The Impact of Local Product Relatedness on Firm Export Performances Cilem Selin Hazir

∗

Flora Bellone† Cyrielle Gaglio

‡

PRELIMINARY DRAFT, NOT FOR QUOTATION October 3, 2015

Abstract This paper studies the effect of product relatedness on firm trade performance by using micro data for France. We investigate the effect of product relatedness at two levels: relatedness of products exported by a firm to products exported by its locality and relatedness of products within the firm own trade basket. Our empirical analysis bases on a sample of manufacturing firms that are continuous exporters over the period 2002-2007 and we define locality of a firm as the region according to NUTS-1 classification. First results point out to a negative relationship between overall export performance and the extent that a firm’s trade basket is related to the products exported by its region. Whereas, within firm diversification towards related goods increases the overall export performance of single product firm. Keywords: Product Space, Networks, Firm Performance, International Trade, France JEL code: L25, F14, C49

1

Introduction

What you export matters. Since this basic claim by Hausmann et al. (2007), the economic literature has increasingly emphasized the role of structural ∗ Corresponding author: Universit´e Nice Sophia Antipolis and GREDEG-CNRS, 250 rue Albert Einstein, Sophia Antipolis, 06560 France. Email: [email protected] † Universit´e Nice Sophia Antipolis, CNRS-GREDEG and OFCE-Science Po. Address at GREDEG: 250 rue Albert Einstein, Valbonne - Sophia Antipolis, 06560 France. ‡ Universit´e Nice Sophia Antipolis and CNRS-GREDEG, 250 rue Albert Einstein, Valbonne - Sophia Antipolis, 06560 France.

1

change as a driver of economic competitiveness. In the literature, it is argued that besides mere technological capabilities, firms ability to change their product portfolio mix in an ”appropriate direction” is a key driver of their performance on international markets, which in turn, qualifies countries ability to benefit from international trade. Beyond this basic claim, lies a theory of complementary capabilities which explain why firms, in a given institutional context, may remain stuck in inappropriate production patterns, unable to diversify towards products entailing high revenues on international markets. For Hausmann and Rodrik (2003) who initiated this theory, the main reason why firms may miss those opportunities is that discovering what you are ”good” at producing (and selling) in an open economy, requires specific investments which are characterized by localized externalities. In this context, the local environment matters and specific institutional arrangements and public policies may be more efficient than other to promote those investments in a given locality. The idea that the local product space matters has been further elaborated by Hausmann and Hidalgo (2011) who develop a theory of the building of production capabilities at the country level. According to this theory, countries do not diversify their production portfolio randomly but proceed step by step by jumping across ”connected” goods, which are goods sharing common production capabilities. This theory has found strong support in the detailed product-level bilateral trade statistics now available for a large variety of countries from the United Nations (UN) Comtrade database (see Hidalgo and Hausmann (2009) for methodology and overview). More direct evidence on how individual firms benefit from connected goods in their local environment is much more scarce as detailed data covering altogether trade, accounting and location information are not widely available at the firm-level. A notable exception is the paper by Poncet and Starosta de Waldemar (2015) which provides first firm-level evidence on Chinese manufacturing firms showing that firms producing goods that are closely related to the goods that are exported in their locality with revealed comparative advantage (RCA) enjoy higher levels of export revenues in the following period. In this literature, several questions are still at stake. One open question is whether this directed structural change is also an important determinant of firms’ international competitiveness in developed countries. A developed country context might make a difference for at least two reasons. On the one hand, in developed countries, production and sales networks are well developed, and it can be easier for firms to supplement missing local facilities by relying on more distant suppliers or sales intermediaries. In this context, the linkages between capabilities available locally and local firms activities could be weaker. On the other hand, profits opportunity for firms operating in developed countries are primarily driven by their innovativeness and innovation activities could be characterized as well by strong localized knowledge 2

externalities. To the extent that technological proximity is correlated with product proximities, one could expect a positive relationship between firms export performance and their technological congruence to their local environment in developed countries. Existing empirical evidence on this issue is still scarce and comprise several studies focusing on Europe and taking the regional or the country as the unit of analysis. Therefore, there exists no systematic evidence that local product space impacts firm export performances in a developed country context1 A second question stems from the fact that relatedness of products might play a role at two different levels: within the firm among its products and between the firm and its locality. Poncet and Starosta de Waldemar (2015) study the impact of product relatedness at these two levels separately. However, for a better comprehension on structural change process and its impact on firm competitiveness, it is important to understand the relative roles played by product relatedness at these two levels. To address these questions, we focus on French regions and explore how product relatedness affects the firms ability to increase their revenues from trade and to change the composition of their export sales. Our analysis covers mono-regional manufacturing firms that have been continuous exporters over the period 2002-2007. We test for the effect of relatedness/unrelatedness at two levels: at the level of firm by taking into account the proximity among products it exports,and at the regional scale by considering proximity of goods produced by the firm to the products exported by the locality. We use the measure proposed by Hidalgo et al. (2007) to quantify bilateral product proximity. Then, to express the extent that a firm produces products that are related to its locality, we employ notions of social network analysis, we build local product networks and compute closeness centralities of products by following the methodology propose by Opsahl et al. (2010). Our first results, contrary to what has been found in the earlier literature on developing countries, reveal that French firms overall export performance decreases with the relatedness of their products to their local production structure. Nonetheless, we show that as compared to firms exporting a single product, firms that have undergone related diversification depict a higher overall export performance. Furthermore, focusing on firms’ relative sales in their new products, we found that those sales increase when the firm diversifies into products more related to their local environment. The contribution of our study could be identified in at least three axis. First, it addresses the literature on economic geography and firm compet1 There exists some evidence on local export spillovers in Europe Koenig et al. (2010); however, the local context is considered as the co-location of exporters by product or by market (product-destination). In our paper we go beyond the conception of ”co-location of exporters” by uncovering more globally the structure (pattern) of local product space, and studying its impact on firm trade performance.

3

itiveness and to the best of our knowledge it attempts to provide the first evidence based on large scale firm-level data on the effect of product relatedness on export performance in developed countries. Second, it relates to the literature on production and social networks which try to uncover relationships across firms besides mere geographical proximities (Powell et al., 1999; Zaheer and Bell, 2005). Third, it relates to the literature on firm structure and international trade, which specifically investigates the relative performance of multi-product firms (see Bernard et al. (2010), Manova (2008), Eckel et al. (2015)) and argues that firms do not diversify their product portfolio randomly. However, while in this strand of literature the main determinant of a firm diversification strategy and competitiveness are argued to be the toughness of competition a firm faces on each in its individual product-destination market taken separately, we provide evidence on the role of common production or export capabilities. In the sequel, we will start with explaining the model specification and discuss the determinants of firm level export performance in detail. Next, we will explain the sample selection, definition of variables, and data sources. Afterwards, we will present estimation results and then discuss how estimation results could be linked to findings from former relevant studies. Finally, we will provide conclusive comments and identify issues for further research.

2

Determinants of Firm Level Export Performance and The Model

Analysing the determinants of firm competitiveness in international markets is a hot research topic. This renewed interest has several causes. First, empirical studies in the 90s, based on newly released large-scale firm level data sets in a large variety of countries, have acknowledged the persistent heterogeneity of firms’ export behavior which raises the issue to know what differentiate firms performing well on international markets from their competitors operating only on their domestic markets. Also, firms international competitiveness has been identified as a key driver of aggregate economic growth at the country level (European Commission (2014), World Bank (2011)) and this in turn raises the issue to know whether specific institutions and/or policies are required to promote the expansion of domestic firms on international markets. Starting from the literature of the mid 90s, we have learned that firm heterogeneity in terms of export performances was closely related to firm heterogeneity in terms of productive abilities. Specifically, within each detailed industry, firms with different technologies co-exist on their domestic market. However, not all of them operate on export markets; only the most productive ones do. Furthermore, those firms are also larger and pay higher wage than their purely domestic counterparts (Bernard and Jensen, 1995, 4

1999). Associated to the evidence of those export premia, Roberts and Tybout (1997) have shown that exporting was a persistent behavior a the firm level. Specifically the main predictor of a firm current exports revenues is its past export revenues. These facts are theoretically rationalized by means of imperfect competition, ex ante firm heterogeneity in marginal costs and specific trade costs (Bernard et al., 2003; Melitz, 2003). In those models, the existence of specific costs related to export activities, such as setting up distribution channels in the destination countries, changing the packaging, advertising the exported products abroad, etc., explains both the existence of export premia (only the most productive firms can afford those costs) and the persistence in export behaviors (firms will not easily enter or exit export markets if those entries and exits entail sunk costs). According to these first models, ex ante productivity differences across firms determine altogether their relative size and their relative export performance. However, in some recent extensions of those models, it has been shown that, besides productivity, a firm size could act as an independent determinant of the export performance of a firm. This is the case, for instance, when capital market or labor market imperfections prevail and export activities require specific external financial support (Manova, 2008) or specific labor requirement (Fajgelbaum, 2013). It can also be the case when firms can invest to endogenously discriminate their product quality and those quality upgrading investments involve fixed costs (Antoniades, 2015). Finally, in this literature the age of the firm can also play a role with two counterbalancing effects. On the one hand, newly created firms usually need to mature on their domestic market before expanding abroad (Albornoz et al., 2012). On the other hand, newer firms embodied more recent technologies and younger labor force and may be more innovative (Huergo and Jaumandreu, 2004) and more export-oriented. In turn, innovation has been shown to be positively related to the ability to sell on international market (Wakelin, 1998) and the average age of the executives in a firm to be negatively related to the export orientation of the firm. As the first effect may matters only for the very first years of existence of a firm, one could expect that the overall impact of age on firm export performance should be negative rather than positive. Among the firm characteristics which may impact their export performances, a more recent literature has emphasized some characteristics specific to multi-product firms. The basic idea is that multi-product firms usually produce core-competence goods (the one for which the have the lowest relative marginal costs) and peripheral goods (the ones for which their competitive advantage is less pronounced). Depending on their relative productivity and on the toughness of competition they face on their destination market, those firms will maximize their export revenues by enlarging more or less their exported product portfolio. For instance, Bernard et al. (2011) shows that facing trade liberalization, firms have interest to re-concentrate 5

on their core-competence goods. On the other hand, Lelarge and Nefussi (2010) show that French firms which are most exposed to low-cost country competition appear to be on average more diversified than firms operating in more sheltered areas. From this literature we except the variety of products a firm is producing to be related to its export performance although the relationship can be positive or negative depending on the competition the firm faces. Besides product portfolios, some papers have also investigated how the destinations portfolio of exporting firms was related to their export performances. As expected, firms which export to several destinations get higher export revenues than the ones which focus on only one or only a few markets (Eaton et al., 2011). They also show higher productivity revealing that export costs are destination specific and somewhat additive (so that only the most productive firms can afford to bear the cost of selling to many different markets). Finally, the specific sector the firm operates in remains a strong predictor of its export revenues (Malmberg et al., 2000). In this paper, we control for broad industries classes differentiated according to their technological intensity. Besides firms and markets characteristics, the literature has also investigated how economic geography could also matter for determining differences in trade performances across firms. In Europe, regions differ in terms of available skills, knowledge, and capabilities, and in terms of the extent that they are related to the individual firms’ activities. Jacobs (1969) suggest that the diversity in the locality would enhance knowledge externalities due to complementaries among firms. Frenken et al. (2007) distinguish, however, two different forms of diversity; i.e., related and unrelated variety, and argue that related variety fosters knowledge externalities and fosters regional growth. Localized knowledge externalities across firms may be key for strategies of product upgrading and diversification. Second, specific capabilities may be required to overcome fixed export costs and their availability in a given locality may depend on the bundle of goods already exported by the locality. Several studies confirmed these arguments empirically by showing that probability that a non-exporting firm starts exporting was positively related to the co-location of exporters in the same employment area (Koenig et al., 2010). Taking stock of these previous findings, we propose the following linear regression model to study whether relatedness of products in the export portfolio of a firm and relatedness of its products to the local product structure have an impact on firm’s export performance along with the abovementioned determinants that earlier literature suggests. yi,t = c + ζyi,t−5 + λpwi,t−5 + γpri,t−5 + Xi,t−5 β + i,t

6

(1)

3

yi,t c yi,t−5 ζ

: : : :

pwi,t−5 λ

: :

pri,t−5

:

γ

:

Xi,t−5

:

β

:

i,t

:

Value of total exports by firm i at time t. Constant. Value of total exports by firm i at time t − 5. The coefficient associated with previous export performance of the firm. Proximity among products exported by firm i at time t − 5. The coefficient associated with proximity of products that a firm exports in the previous period. Proximity of the products that are exported by firm i to the products exported by its locality at time t − 5. The coefficient associated with proximity of products that a firm exports to the products exported by its locality. A 1×k row vector of control variables indicating the following firm characteristics: size, labor productivity, age, number of products, average number of destinations, and industry (high versus low tech). A k × 1 column vector of coefficients associated with the control variables. Identically and independently distributed error term with mean 0 and variance σ 2 .

Empirical Set-Up and Data

We use the econometric specification explained in the previous section to explain export performance of continuous exporters in manufacturing industry in France in 2007. The restriction to manufacturing firms is driven by the fact that we want to focus primarily on producers (rather than on pure intermediaries). Indeed our working assumption is that local product relatedness primarily matters because of production and knowledge complementarities. Also pure intermediary firms sometimes export a very large number of products (over 300) and we want to exclude those outliers in our sample. Whereas, our focus on continuous exporters is due to methodological constraints. As we investigate the relationship between a firm’s export performance at time t and two relatedness measures at t − 5 (relatedness within its export portfolio and relatedness of its export portfolio to the local product space) we need information on the past export performance of the firm. We built the sample using two data sets provided by INSEE2 , the National Institute of Statistics and Economic Studies in France. The first data set is called FICUS and it brings together two types of information on French enterprises: fiscal information and information gathered by the annual en2

INSEE - Institut National de la Statistique et des Etudes Economiques

7

terprise survey (EAE3 ). Whereas the second data set is called CUSTOMS, and it includes information on quantity and value of firm level exports at the product level. We used customs data to determine continuous exporters, which refers to the firms that exported a positive value (in dollars) each year starting from 2002 to 2007 (included). Then, we used FICUS to characterize manufacturing firms and geo-localize them. In terms of location, FICUS includes two types of information: the region, where the headquarters of the enterprise is located, and the extent that effectives of a firm are spatially concentrated. Regarding spatial concentration of its effectives, firms are classified into three: multi-regional enterprises (not more than 80% of its effectives are in the same region), quasi mono-regional enterprises (80% to 100% of its effectives are in the same region), mono-regional enterprises (100% of its effectives are in the same region). For multi-regional and quasi mono-regional enterprises, FICUS enables us to geo-localize only the headquarters although they have multiple establishments in different regions. This means that for these firms it is not possible to geo-localize precisely which products are produced and exported by which production site, and thus from which region. For this reason, we restricted our analysis to mono-regional manufacturing firms, for which the origin of exports can be correctly determined at the regional level. Over the period 2002-2007 total number of exporters vary between 108000− 120000, approximately 48000 firms export continuously, around 34500 of which are mono-regional. Focusing on manufacturing firms, in the end we obtain a sample of 14569 firms. Next, we followed a four step process to build the variables in the econometric specification. This process is illustrated in Figure 1 and abridges the organization of the rest of the section. Hence, in the sequel we will start with measuring proximity among pairs as it is a pre-requisite to build the variables that expresses proximity of products of a firm (to the products exported by the locality pri,t−5 or to each other pwi,t−5 ). Afterwards, we will present definitions of pri,t−5 , pwi,t−5 . Finally, we will summarize the definitions of other variables.

3.1

Measuring Bilateral Product Proximity

Product proximity, or relatedness, might stem from a number of dimensions as they may require similar set of resources, skills, knowledge bases, or institutions. Quantifying each of these dimensions, assigning them weights, and building up a composite indicator of relatedness is not an easy task. Hidalgo et al. (2007) propose adopting an output-based approach instead to quantify relatedness. They argue that as a consequence of relatedness, countries hav3

EAE - l’Enquete Annuelle d’Entreprises.

8

Figure 1: Four Main Steps of Our Methodology

ing a competitive edge in one good could have or develop advantage in the other good. Hence, they measure similarity between two products by means of the conditional probability of having a revealed comparative advantage in one of these products given that the country has a comparative advantage in the other. Obviously, co-occurrence of two products in a country’s export basket might stem from not only overlaps in underlying production processes (like common production factors, input-output relations, common skills, common knowledge base, etc.) but also overlaps in institutions or social and business networks underlying the export process. Therein, the term relatedness here extends beyond a mere connotation of similarity in terms of sophistication of goods but embraces the material and immaterial setting of the production and export process. The product relatedness measure proposed by Hidalgo et al. (2007) has widely been used in later applied work. It is employed to investigate structural transformation and economic development (initially for low diversified countries)in a number of countries 4 . 4

Chile (Hausmann and Klinger, 2007), South Africa (Hausmann and Klinger, 2008b), Colombia (Hausmann and Klinger, 2008a), Ecuador (Hausmann and Klinger, 2010b), the Caribbean Community (Hausmann and Klinger, 2010a), Algeria (Hausmann et al., 2010), the Kyrgyz Republic (Usui and Abdon, 2010), Sub-Saharan Africa (Abdon and Felipe,

9

Country i is said to have a revealed comparative advantage (RCA) in product k at time t if the share of product k in country i’s export basket is larger than its share in the worldwide export basket. In other words, a country having an RCA in a good means that it is a significant exporter (Hidalgo and Hausmann, 2009) of that good. Following (Balassa, 1965), RCAi,k,t can be expressed formally as follows: P  xi,k,t  1 if Pxi,k,t / PiP ∗ xi,k,t xi,k,t > R RCAi,k,t = (2) i k k  0 otherwise where xikt is the value of product k exported by country i at time t, and generally R∗ = 1. The conditional probability (Pt (k | l) that a country has RCA in product k given that it has RCA in product l at time t is given by the ratio of the number of countries with RCA in both products over the number of countries with RCA in only product l. Then, Hidalgo et al. (2007) define relatedness between two products k and l at time t (φklt ) as follows: φk,l,t = min{Pt (k | l), Pt (l | k)} (3) They explain that taking the minimum of these conditional probabilities helps to symmetrize the proximity matrix and avoids that the conditional probability gets a value of one if a country is the sole exporter of a good. To estimate the relatedness between each pair of products, we calculate φ by making use of data provided by the CEPII5 research center. The database is called BACI6 (Gaulier and Zignago, 2010) and it provides bilateral values (in thousands of US dollars) and quantities (in tons) of world trade flows at the Harmonized System (HS) 6-digit product disaggregation, for more than 240 countries and 5,039 products since 1994. BACI is available with versions 1992, 1996, 2002 and 2007 of the HS classification and it is updated every year. In this study we work with version 1992 of the HS product classification. For each year and each product, BACI harmonizes importer and exporter informations. In this research, we only keep manufacturing products (i.e., at the HS6 level, 5,037 products) and 217 countries7 in 2002. To compute φk,l,t , we aggregate the data to HS 4-digit level8 . 2011), Philippine (Bayudan-Dacuycuy, 2012), China (Poncet and Starosta de Waldemar, 2015) and Turkey (Turco and Maggioni, 2014) 5 CEPII - Centre d’Etudes Prospectives et d’Informations Internationales. Accession date: 01/09/2015 6 BACI - Base pour l’Analyse du Commerce International. 7 We drop territories like East Europe, Neutral Zone, Rest of America, Free Zones or Special Categories, etc. because we only focus on individual exporters. 8 For instance, ”Pharmaceutical Products (30)” is a category at the 2-digit level, the breakdown of this category into 4-digits include ”human blood, animal blood, antisera,

10

3.2

Measuring Proximity among Products Exported by a Firm

CUSTOMS data enables us to identify the products exported by each firm at CN 8 digit level9 . CN8 could be considered as the extension of HS4 codes to 8-digit level, meaning that the first four digits of a CN8 code in year 2002 gives the corresponding HS4 code according to 2002 revision. In order to compute product proximity within portfolio of exports of a firm (at HS 4-digit level), product codes in CUSTOMS are harmonized with HS 1992 version10 . We define the proximity within a firm’s portfolio of exports (pwi,t−5 ) as the average bilateral proximity among products exported by firm i at t − 5. Let Pi,t−5 be the set of products exported by firm i at t − 5, i.e., the firm’s export portfolio. Let Ni denote the cardinality of this set and k and l belong to Pi,t−5 . Then, both measures are defined formally as follows:  pwi,t−5 =

3.3

Ni (Ni − 1) 2

−1 X Ni Ni X

φk,l,t−5

(4)

l=k+1 k=1

Measuring Proximity of Products Exported by a Firm to Products Exported by the Locality

We work with two alternative notions of proximity between products exported by a firm and products exported by its locality (22 French regions d c at NUTS 111 regions in this case), pri,t−5 and pri,t−5 . The first one employs the density measure, which is proposed by Hidalgo et al. (2007) and used in a number of empirical studies (Poncet and Starosta de Waldemar, 2015; Boschma et al., 2012b). Whereas the second measure refers to a centrality measure (Freeman, 1979) that we borrow from social networks literature. The two measures rely on different notions of embeddedness of a firm into production structure of its locality. Below, we provide formal definitions of the measures and compare underlying notions. vaccines etc (3002)”, a further breakdown into 6 digits include ”antisera and other blood fractions (300210)”, ”vaccines, human use (300220)”, ”vaccines, foot and mouth disease, veterinary use (300220)”, etc. 9 CN - Combined Nomenclature (Commission Implementing Regulation (EU) No 1101/2014 of 16 October 2014). 10 Conversion tables and detailed technical notes on conversion is available at the website of United Nations Statistics Division. URL: http://unstats.un.org/unsd/trade/ conversions/HS%20Correlation%20and%20Conversion%20tables.htm. These tables provide conversion between HS6 codes, hence we first convert a CN8 code to a HS6 code. Next we use the corresponding conversion table to convert the code to its equivalent in 1992 revision, which we use as a common base for harmonization. Finally, we take the first 4 digits of the HS6 code to convert it to a HS4 code. 11 Nomenclature of territorial units for statistics for European Union

11

3.3.1

Density

The density measure focuses on the products that the locality exports with comparative advantage and bases on the argument that comparative advantage is build upon products that are proximate to the products at which current comparative advantage lies upon. Then, if the locality exports many products that are closely related to products that a firm exports, the firm is said to be more densely connected to the local product structure (Poncet and Starosta de Waldemar, 2015). Let Densk,m,t denote the sensity of product k in its locality m at time t. Let N be the number of all products, and RCAm,l,t be a binary variable indicating whether locality m has revealed comparative advantage in product l at time t, as defined in Equation 2. Then Densk,m,t could be defined formally as follows: N P

Densk,m,t =

RCAm,l,t φk,l,t

l=1,l6=k N P

(5) φk,l,t

l=1,l6=k

Recalling that Pi,t−5 is the set of products exported by firm i at t − 5, Ni denotes the cardinality of this set and letting k ∈ Pi,t−5 ; average density of products that firm i exports at time t − 5 could then be expressed as: d pri,t−5 =

Ni 1 X Densk,m,t−5 Ni

(6)

k=1

d pri,t−5 , therefore, expresses the proximity of products exported by a firm by means of its dense connections to the products that the locality exports with RCA. However, one might claim that knowledge externalities do not emanate solely from the products that the locality exports with RCA. Also, knowledge flows among not only strongly related goods but also loosely related ones. Furthermore, knowledge may not stop flowing once it reaches a destination (say a proximate product), it may flow along a path reaching a number of products. Therein, we propose adopting a network approach to represent local product space and then using closeness centrality as a measure that summarizes the position of a product with respect to all products in the network.

3.3.2

Closeness Centrality

To compute closeness centrality of products exported by a firm, we start with expressing the regional product space by using a valued one-mode network representation. The idea behind building local networks of products by using a valued network representation is to be able to integrate the local 12

pattern (structure) of the product relatedness into the analysis of firm level export performance. In this representation, the nodes (V = 1, 2, ..., N ) refer to the full list of products in HS - 4 digit classification (N = 1241). For all regions and for each year, the number of nodes in the network is kept to be constant to allow cross-regional comparison. Thus, what makes regional product spaces differ from each other and over time is that products that are not exported by the region at time t appear to be an isolated node12 in its regional product space. Whereas, products that are exported by the region correspond to nodes that are connected with each other. An edge, $k,l,t , that connects any two of these nodes (k, l with k 6= l) represents the degree of relatedness between product k and l at time t. In other words we quantify $k,l,t by computing φk,l,t . Closeness centrality is originally defined for non-valued networks, where ties get binary values. It is expressed as the inverse sum of shortest distances from a focal node to all other nodes in the network (Freeman, 1979). Several attempts have been made to define closeness centrality for valued networks, which require identifying the shortest distance when ties have weights other than zero and one. Opsahl et al. (2010) combines these attempts and proposes a generalization. They suggest that the shortest distance from a focal node to any node in the network have two determinants: weights of ties and the number of intermediary nodes. Let dist(k, l)t be the distance between two distinct nodes k and l in a valued network at time t, then:   1 1 distance(k, l)t = min + ... + (7) ($kht )α ($hlt )α where $kht is the weight of a tie that connects node k to an intermediate node h and α is a positive tuning parameter (Opsahl et al., 2010). When α = 0, the weights of ties do not matter any longer and the distance measure yields the binary distance between a pair of nodes. When α = 1, the inverse of tie weights act as the length (cost) of a tie and the distance measure expresses the total length of the path connecting a pair of nodes. When α < 1, paths consisting of weak ties and less number of intermediate nodes are favored against paths consisting of strong ties but greater number of intermediary nodes. Finally, when α > 1, the number of intermediary nodes becomes less and less important, and paths including larger number of intermediate nodes are identified as shortest paths. In this study, we do not make a preference on favoring or penalizing intermediate nodes and take α = 1. Then, closeness centrality of a node (k) in a valued network is expressed as follows (Opsahl et al., 2010): X −1 N C(kt) = distance(k, l)t (8) l 12

An isolated node means a node that is not connected to any other node in the network.

13

Concerning that the distance from an isolated node to any node in the network is infinite, it is not possible to compute closeness centralities in networks with disconnected components. tnet package in R (Opsahl, 2009) proposes an option to handle this issue. While in the original formula inverse of the sum of shortest distances are calculated; in this option, the sum of inversed shortest distances is computed. Since the limit of a number divided by infinity is zero, the infinite distance between disconnected nodes do not pose a problem in computing closeness. Then, the modified closeness centrality measure could be expressed as follows: Cm (kt) =

N X l

1 distance(k, l)t

(9)

We make use of this modified centrality measure to express how well products of a firm are connected to the products in the locality. We define the average closeness centrality of products that are exported by a firm c (pri,t−5 ) as follows: c pri,t−5

3.4

1 = Ni

X N l

1 distance(k, l)t−5

 (10)

Definition of Other Variables

The dependent variable (yit ) is defined as the value of total exports by firm i in 2007 (t). Previous export performance (yi,t−5 ) is measured by the value of total exports by firm i in 2002 (t−5). Labor Productivity (P roductivityi,t−5 ) is defined by value added (before taxes) per labor (average effective number of employees) in 2002. Both value added and labor are harmonized using deflators at the third level of the summary economic classification (INSEE 1994-2007) 13 . Size (Sizei,t−5 ) is measured by the average effective number of employees of firm i in 2002. The variable Agei,t−5 refers to the age of firm i in 2002. The size of the firm’s export portfolio (N umprodi,t−5 ) is defined as the number of products exported by firm i in 2002. The ability of the firm to serve different destinations is measured by the average number of destinations that firm i exports a product in its portfolio in 2002 (AvN umdesti,t−5 ). Finally, hightech is defined as a dummy variable indicating whether the main line of business of the firm belongs to an industry classified as ”high tech” or ”medium high tech” according to the OECD classification of low, medium and high tech industries. 13

Nomenclature Economique de Synth`ese (NES) with 114 positions.

14

4

Results

We express all variables in the specification in logarithm and estimate the linear specification provided in Equation 1. In Model 1, we used the average density to express the proximity of products exported by a firm to the products exported by the locality. In Model 2, however, instead of the density measure we employed the average closeness centrality of the firm’s products in the local product network. Comparing Model 1 and Model 2, we can investigate whether it is only the products giving the region its comparative advantage or the whole and unique pattern of product relatedness in the locality (including products in which the region does not posses RCA) that matters for firm level trade performance. We estimated the two models for the whole sample, then we made separate estimations for two sub-samples, consisting of firms exporting a single product as of 2002 (mono-product firms) and firms exporting at least two products as of 2002 (multi-product firms), respectively. Table 1 summarizes the results for each model and each sample. For all three samples and both models, we find strong statistical evidence on the effect of past export performance and some firm level characteristics (age, size, and labor productivity) on current export performance. The positive and statistically significant coefficient estimate for past level of exports (yi,2002 ) indicates persistence of export activity, which could be explained by associated sunk costs incurred in the past. Again, the positive and statistically significant coefficient estimates for productivity and size are in line with the theoretical rationales suggested by the Melitz model and its extensions (Melitz, 2003; Manova, 2008) as explained in Section, 2 and tell us that increases in firm size or productivity results in an increase in export performance. On the other hand, the negative and statistically significant coefficient estimate for age is in line with the argument that young firms tend to be more innovative and innovation fosters the ability to operate in the international market (Huergo and Jaumandreu, 2004; Wakelin, 1998) We find a positive and statistically significant relationship between the size of the trade basket (number of products) in 2007 and the value of exports in 2007. While former evidence on the direction of this relationship is ambiguous, our finding goes hand in hand with findings of Lelarge and Nefussi (2010) on French exporters and suggests that diversification fosters export performance. The results provide us with mixed evidence on the effect of firm’s ability to serve different markets on its export performance. For all firms and multi-product firms we find positive and statistically significant coefficient estimates indicating that firms will obtain higher export revenues in the succeeding period if they are capable of serving more markets in the current period. One reason for could be that firms’ could take the advantage of their experience and investments for a particular market for a particular good to 15

Table 1: Parameter Estimates (Dependent Variable: Firm’s Total Value of Exports in 2007) All Firms Mono-Product Firms Multi-Product Firms Model 1 Model 2 Model 1 Model 2 Model 1 Model 2 0.676*** 0.672*** 0.581*** 0.575*** 0.719*** 0.716*** (0.009) (0.009) (0.017) (0.017) (0.010) (0.010) 0.522*** 0.459*** 0.465*** 0.509*** 0.515*** P roductivityi,2002 0.517*** (0.020) (0.020) (0.048) (0.048) (0.022) (0.022) 0.421*** 0.421*** 0.448*** 0.452*** 0.388*** 0.388*** Sizei,2002 (0.012) (0.012) (0.024) (0.024) (0.013) (0.013) Agei,2002 -0.183*** -0.183*** -0.325*** -0.328*** -0.147*** -0.147*** (0.019) (0.019) (0.045) (0.045) (0.020) (0.020) 0.060*** 0.071*** 0.038** 0.046*** N umprodi,2002 (0.016) (0.016) (0.017) (0.017) AvN umdesti,2002 0.055*** 0.058*** -0.065** -0.058* 0.110*** 0.113*** (0.018) (0.018) (0.032) (0.032) (0.021) (0.021) 0.047* 0.055** 0.048 0.057 0.034 0.041 hightech (0.025) (0.025) (0.071) (0.071) (0.026) (0.026) pwi,2002 0.140*** 0.152*** -0.065 -0.045 (0.034) (0.034) (0.053) (0.053) d -0.092** -0.152 -0.085** pri,2002 (0.040) (0.101) (0.042) c pri,2002 -0.239*** -0.297*** -0.209*** (0.043) (0.091) (0.048) constant 1.962*** 3.831*** 3.012*** 5.367*** 1.408*** 3.058*** (0.145) (0.341) (0.338) (0.718) (0.164) (0.383) Nobs 14569 3097 11472 R-sqr 0.694 0.694 0.505 0.507 0.712 0.713 *** 99% confidence level, ** 95% confidence level, ** 90% confidence level Heteroscedasticity-robust error estimates are reported in parenthesis. Variable yi,2002

start exporting new or other goods to that market. For mono-product firms, we find a contrasting negative impact of the number of destinations on export revenues. Specifically, we find that exporting to more destinations a unique good, decreases the firm export revenues in the succeeding period. This effect could reveal that mono-product firms may gains at concentrating their international marketing investment efforts on a limited number of markets rather than dissipating those efforts over a too large number of different destinations. Controlling for sectors, high versus low technology sectors, we find mixed evidence for different samples. Neither for mono-product exporters nor for multi-product exporters we do not find enough statistical evidence to confirm a difference between firms, whose main economic activity is in high technology sectors and low technology sectors. However, for the sample covering all firms, we find weak (Model 1) to moderate (Model 2) statistical

16

evidence on that firms operating mainly on high technology sectors perform better than those operating in low technology sectors. This could be due to an interaction effect between diversification and sector type, meaning that diversification in high technology sectors fosters export performance. For the whole sample and in both models, we find strong statistical evidence on that both the number of products that French firms export and the proximity among these products in 2002 affect their export performance in 2007 positively. Note that when we estimate the models for all firms, pwi,2002 takes a value of zero for mono-product firms and a positive value for multiproduct firms, meaning that mono-product firms constitute the reference case. Then, this findings suggest that as compared to mono-product firms, multi-product firms depict a higher export performance as the size of their export portfolio in the preceding period gets larger; and their performance is even higher if the average bilateral proximity among their products are higher. For the sample consisting of only multi-product firms, the positive effect of the size of the portfolio is confirmed but we did not find enough statistical evidence confirming the effect of relatedness among products on export performance. This tells us that as firms diversify the products they export, at the beginning; i.e., when a mono-product exporter becomes a multi-product exporter, related diversity affects export performance in the succeeding period positively; but then although an increase in the size of the portfolio augments the future export performance, relatedness within the portfolio does not matter anymore. Concerning the proximity of products exported by a firm to the products exported by its locality, statistical evidence is mixed across Model 1 and Model 2. As mentioned above, in Model 1 we employ the density measure as an indicator of the extent that the firm is connected to the local product structure, and hence focus more on the congruence of its products with the products that the region exports with RCA. When we estimate Model 1 for the whole sample, we find some statistical evidence on that dense connections to their local product structure in 2002 affect French firms’ export performance in 2007, negatively. The same result holds for multiproduct firms but for mono-product firms we do not have enough statistical evidence to conclude that the local product structure is a determinant of export performance for French firms. While evidence is mixed across samples for Model 1, for Model 2, where we use the closeness centrality measure to express the proximity of products exported by a firm to the products exported by its locality, we have stronger statistical evidence and the evidence holds for all three samples. Unlike Model 1, in Model 2 the extent that the firm is connected to the local product structure is measured by taking the whole pattern of product relatedness in the locality, without a focus on products with local RCA. The closeness centrality measure expresses the position of each product in the local product network by means of its distance to all other products. The 17

c statistically significant and negative coefficient estimates for pri,2002 , therefore suggest that French firms that export products that are more proximate to all products exported by their locality in 2002, export less in 2007.

5

Discussion

The findings presented the in the previous section tells us that regardless of the way we measure it, connectedness of a firm to its local production structure is negatively related to firms’ export performance. This result for the French case for the period 2002-2007 does not simply corroborate earlier firm level evidence suggested by (Poncet and Starosta de Waldemar, 2015), who explains firm performance at the product level14 and defines the locality at the city level. In order to scrutinize this difference, we replicated a part of their study using French data, and found again a negative relationship between density of product k in t−5 and the value of exports of product k by firm i in t. The different evidences on Chinese firms and French firms might simply be due to the different roles played by the locality in developing and developed countries that we discussed in Section 1. Another reason might be the difference at spatial scales that we use to define the locality. Relatedness and unrelatedness might be playing different roles at different spatial scales. Besides, (Poncet and Starosta de Waldemar, 2015) reports evidence on a competition effect within firm’s export basket, meaning that a firm exports less of a particular good if the average density of other products in its export basket increases. Concerning that our specifications take into account density in the region and proximity within the portfolio together and we explain firm exports for all products, overall impact of competition on the export basket might be leading to a negative coefficient estimate for average density of products. A negative relationship between connectedness to local production structure and firm export performance tells us that firms specializing in products that are considerably different from the products exported by the locality would enjoy higher level of exports in the succeeding periods. However, the studies investigating growth and trade performance at the regional level propose that product relatedness affect the direction of regional growth, and they suggest evidence on related diversification at the regional level. We then investigated the direction of diversification at the regional level and corroborated the evidence suggested by (Boschma et al., 2012a) for the French case. At the aggregate level, as depicted by Figure 2 we found also that there is a positive relationship between the average density of prod14 In this study we define our dependent variable as firm’s total exports instead of a product level disaggregation due to the fact that product level export decisions and capabilities are all interdependent on each other, and this may violate the key assumption of a linear regression model that error terms are identically and independently distributed

18

ucts without RCA in 2002 and new products with RCA in 2007 for French regions. Figure 2: Relationship between Average Density of Products without Revealed Comparative Advantage (RCA) in 2002 and the Number of New RCA Products in 2007 for French Regions

The counter result at the firm level might be due to the fact that the dependent variable, being firm’s total exports, does not enable us to see changes at the product level. We, then focus on firm performance regarding new products that they introduce in the successive years. Hence, we investigate the question whether the extent that firms create export revenues by introducing new products depend on the congruence of these products with their existing export portfolio and with the local production structure. In this case, the dependent variable is defined as the ratio of export revenues of firm i from goods that it was not exporting in 2002 but exporting in 2007 to its total export revenues in 2002. Table 2 presents the estimations results for this new dependent variable. For all firms, unlike previous estimations, for both density measure and closeness centrality measure, we find that a firm’s export performance in new products (relative to its past export performance) is positively related to the local production structure. Then the two main axes, namely knowledge externalities and export capabilities, along which we expect the locality to have an impact on firm performance, seem to have a positive effect via diversification. The reason why we do not observe positive effects on total 19

export performance might be reflecting a transition and adjustment in the configuration of the firm’s trade basket. This issue still remains for further investigation. Table 2: Parameter Estimates (Dependent Variable: Export Value of New Products in 2007 / Firm’s Total Value of Exports in 2002) All Firms Mono-Product Firms Multi-Product Firms Model 1 Model 2 Model 1 Model 2 Model 1 Model 2 -0.805*** -0.802*** -0.876*** -0.873*** -0.775*** 0.772*** (0.015) (0.015) ((0.029) (0.029) (0.017) (0.017) P roductivityi,2002 0.383*** 0.378*** 0.494*** 0.490*** 0.343*** 0.337*** (0.037) (0.037) (0.098) (0.098) (0.039) (0.039) Sizei,2002 0.372*** 0.372*** 0.520*** 0.516*** 0.323*** 0.323*** (0.019) (0.019) (0.044) (0.044) (0.021) (0.021) Agei,2002 -0.218*** -0.218*** -0.322*** -0.321*** -0.202*** -0.203*** (0.033) (0.033) (0.085) (0.085) (0.035) (0.035) N umprodi,2002 0.161*** 0.154*** 0.189** 0.181*** (0.029) (0.029) (0.032) (0.032) AvN umdesti,2002 -0.555*** -0.557*** -0.601*** -0.606*** -0.546*** -0.548*** (0.035) (0.035) (0.075) (0.075) (0.039) (0.039) hightech 0.252*** 0.246*** 0.317*** 0.313*** 0.235*** 0.227*** (0.041) (0.041) (0.120) (0.121) (0.044) (0.044) 0.341*** 0.332*** -0.024 -0.056 pwi,2002 (0.060) (0.060) (0.089) (0.090) d pri,2002 0.176*** 0.230 0.180** (0.068) (0.166) (0.074) c pri,2002 0.241*** 0.187 0.301*** (0.074) (0.163) (0.084) constant 6.619*** 4.619*** 6.788*** 5.123*** 6.148*** 3.686*** (0.250) (0.597) (0.644) (1.294) (0.278) (0.686) Nobs 10483 1557 8881 R-sqr 0.475 0.475 0.563 0.563 0.406 0.406 *** 99% confidence level, ** 95% confidence level, ** 90% confidence level Heteroscedasticity-robust error estimates are reported in parenthesis. Variable yi,2002

6

Conclusion

In this study, we aim to understand how product relatedness affects firm level trade competitiveness. We investigate the role of product relatedness along two axis: relatedness of a firm’s products to the production structure in its locality, and relatedness among products in a firm’s export basket. Our empirical analysis bases on a sample of French manufacturing firms that kept exporting continuously over the period 2002-2007. Our first results suggest that as compared to exporters of a single product, firm that are diversified in related products enjoy higher level of exports in the succeeding 20

period. Yet, once diversified the direction of further diversification; i.e., related or unrelated, does not seem to affect performance. Furthermore, we find statistical evidence on a negative relationship between firm’s overall export performance and relatedness of its product to the local production structure. However, focusing on export performance in new products we found evidence on that local production structure have positive effect on the level of export revenues due to diversification relative to the past level of export revenues. However, the results should be taken with care as these are first results of an ongoing study and require further research effort to enable us to make generalizations. One direction for further research is to understand the impact of product relatedness on firm level export performance via upgrading of existing products versus diversification more deeply. A second direction would be to extend this cross-sectional study to a cohort study and observe whether main conclusions hold over a time-window. Another direction would be to explore how export starters are affected by product relatedness so that we can elucidate the micro foundations of regional diversification and competitiveness. Apart this need for further research, some data limitations should also be taken into account. As we do not have plant level data, we are not capable of geo-localizing exports of firms that operate in multiple regions. Hence, we confine our analysis to mono-regional firms, which in turn leads to an underestimation of the local product structure. Still, the study contributes to the relevant literature in several ways. First, to the best of our knowledge it is the first evaluation using micro data of the gains of the consistency of activities at the local level and within the firm in an industrialized country like France. Hence, it tries to address the lack of enough firm-level analysis in the literature and it could be considered as a step forward to enable cross-country comparisons. Second, the study borrows measures from social network analysis and suggest an alternative way to conceptualize and measure embeddedness of firms’ activities to its locality. Namely, closeness centrality measure employed in this analysis tells us that what underlies firm competitiveness in international markets may not be merely the current set of products at which regional comparative advantage relies upon, the particular local pattern of the product structure might indeed be in play.

Acknowledgements We would like to thank Ron Boschma for his comments and suggestions during his visit to Universit´e Nice Sophia Antipolis, CNRS-GREDEG.

21

References Abdon, A and J Felipe (2011), “The product space: What does it say about the opportunities for growth and structural transformation of sub-saharan africa?” Levy Economics Institute Working Paper. Albornoz, Facundo, H´ector F. Calvo Pardo, Gregory Corcos, and Emanuel Ornelas (2012), “Sequential exporting.” Journal of International Economics, 88, 17 – 31. Antoniades, Alexis (2015), “Heterogeneous firms, quality, and trade.” Journal of International Economics, 95, 263 – 273. Balassa, B (1965), “Trade liberalization and revealed comparative advantage.” The Manchester School of Economic and Social Studies, 33, 92– 123. Bayudan-Dacuycuy, C (2012), “The philippine export portfolio in the product space: Potentials, possibilities and policy challenges.” Economics Bulletin. Bernard, A B and J B Jensen (1995), “Exporters, jobs, and wages in u.s. manufacturing: 1976-1987.” Brookings Papers on Economic Activity. Microeconomics, 1995, 67–119. Bernard, A B and J B Jensen (1999), “Exceptional exporter performance: Cause, effect, or both?” Journal of International Economics, 47, 1–25. Bernard, Andrew B., Jonathan Eaton, J. Bradford Jensen, and Samuel Kortum (2003), “Plants and productivity in international trade.” American Economic Review, 93, 1268–1290. Bernard, Andrew B., Stephen J. Redding, and Peter K. Schott (2010), “Multiple-product firms and product switching.” American Economic Review, 100, 70–97. Bernard, Andrew B., Stephen J. Redding, and Peter K. Schott (2011), “Multiproduct firms and trade liberalization.” The Quarterly Journal of Economics, 126, 1271–1318. Boschma, R, A Minondo, and M Navarro (2012a), The emergence of new industries at the regional level in Spain: A proximity approach based on product-relatedness. Boschma, R, A Minondo, and M Navarro (2012b), “Related variety and regional growth in spain.” Papers in Regional Science, 91.

22

Eaton, Jonathan, Samuel Kortum, and Francis Kramarz (2011), “An anatomy of international trade: Evidence from french firms.” Econometrica, 79, 1453–1498. Eckel, C, L Iacovone, B Javorcik, and J. P. Neary (2015), “Multi-product firms at home and away: Cost- versus quality-based competence.” Journal of International Economics, 95, 216–232. European Commission (2014), “European competitiveness report 2014: helping firms grow.” Report, European Commission. Fajgelbaum, Pablo D. (2013), “Labor market frictions, firm growth, and international trade.” NBER Working Papers 19492, National Bureau of Economic Research, Inc. Freeman, L (1979), “Centrality in social networks conceptual clarification.” Social Networks, 1, 215–239. Frenken, K, F G van Oort, and T Verburg (2007), “Related variety, unrelated variety, and regional economic growth.” Regional Studies, 41, 685–697. Gaulier, G and S Zignago (2010), “Baci: International trade database at the product-level. the 1994-2007 version.” CEPII WP. Hausmann, R and B Klinger (2007), “Structural transformation in chile.” Quantum Advisory Group Working Paper. Hausmann, R and B Klinger (2008a), “Achieving export-led growth in colombia.” Harvard Kennedy School Working Paper. Hausmann, R and B Klinger (2008b), “South africa’s export predicament.” Economics of Transition. Hausmann, R and B Klinger (2010a), “Policies for achieving structural transformation in the caribbean.” Inter-American Development Bank Discussion Paper. Hausmann, R and B Klinger (2010b), “Structural transformation in ecuador.” Inter-American Development Bank Policy Brief. Hausmann, R, B Klinger, and J R Lopez-Calix (2010), Export diversification in Algeria. in Lopez-Calix J. R., Diop, N., Walkenhorst, P., Trade competitiveness of the middle East and North Africa, Washington, D.C., The World Bank. Hausmann, Ricardo and C´esar Hidalgo (2011), “The network structure of economic output.” Journal of Economic Growth, 16, 309–342.

23

Hausmann, Ricardo, Jason Hwang, and Dani Rodrik (2007), “What you export matters.” Journal of Economic Growth, 12, 1–25. Hausmann, Ricardo and Dani Rodrik (2003), “Economic development as self-discovery.” Journal of Development Economics, 72, 603–633. Hidalgo, C A and R Hausmann (2009), “The building blocks of economic complexity.” Proceedings of the National Academy of Sciences of the United States of America, 106, 10570–75. Hidalgo, C A, B Klinger, A L Barab´asi, and R Hausmann (2007), “The product space conditions the development of nations.” Science, 317, 482– 487. Huergo, Elena and Jordi Jaumandreu (2004), “Firms’ age, process innovation and productivity growth.” International Journal of Industrial Organization, 22, 541–559. Jacobs, J (1969), The Economy of Cities. Vintage, New York. Koenig, P, F Mayneris, and S Poncet (2010), “Local export spillovers in france.” European Economic Review, 54, 622–641. Lelarge, C and B Nefussi (2010), “Les strat´egies de portefeuille de produits des entreprises industrielles fran¸caises face `a la concurrence des pays `a bas coˆ ut de production.” Economie et statistique, 435, 31–47. Malmberg, A, B Malmberg, and P Lundequist (2000), “Agglomeration and firm performance: Economies of scale, localisation, and urbanisation among swedish export firms.” Environment and Planning A, 32, 305–321. Manova, Kalina (2008), “Credit constraints, equity market liberalizations and international trade.” Journal of International Economics, 76, 33–47. Melitz, M. (2003), “The impact of trade on intra-industry reallocations and aggregate industry productivity.” Econometrica, 71, 1695–1725, URL http://www.nber.org/papers/w8881. Opsahl, Tore (2009), Structure and Evolution of Weighted Networks. University of London (Queen Mary College), London, UK, URL http: //toreopsahl.com/publications/thesis/. Opsahl, Tore, Filip Agneessens, and John Skvoretz (2010), “Node centrality in weighted networks: Generalizing degree and shortest paths.” Social Networks, 32, 245–251. Poncet, Sandra and Felipe Starosta de Waldemar (2015), “Product relatedness and firm exports in china.” The World Bank Economic Review, forthcoming. 24

Powell, W W, K W Koput, L Smith-Doerr, and J Owen-Smith (1999), “Network position and firm performance: Organizational returns to collaborationin biotechnology industry.” Sociology of Organizations, 16, 129–159. Roberts, M J and J R Tybout (1997), “The decision to export in colombia: An empirical model of entry with sunk costs.” The American Economic Review, 87, 545–564. Turco, A Lo and D Maggioni (2014), “On firms’ product space evolution: The role of firm and local product relatedness.” In Universita Politecnica delle Marche Working Paper. Usui, N and A Abdon (2010), “Structural transformation in the kyrgyz republic: Engineering future paths of capability accumulation.” Asian Development Bank Working Paper. Wakelin, Katharine (1998), “Innovation and export behaviour at the firm level.” Research Policy, 26, 829–841. World Bank (2011), Leveraging trade for Development and inclusive growth: the World Bank Group trade strategy, 2011 − 2021. World Bank Group. Zaheer, A and G G Bell (2005), “Benefiting from network position: Firm capabilities, structural holes, and performance.” Strategic MAnagement Journal, 26, 809–825.

25