The pattern of structural change: testing the product space framework

doi:10.1093/icc/dty009

2018-04-05 00:00:00

Abstract The set of available local “capabilities” determines what an economy produces today (its static comparative advantage) and, at the same time, defines the trajectories that the process of structural change may take in the future. The product space (PS) framework developed in recent seminal works by economists and physicists suggests that path-dependence characterizes the evolution of the production basket (Hausmann and Klinger, 2007, Harvard University Center for International Development Working Paper #146; Hidalgo et al., 2007, Science Magazine, 317(5837), 482–487). These authors represent economies as sets of productive capabilities that can be combined in different ways to produce different products. Countries progressively change their production baskets and move toward goods that require capabilities that are already available; on the contrary radical structural change rarely happens. In this article, we analyze the evolution over time of the production baskets in 107 Italian provinces (NUTS 3) and perform the first test on the PS hypothesis of path-dependence. We investigate whether new products entering the provincial production baskets are nonrandomly related to initial production baskets. We confirm the general tendency of path-dependence but highlight at the same time that a sizable share of “new products” are an exception to this general pattern. These “random entries” over the PS are particularly interesting for industrial policy, since they represent radical deviations from the initial comparative advantage. In the final part of the article, we investigate using parametric analysis the product and provincial characteristics that determine these deviations from the PS pattern. 1. Introduction Economies evolve over time in a dynamic process in which available resources are combined to produce a bundle of products (production basket) which reflects the comparative advantages of those economies. The process of structural change may take different paths according to whether marginal or radical changes in the composition of production baskets occur over time. A new wave of intellectual effort in the analysis of the process of economic development has placed structural change at the core of the policy debate (McMillan and Rodrik, 2011; Spence, 2011, Stiglitz et al., 2013). As in early contributions (Kuznets, 1966), structural change is seen as a precondition for sustained economic growth and development, since economic wealth strictly depends on the economic structure and sophistication of the production basket.1 In particular, the product space (PS) framework developed in recent seminal works by economists and physicists suggests that the evolution of the production basket is strongly characterized by path-dependence (Hausmann and Klinger, 2007; Hidalgo et al., 2007). These authors represent economies as sets of productive capabilities that can be combined in different ways to produce different products. Countries progressively change their production baskets and move toward goods that require capabilities that are already at their disposal or easily obtainable; on the contrary, radical structural change rarely happens.2 Since capabilities cannot be easily identified, measured, and observed, these authors employ an “agnostic approach” and use an outcome-based measure which relies on the idea that if two goods are “related” (i.e., produced and exported in tandem), they use production factors that are “common.” Unrelated goods, i.e., those goods that are unlikely to be produced and exported by the same country, do not share a similar set of productive factors. The PS was first presented in Hausmann and Klinger (2007) and Hidalgo et al. (2007) as a network of relatedness between 774 globally produced and exported products. The PS has been represented effectively using a map (reported in Figure 1) of global production in which each node represents a product, and connections between nodes represent the degree of proximity between them.3 The authors assert that goods entering a country’s export basket are those highly connected with the set of products that were previously exported. 4 Figure 1. View largeDownload slide Hidalgo et al. (2007)—representation of the network of relatedness between goods. Figure 1. View largeDownload slide Hidalgo et al. (2007)—representation of the network of relatedness between goods. It is important to notice that alternative measures of relatedness have been proposed in the literature. Zaccaria et al. (2014) build a hierarchical network of relatedness which represents “causal” relationships between products. The links between products rather than capturing the use of similar capabilities—as postulated by the approach of Hidalgo et al. (2007)—measure the probability that the current production of good a will induce the production of good b in the future. Using world export data the authors embed in their proximity matrix information on recurrent paths of evolution of countries comparative advantage. Related methodological approaches could be also found in the studies on “corporate coherence” which investigate the determinants and effects of related activities within firms (see Teece et al., 1994). For instance, Nesta and Saviotti (2005) employ a metric of relatedness using US biotechnology patent data to investigate the effect of the coherence of firm’s knowledge capabilities on innovative performance. The authors measure technological relatedness as “the frequency with which two technology classes are jointly assigned to the same patent application” (p. 128). The idea of relatedness in this work is based—following the methodology of Teece et al. (1994)—on the co-occurrences of the use of a set of 30 different technologies in actual biotech patents data relative to their expected values (i.e., random assignment of these technologies to the same pool of patents). Bottazzi and Pirino (2010) shed light on some important drawbacks of previous studies and propose the use of P-scores as more robust measures of relatedness.5 According to many observers, these recent contributions add new “weapons” to the arsenal of industrial policies, since the network of relatedness provides a guide for policymakers in terms of which products/sectors are likely to be successfully developed in a country or region (latent comparative advantage). In fact, most industrial policies that aim to implement ambitious projects have failed because of the existence of capability constraints to “big leaps.” In this light, the PS poses limits to overoptimistic and “comparative advantage defying” policies and suggests a step-by-step approach featuring “small leaps” toward these products where countries may have a latent comparative advantage. Although the PS framework has spurred considerable interest among the academia6 and policymakers7, to date, to the best of our knowledge, there is no systematic empirical test which shows whether the pattern of specialization of countries or regions follows its predictions. The aim of the analysis performed in this work is to fill this gap by providing a new methodological approach for testing the validity of one of the key hypotheses of the PS: specialization in new products does not follow a random process but is likely to occur in products that are strongly related (or connected) to the ones that are already produced. We develop a “dart-board” approach which allows us to compare the actual short-term evolution of the export baskets in 107 Italian provinces (NUTS 3 classification)8 with randomly generated counterfactuals. After presenting the methodology, which can be easily applied to other countries/regions in the world, we show that although the overall evolution of the Italian export basket shows a significant degree of path-dependence—as predicted by the PS framework—more radical changes do often occur. To assess the impact of the recent crisis, we identify two periods: (i) precrisis, 2002–2006; (ii) crisis 2007–2011. Interestingly, we find evidence in both periods of a large heterogeneity in terms of frequency of these “big leaps” over the PS both across provinces (NUTS 3 areas) and across sectors (Harmonized System at 6-digit trade classification).9 From a policy perspective, these deviations from the hypothesis of path-dependence are the most interesting ones in our opinion. In fact, the development of products that are unrelated to the preexisting export basket signals the ability of the economic system to combine old and new capabilities in a way that allows production to be diversified away from the static comparative advantage. As argued in Castaldi et al. (2015) in the context of regional innovation, technological breakthroughs are the result of the combination of knowledge from “unrelated” technological capabilities and allow economies to follow new technological trajectories (Dosi, 1982). Some successful and rather emblematic “jumps” over the PS network have been hotly discussed in the development literature. The rise of the aircraft industry in Brazil as well as the ascent of the automotive industry in Korea are notable examples.10 The rise of the 64 Kbit DRAM sector in Korea is another emblematic case which defeats the gravity of the PS (Lin and Chang, 2009).11 In both cases, the role played by public actors in supporting industrial competitiveness has been determinant. The PS framework is not able to explain why these jumps occur; quite the contrary, the framework predicts small-distance and gradual jumps toward related goods. In the last part of our work, we investigate—using probit models—which provincial features are associated with the likelihood of observing these more radical structural changes. Our results show that the diversification of provincial productions away from the initial comparative advantage is more likely that the more sophisticated the initial production basket is, the higher the mix of unrelated varieties produced and the more open and skilled intensive provincial economies are.12 The remainder of the article is structured as follows. In Section 2 we discuss recent contributions to the economic literature on PS. Then, in Section 3 we describe the data and the methodology used for computing the econometric strategy to test the PS theory on Italian provinces. In Section 4, we present the main evidence of analysis and investigate the determinants of path-dependency in the evolution of PS in Italy distinguishing the precrisis period from crisis one. Finally, we conclude with some policy remarks. 2. Specialization and path-dependence: a brief review of the PS framework The PS framework briefly outlined above provides a powerful prediction of path-dependence in the evolution of countries or regional specialization over time. In fact, the inclusion of new products in the export basket of an economy is likely to be strictly related to the preexisting specialization. The economic intuition is the following: products that are closely connected in the PS (i.e., high degree of proximity) require a similar set of production capabilities. If an economy has a comparative advantage in a given product, then it is relatively simple for that economy to also develop a comparative advantage in products requiring the same set of capabilities. In recent years, an increasing number of studies based on the PS framework have investigated the existence of path-dependence in the process of structural transformation. As in the original contribution by Hidalgo et al. (2007), these studies generally use trade specialization—measured by revealed comparative advantage (RCA)—as a proxy of production specialization and analyze the pattern of specialization across the PS over time. An important contribution made by this approach is the evidence that countries at a different level of development tend to be positioned differently in the PS. While industrialized countries are mainly specialized in the production of “central goods,” i.e., goods with higher average connections to others and higher sophisticatedness13, low income countries have most of their export baskets located in the periphery of the PS. Hausmann and Klinger (2010) and Hidalgo (2012) show how the export baskets of Ecuador and a pool of African countries (Kenya, Mozambique, Rwanda, Tanzania and Zambia) respectively mostly consist of peripheral products and highlight a rather strong persistence of position on the PS over time. Felipe et al. (2013a, 2013b) perform single country analyses on a long-term perspective (from the 1960s to the 2000s) for two important emerging economies, China and India. Their works suggests that the process of development in these two countries is accompanied by a gradual and continuous increase in export sophisticatedness. Further studies based on the PS approach have focused attention on the nexus between centrality in the PS and trade diversification. Minondo (2010) in a study on a set of 91 countries shows that the average connectedness of countries’ export baskets (i.e., the degree of centrality in the PS) is a strong predictor of actual diversification level. In a related study, Boschma and Capone (2016) analyze the process of trade diversification for EU-27 and European Neighbourhood Policy (ENP) countries between 1995 and 2010. The authors find evidence of path-dependence as countries developed RCA at time t in products related to those in which they were already specialized at time t-3/t-5.14 So far only a few contributions have analyzed the pattern of trade diversification at the sub-national level which is likely to be the most significant since capabilities are lumpy across space and have a strong local dimension. Using US States data in the period 2002-2012, Donoso and Martin (2016) show that only the local capabilities have a role in the path-dependence process of industrial structure dynamics whereas the industrial structure at the national level has a negative effect on States’ export diversification. The authors also find that the higher internal migration, firm cluster strength and R&D spending over gross domestic product are, the stronger the effect of current structure on the probability of diversifying a State’s production. The importance of looking at sub-national areas is confirmed by the contribution of Boschma et al. (2013). The authors show that during the period 1988-2008, Spanish regions diversified into those new sectors that were related to the existing set of industries. Moreover, Boschma et al. (2013) find strong evidence that capabilities available at the regional level played a larger role than capabilities available at the country level in the emergence and development of new industries. A small but growing number of works have investigated the path-dependence of structural transformation using firm-level data (Neffke et al., 2011; Cirera et al., 2012; Lo Turco and Maggioni 2016). Using plant-level data for 70 Swedish regions in the period 1962–2002, Neffke et al. (2011) find evidence of path-dependence in the evolution of long-term production diversification, since industries that are technologically related to preexisting ones have a higher probability of entering the region’s production portfolio, whereas unrelated ones have a higher probability of exiting. Analogous results are found by Cirera et al. (2012) in Brazil for the period 2000–2009. The authors document that trade diversification mostly stems from related sectors. Diversification in sectors that are unrelated to the preexisting production basket is limited and mainly concerns vertically integrated firms which specialize in one or few stages in a specific value chain. Lo Turco and Maggioni (2016) using Turkish firm-level data show that the introduction of new products by manufacturing firms is significantly higher if related products are produced by the same firm or by other firms in the affected province. In this study, relatedness is also measured using “density” variables à laHidalgo et al. (2007). The local set of available capabilities is important—although less than internal (firm-specific) resources—in explaining what firms produce.15 All these studies confirm the importance of the set of available local capabilities in guiding the evolution of the comparative advantage of countries and/or regions and in shaping the introduction of new goods and products by firms. It is easier to produce new goods requiring capabilities that are already present in the local economy rather than those requiring different sets of capabilities. In other words, according to the capability approach, jumps over the PS are unlikely, and the process of structural change—measured as the evolution of the basket exported with RCA—is path-dependent. These studies also suggest that it is important to consider the geographical dimension of the changes in the production basket, since capabilities are not uniformly distributed within a country. Although the contributions described above document a large extent of path-dependence in the evolution of the production basket, these studies cannot be considered as a formal test of the PS framework, since they do not discriminate between the relatedness due to shared production capabilities (as the framework suggests) and spurious relatedness which is the result of a random process. Our analysis—using different definitions of “relatedness” and different “new entry” identification methods—aims at testing whether new products in Italian provinces are nonrandomly related to those previously exported with RCA. While other studies employ measures of “density”16 or “open-forest” indexes à laHausmann and Klinger (2007) for assessing path-dependence, we develop a test for inferring whether new entries in the export basket are related in a statistically significant way compared with randomly generated ones. The methodology developed in the article allows us to measure the extent to which structural change deviates from the hypothesis of path-dependence and, in turn, to shed light on the provincial characteristics that are significantly associated with “big leaps” over the PS (i.e., rather radical changes in the composition of the production basket). 3. Data and methodology 3.1 A general test of path-dependence: a dartboard approach What countries/regions produce and export changes over time as new products enter the production baskets. In this section, our aim is to test—using Italian NUTS 3 data (provinces)—whether new products that enter the export baskets at time t1 are related à laHidalgo et al. (2007) with the preexisting comparative advantage at time t0. As in the seminal contribution of these authors, the relatedness between any two products is measured using their proximity in the PS, i.e., the minimum of the pairwise conditional probability of being co-exported. We develop a “dart-board approach” to test the nonrandomness of the development of provincial production space over time during the period 2002–2011.17 Given the important role that the crisis played in reshaping the provincial pattern of trade (see Coniglio et al., 2016), we consider separately the precrisis period (2002–2006) from the crisis one (2007–2011).18 For this purpose, we select two base years as t0 (2002 and 2007) and two as t1 (2006 and 2011). We allow for a lag of 4 years between t0 and t1 to investigate short-term changes in the structure of provincial production.19 Data on Italian provinces” exports are provided by the A.D.ELE. Laboratory at six-digit whereas data on country exports used to create proximity matrixes are obtained from the UNCOMTRADE data set. In the first step of our analysis we need to define “new entries” as those goods that are not part of the production basket at time t0 and enter the provincial export basket at time t1. We recur to the standard definition of RCA and define the set of goods in the export basket as those with a Balassa index that is larger than 1; i.e., the ratio between the provincial export share and the world export share for each good is higher than unity. More precisely, in our study a new entry is a product with an RCA lower than 0.5 at t0 and higher than unity at t1.20 For each province k∈K we identify the set of new entries n∈Nk in both subperiods. In the second step, we compute—as in Hidalgo et al. (2007)—a MxM matrix containing the relatedness measures between any pair of goods ij exported in the world ( i,j ∈Wt1 where W is the set of goods exported in year t1, 2006 and 2011, respectively).21 More precisely the matrix is built as follows. For each country in the World, c, and for each of the two years, we denote xic as 1 if country c has an RCA in the production of good i and 0 otherwise: xic=1 if RCAic>10 otherwise, (1) where RCAicis the standard Balassa (1965) index employed as a measure of export specialization. Thus, after creating the country–product matrixes of RCAs, following Hausmann and Klinger (2007), we compute the distances between each couple of goods i and j as the minimum of the pairwise conditional probability of being co-exported: φij=min{P(xi|xj),(Pxj|xi)}, (2) where φi,j represents the proximity between any good i and j. In the third step, we denote with Bk,t the set of goods exported with RCA by province k at time t. We then define Di,k, a MxK matrix of relatedness measures between the new Nk products (entering the export basket between t0 and t1) and the preexisting export basket, for each province k ∈ K, as follows: Di,k=dik⁡φi,j when j∈Bk,t0, i∈Nkno value, (3) where dik⁡φi,j is a measure of proximity of the new product i with the preexisting export basket in province k. Given that the export basket at time t0 typically consists of several goods belonging to a variety of sectors (and hence positioned in different branches of the PS), the concept of relatedness can be specified either in an absolute term (i.e., the distance in the PS of each new product i with each of all the products already in the export basket) or in a relative term (i.e., the distance of new products relative to the overall preexisting basket). For this reason, we employ three alternative measures of relatedness: Maximum proximity: dik,m⁡φi,j=maxφi,j. (4) Average proximity: dik,a⁡φi,j=Σjφi,jJk. (5) Weighted average proximity:dik,wa⁡φi,j=Σj expjk,t0Σj expjk,t0φi,jJk, (6) with Jk being the number of goods in the export basket of province k at t0. Equation (4) represents relatedness of a new entrant product i with the set of products Jk exported with RCA at t0 as the maximum value among the proximities between i and all j∈Jk; in other words, distance is measured with respect to the closest product in the provincial PS that is already exported. As an alternative, the measure computed using equation (5) identifies the distance as the average proximity between good n and all the goods j∈Jk. Finally, equation (6) computes distance as the weighted average proximity with weights represented by the export share of goods in Jk at time t0. The relatedness (or unrelatedness) of new products can be easily appreciated with a simple graphical example. Figure 2 represents a subset of the export basket of a fictitious province at t0 using a simplified representation of the PS in which node A identifies a new product that enters the export basket at t1. In Panel A, the maximum proximity measure (equation (4)) is employed, thus the relevant preexisting product is the closest one. This measure represents an absolute dimension of proximity which probably better captures the role of available capabilities in shaping the path-dependence of product diversification. In Panel B, the relatedness takes into consideration all products exported with RCA in time t0 (respectively equation (5) if no weights are employed and 6 otherwise). These two measures can provide information on the relative degree of proximity of new entries. Figure 2. View largeDownload slide Relatedness of new entries in the export basket: an illustration. Blue dots represent products exported with RCA at time t0, while red ones represent new entrants at time t1. Figure 2. View largeDownload slide Relatedness of new entries in the export basket: an illustration. Blue dots represent products exported with RCA at time t0, while red ones represent new entrants at time t1. Once new entries and proximities have been defined, the subsequent step is to perform a formal test that allows us to reject the hypothesis that the new entries in province k are randomly related to the initial export basket of that province. Our idea is that if new entries are driven by path-dependence—as the PS framework asserts—we should observe that the distribution of relatedness based on the observed new entries ( Nk) significantly differs from that of randomly generated new entries of identical size and, more precisely that it is significantly more concentrated at high proximity levels compared to the counterfactual distribution. By drawing a parallel between the PS and a dartboard, each new entry is equivalent to a dart and will be localized in a given place on the board. The actual data will tell us where the nk∈Nk darts are localized, but we need a counterfactual distribution of localization of darts on the board to say if the localization of actual data is significantly different. We build a counterfactual distribution of relatedness using 1000 random draws of size equal to Nk for each province k from the set of products not exported with comparative advantage at t0. We then reject the null hypothesis of random relatedness when the relatedness of actual new entries produces a pattern that is different in a statistically significant way from the random counterfactual. More specifically, we implement a Kernel smoothed density estimation of relatedness of new entries in provincial export baskets.22 Like Duranton and Overman (2005), we estimate a smoothed Kernel density function of relatedness for any level of proximity, d, defined as: K^d≡1Σi=1MΣk=1KIi,khΣi=1MΣk=1Kfd-di,kh, (7) with densities calculated nonparametrically using a Gaussian Kernel function with bandwidth h set according to Silverman’s optimal rule of thumb (Silverman, 1986), where di,k is measured using one of the three alternative definitions of relatedness reported in equations (4)–(6), while Ii, k is a product by province matrix of size MxK which has values of 1 for each new entrant product for each province and 0 otherwise. Therefore, Σi=1MΣk=1KIi,k is equal to the total number of new entries across all provinces from t0 to t1. Finally, we build a counterfactual distribution of relatedness and compare it with the actual one obtained from equation (7). The counterfactual density function is based on simulated relatedness computed from 1000 random draws of size Σi=1MΣk=1KIi,k (total number of entries).23 For each definition of relatedness, using a Kolmogorov–Smirnov test for first-order stochastic dominance, we compare the distribution of our actual data with the randomly generated one and assess whether: (i) the former significantly diverge from the latter (i.e., if the two distributions cannot be generated by the same random process); (ii) the relatedness of actual data is significantly more concentrated at higher level of proximities (i.e., the cumulative distribution function of the actual data distributions lies below the mean values counterfactual one). This simple test can be considered as a global test of path-dependence.24 Moreover, employing the methodology from Duranton and Overman (2005), it is possible to develop a statistical test for nonrandom concentration of actual new entrants compared to the counterfactual at each level of proximity d∈0, 1, i.e. a local test ofpath-dependence. Also this approach is simple and intuitive. From the random draws we obtain 1000 density estimates for each level of proximity, d. At each particular proximity between 0 and 1, we can construct a confidence interval that contains 90% of all estimated values (upper limit the 95th percentile, lower limit the 5th percentile). Local path-dependence is found if at high levels of proximity the actual data kernel distribution lies above the upper boundary. At a particular level of proximity, if the kernel density of actual data lies below the upper bound, we cannot reject the hypothesis of random relatedness. Besides the intuitive and direct Kolmogorov–Smirnov test on the overall distribution of actual and counterfactual data, we believe that applying the Duranton and Overman methodology for detecting nonrandom concentration of new entrant products at different levels of relatedness (our inverse measure of distance) gives added value and originality to the present work and allows us to identify—for each definition of distance employed—the threshold value beyond which new products can be said to be the outcome of a path-dependent process. 3.2 New entry in the export basket: a single product’s test of path-dependence The methodology explained above allows us to test the “aggregate” (or general) compliance of the evolution of provincial specialization with the predictions of the PS framework. In this section we describe a Monte Carlo methodology which allows us to shift the analysis to each of the new products that enters provinces’ export baskets at time t1. For each Italian province k∈K, we randomly draw, from the set of products that were not present in the export basket at time t025, a number of products equal to the actual number of new goods that enter in the export basket at time t1, Jk and compute proximities using equations (4)–(6) and generate an average value per draw. The random draw is carried out 2000 times to compute a distribution of random average proximities which represent our province-specific statistical counterfactual. From these counterfactual distributions of proximities, we identify the 95th percentile values. In this way, for each new product entering the Italian provinces' export basket between t0 and t1, we can test its compliance to the PS (statistically nonrandom relatedness). 4. Results 4.1 Testing nonrandom concentration in the PS As described in the methodology section, the null hypothesis of our global test is that new products are randomly located in the PS; a rejection of the null hypothesis implies that the distribution of actual data is different from the distribution of the counterfactual and significantly more concentrated at high levels of proximities with respect to the initial comparative advantage. In Table 1 we report the main descriptive statistics regarding the new entries in the provincial export baskets for the two subperiods considered in our analysis, respectively 2002–2006 (precrisis period) and 2007–2011 (crisis period). The number of new products that enter the Italian provinces’ export baskets in the two periods is slightly increased, 13,024 and 14,340, respectively.26 As expected, the mean values of proximities differ according to the employed definition. When using the maximum proximity, the “distance” between the new products and the initial specialization is lower. When using the relative measured of proximity (equations (5)–(6)), these distances are higher, since the proximity is computed with respect to all goods exported with an RCA larger than unity at time t0. Interestingly, from Table 1 it is clear that the mean values of proximities—regardless of the definition of new products or the measure of proximity adopted—increase during the second subperiod. It is important to note that we cannot infer evidence of increasing relatedness from this change in average levels of proximities, since the network structure of the provincial baskets is different in the two subperiods. For this reason, a methodology that explicitly takes into account how provincial export baskets evolve is fundamental for testing the dynamics of specialization. Table 1. New products in the export baskets of Italian provinces: descriptive statistics Number of jumps Mean Standard deviation Minimum Maximum 2002–2006 Maximum proximity 13,024 0.5085 0.0906 0.125 1 Average proximity 13,024 0.1876 0.0402 0.0607 0.396 Weighted average proximity 13,024 0.1858 0.0569 0.0042 0.4841 2007–2011 Maximum proximity 14,340 0.5119 0.0942 0.125 1 Average proximity 14,340 0.1936 0.0416 0.0503 0.3714 Weighted average proximity 14,340 0.1945 0.0592 0.0058 0.4576 Number of jumps Mean Standard deviation Minimum Maximum 2002–2006 Maximum proximity 13,024 0.5085 0.0906 0.125 1 Average proximity 13,024 0.1876 0.0402 0.0607 0.396 Weighted average proximity 13,024 0.1858 0.0569 0.0042 0.4841 2007–2011 Maximum proximity 14,340 0.5119 0.0942 0.125 1 Average proximity 14,340 0.1936 0.0416 0.0503 0.3714 Weighted average proximity 14,340 0.1945 0.0592 0.0058 0.4576 Table 1. New products in the export baskets of Italian provinces: descriptive statistics Number of jumps Mean Standard deviation Minimum Maximum 2002–2006 Maximum proximity 13,024 0.5085 0.0906 0.125 1 Average proximity 13,024 0.1876 0.0402 0.0607 0.396 Weighted average proximity 13,024 0.1858 0.0569 0.0042 0.4841 2007–2011 Maximum proximity 14,340 0.5119 0.0942 0.125 1 Average proximity 14,340 0.1936 0.0416 0.0503 0.3714 Weighted average proximity 14,340 0.1945 0.0592 0.0058 0.4576 Number of jumps Mean Standard deviation Minimum Maximum 2002–2006 Maximum proximity 13,024 0.5085 0.0906 0.125 1 Average proximity 13,024 0.1876 0.0402 0.0607 0.396 Weighted average proximity 13,024 0.1858 0.0569 0.0042 0.4841 2007–2011 Maximum proximity 14,340 0.5119 0.0942 0.125 1 Average proximity 14,340 0.1936 0.0416 0.0503 0.3714 Weighted average proximity 14,340 0.1945 0.0592 0.0058 0.4576 Figure 3 represents the Kernel smoothed density estimates and the cumulative density functions (CDFs) for our three definitions of proximity in the subperiod 2002–2006.27 The horizontal axes measure the proximity between new products at t1 and the bundle of goods exported at time t0. The higher the value of our measure of proximity is, the closer the new entrant product to the export basket at time t0. In all these estimates, we can see that for high levels of relatedness, the distribution of proximities in the actual data is above that of the counterfactual. Such outcome is confirmed also looking at the plots of the empirical CDFs for all our measures of relatedness (Panels d, e, and f) and in particular by performing a Kolmogorov–Smirnov Test for first-order stochastic dominance. The results are reported in Table 2 and strongly reject the null hypothesis that actual data and counterfactuals are drawn from the same distribution, accepting the alternative that the CDF of actual data is dominated by the counterfactual one. Our global test confirms a statistically significant degree of path-dependence in the evolution of provincial export baskets confirming the PS hypothesis. Table 2. Kolmogorov–Smirnov test for first-order stochastic dominance over new entries' relatedness distributions, precrisis Years Relatedness measured as One sided 2002–2006 Maximum proximity Dn 0.1636 P-value (0.000) Average proximity Dn 0.2436 P-value (0.000) Weighted average proximity Dn 0.2134 P-value (0.000) Years Relatedness measured as One sided 2002–2006 Maximum proximity Dn 0.1636 P-value (0.000) Average proximity Dn 0.2436 P-value (0.000) Weighted average proximity Dn 0.2134 P-value (0.000) Table 2. Kolmogorov–Smirnov test for first-order stochastic dominance over new entries' relatedness distributions, precrisis Years Relatedness measured as One sided 2002–2006 Maximum proximity Dn 0.1636 P-value (0.000) Average proximity Dn 0.2436 P-value (0.000) Weighted average proximity Dn 0.2134 P-value (0.000) Years Relatedness measured as One sided 2002–2006 Maximum proximity Dn 0.1636 P-value (0.000) Average proximity Dn 0.2436 P-value (0.000) Weighted average proximity Dn 0.2134 P-value (0.000) Figure 3. View largeDownload slide Kernel density estimates and CDFs for actual and counterfactual data in the precrisis period. Figure 3. View largeDownload slide Kernel density estimates and CDFs for actual and counterfactual data in the precrisis period. The local test of path-dependence adds information on the range of proximities for which we can exclude, given the comparison between actual and counterfactual data, the null hypothesis of random relatedness. Using absolute measure of proximity, equation (4), the hypothesis of random relatedness is rejected for proximities ranging from 0.46 to 0.84; in this range, where 70.8% of new entries fall, the actual data’s distribution lies above our counterfactual 95th percentile’s threshold (90% confidence interval upper boundary) (see Figure 3, Panel a). Using alternative measures of proximity, equations (5) and (6), produce highly similar results. In Panelb of Figure 3, we report the Kernel densities when we use the average proximity of new products with all the products composing the export basket at time t0. The Kernel distribution of actual data is above the counterfactual 95th percentile distribution (our randomness threshold) for values of proximity between 0.17 and 0.4. The 70.35% of new products that enter the Italian export basket in 2006 fall in this range. Similar results are obtained when using the weighted average relatedness specified in equation (6). In this case, the range of nonrandom proximities is between 0.16 and 0.49 which represents 71.34% of new entries between 2002 and 2006.28 Overall our results thus confirm that using alternative measures of proximities more than 70% of the new products is related in the PS to those already included in the provincial export baskets. Is this result of a relatively strong path-dependence confirmed for new products that enter the export basket during the crisis period? The Kernel density estimations and the CDFs for the period 2007–2011 are reported in the panels of Figure 4. For both absolute and relative measures of proximities, the strong path-dependence in the evolution of the export basket is confirmed by the Kernel comparison, the plot of estimated CDFs and the results of the Kolmogorov–Smirnov Test which are reported in Table 3. Also in this case, we accept the alternative hypothesis that the estimated CDFs of actual data are dominated by the counterfactual ones (i.e., significantly higher concentration of actual data at high level of proximities). Nevertheless, we note a slight shift of the threshold values to the right, i.e., the null hypothesis of randomness is rejected for slightly higher values of proximity. In line with this result, we find that on average a smaller percentage of new entries is significantly related to the preexisting export basket (respectively, 69.6%, 68.9%, and 72.5% for the three definitions of proximities employed). Table 3. Kolmogorov–Smirnov test for first-order stochastic dominance over new entries' relatedness distributions, crisis Years Relatedness measured as One sided 2007–2011 Maximum proximity Dn 0.1729 P-value (0.000) Average proximity Dn 0.2466 P-value (0.000) Weighted average proximity Dn 0.2142 P-value (0.000) Years Relatedness measured as One sided 2007–2011 Maximum proximity Dn 0.1729 P-value (0.000) Average proximity Dn 0.2466 P-value (0.000) Weighted average proximity Dn 0.2142 P-value (0.000) Table 3. Kolmogorov–Smirnov test for first-order stochastic dominance over new entries' relatedness distributions, crisis Years Relatedness measured as One sided 2007–2011 Maximum proximity Dn 0.1729 P-value (0.000) Average proximity Dn 0.2466 P-value (0.000) Weighted average proximity Dn 0.2142 P-value (0.000) Years Relatedness measured as One sided 2007–2011 Maximum proximity Dn 0.1729 P-value (0.000) Average proximity Dn 0.2466 P-value (0.000) Weighted average proximity Dn 0.2142 P-value (0.000) Figure 4. View largeDownload slide Kernel density estimates and CDFs for actual and counterfactual data in the crisis period. Figure 4. View largeDownload slide Kernel density estimates and CDFs for actual and counterfactual data in the crisis period. In other words, although the crisis has not halted the development of new products in quantitative terms, the number of “new apples that fall closer to the tree” is slightly lower compared to the precrisis period. This result seems to go in the same direction as the Schumpeterian process of creative destruction during crisis confirming the role of such shocks as engines of structural change (Schumpeter, 1942), at least in the very short term. The analysis of the Kernel smoothed densities shows us that, regardless of the definition used to compute our measure of proximity and the definition of new products (see Appendix 1), our general test of nonrandomness highlights a clear pattern: new products entering the export baskets of Italian provinces, both in the precrisis and the crisis periods, are not randomly distributed over the network of relatedness between goods developed by Hidalgo et al. (2007). Indeed, a large share of new entries is related in a nonrandom manner to the initial export basket. 4.2 Test on single products In the previous paragraph, we have shown that the evolution of Italian provincial export baskets follows a nonrandom pattern that confirms the existence of strong path-dependence for most of the (new) products in which provinces develop an RCA. Our local test of non-random relatedness also reveals that a significant share of new products—over 30%—defeats the “static” comparative advantage contrary to the PS framework’s dictates.29 In our opinion, these “apples that fall far from trees” are probably the most interesting ones for informing the current debate on structural change and industrial policy. It is thus interesting to analyze whether there is any province- or product-specific characteristics that are systematically associated with these “long distance” jumps over the PS of Italian provinces. In Table 4, we report the main results from our test of nonrandom relatedness performed on each single product that entered the Italian export basket in the two periods considered (see previous section). In particular, we show the percentage of new goods by macro-sector and macro-geographical areas for which we confirm the hypothesis of nonrandom relatedness (i.e., path-dependence). For completeness of the analysis we report also the average measures of complexity—measured using ProdY—by macro-sector and geographical areas (equivalent to NUTS 1). Table 4. Test on single products Northwest Northeast Center South and Isles Italy Number of new entries Maximum proximity (%) Number of new entries Maximum proximity (%) Number of new entries Maximum proximity (%) Number of new entries Maximum proximity (%) Number of new entries Maximum proximity (%) New entrants’ average ProdY Precrisis Animal and animal products 97 53 93 45 41 49 134 42 365 46 16,260.42 Vegetable products 88 35 104 33 84 32 202 50 478 40 11,260.73 Foodstuffs 100 33 103 52 80 59 207 61 490 53 12,503.81 Mineral products 56 16 47 21 32 22 48 13 183 17 10,347.15 Chemicals and allied industries 340 53 239 50 154 49 260 50 993 51 21,347.59 Plastics/rubbers 136 71 132 64 90 63 144 68 502 67 18,775.04 Raw hides, skins, leather, and furs 60 58 85 39 85 47 92 50 322 48 10,038.09 Wood and wood products 145 39 143 57 99 62 99 73 486 56 17,926.95 Textiles 980 69 871 73 713 74 871 75 3435 73 10,867.97 Footwear/headgear 31 52 43 60 50 52 46 52 170 54 8,705.37 Stone/glass 84 58 105 57 93 62 143 59 425 59 15,365.29 Metals 432 57 398 60 220 53 341 58 1391 57 18,513.54 Machinery/electrical 759 65 699 65 551 63 738 62 2747 64 22,508.25 Transportation 63 43 69 43 51 43 94 48 277 45 17,672.45 Miscellaneous 204 54 212 51 171 50 173 61 760 54 20,778.79 New entrants’ average ProdY 17,117.69 16,751.56 16,455.66 15,855.12 16,547.71 Crisis Animal and animal products 94 43 102 55 43 40 179 42 418 45 17,468.08 Vegetable products 109 31 150 34 95 35 287 46 641 39 10,512.61 Foodstuffs 128 50 104 43 100 48 276 60 608 53 12,407.51 Mineral products 64 25 45 22 48 25 62 35 219 27 11,498.25 Chemicals and allied industries 398 54 235 46 164 57 245 59 1042 54 20,747.07 Plastics/rubbers 182 68 104 70 110 67 174 69 570 69 19,069.08 Raw hides, skins, leather, and furs 62 42 102 43 90 58 84 46 338 48 11,962.28 Wood and wood products 182 52 196 61 130 62 171 59 679 58 16,138.75 Textiles 808 67 789 74 683 74 677 74 2957 72 11,072.29 Footwear/headgear 42 43 53 77 41 63 48 60 184 62 9,410.77 Stone/glass 171 63 139 68 123 63 195 69 628 66 15,487.23 Metals 544 60 443 63 273 66 432 64 1692 63 18,635.83 Machinery/electrical 872 70 813 71 676 69 756 63 3117 69 22,667.66 Transportation 97 57 75 44 61 52 134 57 367 53 17,583.94 Miscellaneous 269 58 205 68 195 59 211 54 880 60 21,636.36 New entrants’ average ProdY 17,461.96 17,180.05 16,824.66 16,411.97 16,978.38 Northwest Northeast Center South and Isles Italy Number of new entries Maximum proximity (%) Number of new entries Maximum proximity (%) Number of new entries Maximum proximity (%) Number of new entries Maximum proximity (%) Number of new entries Maximum proximity (%) New entrants’ average ProdY Precrisis Animal and animal products 97 53 93 45 41 49 134 42 365 46 16,260.42 Vegetable products 88 35 104 33 84 32 202 50 478 40 11,260.73 Foodstuffs 100 33 103 52 80 59 207 61 490 53 12,503.81 Mineral products 56 16 47 21 32 22 48 13 183 17 10,347.15 Chemicals and allied industries 340 53 239 50 154 49 260 50 993 51 21,347.59 Plastics/rubbers 136 71 132 64 90 63 144 68 502 67 18,775.04 Raw hides, skins, leather, and furs 60 58 85 39 85 47 92 50 322 48 10,038.09 Wood and wood products 145 39 143 57 99 62 99 73 486 56 17,926.95 Textiles 980 69 871 73 713 74 871 75 3435 73 10,867.97 Footwear/headgear 31 52 43 60 50 52 46 52 170 54 8,705.37 Stone/glass 84 58 105 57 93 62 143 59 425 59 15,365.29 Metals 432 57 398 60 220 53 341 58 1391 57 18,513.54 Machinery/electrical 759 65 699 65 551 63 738 62 2747 64 22,508.25 Transportation 63 43 69 43 51 43 94 48 277 45 17,672.45 Miscellaneous 204 54 212 51 171 50 173 61 760 54 20,778.79 New entrants’ average ProdY 17,117.69 16,751.56 16,455.66 15,855.12 16,547.71 Crisis Animal and animal products 94 43 102 55 43 40 179 42 418 45 17,468.08 Vegetable products 109 31 150 34 95 35 287 46 641 39 10,512.61 Foodstuffs 128 50 104 43 100 48 276 60 608 53 12,407.51 Mineral products 64 25 45 22 48 25 62 35 219 27 11,498.25 Chemicals and allied industries 398 54 235 46 164 57 245 59 1042 54 20,747.07 Plastics/rubbers 182 68 104 70 110 67 174 69 570 69 19,069.08 Raw hides, skins, leather, and furs 62 42 102 43 90 58 84 46 338 48 11,962.28 Wood and wood products 182 52 196 61 130 62 171 59 679 58 16,138.75 Textiles 808 67 789 74 683 74 677 74 2957 72 11,072.29 Footwear/headgear 42 43 53 77 41 63 48 60 184 62 9,410.77 Stone/glass 171 63 139 68 123 63 195 69 628 66 15,487.23 Metals 544 60 443 63 273 66 432 64 1692 63 18,635.83 Machinery/electrical 872 70 813 71 676 69 756 63 3117 69 22,667.66 Transportation 97 57 75 44 61 52 134 57 367 53 17,583.94 Miscellaneous 269 58 205 68 195 59 211 54 880 60 21,636.36 New entrants’ average ProdY 17,461.96 17,180.05 16,824.66 16,411.97 16,978.38 Note: Relatedness measured as “maximum proximity.” Table 4. Test on single products Northwest Northeast Center South and Isles Italy Number of new entries Maximum proximity (%) Number of new entries Maximum proximity (%) Number of new entries Maximum proximity (%) Number of new entries Maximum proximity (%) Number of new entries Maximum proximity (%) New entrants’ average ProdY Precrisis Animal and animal products 97 53 93 45 41 49 134 42 365 46 16,260.42 Vegetable products 88 35 104 33 84 32 202 50 478 40 11,260.73 Foodstuffs 100 33 103 52 80 59 207 61 490 53 12,503.81 Mineral products 56 16 47 21 32 22 48 13 183 17 10,347.15 Chemicals and allied industries 340 53 239 50 154 49 260 50 993 51 21,347.59 Plastics/rubbers 136 71 132 64 90 63 144 68 502 67 18,775.04 Raw hides, skins, leather, and furs 60 58 85 39 85 47 92 50 322 48 10,038.09 Wood and wood products 145 39 143 57 99 62 99 73 486 56 17,926.95 Textiles 980 69 871 73 713 74 871 75 3435 73 10,867.97 Footwear/headgear 31 52 43 60 50 52 46 52 170 54 8,705.37 Stone/glass 84 58 105 57 93 62 143 59 425 59 15,365.29 Metals 432 57 398 60 220 53 341 58 1391 57 18,513.54 Machinery/electrical 759 65 699 65 551 63 738 62 2747 64 22,508.25 Transportation 63 43 69 43 51 43 94 48 277 45 17,672.45 Miscellaneous 204 54 212 51 171 50 173 61 760 54 20,778.79 New entrants’ average ProdY 17,117.69 16,751.56 16,455.66 15,855.12 16,547.71 Crisis Animal and animal products 94 43 102 55 43 40 179 42 418 45 17,468.08 Vegetable products 109 31 150 34 95 35 287 46 641 39 10,512.61 Foodstuffs 128 50 104 43 100 48 276 60 608 53 12,407.51 Mineral products 64 25 45 22 48 25 62 35 219 27 11,498.25 Chemicals and allied industries 398 54 235 46 164 57 245 59 1042 54 20,747.07 Plastics/rubbers 182 68 104 70 110 67 174 69 570 69 19,069.08 Raw hides, skins, leather, and furs 62 42 102 43 90 58 84 46 338 48 11,962.28 Wood and wood products 182 52 196 61 130 62 171 59 679 58 16,138.75 Textiles 808 67 789 74 683 74 677 74 2957 72 11,072.29 Footwear/headgear 42 43 53 77 41 63 48 60 184 62 9,410.77 Stone/glass 171 63 139 68 123 63 195 69 628 66 15,487.23 Metals 544 60 443 63 273 66 432 64 1692 63 18,635.83 Machinery/electrical 872 70 813 71 676 69 756 63 3117 69 22,667.66 Transportation 97 57 75 44 61 52 134 57 367 53 17,583.94 Miscellaneous 269 58 205 68 195 59 211 54 880 60 21,636.36 New entrants’ average ProdY 17,461.96 17,180.05 16,824.66 16,411.97 16,978.38 Northwest Northeast Center South and Isles Italy Number of new entries Maximum proximity (%) Number of new entries Maximum proximity (%) Number of new entries Maximum proximity (%) Number of new entries Maximum proximity (%) Number of new entries Maximum proximity (%) New entrants’ average ProdY Precrisis Animal and animal products 97 53 93 45 41 49 134 42 365 46 16,260.42 Vegetable products 88 35 104 33 84 32 202 50 478 40 11,260.73 Foodstuffs 100 33 103 52 80 59 207 61 490 53 12,503.81 Mineral products 56 16 47 21 32 22 48 13 183 17 10,347.15 Chemicals and allied industries 340 53 239 50 154 49 260 50 993 51 21,347.59 Plastics/rubbers 136 71 132 64 90 63 144 68 502 67 18,775.04 Raw hides, skins, leather, and furs 60 58 85 39 85 47 92 50 322 48 10,038.09 Wood and wood products 145 39 143 57 99 62 99 73 486 56 17,926.95 Textiles 980 69 871 73 713 74 871 75 3435 73 10,867.97 Footwear/headgear 31 52 43 60 50 52 46 52 170 54 8,705.37 Stone/glass 84 58 105 57 93 62 143 59 425 59 15,365.29 Metals 432 57 398 60 220 53 341 58 1391 57 18,513.54 Machinery/electrical 759 65 699 65 551 63 738 62 2747 64 22,508.25 Transportation 63 43 69 43 51 43 94 48 277 45 17,672.45 Miscellaneous 204 54 212 51 171 50 173 61 760 54 20,778.79 New entrants’ average ProdY 17,117.69 16,751.56 16,455.66 15,855.12 16,547.71 Crisis Animal and animal products 94 43 102 55 43 40 179 42 418 45 17,468.08 Vegetable products 109 31 150 34 95 35 287 46 641 39 10,512.61 Foodstuffs 128 50 104 43 100 48 276 60 608 53 12,407.51 Mineral products 64 25 45 22 48 25 62 35 219 27 11,498.25 Chemicals and allied industries 398 54 235 46 164 57 245 59 1042 54 20,747.07 Plastics/rubbers 182 68 104 70 110 67 174 69 570 69 19,069.08 Raw hides, skins, leather, and furs 62 42 102 43 90 58 84 46 338 48 11,962.28 Wood and wood products 182 52 196 61 130 62 171 59 679 58 16,138.75 Textiles 808 67 789 74 683 74 677 74 2957 72 11,072.29 Footwear/headgear 42 43 53 77 41 63 48 60 184 62 9,410.77 Stone/glass 171 63 139 68 123 63 195 69 628 66 15,487.23 Metals 544 60 443 63 273 66 432 64 1692 63 18,635.83 Machinery/electrical 872 70 813 71 676 69 756 63 3117 69 22,667.66 Transportation 97 57 75 44 61 52 134 57 367 53 17,583.94 Miscellaneous 269 58 205 68 195 59 211 54 880 60 21,636.36 New entrants’ average ProdY 17,461.96 17,180.05 16,824.66 16,411.97 16,978.38 Note: Relatedness measured as “maximum proximity.” The results in Table 4 highlight a heterogeneous pattern across sectors (HS sections) and Italian macro-regions. The percentages reported represent the number of products for which the null hypothesis of random relatedness is rejected over the total number of new entries for each region and each product section. The higher the share, the more a provincial export basket follows a path-dependent pattern. Textile products are, on average, those with the highest number of new entries and percentages of randomness rejection, irrespective of the region that is taken into account (overall ratio of 73% and 72% in the precrisis and crisis periods, respectively). Note that these new entries have on average a rather low level of complexity as measured by the (unweighted) average Prody (between 10.9 and 11 thousands US$). High frequencies of new entries are recorded also for machinery and electrical products with a total number of new entries increasing in the second subperiod from 2747 to 3117. In this sector, the degree of path-dependence is also quite high and the percentage of relatedness ranges from 62% in the precrisis period for southern provinces to 71% in the crisis period for northeastern provinces. The level of complexity of the new products in machinery and electronics is the highest by macro-sector. A low degree of path-dependence is found, as expected, given the weak role of local capabilities in resource extractive industries, for mineral products (17% and 27% in precrisis and crisis periods, respectively). Heterogeneity within sectors is evident in food industries during the 2002–2006 period; results show path-dependence for 33% of products entering the export basket of northwestern provinces with 61% for southern regions. A similar outcome is shown—for the crisis period—in the Footwear section for which we find 43% of related entries for the provinces in the northwestern and 77% for northeastern regions. The number of new entries in the South and Islands is larger than in other Italian macro-regions (new entries in South and Islands represent the 27.6% and 27.4% of total new entries in the precrisis and crisis period, respectively). This is likely to be the result of the initial lower degree of export diversification of southern provinces. Note also that in both periods the average complexity of new exported goods in South and Islands is lower than the national average. The high degree of heterogeneity highlighted in Table 4 suggests that the characteristics of local economies might play an important role in influencing the pattern of structural change. Furthermore product specificities might influence the degree of path-dependence as the sectoral differences seem to suggest. It is interesting to notice that a higher degree of path-dependency is not necessarily more likely to happen in more high-tech sectors. In fact we find high shares of related new entries both for machinery and electronics—the macro-sector with the highest complexity—as well textile—one of the sectors with the lowest level of complexity. In what follows we explore these diverging patterns using a parametric approach. 4.3 Which provinces “defeat” the static comparative advantage? A probit analysis What economies produce matters for growth as emphasized by previous studies (Hausmann et al., 2007; Minondo, 2010; Jarreau and Poncet, 2012; Kadochnikov and Fedyunina, 2013; Ferrarini and Scaramozzino, 2016).30 The evolution of the production basket crucially depends on local capabilities. The transfer of productive resources to new sectors and the development of new production capabilities or a recombination of existing ones are quintessential to structural transformation and growth. In this section, we focus on a particularly important research question: What drives radical (unrelated) changes in the composition of provincial production baskets?31 To this purpose we estimate a probit model with the aim of investigating which factors are associated with a higher ability of provinces to diversify away from the initial comparative advantage in both precrisis and crisis periods. Our dependent variable, random_entry, is a dummy which is equal to 1 if the new product i entering the production space of province k at time t1 is statistically unrelated to preexisting economic specialization as defined in the previous section, and 0 otherwise. We define new entries using our preferred measure of distance as defined in equation (4), i.e., maximum proximity.32 Our estimated model is the following: Prob(random_entryik)=αik+βXXk+βZZi+βINDINDi+εik. (8) where Xk includes our main province-level covariates, Zi includes product-level controls, and INDi is a set of (macro)sector fixed effects.33 We employ the measure developed by Hausmann et al. (2007), ExpY, as a proxy for the level of sophisticatedness of the export basket.34 We expect that the more complex the degree of sophisticatedness of the production basket is, the higher the probability there will be radical changes. In fact, since more complex goods require a broader set of capabilities, it would be relatively easier in these economies to redeploy these capabilities to develop new unrelated products. Following Boschma and Iammarino (2009), we include a measure of export diversification variety in our specification as defined in Frenken (2007).35 We expect more differentiated economies producing a large number of varieties to be endowed with a relatively larger and broader set of production capabilities which allows the provincial economy to diversify away from the initial production basket. This positive association is expected to be stronger when the diversification in terms of unrelated varieties is higher, which presumably implies the use of heterogeneous local capabilities (as in Castaldi et al., 2015 in the context of the emergence of technological breakthroughs). Conversely, economies that are diversified within a specific sector (related varieties) may experience a “lock-in” effect which hampers diversification into other areas of the PS. Moreover, we include the variable trade openness to control for the provincial exposure to international trade and international knowledge transfers which may affect the ability of the provincial economy to diversify its production bundle “away” from the preexisting one. As a proxy of human capital, we also include a variable measuring the number of students enrolled in tertiary education over the total active population (tertiary education) and the patent rate measured as the number of patent per million inhabitants. The concentration of economic activities in the geographical space might affect the emergence of new product specializations and the acquisition of new technological capabilities (see Boschma et al., 2015). The effect on the probability of radical changes might be ambiguous since agglomeration boosts both information flows and the potential pooling of production capabilities but, on the other hand, might favor a strong degree of specialization in a few industries or sectors. As a proxy of agglomeration of economic activities we include population density in our model. Finally, we control for the size of the provincial population, as we expect that a larger population has, ceteris airbus, a stronger potential of generating radical changes in the production basket. Our specifications include two product-level controls. First, we include ProdY as a proxy of the sophisticatedness of the new products that enter the provincial production basket at time t1. Second, a measure of the “centrality” of the new products in the network of relatedness—equals to its average proximity—is included. Both measures, together with the macro-industry fixed effects, capture product-specific characteristics which affect the dependent variable but are unrelated to provincial features (our main variables of interest). Finally, to control for the macro-areas, we include three dummies variables (Northwest, Northeast, and Center).36 Descriptive statistics of the dependent variable and covariates for both the precrisis and crisis periods are reported in Table 5. Table 5. Descriptive statistics Variable Observed Mean Standard deviation Minimum Maximum Description Precrisis: 2002–2006 PS jump 13,024 0.3997236 0.4898603 0 1 Dummy variable: 1 if new entry is random, 0 otherwise ProdY t1(log) 13,024 9.505272 0.7283572 3.85543 11.256 See Appendix 2 Centrality t1 13,024 0.1528922 0.0306515 0.0269282 0.231003 Average relatedness between each product and ALL the other products ExpY t0 (log) 13,024 9.866243 0.1597781 9.294399 10.22813 See Appendix 2 Variety t0 (log) 13,024 1.846666 0.2439395 0.6236902 2.238506 See Appendix 2 Related variety t0 (log) 13,024 1.876836 0.7684997 −2.957147 3.2948 See Appendix 2 Unrelated variety t0 (log) 13,024 5.333217 0.8126831 1.619947 6.468346 See Appendix 2 Trade openness t0 (log) 13,024 0.3756517 0.2102692 0.0178385 1.284592 (Import + export)/gross value added Tertiary education t0 12,848 0.038368 0.037405 2.19E-05 0.1934 Number of students enrolled in university over population 15+ Patent_rate t0 13,024 85.50854 73.19639 1 332.048 Number of patents over million inhabitants. Population density t0 13,024 0.2898346 0.3896068 0.0367315 2.611597 Number of inhabitants over thousand of squre kilometers Population t0 (log) 13,024 13.0173 0.7562106 11.40497 15.1265 Provincial population in log Crisis: 2007–2011 PS jump 14,340 0.3760112 0.4843998 0 1 Dummy variable: 1 if new entry is random, 0 otherwise ProdY t1(log) 14,340 9.539061 0.7151328 5.580651 11.24809 See Appendix 2 Centrality t1 14,340 0.1536002 0.031584 0.0077109 0.229469 Average relatedness between each product and ALL the other products ExpY t0 (log) 14,340 10.05196 0.1404983 9.602266 10.51017 See Appendix 2 Variety t0 (log) 14,340 1.828332 0.2639739 −0.015982 2.267477 See Appendix 2 Related variety t0 (log) 14,340 1.734126 0.90491 −4.60517 3.160912 See Appendix 2 Unrelated variety t0 (log) 14,340 5.287507 0.9041724 0.3585037 6.450047 See Appendix 2 Trade openness t0 (log) 14,340 0.4682068 0.2732955 0.0012226 2.652769 (Import + export)/gross value added Tertiary education t0 14,340 0.036236 0.033688 0 0.180167 Number of of students enrolled in university over population 15+. Patent_rate t0 14,340 74.15224 68.82898 1 497.0236 Number of patents over million inhabitants. Population density t0 14,340 0.286271 0.40032 0.0311 2.593696 Number of inhabitants over thousand of sq.kms Population t0 (log) 14,340 12.99877 0.755691 10.9626 15.1629 Variable Observed Mean Standard deviation Minimum Maximum Description Precrisis: 2002–2006 PS jump 13,024 0.3997236 0.4898603 0 1 Dummy variable: 1 if new entry is random, 0 otherwise ProdY t1(log) 13,024 9.505272 0.7283572 3.85543 11.256 See Appendix 2 Centrality t1 13,024 0.1528922 0.0306515 0.0269282 0.231003 Average relatedness between each product and ALL the other products ExpY t0 (log) 13,024 9.866243 0.1597781 9.294399 10.22813 See Appendix 2 Variety t0 (log) 13,024 1.846666 0.2439395 0.6236902 2.238506 See Appendix 2 Related variety t0 (log) 13,024 1.876836 0.7684997 −2.957147 3.2948 See Appendix 2 Unrelated variety t0 (log) 13,024 5.333217 0.8126831 1.619947 6.468346 See Appendix 2 Trade openness t0 (log) 13,024 0.3756517 0.2102692 0.0178385 1.284592 (Import + export)/gross value added Tertiary education t0 12,848 0.038368 0.037405 2.19E-05 0.1934 Number of students enrolled in university over population 15+ Patent_rate t0 13,024 85.50854 73.19639 1 332.048 Number of patents over million inhabitants. Population density t0 13,024 0.2898346 0.3896068 0.0367315 2.611597 Number of inhabitants over thousand of squre kilometers Population t0 (log) 13,024 13.0173 0.7562106 11.40497 15.1265 Provincial population in log Crisis: 2007–2011 PS jump 14,340 0.3760112 0.4843998 0 1 Dummy variable: 1 if new entry is random, 0 otherwise ProdY t1(log) 14,340 9.539061 0.7151328 5.580651 11.24809 See Appendix 2 Centrality t1 14,340 0.1536002 0.031584 0.0077109 0.229469 Average relatedness between each product and ALL the other products ExpY t0 (log) 14,340 10.05196 0.1404983 9.602266 10.51017 See Appendix 2 Variety t0 (log) 14,340 1.828332 0.2639739 −0.015982 2.267477 See Appendix 2 Related variety t0 (log) 14,340 1.734126 0.90491 −4.60517 3.160912 See Appendix 2 Unrelated variety t0 (log) 14,340 5.287507 0.9041724 0.3585037 6.450047 See Appendix 2 Trade openness t0 (log) 14,340 0.4682068 0.2732955 0.0012226 2.652769 (Import + export)/gross value added Tertiary education t0 14,340 0.036236 0.033688 0 0.180167 Number of of students enrolled in university over population 15+. Patent_rate t0 14,340 74.15224 68.82898 1 497.0236 Number of patents over million inhabitants. Population density t0 14,340 0.286271 0.40032 0.0311 2.593696 Number of inhabitants over thousand of sq.kms Population t0 (log) 14,340 12.99877 0.755691 10.9626 15.1629 Table 5. Descriptive statistics Variable Observed Mean Standard deviation Minimum Maximum Description Precrisis: 2002–2006 PS jump 13,024 0.3997236 0.4898603 0 1 Dummy variable: 1 if new entry is random, 0 otherwise ProdY t1(log) 13,024 9.505272 0.7283572 3.85543 11.256 See Appendix 2 Centrality t1 13,024 0.1528922 0.0306515 0.0269282 0.231003 Average relatedness between each product and ALL the other products ExpY t0 (log) 13,024 9.866243 0.1597781 9.294399 10.22813 See Appendix 2 Variety t0 (log) 13,024 1.846666 0.2439395 0.6236902 2.238506 See Appendix 2 Related variety t0 (log) 13,024 1.876836 0.7684997 −2.957147 3.2948 See Appendix 2 Unrelated variety t0 (log) 13,024 5.333217 0.8126831 1.619947 6.468346 See Appendix 2 Trade openness t0 (log) 13,024 0.3756517 0.2102692 0.0178385 1.284592 (Import + export)/gross value added Tertiary education t0 12,848 0.038368 0.037405 2.19E-05 0.1934 Number of students enrolled in university over population 15+ Patent_rate t0 13,024 85.50854 73.19639 1 332.048 Number of patents over million inhabitants. Population density t0 13,024 0.2898346 0.3896068 0.0367315 2.611597 Number of inhabitants over thousand of squre kilometers Population t0 (log) 13,024 13.0173 0.7562106 11.40497 15.1265 Provincial population in log Crisis: 2007–2011 PS jump 14,340 0.3760112 0.4843998 0 1 Dummy variable: 1 if new entry is random, 0 otherwise ProdY t1(log) 14,340 9.539061 0.7151328 5.580651 11.24809 See Appendix 2 Centrality t1 14,340 0.1536002 0.031584 0.0077109 0.229469 Average relatedness between each product and ALL the other products ExpY t0 (log) 14,340 10.05196 0.1404983 9.602266 10.51017 See Appendix 2 Variety t0 (log) 14,340 1.828332 0.2639739 −0.015982 2.267477 See Appendix 2 Related variety t0 (log) 14,340 1.734126 0.90491 −4.60517 3.160912 See Appendix 2 Unrelated variety t0 (log) 14,340 5.287507 0.9041724 0.3585037 6.450047 See Appendix 2 Trade openness t0 (log) 14,340 0.4682068 0.2732955 0.0012226 2.652769 (Import + export)/gross value added Tertiary education t0 14,340 0.036236 0.033688 0 0.180167 Number of of students enrolled in university over population 15+. Patent_rate t0 14,340 74.15224 68.82898 1 497.0236 Number of patents over million inhabitants. Population density t0 14,340 0.286271 0.40032 0.0311 2.593696 Number of inhabitants over thousand of sq.kms Population t0 (log) 14,340 12.99877 0.755691 10.9626 15.1629 Variable Observed Mean Standard deviation Minimum Maximum Description Precrisis: 2002–2006 PS jump 13,024 0.3997236 0.4898603 0 1 Dummy variable: 1 if new entry is random, 0 otherwise ProdY t1(log) 13,024 9.505272 0.7283572 3.85543 11.256 See Appendix 2 Centrality t1 13,024 0.1528922 0.0306515 0.0269282 0.231003 Average relatedness between each product and ALL the other products ExpY t0 (log) 13,024 9.866243 0.1597781 9.294399 10.22813 See Appendix 2 Variety t0 (log) 13,024 1.846666 0.2439395 0.6236902 2.238506 See Appendix 2 Related variety t0 (log) 13,024 1.876836 0.7684997 −2.957147 3.2948 See Appendix 2 Unrelated variety t0 (log) 13,024 5.333217 0.8126831 1.619947 6.468346 See Appendix 2 Trade openness t0 (log) 13,024 0.3756517 0.2102692 0.0178385 1.284592 (Import + export)/gross value added Tertiary education t0 12,848 0.038368 0.037405 2.19E-05 0.1934 Number of students enrolled in university over population 15+ Patent_rate t0 13,024 85.50854 73.19639 1 332.048 Number of patents over million inhabitants. Population density t0 13,024 0.2898346 0.3896068 0.0367315 2.611597 Number of inhabitants over thousand of squre kilometers Population t0 (log) 13,024 13.0173 0.7562106 11.40497 15.1265 Provincial population in log Crisis: 2007–2011 PS jump 14,340 0.3760112 0.4843998 0 1 Dummy variable: 1 if new entry is random, 0 otherwise ProdY t1(log) 14,340 9.539061 0.7151328 5.580651 11.24809 See Appendix 2 Centrality t1 14,340 0.1536002 0.031584 0.0077109 0.229469 Average relatedness between each product and ALL the other products ExpY t0 (log) 14,340 10.05196 0.1404983 9.602266 10.51017 See Appendix 2 Variety t0 (log) 14,340 1.828332 0.2639739 −0.015982 2.267477 See Appendix 2 Related variety t0 (log) 14,340 1.734126 0.90491 −4.60517 3.160912 See Appendix 2 Unrelated variety t0 (log) 14,340 5.287507 0.9041724 0.3585037 6.450047 See Appendix 2 Trade openness t0 (log) 14,340 0.4682068 0.2732955 0.0012226 2.652769 (Import + export)/gross value added Tertiary education t0 14,340 0.036236 0.033688 0 0.180167 Number of of students enrolled in university over population 15+. Patent_rate t0 14,340 74.15224 68.82898 1 497.0236 Number of patents over million inhabitants. Population density t0 14,340 0.286271 0.40032 0.0311 2.593696 Number of inhabitants over thousand of sq.kms Population t0 (log) 14,340 12.99877 0.755691 10.9626 15.1629 In Table 6 the marginal effects of the probit regressions are reported separately for the two subperiods, precrisis and crisis. In Models 1 and 5, we employ a parsimonious model of the probability of unrelated new entries where we include our measure of overall provincial export basket sophisticatedness, ExpY (in log), and a measure of production diversification, Variety (in log), controlling for macro-area, industry, and product fixed effects. As expected, provinces characterized by a higher degree of sophisticatedness are found to be positively associated with the probability of experiencing more radical structural changes in the composition of the export basket. We interpret this result as further evidence of the important role of the “complexity” of what economies produce and export. Our result show that complexity has a direct effect on growth performance as in Hausmann et al. (2007) as well as an effect on the ability of an economy to diversify away from the initial comparative advantage. Table 6. Probability of experiencing unrelated new entries in the export basket (probit model) Variables 2002–2006 2007–2011 (1) (2) (5) (10) (1) (2) (5) (10) Margin Margin Margin Margin Margin Margin Margin Margin ExpY t0 (log) 0.111*** 0.0868*** 0.0847*** 0.0830*** 0.0899*** 0.0788*** 0.0773*** 0.0820*** (0.0272) (0.0292) (0.0303) (0.0303) (0.0271) (0.0272) (0.0275) (0.0275) Variety t0 (log) 0.0221 −0.00779 (0.0202) (0.0168) Related variety t0 (log) −0.0155* −0.0183** −0.0150* −0.0216*** −0.0240*** −0.0217*** (0.00857) (0.00865) (0.00868) (0.00723) (0.00726) (0.00729) Unrelated variety t0 (log) 0.0288*** 0.0229** 0.00494 0.0182** 0.0156* −0.000166 (0.0101) (0.0110) (0.0118) (0.00870) (0.00943) (0.0101) Tertiary education t0 0.143 0.0813 0.248** 0.130 (0.124) (0.125) (0.124) (0.127) Trade openness t0(log) −0.0104 0.00350 0.0475*** 0.0466** (0.0265) (0.0268) (0.0184) (0.0184) Patent rate t0 0.000288*** 0.000227*** 2.40e-06 −2.52e-06 (8.39e-05) (8.51e-05) (7.33e-05) (7.34e-05) Population density t0 −0.0182 −0.0384*** −0.0169* −0.0372*** (0.0112) (0.0122) (0.00977) (0.0109) Population t0 (log) 0.0297*** 0.0289*** (0.00711) (0.00686) Northwest 0.0104 −0.00343 −0.0147 0.0110 0.0243** 0.0185 0.0115 0.0293* (0.0138) (0.0143) (0.0169) (0.0180) (0.0120) (0.0126) (0.0150) (0.0156) Northeast −0.00118 −0.0175 −0.0421** −0.0126 0.00483 −0.00139 −0.0131 0.00676 (0.0134) (0.0146) (0.0172) (0.0187) (0.0123) (0.0133) (0.0157) (0.0164) Center 0.000994 −0.00752 −0.0152 0.00175 0.00433 0.00516 −0.00554 0.00860 (0.0120) (0.0127) (0.0137) (0.0143) (0.0109) (0.0115) (0.0123) (0.0128) Observations 13,024 13,024 13,024 13,024 14,340 14,340 14,340 14,340 Industry FE Yes Yes Yes Yes Yes Yes Yes Yes Product FE Yes Yes Yes Yes Yes Yes Yes Yes Pseudo R-squared 0.177 0.177 0.178 0.179 0.204 0.205 0.205 0.206 Variables 2002–2006 2007–2011 (1) (2) (5) (10) (1) (2) (5) (10) Margin Margin Margin Margin Margin Margin Margin Margin ExpY t0 (log) 0.111*** 0.0868*** 0.0847*** 0.0830*** 0.0899*** 0.0788*** 0.0773*** 0.0820*** (0.0272) (0.0292) (0.0303) (0.0303) (0.0271) (0.0272) (0.0275) (0.0275) Variety t0 (log) 0.0221 −0.00779 (0.0202) (0.0168) Related variety t0 (log) −0.0155* −0.0183** −0.0150* −0.0216*** −0.0240*** −0.0217*** (0.00857) (0.00865) (0.00868) (0.00723) (0.00726) (0.00729) Unrelated variety t0 (log) 0.0288*** 0.0229** 0.00494 0.0182** 0.0156* −0.000166 (0.0101) (0.0110) (0.0118) (0.00870) (0.00943) (0.0101) Tertiary education t0 0.143 0.0813 0.248** 0.130 (0.124) (0.125) (0.124) (0.127) Trade openness t0(log) −0.0104 0.00350 0.0475*** 0.0466** (0.0265) (0.0268) (0.0184) (0.0184) Patent rate t0 0.000288*** 0.000227*** 2.40e-06 −2.52e-06 (8.39e-05) (8.51e-05) (7.33e-05) (7.34e-05) Population density t0 −0.0182 −0.0384*** −0.0169* −0.0372*** (0.0112) (0.0122) (0.00977) (0.0109) Population t0 (log) 0.0297*** 0.0289*** (0.00711) (0.00686) Northwest 0.0104 −0.00343 −0.0147 0.0110 0.0243** 0.0185 0.0115 0.0293* (0.0138) (0.0143) (0.0169) (0.0180) (0.0120) (0.0126) (0.0150) (0.0156) Northeast −0.00118 −0.0175 −0.0421** −0.0126 0.00483 −0.00139 −0.0131 0.00676 (0.0134) (0.0146) (0.0172) (0.0187) (0.0123) (0.0133) (0.0157) (0.0164) Center 0.000994 −0.00752 −0.0152 0.00175 0.00433 0.00516 −0.00554 0.00860 (0.0120) (0.0127) (0.0137) (0.0143) (0.0109) (0.0115) (0.0123) (0.0128) Observations 13,024 13,024 13,024 13,024 14,340 14,340 14,340 14,340 Industry FE Yes Yes Yes Yes Yes Yes Yes Yes Product FE Yes Yes Yes Yes Yes Yes Yes Yes Pseudo R-squared 0.177 0.177 0.178 0.179 0.204 0.205 0.205 0.206 Notes: Dependent variable: probability of experiencing a random new entry. Relatedness measured as the maximum proximity between new entry and export basket at time t0. Standard errors in parentheses. *** P < 0.01; ** P < 0.05; *P < 0.1. Table 6. Probability of experiencing unrelated new entries in the export basket (probit model) Variables 2002–2006 2007–2011 (1) (2) (5) (10) (1) (2) (5) (10) Margin Margin Margin Margin Margin Margin Margin Margin ExpY t0 (log) 0.111*** 0.0868*** 0.0847*** 0.0830*** 0.0899*** 0.0788*** 0.0773*** 0.0820*** (0.0272) (0.0292) (0.0303) (0.0303) (0.0271) (0.0272) (0.0275) (0.0275) Variety t0 (log) 0.0221 −0.00779 (0.0202) (0.0168) Related variety t0 (log) −0.0155* −0.0183** −0.0150* −0.0216*** −0.0240*** −0.0217*** (0.00857) (0.00865) (0.00868) (0.00723) (0.00726) (0.00729) Unrelated variety t0 (log) 0.0288*** 0.0229** 0.00494 0.0182** 0.0156* −0.000166 (0.0101) (0.0110) (0.0118) (0.00870) (0.00943) (0.0101) Tertiary education t0 0.143 0.0813 0.248** 0.130 (0.124) (0.125) (0.124) (0.127) Trade openness t0(log) −0.0104 0.00350 0.0475*** 0.0466** (0.0265) (0.0268) (0.0184) (0.0184) Patent rate t0 0.000288*** 0.000227*** 2.40e-06 −2.52e-06 (8.39e-05) (8.51e-05) (7.33e-05) (7.34e-05) Population density t0 −0.0182 −0.0384*** −0.0169* −0.0372*** (0.0112) (0.0122) (0.00977) (0.0109) Population t0 (log) 0.0297*** 0.0289*** (0.00711) (0.00686) Northwest 0.0104 −0.00343 −0.0147 0.0110 0.0243** 0.0185 0.0115 0.0293* (0.0138) (0.0143) (0.0169) (0.0180) (0.0120) (0.0126) (0.0150) (0.0156) Northeast −0.00118 −0.0175 −0.0421** −0.0126 0.00483 −0.00139 −0.0131 0.00676 (0.0134) (0.0146) (0.0172) (0.0187) (0.0123) (0.0133) (0.0157) (0.0164) Center 0.000994 −0.00752 −0.0152 0.00175 0.00433 0.00516 −0.00554 0.00860 (0.0120) (0.0127) (0.0137) (0.0143) (0.0109) (0.0115) (0.0123) (0.0128) Observations 13,024 13,024 13,024 13,024 14,340 14,340 14,340 14,340 Industry FE Yes Yes Yes Yes Yes Yes Yes Yes Product FE Yes Yes Yes Yes Yes Yes Yes Yes Pseudo R-squared 0.177 0.177 0.178 0.179 0.204 0.205 0.205 0.206 Variables 2002–2006 2007–2011 (1) (2) (5) (10) (1) (2) (5) (10) Margin Margin Margin Margin Margin Margin Margin Margin ExpY t0 (log) 0.111*** 0.0868*** 0.0847*** 0.0830*** 0.0899*** 0.0788*** 0.0773*** 0.0820*** (0.0272) (0.0292) (0.0303) (0.0303) (0.0271) (0.0272) (0.0275) (0.0275) Variety t0 (log) 0.0221 −0.00779 (0.0202) (0.0168) Related variety t0 (log) −0.0155* −0.0183** −0.0150* −0.0216*** −0.0240*** −0.0217*** (0.00857) (0.00865) (0.00868) (0.00723) (0.00726) (0.00729) Unrelated variety t0 (log) 0.0288*** 0.0229** 0.00494 0.0182** 0.0156* −0.000166 (0.0101) (0.0110) (0.0118) (0.00870) (0.00943) (0.0101) Tertiary education t0 0.143 0.0813 0.248** 0.130 (0.124) (0.125) (0.124) (0.127) Trade openness t0(log) −0.0104 0.00350 0.0475*** 0.0466** (0.0265) (0.0268) (0.0184) (0.0184) Patent rate t0 0.000288*** 0.000227*** 2.40e-06 −2.52e-06 (8.39e-05) (8.51e-05) (7.33e-05) (7.34e-05) Population density t0 −0.0182 −0.0384*** −0.0169* −0.0372*** (0.0112) (0.0122) (0.00977) (0.0109) Population t0 (log) 0.0297*** 0.0289*** (0.00711) (0.00686) Northwest 0.0104 −0.00343 −0.0147 0.0110 0.0243** 0.0185 0.0115 0.0293* (0.0138) (0.0143) (0.0169) (0.0180) (0.0120) (0.0126) (0.0150) (0.0156) Northeast −0.00118 −0.0175 −0.0421** −0.0126 0.00483 −0.00139 −0.0131 0.00676 (0.0134) (0.0146) (0.0172) (0.0187) (0.0123) (0.0133) (0.0157) (0.0164) Center 0.000994 −0.00752 −0.0152 0.00175 0.00433 0.00516 −0.00554 0.00860 (0.0120) (0.0127) (0.0137) (0.0143) (0.0109) (0.0115) (0.0123) (0.0128) Observations 13,024 13,024 13,024 13,024 14,340 14,340 14,340 14,340 Industry FE Yes Yes Yes Yes Yes Yes Yes Yes Product FE Yes Yes Yes Yes Yes Yes Yes Yes Pseudo R-squared 0.177 0.177 0.178 0.179 0.204 0.205 0.205 0.206 Notes: Dependent variable: probability of experiencing a random new entry. Relatedness measured as the maximum proximity between new entry and export basket at time t0. Standard errors in parentheses. *** P < 0.01; ** P < 0.05; *P < 0.1. It is interesting to note that the magnitude of the effect is rather stable in the considered time span and slightly smaller during the crisis period. Furthermore, we find a weak positive effect of the degree of export diversification (LnVariety) on the probability of unrelated new entries only in the precrisis period. In Models 2 and 6, we test for heterogeneous effects of related versus unrelated varieties. We find that the number of unrelated varieties is positively associated with the probability of unrelated new entries. On the contrary, a higher provincial diversification within the same macro-sector (i.e., product diversification within a four-digit sector) is negatively associated with unrelated entries in the provincial export basket. This result provides evidence of the lock-in effect that “dense” sectoral specialization may represent. For instance, the presence of consolidated industrial districts may hamper the diversification of the provincial economy over the PS. In the ascendant phase of development of a district, this path-dependence may reinforce growth, but in the maturity or decline phases, this pattern may represent a less desirable feature of the local economy. In the specifications reported in Models 3 and 7, we introduce additional covariates, tertiary education, patent rate (proxies for provincial human capital and innovation potential, respectively), provincial trade openness and population density. It is interesting to observe that openness to trade and (although less strongly) a larger endowment of human capital are positively associated with the probability of unrelated new entries only in the period of crisis. Intuitively, a larger market access amplifies the value of producing and the ability to adapt to new products. The effect may also be driven by the presence of high-productivity firms in more open economies which are in turn more able to react to market difficulties by diversifying production.37Neffke et al. (2014) show that firms with a higher degree of internationalization represent crucial agents of structural change, since nonlocal firms and entrepreneurs tend to diversify in sectors that are less related to preexisting regional production bundles. The variable patent rate is positive but statistically significant only in the precrisis period. Finally we find that a higher population density is weakly associated to a lower probability of experiencing radical changes. Boschma et al. (2015) find an ambiguous effect of population density on the probability that a new technology is acquired by a US metropolitan area. Our result seems to suggest that density promotes the development of related varieties rather than unrelated ones. As a robustness check, we replicate the estimates using alternative and more restrictive definitions of unrelated new entries in the provincial export basket. We consider a new entry as unrelated (i.e., our dependent variable equals 1) if the new good enters the provincial export basket in a statistically unrelated way according to all three measures of relatedness (maximum, average, and weighted average; equations (4)–(6)). The results reported in Table 7 confirm the evidence described above. All variables show lower values of the marginal effects, as expected, with a narrow definition of the dependent variable, but the main covariates, EXPY and Unrelated Variety, retain a positive and statistically significant effect on the probability of more radical changes in both time spans considered. Table 7. Probability of experiencing unrelated new entries in the export basket (robustness check: Alternative definition of relatedness) Variables 2002–2006 2007–2011 (1) (2) (5) (10) (1) (2) (5) (10) Margin Margin Margin Margin Margin Margin Margin Margin ExpY t0(log) 0.0354** 0.0255 0.0381* 0.0373* 0.0436** 0.0420** 0.0404** 0.0421** (0.0180) (0.0194) (0.0213) (0.0213) (0.0179) (0.0180) (0.0182) (0.0182) Variety t0 (log) 0.0251* −0.00915 (0.0135) (0.0109) Related variety t0 (log) −0.00475 −0.00720 −0.00904 −0.00677 −0.00781 −0.00708 (0.00575) (0.00708) (0.00742) (0.00475) (0.00478) (0.00479) Unrelated variety t0(log) 0.0104 0.0163* 0.0175** 0.00436 0.00402 −0.00230 (0.00676) (0.00869) (0.00883) (0.00572) (0.00617) (0.00661) Tertiary education t0 0.00490 −0.00311 0.164* 0.115 (0.0866) (0.0870) (0.0838) (0.0862) Trade openness t0(log) −0.0202 −0.0238 0.0191 0.0185 (0.0171) (0.0177) (0.0122) (0.0122) Patent rate t0 0.00607* 0.00897* −4.91e-05 −5.29e-05 (0.00365) (0.00502) (5.24e-05) (5.26e-05) Population density t0 −0.0299*** −0.0271*** −0.00887 −0.0172** (0.00750) (0.00817) (0.00656) (0.00732) Population t0 (log) −0.00548 0.0120*** (0.00653) (0.00460) Northwest 0.00490 0.00577 −0.00440 −0.0106 0.0187** 0.0160* 0.0181* 0.0258** (0.00906) (0.00938) (0.0112) (0.0134) (0.00803) (0.00842) (0.00999) (0.0104) Northeast 0.00449 0.00397 −0.00925 −0.0155 0.0133 0.0106 0.0108 0.0195* (0.00879) (0.00959) (0.0112) (0.0134) (0.00820) (0.00892) (0.0105) (0.0110) Center −0.00435 −0.00516 −0.0130 −0.0172* −0.00466 −0.00443 −0.00742 −0.00119 (0.00795) (0.00838) (0.00918) (0.0104) (0.00743) (0.00787) (0.00836) (0.00870) Observations 13,024 13,024 12,693 12,693 14,340 14,340 14,340 14,340 Industry FE Yes Yes Yes Yes Yes Yes Yes Yes Product FE Yes Yes Yes Yes Yes Yes Yes Yes Pseudo R-squared 0.415 0.415 0.418 0.418 0.449 0.450 0.450 0.450 Variables 2002–2006 2007–2011 (1) (2) (5) (10) (1) (2) (5) (10) Margin Margin Margin Margin Margin Margin Margin Margin ExpY t0(log) 0.0354** 0.0255 0.0381* 0.0373* 0.0436** 0.0420** 0.0404** 0.0421** (0.0180) (0.0194) (0.0213) (0.0213) (0.0179) (0.0180) (0.0182) (0.0182) Variety t0 (log) 0.0251* −0.00915 (0.0135) (0.0109) Related variety t0 (log) −0.00475 −0.00720 −0.00904 −0.00677 −0.00781 −0.00708 (0.00575) (0.00708) (0.00742) (0.00475) (0.00478) (0.00479) Unrelated variety t0(log) 0.0104 0.0163* 0.0175** 0.00436 0.00402 −0.00230 (0.00676) (0.00869) (0.00883) (0.00572) (0.00617) (0.00661) Tertiary education t0 0.00490 −0.00311 0.164* 0.115 (0.0866) (0.0870) (0.0838) (0.0862) Trade openness t0(log) −0.0202 −0.0238 0.0191 0.0185 (0.0171) (0.0177) (0.0122) (0.0122) Patent rate t0 0.00607* 0.00897* −4.91e-05 −5.29e-05 (0.00365) (0.00502) (5.24e-05) (5.26e-05) Population density t0 −0.0299*** −0.0271*** −0.00887 −0.0172** (0.00750) (0.00817) (0.00656) (0.00732) Population t0 (log) −0.00548 0.0120*** (0.00653) (0.00460) Northwest 0.00490 0.00577 −0.00440 −0.0106 0.0187** 0.0160* 0.0181* 0.0258** (0.00906) (0.00938) (0.0112) (0.0134) (0.00803) (0.00842) (0.00999) (0.0104) Northeast 0.00449 0.00397 −0.00925 −0.0155 0.0133 0.0106 0.0108 0.0195* (0.00879) (0.00959) (0.0112) (0.0134) (0.00820) (0.00892) (0.0105) (0.0110) Center −0.00435 −0.00516 −0.0130 −0.0172* −0.00466 −0.00443 −0.00742 −0.00119 (0.00795) (0.00838) (0.00918) (0.0104) (0.00743) (0.00787) (0.00836) (0.00870) Observations 13,024 13,024 12,693 12,693 14,340 14,340 14,340 14,340 Industry FE Yes Yes Yes Yes Yes Yes Yes Yes Product FE Yes Yes Yes Yes Yes Yes Yes Yes Pseudo R-squared 0.415 0.415 0.418 0.418 0.449 0.450 0.450 0.450 Notes: Dependent variable: probability of experiencing a random new entry. Randomjump =1 if the new entry is simultaneously random according to our three measures of relatedness t0. Standard errors in parentheses. *** P < 0.01; **P < 0.05; *P < 0.1. Table 7. Probability of experiencing unrelated new entries in the export basket (robustness check: Alternative definition of relatedness) Variables 2002–2006 2007–2011 (1) (2) (5) (10) (1) (2) (5) (10) Margin Margin Margin Margin Margin Margin Margin Margin ExpY t0(log) 0.0354** 0.0255 0.0381* 0.0373* 0.0436** 0.0420** 0.0404** 0.0421** (0.0180) (0.0194) (0.0213) (0.0213) (0.0179) (0.0180) (0.0182) (0.0182) Variety t0 (log) 0.0251* −0.00915 (0.0135) (0.0109) Related variety t0 (log) −0.00475 −0.00720 −0.00904 −0.00677 −0.00781 −0.00708 (0.00575) (0.00708) (0.00742) (0.00475) (0.00478) (0.00479) Unrelated variety t0(log) 0.0104 0.0163* 0.0175** 0.00436 0.00402 −0.00230 (0.00676) (0.00869) (0.00883) (0.00572) (0.00617) (0.00661) Tertiary education t0 0.00490 −0.00311 0.164* 0.115 (0.0866) (0.0870) (0.0838) (0.0862) Trade openness t0(log) −0.0202 −0.0238 0.0191 0.0185 (0.0171) (0.0177) (0.0122) (0.0122) Patent rate t0 0.00607* 0.00897* −4.91e-05 −5.29e-05 (0.00365) (0.00502) (5.24e-05) (5.26e-05) Population density t0 −0.0299*** −0.0271*** −0.00887 −0.0172** (0.00750) (0.00817) (0.00656) (0.00732) Population t0 (log) −0.00548 0.0120*** (0.00653) (0.00460) Northwest 0.00490 0.00577 −0.00440 −0.0106 0.0187** 0.0160* 0.0181* 0.0258** (0.00906) (0.00938) (0.0112) (0.0134) (0.00803) (0.00842) (0.00999) (0.0104) Northeast 0.00449 0.00397 −0.00925 −0.0155 0.0133 0.0106 0.0108 0.0195* (0.00879) (0.00959) (0.0112) (0.0134) (0.00820) (0.00892) (0.0105) (0.0110) Center −0.00435 −0.00516 −0.0130 −0.0172* −0.00466 −0.00443 −0.00742 −0.00119 (0.00795) (0.00838) (0.00918) (0.0104) (0.00743) (0.00787) (0.00836) (0.00870) Observations 13,024 13,024 12,693 12,693 14,340 14,340 14,340 14,340 Industry FE Yes Yes Yes Yes Yes Yes Yes Yes Product FE Yes Yes Yes Yes Yes Yes Yes Yes Pseudo R-squared 0.415 0.415 0.418 0.418 0.449 0.450 0.450 0.450 Variables 2002–2006 2007–2011 (1) (2) (5) (10) (1) (2) (5) (10) Margin Margin Margin Margin Margin Margin Margin Margin ExpY t0(log) 0.0354** 0.0255 0.0381* 0.0373* 0.0436** 0.0420** 0.0404** 0.0421** (0.0180) (0.0194) (0.0213) (0.0213) (0.0179) (0.0180) (0.0182) (0.0182) Variety t0 (log) 0.0251* −0.00915 (0.0135) (0.0109) Related variety t0 (log) −0.00475 −0.00720 −0.00904 −0.00677 −0.00781 −0.00708 (0.00575) (0.00708) (0.00742) (0.00475) (0.00478) (0.00479) Unrelated variety t0(log) 0.0104 0.0163* 0.0175** 0.00436 0.00402 −0.00230 (0.00676) (0.00869) (0.00883) (0.00572) (0.00617) (0.00661) Tertiary education t0 0.00490 −0.00311 0.164* 0.115 (0.0866) (0.0870) (0.0838) (0.0862) Trade openness t0(log) −0.0202 −0.0238 0.0191 0.0185 (0.0171) (0.0177) (0.0122) (0.0122) Patent rate t0 0.00607* 0.00897* −4.91e-05 −5.29e-05 (0.00365) (0.00502) (5.24e-05) (5.26e-05) Population density t0 −0.0299*** −0.0271*** −0.00887 −0.0172** (0.00750) (0.00817) (0.00656) (0.00732) Population t0 (log) −0.00548 0.0120*** (0.00653) (0.00460) Northwest 0.00490 0.00577 −0.00440 −0.0106 0.0187** 0.0160* 0.0181* 0.0258** (0.00906) (0.00938) (0.0112) (0.0134) (0.00803) (0.00842) (0.00999) (0.0104) Northeast 0.00449 0.00397 −0.00925 −0.0155 0.0133 0.0106 0.0108 0.0195* (0.00879) (0.00959) (0.0112) (0.0134) (0.00820) (0.00892) (0.0105) (0.0110) Center −0.00435 −0.00516 −0.0130 −0.0172* −0.00466 −0.00443 −0.00742 −0.00119 (0.00795) (0.00838) (0.00918) (0.0104) (0.00743) (0.00787) (0.00836) (0.00870) Observations 13,024 13,024 12,693 12,693 14,340 14,340 14,340 14,340 Industry FE Yes Yes Yes Yes Yes Yes Yes Yes Product FE Yes Yes Yes Yes Yes Yes Yes Yes Pseudo R-squared 0.415 0.415 0.418 0.418 0.449 0.450 0.450 0.450 Notes: Dependent variable: probability of experiencing a random new entry. Randomjump =1 if the new entry is simultaneously random according to our three measures of relatedness t0. Standard errors in parentheses. *** P < 0.01; **P < 0.05; *P < 0.1. 5. Concluding remarks In this work we have analyzed the evolution of the export basket of Italian provinces between 2002 and 2011 to test its conformity with the prediction of path-dependence which is a cornerstone of the PS framework. According to the approach of the network of relatedness between goods developed by the seminal contributions of Hausmann and Klinger (2007) and Hidalgo et al. (2007), the goods that have higher probabilities of entering the export portfolio are those sharing common local capabilities with those previously produced. Hence, local capabilities determine the direction of structural change and, at the same time, constrain the evolution of the comparative advantage of nations and regions to those products that are strongly related to the ones already produced. These predictions have important implications for industrial and innovation policies, since they suggest the implementation of selective policies targeted to sectors related to the current comparative advantage. Our results show that both in precrisis and crisis periods, the goods that Italian provinces started exporting with RCA tend to be highly related to the set of goods exported 4 years before, thus confirming a general pattern of path-dependence. This article contributes to the existing literature by providing a methodological approach for testing the nonrandomness of the evolution of structural change along the PS. To our knowledge, this is the first study that moves beyond a simple description of the dynamics of changes in the bundle of goods produced by economies over time. We focus on subnational areas, since capabilities—i.e., technologies, capital, skills, and institutions—have a strong local dimension and are unequally distributed over space, in particular in countries such as Italy with highly differentiated and heterogeneous economic areas. Although we confirm the general tendency of path-dependence, we find that on average approximately 30% of new goods that enter the export basket of Italian provinces are largely unrelated with the preexisting comparative advantage. These apples that fall far from the tree are the most interesting from a policy perspective, since they represent cases of more radical structural change. The significant deviations from the pattern of path-dependence observed in Italian provinces and its high degree of geographical heterogeneity suggest that caution should be exercised in using the PS as a map for identifying the “latent comparative advantage” of countries and regions. Structural change may take a different, often unpredictable, path. Interestingly, we find that the provinces that are more likely to “defeat” the initial static comparative advantage are those characterized by a relatively higher production sophisticatedness, a higher initial level of product diversification in unrelated sectors, with relatively more open economies, and (although the evidence is less robust) better endowment of human capital. Those provinces with higher average production sophisticatedness and more complex and diversified sets of local capabilities are less constrained by path-dependence and have a higher probability of experiencing long jumps over the PS. In addition, our finding of a relatively lower degree of path-dependence in the evolution of structural change during the crisis seems to go in the same direction as the Schumpeterian process of creative destruction during large and pervasive shocks, at least in the short term (Schumpeter, 1942). The findings summarized above on the determinants of radical changes in the export basket are obtained from a cross-section regression approach. Although this approach allows us to shed some lights on the observed patterns, admittedly we are not able to fully control for the unobserved heterogeneity at the province level. An additional limit of our analysis is the focus on short-term changes in the export baskets due to the nature of the data and our interest in the role played by the crisis. In this respect it is important to underline that our methodology can be easily used to analyze the pattern of structural change in other countries or regions and over different time horizons. It would be interesting to analyze the evolution of the production bundle once the Italian (and global) economy regains momentum to test whether our results are confirmed. Another important question is the role played by spatial spillovers in the process of provincial structural change. These interesting analyses are left to future research. Supplementary Material Supplementary material is available at Industrial and Corporate Change online. Footnotes 1 An important difference between this wave of “structural economics” and the early one is rooted in the role of the State and normative implications in general. The first wave of structural economics was based on a firm belief that structural differences were essentially the result of market failures which required pervasive and often highly distortionary Government interventions. This “dirigiste dogma” led to the widespread adoption of quantitative restrictions to international trade flows and the heavy use of currency manipulations which caused several crises that paved the way to another extreme, the “market dogma.” The new wave of structural economics can be seen as a “market-State” blend that is perfectly represented by the words of one of its main exponents, Justin Yifu Lin “the market should be the basic mechanism for resource allocation, but that government must play an active role in coordinating investments for industrial upgrading and diversification and in compensating for externalities generated by first movers in the dynamic growth process” (Lin, 2012). 2 Boschma et al. (2015) using a similar approach introduce the concept of technological space and show that the acquisition of new technological capabilities in 366 US cities is more likely if related technologies have already been acquired. This result suggests that a strong path-dependence not only affects the introduction of new products but also characterizes the development of local productive capabilities. 3 The proximity between each couple of goods is given by the minimum of the pairwise conditional probability of being co-exported. In other words, products are connected or related if they tend to be exported by the same economies. 4 Hidalgo et al. (2007) argue that where a country's export basket is “located” in the product space matters for economic development. As new industries develop from existing ones, countries that produce goods that are better connected are more likely to develop more sophisticated goods. On the contrary, countries specialized in goods that are located in the periphery of the product space are more likely to be trapped in development “dead corners” and face higher difficulties in kick-starting new more complex and sophisticated industries. Several contributions, starting from the work of Hausmann et al. (2007), have shown that “what you produce matters!”, the complexity and sophisticatedness of what an economy produces enhances its future growth (Berg et al., 2012, Felipe et al., 2012, and Ferrarini and Scaramozzino 2016 on a sample of world economies; subnational evidence is provided for a panel of Chinese cities by Poncet and Starosta de Waldemar, 2013, and on Spanish, Chinese, and Russian regions by—respectively—Minondo, 2010, Jarreau and Poncet, 2012, and Kadochnikov and Fedyunina, 2013). 5 While we apply the methodology developed in our article to the network of relatedness à la Hidalgo et al. (2007)—given the importance of this study and our research questions—alternative metrics for computing the matrix of products relatedness can be used. This task is left for future research. 6 Cfr. next section for details. 7 Industrial policy is back in the agenda of many countries around the world. The framework developed by Hausmann et al., 2007 has received a great deal of attention from several countries which are seeking the support of experts—for instance, the Center for International Development, CID, based at Harvard University and led by Ricardo Hausmann—to design their industrial strategies. The list of countries inspired by this approach is expanding and includes Albania, Colombia, and Mexico, among others. 8 Nomenclature des unités territoriales statistiques, in english Classification of Territorial Units for Statistics. 9 We find that the share of new goods that is statistically unrelated to the initial export basket ranges from a minimum of 17% in the province of Isernia in the crisis period to a maximum of 75% in the province of Siracusa in the precrisis period. Such heterogeneity is also confirmed between sectors: only 27% of new entries belonging to the textile sector are found to be unrelated against 83% for the mineral sector. 10 Cfr. Altenburg (2011) 11 According to H. -J. Chang, such a successful production specialization decision supported by active industrial policy in Korea is the proof that defying a country’s comparative advantage (in that period the economy was mainly specialized in the production of labor-intensive goods) allows “learnable-by-doing” competences that then made Korea one of the major producers of electronic components to be developed. On the other hand, J. Y. Lin asserts that the kind of electronic components produced at that time in Korea did not require very high skills since 64 Kbit DRAM was no longer at the technology frontier (Lin and Chang, 2009). Transposing these two views to the context of the network of relatedness between goods implies either that an economy is able—under certain conditions—to specialize in products that are not very proximal to the preexisting export basket or that the product space is dynamic and that links connecting nodes change over time. 12 Using US patent citation data, Castaldi et al. (2015) find that technological breakthroughs—i.e., radical innovations—are more likely to happen in US states endowed with a large set of unrelated varieties. 13 The degree of sophisticatedness of a product is generally proxied by the ProdY index originally presented in Hausmann et al. (2007). The ProdY index represents the productivity level associated with the production of a certain product (see Appendix 2 for details). 14 The authors use the “density” measure developed by Hidalgo et al. (2007) in their parametric analysis of the probability that (new) goods enter the export basket of a country, computed as the average proximity of a new potential product to a country’s current productive capability. 15 Using Chinese firm-level data, Poncet and Starosta de Waldemar (2013) show that “domestic capabilities” matter not only for explaining what firms produce but also for the growth enhancement effects of new products and new technologies. 16 Previous studies (Boschma et al., 2013; Boschma and Capone, 2016; Donoso and Martin, 2016; Lo Turco and Maggioni, 2016) have used measures of “density” as a predictor of the entry of a given product that was not previously exported. 17 Our approach has some similarity with the one employed by Duranton and Overman (2005) to measure the nonrandomness of the geographical concentration of industrial plants in the UK. 18 Between the two subperiods, four new provinces have been formed (in 2005); hence the total number of provinces used in the analysis is 103 and 107, respectively. We do not have data in both periods for three provinces which are excluded from the analysis (Barletta-Andria-Trani, Fermo and Monza-Brianza). 19 For robustness, different base and term years have been used and are available upon request from the authors. Note that the split of the two subperiods reported in the article is also preferred because it allows us to use the same nomenclature for international and national trade statistics (Harmonized System revisions H2 and H3 have been issued in 2002 and 2007, respectively) between t0 and t1, hence avoid the use of correspondence tables that may result in a less precise conversion of the data. 20 Since this choice of RCA thresholds is arbitrary, for robustness we identify a new entry using three additional alternative thresholds. We use one definition of a new entry that is less restrictive that the one presented in the article ( RCAt0<1 and RCAt1≥1) and two definitions that are more restrictive, respectively, RCAt0 lower than 0.1 and lower than 0.2 and RCAt1≥1. These range from 8568 “new entries” in the precrisis period for the most restrictive definition to 18,656 “new entries” in the crisis period for the less restrictive definition. The results are qualitatively similar and are available upon request from the authors. 21 We obtain a 5222-by-5222 and a 5050-by-5050 matrix for 2006 and 2011, respectively. Note that we use a more detailed network of relatedness than the original version (Hidalgo et al., 2007, 774 goods in the SITC rev.4 Nomenclature) which, in our opinion, allows us to obtain a more precise representation of the evolution of export baskets. 22 A vector of distances for each of the four definitions of new entries and for each alternative measure of relatedness is created to ensure the robustness of our results to the definition of these two key elements. In the article, we only present, for the sake of brevity, the results for one definition of a new entry ( RCAt0<0.5 and RCAt1≥1). 23 In every simulation, for each province we randomly draw a number of new entries from the products not in the basket at time t0 which is identical to the number of effective ones. In other words, our counterfactual exercise takes explicit account of the province-specific distribution of new entries. 24 We would like to thank an anonymous referee for suggesting this option. 25 In other words, we draw random samples from all goods i∈[Wt1−Bk,t0], where W is the set of all goods exported in the world at time t1 with t1=2006, 2011. 26 The increase is not due to the slightly higher number of provinces in the crisis period. In fact, the new Italian provinces (Carbonia-Iglesias, Medio-Campidano, Ogliastra and Olbia-Tempio) are all located in Sardinia and present a low number of new products in the export basket. 27 For both subperiods, we report figures representing the Kernel density distributions for the alternative identification strategies of new entries in Appendix 1. The results confirm the findings reported in this paragraph. 28 Note that for proximity values that are in the upper tail of the distribution (above the threshold values), we cannot statistically reject the null hypothesis, since few observations both in actual and simulated data fall in this area. It is important to underline that the percentage of actual proximities falling in the upper tails of all the kernel distributions in Figures 3 and 4 is higher than the simulated ones; we interpret this as an indication of nonrandom relatedness (although statistically not significant). 29 In other words, a significant number of new products enter the export baskets at t1 in areas of the product space that are ideally “far away” from those where the export basket at t0 lies. 30 See also Coniglio et al. (2016) for a detailed analysis of provincial growth and sophisticatedness in Italy. 31 The effect of radical changes on economic performance is an important related question. In our data we find a weak correlation between the share of unrelated products and subsequent provincial growth. A methodologically robust analysis on this fundamental question would require a longer time span and a panel approach. We refer the reader to the recent work of Content and Frenken (2016) on the nexus between related/unrelated variety and economic performance at different geographical scales. These authors report very mixed results in line with the weak correlation found in our data. 32 The test will be on data with distances measured as the maximum among the proximities between new entrant goods and those already part of the export basket at time t0. Among the three methods reported above, maximum proximity is the one that we believe better captures the concept of product relatedness in the product space framework (i.e., the fact that related products share common productive capabilities). 33 We include industrial sectors’ fixed effects using the 21 sections of HS nomenclature to consider heterogeneity across macro industries. All province-specific variables refer to the year t0 whereas all product-specific variables refer to year t1. 34 The index of export basket sophistication was first introduced by Hausmann et al. (2007), and it is computed as the weighted sum of ProdYs of the products exported by a province with weights represented by the export shares. For the definition of ProdY see Appendix 2. 35 To simplify the interpretation of results, we include all the “diversification” indexes in logarithms. Such measures are computed as reported in Appendix 2. 36 South is excluded as in Boschma and Iammarino (2009). 37 Recent contributions in the international trade literature have emphasized the structural differences between firms exporting and investing abroad and purely domestic ones in terms of productivity, wages, size, markups, and other crucial firm-level characteristics (Melitz, 2003; Bernard et al., 2007; Melitz and Ottaviano, 2008). References Altenburg T. ( 2011 ), ‘ Industrial Policy in Developing Countries ’, DIE-German Development Institute Discussion Paper no. 4/2011 . Google Scholar CrossRef Search ADS Balassa B. ( 1965 ), ‘ Trade liberalisation and “revealed” comparative advantage ,’ The Manchester School , 33 ( 2 ), 99 – 123 . Google Scholar CrossRef Search ADS Berg A. , Ostry J. D. , Zettelmeyer J. ( 2012 ). ‘ What makes growth sustained? ,’ Journal of Development Economics , 98 ( 2 ), 149 – 166 . Google Scholar CrossRef Search ADS Bernard A. B. , Jensen J.B. , Redding S.J. , Schott P.K. ( 2007 ), ‘ Firms in international trade ,’ Journal of Economic Perspectives , 21 ( 3 ), 105 – 130 . Google Scholar CrossRef Search ADS Boschma R. , Capone G. ( 2016 ), ‘ Relatedness and diversification in the European Union (EU-27) and European Neighbourhood Policy countries ,’ Environment and Planning C: Government and Policy , 34 ( 4 ), 617 – 637 . Google Scholar CrossRef Search ADS Boschma R. , Iammarino S. ( 2009 ), ‘ Related variety, trade linkages, and regional growth in Italy ,’ Economic Geography , 85 , 289 – 311 . Google Scholar CrossRef Search ADS Boschma R. , Balland P. , Kogler D. F. ( 2015 ), ‘ Relatedness and technological change in cities: the rise and fall of technological knowledge in US metropolitan areas from 1981 to 2010 ,’ Industrial and Corporate Change , 24 ( 1 ), 223 – 250 . Google Scholar CrossRef Search ADS Boschma R. , Minondo A. , Navarro M. ( 2013 ), ‘ The emergence of new industries at the regional level in Spain: a proximity approach based on product relatedness ,’ Economic Geography, Clark University , 89 ( 1 ), 29 – 51 . Google Scholar CrossRef Search ADS Bottazzi G. , Pirino D. ( 2010 ), ‘Measuring industry relatedness and corporate coherence,’ LEM Paper Series 2010/10, Sant'Anna School of Advanced Studies: Pisa, Italy. Castaldi C. , Frenken K. , Los B. ( 2015 ), ‘ Related variety, unrelated variety and technological breakthroughs: an analysis of US state-level patenting ,’ Regional Studies , 49 ( 5 ), 767 – 781 . Google Scholar CrossRef Search ADS Cirera X. , Marin A. , Markwald R. ( 2012 ), ‘ Firm behaviour and the introduction of new exports: evidence from Brazil ,’ IDS - Institute of Development Studies Working Paper n. 390 , pp. 1 – 105 . Coniglio N. D. , Lagravinese R. , Vurchio D. ( 2016 ), ‘ Production ‘sophisticatedness’ and growth: evidence from Italian Provinces before and during the crisis, 1997-2013 ,’ Cambridge Journal of Regions, Economy and Society , 9 ( 2 ), 423 – 442 . Google Scholar CrossRef Search ADS Content J. , Frenken K. ( 2016 ), ‘ Related and economic development: a literature review ,’ European Planning Studies , 24 ( 12 ), 2097 – 2112 . Google Scholar CrossRef Search ADS Donoso V. , Martin V. ( 2016 ), ‘ Product relatedness and economic diversification in the USA: an analysis at the state level ,’ The Annals of Regional Science , 56 ( 2 ), 449 – 471 . Google Scholar CrossRef Search ADS Dosi G. ( 1982 ), ‘ Technological paradigms and technological trajectories ,’ Research Policy , 11 , 147 – 162 . Google Scholar CrossRef Search ADS Duranton G. , Overman H. G. ( 2005 ), ‘ Testing for localization using micro-geographic data ,’ Review of Economic Studies , 72 ( 4 ), 1077 – 1106 . Google Scholar CrossRef Search ADS Felipe J. , Kumar U. , Abdon A. ( 2013a ), ‘ Exports, capabilities, and industrial policy in India ,’ Journal of Comparative Economics , 41 ( 3 ), 939 – 956 . ISSN 0147-5967. Google Scholar CrossRef Search ADS Felipe J. , Kumar U. , Abdon A. , Bacate M. ( 2012 ), ‘ Product complexity and economic development ,’ Structural Change and Economic Dynamics , 23 ( 1 ), 36 – 68 . Google Scholar CrossRef Search ADS Felipe J. , Kumar U. , Usui N. , Abdon A. ( 2013b ), ‘ Why has China succeeded? And why it will continue to do so ,’ Cambridge Journal of Economics , 37 ( 4 ), 791 – 818 . Google Scholar CrossRef Search ADS Ferrarini B. , Scaramozzino P. ( 2016 ), ‘ Production complexity, adaptability and economic growth ,’ Structural Change and Economic Dynamics , 37 , 52 – 61 . Google Scholar CrossRef Search ADS Frenken K. ( 2007 ), ‘Entropy statistics and information theory,’ in Hanusch H. , Pyka A. (eds), The Elgar Companion to Neo-Schumpeterian Economics . Edward Elgar : Cheltenham, UK; Northampton MA , pp. 544 – 555 . Hausmann R. , Hwang J. , Rodrik D. ( 2007 ), ‘ What you export matters ,’ Journal of Economic Growth , 12 ( 1 ), 1 – 25 . Google Scholar CrossRef Search ADS Hausmann R. , Klinger B. ( 2010 ), ‘Structural transformation in ecuador,’ Policy Brief. No. 1112, Inter-American Development Bank: Washington, DC. Hausmann R. , Klinger B. ( 2007 ), ‘The structure of the product space and the evolution of comparative advantage,’ Harvard University Center for International Development Working Paper #146. Hidalgo C. ( 2012 ), ‘ Discovering East Africa’s industrial opportunities ,’ Papers 1203.0163, arXiv.org. <https://ideas.repec.org/p/arx/papers/1203.0163.html>. Hidalgo C. , Klinger B. , Barabasi A. , Hausmann R. ( 2007 ), ‘ The product space conditions the development of nations ,’ Science Magazine , 317 ( 5837 ), 482 – 487 . Jarreau J. , Poncet S. ( 2012 ) ‘ Export sophistication and economic growth: evidence from China ,’ Journal of Development Economics , 97 ( 2 ), 281 – 292 . Google Scholar CrossRef Search ADS Kadochnikov S. , Fedyunina A. ( 2013 ), ‘Export diversification in the product space and regional growth: Evidence from Russia,’ Papers in Evolutionary Economic Geography (PEEG) 1327, Utrecht University, Section of Economic Geography. Kuznets S. ( 1966 ), Modern Economic Growth . Yale University Press : New Haven, CT . Lin J. Y. ( 2012 ) ‘ From flying geese to leading dragons: new opportunities and strategies for structural transformation in developing countries ,’ Global Policy , 3 , 397 – 409 . Google Scholar CrossRef Search ADS Lin J. , Chang H. J. ( 2009 ), ‘ Should industrial policy in developing countries conform to comparative advantage or defy it? A debate between Justin Lin and Ha-Joon Chang ,’ Development Policy Review , 27 ( 5 ), 483 – 502 . Google Scholar CrossRef Search ADS Lo Turco A. , Maggioni D. ( 2016 ), ‘ On firms' product space evolution: the role of firm and local product relatedness ,’ Journal of Economic Geography , 16 ( 5 ), 975 – 1006 . Google Scholar CrossRef Search ADS McMillan M. S. , Rodrik D. ( 2011 ), ‘Globalization, structural change and productivity growth,’ NBER Working Papers 17143, National Bureau of Economic Research, Inc. Melitz M. J. ( 2003 ), ‘ The impact of trade on intra-industry reallocations and aggregate industry productivity ,’ Econometrica , 71 ( 6 ), 1695 – 725 . Google Scholar CrossRef Search ADS Melitz M. J. , Ottaviano G. ( 2008 ), ‘ Market size, trade, and productivity ,’ Review of Economic Studies , 75 ( 1 ), 295 – 316 . Google Scholar CrossRef Search ADS Minondo A. ( 2010 ), ‘‘ Exports’ productivity and growth across Spanish provinces ,’ Regional Studies , 44 ( 5 ), 569 – 577 . Google Scholar CrossRef Search ADS Neffke F. , Hartog M. , Boschma R. , Henning M. ( 2014 ), ‘Agents of structural change. The role of firms and entrepreneurs in regional diversification,’ Papers in Evolutionary Economic Geography (PEEG) 1410, Utrecht University, Section of Economic Geography, revised Apr 2014. Neffke F. , Henning M. , Boschma M.R. ( 2011 ), ‘ How do regions diversify over time? Industry relatedness and the development of new growth paths in regions ,’ Economic Geography , 87 ( 3 ), 237 – 265 . Google Scholar CrossRef Search ADS Nesta L. , Saviotti P. P. ( 2005 ), ‘ Coherence of the knowledge base and the firm’s innovative performance: evidence from the U.S. pharmaceutical industry ,’ The Journal of Industrial Economics , 53 ( 1 ), 123 – 142 . Google Scholar CrossRef Search ADS Poncet S. , Starosta de Waldemar F. ( 2013 ), ‘ Export upgrading and growth: the prerequisite of domestic embeddedness ,’ World Development , 51 , 104 – 118 . Google Scholar CrossRef Search ADS Schumpeter J. A. ( 1942 ), Capitalism, Socialism, and Democracy . Harper and Brothers : New York, NY . Silverman B. W. ( 1986 ), Density Estimation for Statistics and Data Analysis . Chapman and Hall : New York, NY . Google Scholar CrossRef Search ADS Spence M. ( 2011 ), The Next Convergence: The Future of Economic Growth in a Multispeed World . Farrar, Straus and Giroux : New York, NY . Stiglitz J. , Lin J. , Monga C. , Patel E. ( 2013 ), ‘Industrial policy in the African context,’ Policy Research Working Paper Series 6633, The World Bank. Teece D. , Rumelt R. , Dosi G. , Winter S. ( 1994 ), ‘ Understanding corporate coherence: theory and evidence ,’ Journal of Economic Behaviour and Organization , 23 ( 1 ), 1 – 30 . Google Scholar CrossRef Search ADS Zaccaria A. , Cristelli M. , Tacchella A. , Pietronero L. ( 2014 ), ‘ How the taxonomy of product drives the economic development of countries ,’ PLoS One , 9 ( 12 ), e113770-17 . Google Scholar CrossRef Search ADS © The Author(s) 2018. Published by Oxford University Press on behalf of Associazione ICC. All rights reserved. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png

Industrial and Corporate Change Oxford University Press

http://www.deepdyve.com/lp/oxford-university-press/the-pattern-of-structural-change-testing-the-product-space-framework-bgOXjoAccu

Loading next page...

Publisher: Oxford University Press
Copyright: © The Author(s) 2018. Published by Oxford University Press on behalf of Associazione ICC. All rights reserved.
ISSN: 0960-6491
eISSN: 1464-3650
DOI: 10.1093/icc/dty009
Publisher site: See Article on Publisher Site

Abstract

Journal

Industrial and Corporate Change – Oxford University Press

Published: Apr 5, 2018

References

Industrial Policy in Developing Countries

Altenburg
Trade liberalisation and “revealed” comparative advantage

Balassa
What makes growth sustained?

Berg
Firms in international trade

Bernard
Relatedness and diversification in the European Union (EU-27) and European Neighbourhood Policy countries

Boschma
Related variety, trade linkages, and regional growth in Italy

Boschma
Relatedness and technological change in cities: the rise and fall of technological knowledge in US metropolitan areas from 1981 to 2010

Boschma
The emergence of new industries at the regional level in Spain: a proximity approach based on product relatedness

Boschma

Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

The pattern of structural change: testing the product space framework

The pattern of structural change: testing the product space framework

Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

The pattern of structural change: testing the product space framework

The pattern of structural change: testing the product space framework

Abstract

Journal

Recommended Articles

References