity of imports from China of goods produced by those very same industries over a period starting from 1990 and ending in 2007. Secondly, they “take the weighted sum of the change in imports per worker across industries, where the weights capture the relative importance of each industry in a given region” (Colantone and Stanig, 2018), making use of the NUTS-3 (French acronym for ““Nomenclature of Territorial Units for Statistics”, an official classification system of territorial units within the European Union) regional subdivision. The rationale behind such reasoning is that the shock caused by Chinese import competition will be stronger in those areas with a higher concentration of workers employed in those sectors which produce goods similar to the cheaper ones being imported, thus following a similar process to the one adopted in the analysis of the US. The result is a new variable termed “Import Shock”, for which an increase of 0.01 in its value equates to a growth in imports from China of 10 real euros per worker, meaning that higher values will represent a larger increase in imports in that specific region and, consequently, increased foreign competition. In

Figure 4 the values of such variable are neatly graphically summarized (Northern Ireland is excluded

for lack of data availability). Once again, it is possible to notice how areas with high concentrations of import-competing industries, represented by a darker shade in the map, tend to cluster around certain “macro-regions”, as it was in the case of the US. Specifically, areas corresponding to North-Eastern and Central England, as well as the east of Wales, have suffered the entry of Chinese imports the most, with Leicester showing the largest value of “import shock”, while most of Scotland and,

unsurprisingly, Central London have remained relatively shielded from the side effects of trade liberalization.

How does this compare to the distribution of shocks following the advent of the Great Recession? Once again, the macroeconomic indicator that better provides us with a clear general snapshot of the crisis’ impact at a sub-national level is aggregate real GDP and its respective growth rate. However, unlike in the case of the US, only annual data is available at the desired level of regional subdivision, i.e. NUTS-3, potentially limiting the precision of the analysis.

(29)

28

Figure 4: import shock at NUTS-3 regional subdivision. Source: Colantone and Stanig (2018, p. 205)

On the contrary, the availability of a better structured index of import sensitivity (i.e. the “import shock” variable from Colantone and Stanig (2018)) allows us to obtain more statistically precise estimates of its interplay with crisis effects. Nevertheless, we will proceed in the same way as above, i.e. by first performing a time trend analysis of the real gross value added in the years immediately before and after the Crisis, then analyzing the actual distribution of negative shock intensity and, finally, by assessing the correlation between areas exposed to import competition and those more adversely affected by the Great Recession. Firstly, since the data allows us to perform such operation, a quick OLS regression will allow us to better identify the correlation between sensitivity to import shocks and the negative impact of crisis. Results of such regressions are shown in Table 4: the first column clearly indicates the presence of a negative and statistically significant relationship between import sensitivity and aggregate (all industries) real GDP growth for the year 2009, i.e. the most critical period of the Great Recession, meaning that on average regions with a higher concentration of workers employed in import-competing sectors suffered the initial impact of the crisis more than the rest of the country. More precisely, an increase of one standard deviation (0.14) in the level of import sensitivity leads to a decrease of 0.48% in the GDP growth rate. Coefficients for aggregate growth in years 2014, half a decade after the end of the crisis, and 2016, the last year available in the dataset (we do not

(30)

29

consider 2017 due to possible post-Brexit referendum distortions) show both a negative sign, though they are not statistically significant. This implies for the moment a less clear relationship for recovery and post-recovery years.

To provide a clearer picture of these relationships and of inter-regional differences, we then focus on a few specific examples as in the above case. Data is summarized in Table 5. We first start with

Leicester, which has been individuated as the area that suffered the highest degree of import shock in the seventeen years preceding the crisis (value of 0.75). In 2009 its total output decreased by the 4.10% compared to that of the previous year and that it did not recover to 2008 levels until 2012. In addition, while in 2014 and in 2015 real GDP growth experienced a brief surge, with positive growth rates of 5.74% and 5.32% respectively, it slowed down once again during 2016, growing only by 0.21%, while in 2017 it contracted by 1.52%. However, as previously noted, such figures may be a byproduct of the uncertain political climate that emerged following the “Brexit” referendum of June 2016.

Table 4: effect of import sensitivity on real GDP growth

(1) (2) (3) VARIABLES 2009 2014 2016 Import shock -0.0364** -0.00366 -0.00260 (0.0167) (0.0146) (0.0556) Constant -0.0311*** 0.0301*** 0.0204 (0.00576) (0.00506) (0.0192) Observations 151 151 151 R-squared 0.031 0.000 0.000

Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Second on the list is the area of Blackburn and Darwen (import shock of 0.67), which followed a similar pattern, losing a whopping 8.21% in 2001 and not recovering until 2013, as well as similarly experiencing a momentary surge in 2014 (+4.66%) but again slowing down in 2016 (+2.76%). The “Central Valleys” region immediately follows, having sustained an import shock of 0.64. Here,

surprisingly, we can notice the presence of a different pattern compared to the previous two examples: while the region did not experience a serious negative shock in 2009 (-1.96%) and already recovered to 2007 levels by 2011, its total real output remained relatively stagnant during the following years, even decreasing once again in 2014 (-0.98%) and only growing by 1.97% in 2016. Even with all the