Export-led innovation: the role of export destinations

Export-led innovation: the role of export destinations Abstract This article analyzes the effect of exporting activity on the innovative performances of firms in France, Germany, Italy, Spain, and UK. It argues that the positive effect of exporting on innovation usually found in the literature varies according to the specific destinations of exports, and it identifies two dimensions along which export destinations might differ: the level of foreign technological spillovers available to exporting firms (the technological learning effect) and the type of foreign demand that exporting firms are able to access (the foreign demand effect). The empirical analysis, which takes advantage of firm-level information about the export destinations of exporters, shows that while the technological learning effect increases mainly the incentives to introduce brand new product innovations, the foreign demand effect fosters the adoption of process innovations. 1. Introduction Empirical contributions in both the trade literature and the technology gap tradition have provided extensive evidence that innovative capabilities are important factors able to explain the export participation and the export success of firms (Posner, 1961; Fagerberg, 1988; Wakelin, 1998). Indeed innovative activities, even more than cost advantages, are able to increase the overall competitiveness of firms and allow them to compete in international markets (Basile, 2001; Cassiman et al., 2010; Dosi et al., 2015). While the fact that innovation is an important boost for export activities is generally accepted as an established fact, recent empirical evidence has suggested that the relationship between innovation and exports can also work in the opposite direction: exporters might be induced to innovate precisely because they are active in export markets. The intuition is that by operating on international markets, firms are induced to implement new organizational and technological routines that eventually allow them to increase productivity levels. A growing literature has analyzed the causal link existing between exporting activity and innovation (Liu and Buck, 2007; Andersson and Lööf, 2009; Lileeva and Trefler, 2010; Bustos, 2011), finding, in most cases, that exporting activity indeed increases the probability to introduce innovations at the firm level. While these studies agree on the existence of a generalized positive effect of exporting activity on innovation performances, there is instead no unanimity concerning which specific types of innovations are mostly affected. The results differ according to the specific national setting of the study: positive effects have been found in different empirical studies on, respectively, product innovation (Bratti and Felice, 2012), process innovation (Damijan et al., 2010), or patent applications (Salomon and Shaver, 2005). A possible explanation of the heterogeneity of these results is that in most of the cases the existing literature has considered exporting as an undifferentiated activity, regardless of the specific export destinations of firms. Quite on the contrary export destinations can differ a lot along several dimensions (level of competition; geographical and cultural distance; institutional setting), and each of them might induce different types of innovation outcomes. Studies that do not acknowledge this factor are likely to find different results, according to the specific destination of exports of the firms analyzed. In this article I focus on two important dimensions along which export destinations can differ and which might affect innovation outcomes: one is the level of foreign technological spillovers available to exporting firms, and the other is the type of foreign demand that exporting firms can have access to. Firms might benefit from exporting either because they are active in very technologically advanced foreign markets, where they learn to innovate from foreign clients and competitors, or because they export to foreign markets with a strong demand growth, which increases the demand for the firms’ products and also the profitability of introducing new technologies. Throughout the article I will refer to the first effect as the technological learning effect and to the second as the foreign demand effect. I argue that each of these two effects, which directly depend on export destinations, will exert different incentives on the introduction of innovations: the technological learning effect mainly reduces the costs of developing innovations, while the foreign demand effect increases the firms’ output and hence the profitability of introducing innovations. To identify the differentiated impact of each of these two effects on innovation outcomes, I examine two very different types of innovations—brand new product innovations and process innovations—and explore how the technological learning and foreign demand effect influence the introduction of these two innovation outputs. I take advantage of the EU-EFIGE/Bruegel-UniCredit data set (European firms in a Global Economy), which includes detailed firm-level information about export activity and innovation performances of a large number of firms active in the five largest European economies (Germany, France, Italy, Spain and UK) in the period 2007–2009. The main advantage of the data set is that it includes information on the most important export destinations of firms. I use this information to build two indexes that measure how much an exporting firm is exposed to each of the two effects. I proxy the technological learning effect with the R&D intensity of the countries in which a firm exports, and the foreign demand effect with the growth of imports of the markets in which an exporting firm is active. Since I expect firms to be mainly affected by the features and dynamics of their sector of affiliation, rather than by country-level ones, in both cases I use the R&D intensity and import growth of the sector in which exporting firms are active. I include these two indices in a model that explains the introduction of, respectively, brand new product innovations and process innovations. I test the robustness of the results through the inclusion of different productivity measures, through the use of instrumental variable (IV) estimations and through different methods of building the indexes of the technological learning and foreign demand effect. The results of the empirical analyses show that indeed the technological learning and foreign demand effects induce different innovation outcomes. The technological learning effect only affects brand new product innovations. The foreign demand effect is positively correlated with both process innovation and brand new product innovation, but the IV analysis points to a causal effect only on process innovations. The article contributes to the literature in a number of ways: first, in line with other recent studies (Golovko and Valentini, 2014), the effect of exporting activity on innovation is not anymore assessed on a generic measure of innovation, but rather on specific types of innovation outcomes. Second, the article shows that export destinations are an important moderator factor, able to drive the effect of exporting activity on different types of innovations. In doing so, it also provides a more precise measurement of the role of export destinations: indeed so far existing studies that took into consideration export destinations, only distinguished between the effect of exporting to high-income or Organisation for Economic Co-operation and Development (OECD) countries with respect to low- or middle-income countries (Salomon, 2006; Trofimenko, 2008). The present analysis instead not only takes into account the features of the specific country of destination but also of the sector of activity of each firm. Finally, the article introduces two different dimensions along which export destinations might differ: the availability of technological spillovers (the technological learning effect) and the access to foreign expanding markets (the foreign demand effect). While the former is often acknowledged in the existing literature, the latter has been largely overlooked: however, both seems relevant to understand the effect of exports on innovation. The article is organized as follows: Section 2 presents the relevant literature and introduces the hypotheses concerning the role of the technological learning and foreign demand effect on different types of innovations; Section 3 presents the empirical strategy and the data used; Section 4 presents the results, while Section 5 draws some conclusions and implications for policy and management. 2. The effect of exports on innovation: technological learning and foreign demand effects The evidence that exporters are generally more productive than non-exporters is usually explained by two main hypotheses: either by the fact that more productive firms self-select into export or, alternatively, by the fact that being active in international markets allows firms to upgrade their technologies. The self-selection hypothesis explains the productivity differentials among exporters and non-exporters with the existence of additional costs related to the sale of goods in foreign countries that act as entry barriers: only highly productive firms will be able to overcome such costs and hence decide to export (Bernard and Jensen, 1999; Melitz, 2003). The alternative hypothesis instead suggests that exporters become more productive precisely because they are in international markets: an increasing number of studies report results highlighting an ex post positive effect of exporting both on the productivity levels (Lileeva and Trefler, 2010; Bustos, 2011) and on the innovation outcomes (Damijan et al., 2010) of exporting firms. These results are often explained by the fact that, through interactions with foreign customers, exporters benefit from international technological spillovers in the forms of know-how and new technological capabilities: hence, higher productivity is a consequence of exporting, rather than a prerequisite. In this article I refer to this as the “technological learning effect.” In the literature that studies the mechanism by which firm self-select into exporting, the role of export destinations has been explored in several theoretical and empirical contributions, which showed that the level of productivity required for firms to start to export strongly depends on destination-specific factors, such as distance, the level of income of export destinations, the size of the market, and the exporters’ familiarity with institutions (Blanes-Cristóbal et al. 2008; Serti and Tomasi, 2012). On the contrary the literature focused on the effect of exporting on innovation and productivity has largely overlooked the role that different export destinations may exert on exporters. Nevertheless, the specific characteristics of destination markets are likely to be extremely important also to understand the impact of exporting activities on innovation and productivity. Indeed the learning by exporting mechanism is implicitly related with the specific destination markets of exporters, since to learn firms should be active in markets that are more advanced than the domestic one. In line with this, most of the existing studies find a positive effect of exporting on innovation and productivity among firms in middle-income and emerging countries, which export to high-income countries. Verhoogen (2008) finds that increased access to foreign markets induces Mexican exporters to upgrade their quality. Similar findings are provided by Liu and Buck (2007) for Chinese firms, by Van Biesebroeck (2005) for Sub-Saharan African companies, and by Fafchamps et al. (2008) for Moroccan firms exporting to Europe. However only Trofimenko (2008) explicitly measures the effect of different export destinations, although at a very aggregate level: in her analysis on Colombian firms, she finds that only firms which export to OECD countries truly benefit from exporting activity. The positive effect of exports on innovation and productivity is also found among firms in high-income economies. Salomon and Shaver (2005) report a positive effect of exporting on patenting and product innovation among Spanish firms; Crespi et al. (2008) find that UK firms able to access relevant information from foreign buyers are also able to increase their productivity. Damijan et al. (2010) and Bratti and Felice (2012) find a positive effect of exporting activity on innovation among, respectively, Slovenian and Italian firms: however while the former report a positive effect on process innovation, the latter show that exporting increases product innovation. Also in these studies the specific destinations of exports are usually not considered, implicitly assuming that technological spillovers occur, regardless of the specific foreign markets in which firms are active. The only exception is Salomon (2006), who finds a positive effect of exports to OECD countries on patents among Spanish firms. Overall the literature that explains the positive effect of exports on innovation through the access to foreign technological spillovers (the “technological learning effect”) has largely overlooked the role of export destinations. Moreover when export destinations have been introduced, this was limited to the broad distinction between high- and low-income countries. However it seems likely that export destinations should indeed matter for learning processes, since these only occur when foreign buyers are more sophisticated than domestic ones. Moreover the literature is also not unanimous about which types of innovation are affected by the technological learning effect, whether patents, product, or process innovations. A second explanation of the positive effect of exporting activity on innovation and productivity that is found in the literature is related to the role of foreign demand: in this article I will refer to it as “the foreign demand effect.” In the tradition of Schmookler (1966), demand growth has often been considered as an important determinant of innovation. The demand pull literature has largely emphasized the positive effect of the growth of markets on the innovative efforts put in place by firms (Geroski and Walter, 1995; Brouwer and Kleinknecht, 1999). Therefore also the growth of foreign markets is likely to induce exporting firms to invest in new technologies, since the expected profits related to the introduction of innovations will increase with the level of firms’ sales. The evidence on the effect of foreign demand on exporters’ innovative activities is quite limited and not clear-cut. Piva and Vivarelli (2007) measure the impact of demand on the investments in R&D of Italian firms and find that the positive effect of demand is stronger among firms with higher export intensity. Woerter and Rope (2010) show that foreign demand does not have any effect on the innovation outputs of Irish firms, while it has a positive, but quite limited, effect on product and process innovation among Swiss firms. Both studies however adopt aggregate measures of foreign demand growth which do not distinguish between firms exporting to markets with high or low growth of demand; also in this case, the role of export destinations is not accounted for. However it seems plausible that the foreign demand effect should have a great importance in fostering innovation strategies only for firms exposed to a high growth of foreign sales. Moreover also in this case it is likely that the foreign demand effect might foster some specific types of innovations, those especially affected by the growth of demand and sales, rather than others. Summing up existing studies suggest that the positive effect of exporting activities on innovation might be due to (at least) two different underlying mechanisms: a technological learning effect that allows firms to benefit from foreign technological spillovers and a foreign demand effect that creates incentives for firms to introduce innovations due to the increase of their foreign markets. However both effects will strongly depend on the specific export destinations of each firm. Moreover each of these two effects might also induce different types of innovations. So far the existing literature has rather overlooked these last two points: in the next section, I will specifically investigate whether the technological and foreign demand effects, which depend on export destinations, also lead to different innovative strategies. 2.1 The effect of exporting activities on different innovation outcomes The technological learning and foreign demand effects are two factors able to explain the positive effect of exporting activity on innovation, and their impact depending on the specific export destinations of each exporting firm. However their impact on innovation is also likely to differ according to the specific innovative output considered. Indeed the technological learning effect allows firms to benefit from foreign knowledge spillovers which decrease the internal research costs necessary to develop new innovations. The foreign demand effect instead increases firms’ potential output and the number of units sold. Therefore different types of innovations will be affected, according to how much these factors affect the profitability of their introduction. In this article I will focus on the impact of the technological learning and foreign demand effects on two specific types of innovation outcomes: process innovations and brand new product innovations. The reason is that these two types of innovations are sufficiently different to allow checking if the technological learning and foreign demand effects affect different innovation types. Indeed while process innovations are a quite common type of innovation, available also to firms with little technological capabilities, brand new product innovations are instead introduced only by leading firms able to actually shift the technological frontier. If the two effects influence different types of innovation outcomes, this might explain why in the existing literature there is no unanimity on which innovation outputs are more affected by exporting activity: according to the specific destinations of exports (and to the main effect at stake), different types of innovations will be affected. 2.1.1 Process innovations Process innovations increase the efficiency of the productive processes, that is, they decrease the cost of inputs, given a certain quantity of output, typically leading to increase in productivity levels. Firms might be induced to introduce process innovation both by the foreign demand and by the technological learning effect. The foreign demand effect might foster process innovation: in line with Schmookler tradition (1954), the incentives to introduce process innovations should increase with the quantity of output produced, since the efficiency gains on each unit produced will be multiplied by a larger number of units. Scherer (1991) and Cohen and Klepper (1996) provide theoretical and empirical evidence that the increase in the number of units sold induces firms to dedicate more research efforts toward process innovations. Desmet and Parente (2010) develop a theoretical model to show that an increase of international sales for an exporting firm will mainly foster process innovations. Firms exporting to markets with a growing demand should hence have a strong incentive to introduce process innovations and increase the efficiency of their expanding production. It is hence possible to spell out Hypothesis 1a: Hypothesis 1a: The foreign demand effect of export activity increases the incentives for firms to introduce process innovations. At the same time process innovations might also be induced by the technological learning effect, i.e. firms in very advanced markets might benefit from spillovers that eventually lead them to introduce process innovations. Indeed Damijan et al. (2010) find that Slovenian firms exporting to advanced European markets tend to introduce more process innovations. Interacting with foreign customers, which require higher-quality standards, might induce exporters to upgrade their productive processes and introduce process innovations. Also the need to comply with specific regulations in advanced markets concerning production processes might induce exporters to introduce process innovations. Accordingly it is possible that also the technological learning effect might induce process innovations, as stated in Hypothesis 1b: Hypothesis 1b: The technological learning effect of export activity increases the incentives for firms to introduce process innovations. 2.1.2 Brand new product innovations The introduction of brand new product innovations able to shift the technological frontier allows firms to earn a temporary monopolistic profit on the products sold. However, it also entails very high research costs and long development processes, which not all firms are able to undertake. Both the technological learning effect and the foreign demand effect might increase the incentives for exporters to introduce brand new product innovations. The technological learning effect allows firms to decrease research costs through knowledge spillovers stemming from foreign users or foreign competitors. Since research costs are of crucial importance for brand new innovative strategies, the technological learning effect, which decreases such costs, is likely to have a positive impact on this innovative strategy. The literature that focuses on the role of users in the innovative process (von Hippel, 1986, 2005; Malerba et al., 2007) shows that interactions with users that can increase the firms’ competences typically lead to brand new product innovations. Moreover, a growing literature has found that interactions with international customers have a high probability to increase the novelty of a firm’s innovation output and eventually lead to truly new product innovations (Laursen, 2011; Fitjar and Rodriguez-Pose, 2012; Harirchi and Chaminade, 2014). Accordingly Hypothesis 2a is as follows: Hypothesis 2a: The technological learning effect of export activity increases the incentives for firms to introduce brand new product innovations. At the same time, also the foreign demand effect might increase the incentives for firms to introduce brand new product innovations, since this type of innovations allows for temporary monopolistic rents which are especially profitable when a market is growing. Firms active in foreign expanding markets might have much larger incentives to invest resources and introduce truly innovative products, since the costs associated with the development of the new products will be more than compensated by the markups typically associated to these products. Indeed in most cases truly new products are the outcome of the investments in R&D activities and, as shown by Hall et al. (1999), the elasticity of R&D expenditures to sales is quite high. Garcia-Quevedo et al. (2017) also find evidence that demand dynamics strongly influence R&D investments. Moreover Piva and Vivarelli (2007) show that the positive effect of sales on R&D expenditures is especially strong for exporting firms. Based on these considerations, it is possible to expect also a positive effect of foreign demand on brand new product innovation and hence Hypothesis 2b goes as follows: Hypothesis 2b: The foreign demand effect of export activity increases the incentives for firms to introduce brand new product innovations. 3. The empirical strategy 3.1. A simple model To test the hypotheses about the effect of exports on firms’ innovative outputs, I introduce the following model: the probability to introduce any of the different innovation strategies y of firm i is a linear function of the firm’s past exporting activity.   yis=c+β EXPOi{t−1}+δXi+μj+νr+ρc+ui, (1) where s= process innovation, brand new product innovation yi indicates whether firm i implemented an innovation s. Therefore I introduce an equation for each of the two possible innovation outputs. EXPOi{t−1} is a dummy equal to 1 if a firm exported in the previous period t−1 and equal to zero if the firm did not export in time t−1. Xi includes a set of firm-level control variables, while μj, νr, and ρc control, respectively, for sector, regional, and country effects. The idiosyncratic error term is denoted by uit. While the literature so far has only focused on the size and sign of the β coefficient of being an exporter, here the hypothesis is that for each firm i the marginal effect of exporting on innovation activities is a linear function of the technological learning effect L and the foreign demand effect D of exporting, which on their turn depend on the specific export destinations of each firm. Accordingly it is possible to write:   βi=γ1Li{t−1}+γ2Di{t−1}. (2) Where for each firm the coefficient of the export dummy depends on the specific impact of the two identified effects. Substituting (2) into (1) I obtain the following specification:   yis=c+γ1(Li{t−1}* EXPOi{t−1})+γ2(Di{t−1}* EXPOi{t−1})+δXi+μj+νr+ρc+ui. (3) To ease the notation, the interaction terms will be simply denoted as TL and TD, suppressing the time indicators, as follows:   yis=c+γ1TL+γ2TD+δ Xi+μj+νr+ρc+ui. (4) Hence the two variables of interest are TL=0 if a firm did not export in t−1 and TL=  Li{t−1} if the firm exported in t−1. Also TD will be equal to zero if a firm did not export in time t−1 and TD=  Di{t−1} if the firm exported in the previous time period. According to the hypotheses spelled out in Section 2.1, the two coefficients γ1 and γ2 are likely to differ according to the type of innovation output considered. 3.2. Data The data used are the EU-EFIGE/Bruegel-UniCredit data set, a unique firm-level database collected within the EFIGE project (European Firms in a Global Economy), coordinated by Bruegel, which includes detailed firm-level information about the destinations of exports and the innovation performances of representative samples of manufacturing firms (with a lower threshold of 10 employees) in seven European countries in the period 2007–2009: the survey includes around 3000 firms for France, Germany, Italy, and Spain, and 2000 for the UK.1 The survey followed a proper stratification strategy of the sample to ensure representativeness of the collected data for each country, on the basis of industries, regions, and size classes; however in the stratification strategy, large firms have been slightly oversampled (Altomonte and Aquilante, 2012). The EFIGE data set is an extremely rich data set with harmonized information across the different countries about firms’ structural information (size; group affiliation; ownership structure), as well as information about the labor force, the innovative investments, and the internationalization strategies. The great advantage of the EFIGE data set is that it has detailed information on both the innovation strategies adopted by firms and on the specific destinations of their exports. It is hence possible to know what type of innovation strategies were implemented by each firm and, for the firms who exported, the main markets of destination of their exports. The main limit of the data set is that it is a cross-section, which makes it more difficult to address causality issues: however even if there is only one observation per firm, the questions concerning exporting activity cover also past years, so they allow to introduce suitable lags in the empirical analysis. 3.2.1 Dependent variables To identify the possible types of innovation strategies, two dependent variables will be used: each of them indicating a specific innovation type, as outlined in Section 2.1. Process innovation is proxied by a dummy that is equal to 1 if a firm introduced a process innovation in the period 2007–2009 and 0 otherwise. Brand new product innovation is proxied by a dummy variable that is equal to 1 if in the period 2007–2009, a firm introduced a product innovation that is new to the market and it also applied for a patent. This specific combination assures that the firm which introduces the product innovation is also able to introduce truly novel technologies. While this definition is very restrictive and cannot be compared with the usual definition of product innovation, it is useful to identify product innovations introduced by leading firms actually able to shift the technological frontier, as proxied by the application for a patent (which is by definition associated with a technological novelty). Accordingly, only a restricted number of highly competitive and technologically advanced firms will be able to introduce this type of product innovations, due to the high research costs. This strategy seems especially appropriate to identify brand new product innovations when domestic and exporting firms are compared, since the simple notion of “product new to the market” does not clearly indicate which markets firms are referring to.2 3.2.2 Independent variables The EFIGE survey asks firms if they export. To the exporters it also asks to indicate their three main export destinations in 2008 and to specify if they were already active in those countries in 2003. To decrease as much as possible the problems of simultaneity, I only consider export destinations in which the firm was already active in the 5 years from 2003 up to 2007: in this way I introduce a relevant time lag between exporting activity (2003–2007) and the period considered for the introduction of innovations (2007–2009). This also allows me to identify long-term export destinations, which have a high degree of persistence for the firms, since usually the positive effect of export on innovation is found for persistent exporters (Andersson and Lööf, 2009) (Figure 1). Figure 1. View largeDownload slide The time lag between exporting and innovation. Figure 1. View largeDownload slide The time lag between exporting and innovation. Combining this information with the sectoral affiliation of each firm, I can build the two main indices that measure the technological learning effect and the foreign demand effect according to the specific export destinations of each firm: to each exporting firm, I associate the level of foreign market growth and technological advancement of the countries in which the firm exports and specifically in the sector in which the firm is active (see Figure 2). The main assumption behind this approach is that the possibility to learn through exporting activity (technological learning) and to benefit from the increase of the foreign markets does not depend on the features and dynamics of the overall economy of the countries of destination, but only by the characteristics of the same sector in which the firm is active. As a matter of example this implies that for a German firm active in the electronic industry which exports to the United States, the technological learning and foreign demand effects will depend on the characteristics and the dynamics of the electronic industry in the United States and not on the overall dynamics of the US economy. This approach seems legitimate, since firms, especially small- and medium-sized firms, are often working in a specific market niche; therefore, the features of the economy at the aggregate level may have little or no influence at all on their economic decisions. Figure 2. View largeDownload slide The construction of the technological learning and foreign demand indices. Figure 2. View largeDownload slide The construction of the technological learning and foreign demand indices. However the advantages of this sectoral strategy increase only up to a certain threshold: if the sectoral disaggregation is too thin, there is the risk to miss important inter-sectoral effects. Indeed a firm necessarily sells outputs to other firms that perform slightly different economic activities along the vertical supply chains. Restricting the sectoral focus too much may result in losing these interactions occurring with foreign buyers. To take into account both these effects, I use the two-digit (ISIC. Rev. 3) sectoral aggregation: this classification distinguishes between manufacturing firms that do completely different economic activities (such as the pharmaceutical industry and the automotive sector), but at the same time, it aggregates across similar economic activities (such as the production of basic chemicals and the production of plastic products). Technological learning effect index. The technological learning effect can be proxied by the level of technological sophistication of the country in which a firm is exporting, in the specific two-digit sector in which the firm is active. The higher is the level of technological advancement of the markets/sectors of destination, the higher will be the possibility for the exporting firm to acquire new knowledge and new useful routines to be eventually incorporated in new products or new processes. The share of Research and Development (R&D) expenditures over the total value added of a sector can be considered a reliable proxy of the general level of technological advancement of a sector in a country.3 For each national sector indicated as a long-term export destination by the firms in the EFIGE sample, I calculate the level of business R&D intensity over value added using data from the OECD-STAN, integrating it with data from the UNIDO and the World Bank: for each country-sector, I use the average value of R&D intensity for the years between 2003–2007. In this way the technological intensity of export destinations corresponds to the period to which firms refer when they indicate their export markets. As shown in Figure 2, the technological learning effect L hence corresponds to the average level of R&D intensity in sector j among the three main countries of destinations d indicated by firm i, conditional on the fact that the firm was already exporting in those markets in 2003.   Li{t−1}=∑d=13(avg R&D  03−07jd)/3, (5) where d = 1, …, 3. Foreign demand effect index. Contrary to the technological learning effect in the literature, there are already some attempts to measure the effect of foreign demand on the innovative performances of exporting firms: Bratti and Felice (2012) use the level of gross domestic product (GDP) per capita of export destinations weighted by the relative distance. Accetturo et al. (2013) instead use import growth as a proxy of the growth of demand. Here I follow the second strategy and build an index that is equal to the average rate of growth of imports in the period 2003–2007 in each specific two-digit sector in the three export destinations of each exporting firm. The data come from COMTRADE and are calculated in US dollars. Since I am only considering long-term export destinations in which firms were already active in 2003 and were still active in 2007, I can be sure that from 2003 to 2007, these firms have been continuously exporting to that specific country c which experienced that rate of growth of imports in sector j. As also shown in Figure 2, the foreign demand effect is:   Di{t−1}=∑d=13(impjd2007−impjd2003)/3, (6) where d = 1, …,3. And imp is the log of imports from country d and sector j in time t. This measure is able to capture the extent to which the markets in which the firm was exporting have grown in the period before the decision to adopt any of the innovative output identified above. Also here I adopt a lag specification to restrict the focus on the sectoral import growth for the period 2003–2007 of the markets in which firms were already operating in 2003. National R&D and market growth. The same level of R&D intensity in a foreign market might have different effects for a firm in a highly advanced country as compared to a firm in a less advanced one. Exporting to the United States might substantially increase knowledge spillovers for an Italian firm active in a low competitive domestic market, but not necessarily for a German firm operating in a very competitive national market. For this reason, I also include the level of R&D intensity in the national two-digit sector of affiliation of each firm (OECD-STANBERD data). For the same reason, I also include as a further control a proxy for the growth of the internal markets, measured by the growth of value added in the national two-digit sector of affiliation of each firm (OECD-STAN data). Structural variables. The model includes controls for structural characteristics of the firms such as employment size, age of the firm, group affiliation, and the type of ownership control (family versus non-family business). Innovative capacity. The innovative capacity of the firm is measured by the share of R&D expenses over turnover in 2007–2009. The level of human capital is controlled by a dummy equal to 1 if the firm has a higher share of graduate employees with respect to the national average (Altomonte, Aquilante, 2012) and a variable that measures the share of employees with a fixed-term contract, assuming that fixed-term contracts are associated to a lower quality of the employees. Internationalization activity. The model also controls if the firm runs part of its production activity in another country through direct investments or through contracts and arms’ length agreements and whether it has foreign affiliates. I also introduce a set of dummies indicating the geographic localization of the main competitors and the level of vertical integration of firms, since the possibility to learn from foreign customers will change a lot if the firm sells directly to final consumers or to other firms. The model also controls for country effects, two-digit sector effects, and regional effects at the nuts-2 level. 3.2.3 Descriptive statistics Table 1 presents the aggregate descriptive statistics of the main variables in the whole sample that includes French, German, Italian, Spanish, and UK firms. As expected the most common innovation output is process innovation, which is adopted by more than 40% of firms, while brand new product innovations, which entail high research costs, are implemented only by 11% of firms. In total, 7% of firms implement both innovative strategies. Firms with up to 50 employees represent the large majority of the overall sample (75%). Only a small share of firms belongs to national or foreign groups, respectively, 13% and 8%. The variables related with internationalization strategies show that only a limited fraction of firms (5%) has foreign direct investments abroad. The majority of firms considers domestic competitors as the most important, followed by European competitors (43%) and competitors in other geographical areas (27%). Table 1. Descriptive statistics Variable  Mean  Standard deviation  Minimum  Maximum  Dependent variables           Process innovation  0.438  0.496  0  1   Brand new product innovation  0.114  0.317  0  1   Process innovation and brand new product innovation  0.070  0.256  0  1  Independent variables           Export activities            Export in 2003  0.411  0.492  0  1    Foreign demand effect  0.058  0.077  −0.164  0.610    Technological learning effect  0.019  0.049  0  0.735   Alternative indexes            Foreign demand effect (2005–2007)  0.063  0.089  −0.844  0.609    Technological learning effect (2005–2007)  0.019  0.049  0  0.732    Foreign demand effect (largest market)  0.057  0.079  −0.307  0.666    Technological learning effect (highest R&D intensity)  0.026  0.066  0  0.735    Foreign demand effect (weighted by export shares)  0.033  0.053  −0.073  0.602    Technological learning effect (weighted by export shares)  0.010  0.030  0  0.735  Structural variables           Labour productivity            TFP            Employment (≤25)  0.470  0.499  0  1    Employment (>25 and ≤50)  0.283  0.450  0  1    Employment (>50 and ≤100)  0.111  0.314  0  1    Employment (>100 and ≤150)  0.041  0.197  0  1    Employment (>150 and ≤250)  0.033  0.180  0  1    Employment (>250 and <500)  0.037  0.189  0  1    Employment (≥500)  0.026  0.160  0  1    Share of fixed-term contracts  26.773  38.902  0  100    Firm age (<6 years)  0.338  0.473  0  1    Firm age (6–20 years)  0.338  0.473  0  1    Firm age (>20 years)  0.594  0.491  0  1    National group  0.137  0.344  0  1    Foreign group  0.081  0.273  0  1    Family member as CEO  0.631  0.482  0  1   Innovative capacities           Share of R&D  0.037  0.076  0  1   Skilled labor force  0.281  0.449  0  1   ICT access  0.914  0.280  0  1  Internationalization variables           Foreign direct investments  0.049  0.216  0  1   Arms’ length foreign production  0.040  0.197  0  1   Domestic affiliates  0.133  0.339  0  1   Foreign affiliates  0.075  0.263  0  1   Domestic competitors  0.855  0.352  0  1   Competitors in EU  0.431  0.495  0  1   Competitors in the United States  0.126  0.332  0  1   Competitors other geo areas  0.273  0.445  0  1   Vertical integration            Sales to order share (1–30%)  0.120  0.325  0  1    Sales to order share (30%–70%)  0.088  0.284  0  1    Sales to order share (>70%)  0.662  0.473  0  1   Domestic effects           Growth of domestic sector  0.116  0.126  −0.646  0.673   R&D intensity domestic sector (%)  3.392  5.996  0.106  51.061  National composition  Number of firms  (%)       France  2723  21.1       Germany  2827  21.91       Italy  2950  22.86       Spain  2728  21.14       UK  1677  12.99      Total number of observations  12,905  100      Variable  Mean  Standard deviation  Minimum  Maximum  Dependent variables           Process innovation  0.438  0.496  0  1   Brand new product innovation  0.114  0.317  0  1   Process innovation and brand new product innovation  0.070  0.256  0  1  Independent variables           Export activities            Export in 2003  0.411  0.492  0  1    Foreign demand effect  0.058  0.077  −0.164  0.610    Technological learning effect  0.019  0.049  0  0.735   Alternative indexes            Foreign demand effect (2005–2007)  0.063  0.089  −0.844  0.609    Technological learning effect (2005–2007)  0.019  0.049  0  0.732    Foreign demand effect (largest market)  0.057  0.079  −0.307  0.666    Technological learning effect (highest R&D intensity)  0.026  0.066  0  0.735    Foreign demand effect (weighted by export shares)  0.033  0.053  −0.073  0.602    Technological learning effect (weighted by export shares)  0.010  0.030  0  0.735  Structural variables           Labour productivity            TFP            Employment (≤25)  0.470  0.499  0  1    Employment (>25 and ≤50)  0.283  0.450  0  1    Employment (>50 and ≤100)  0.111  0.314  0  1    Employment (>100 and ≤150)  0.041  0.197  0  1    Employment (>150 and ≤250)  0.033  0.180  0  1    Employment (>250 and <500)  0.037  0.189  0  1    Employment (≥500)  0.026  0.160  0  1    Share of fixed-term contracts  26.773  38.902  0  100    Firm age (<6 years)  0.338  0.473  0  1    Firm age (6–20 years)  0.338  0.473  0  1    Firm age (>20 years)  0.594  0.491  0  1    National group  0.137  0.344  0  1    Foreign group  0.081  0.273  0  1    Family member as CEO  0.631  0.482  0  1   Innovative capacities           Share of R&D  0.037  0.076  0  1   Skilled labor force  0.281  0.449  0  1   ICT access  0.914  0.280  0  1  Internationalization variables           Foreign direct investments  0.049  0.216  0  1   Arms’ length foreign production  0.040  0.197  0  1   Domestic affiliates  0.133  0.339  0  1   Foreign affiliates  0.075  0.263  0  1   Domestic competitors  0.855  0.352  0  1   Competitors in EU  0.431  0.495  0  1   Competitors in the United States  0.126  0.332  0  1   Competitors other geo areas  0.273  0.445  0  1   Vertical integration            Sales to order share (1–30%)  0.120  0.325  0  1    Sales to order share (30%–70%)  0.088  0.284  0  1    Sales to order share (>70%)  0.662  0.473  0  1   Domestic effects           Growth of domestic sector  0.116  0.126  −0.646  0.673   R&D intensity domestic sector (%)  3.392  5.996  0.106  51.061  National composition  Number of firms  (%)       France  2723  21.1       Germany  2827  21.91       Italy  2950  22.86       Spain  2728  21.14       UK  1677  12.99      Total number of observations  12,905  100      About 40% of firms were already exporting in 2003: for each of them, it was possible to calculate their respective index of technological learning and foreign demand effects—as proxied by the intensity of R&D expenditures and import growth of the sectors and markets in which they were exporting in 2003. 4. Results As in some of the existing literature on export and innovation (Lileeva and Trefler, 2007; Bustos, 2011; Bratti and Felice, 2012), I estimate the equations presented in Section 3.1 with a linear probability model (LPM) instead of nonlinear estimators such as probit or logit. The main reason is that LPM is especially recommended when IVs strategies are implemented (Angrist, 2001), and since in this section I will also introduce IV estimates obtained with LPM, I use this estimator throughout the article to ease the comparisons of the results across different specifications. An additional advantage of LPM with respect to nonlinear estimators is that it yields unbiased and consistent estimates with no assumptions on the distribution of the error term and is therefore suggested if one is only interested in measuring average treatment effects. Of course this choice would not be appropriate were I interested in the specific distribution of outcomes.4 Before estimating the impact of the technological learning and demand effect on firms’ different innovative strategies, I start with the ordinary least squares (OLS) estimation of the LPMs that explain the implementation of the two possible innovation outputs, using the fact of being an exporter in 2003 as the main independent variable. This is a useful benchmark with respect to the previous literature. The results in Table 2 show that indeed exporting activity has always a positive effect on innovation and that the size of such effect is broadly equal for process and brand new product innovations. The results also show that the inclusion of further controls in the model decreases by roughly two-third the coefficient of export activity in both types of innovation. Table 2. The effect of exports on innovation strategies   (1)  (2)  (3)  (4)  (5)  (6)  Process innovation  Brand new product innovation  Export  0.096***  0.052***  0.037***  0.093***  0.054***  0.034***  (0.009)  (0.010)  (0.010)  (0.006)  (0.006)  (0.006)  Share of R&D    0.908***  0.884***    0.742***  0.683***    (0.069)  (0.069)    (0.056)  (0.056)  Skilled labor force    0.048***  0.043***    0.035***  0.028***    (0.010)  (0.010)    (0.006)  (0.006)  National group    0.006  0.002    0.011  0.001    (0.014)  (0.014)    (0.009)  (0.009)  Foreign group    0.006  0.002    0.029**  0.017    (0.018)  (0.018)    (0.014)  (0.014)  Employment (>25 and ≤50)    0.076***  0.072***    0.027***  0.022***    (0.010)  (0.010)    (0.006)  (0.006)  Employment (>50 and ≤100)    0.125***  0.117***    0.074***  0.056***    (0.015)  (0.015)    (0.010)  (0.010)  Employment (>100 and ≤150)    0.156***  0.143***    0.134***  0.104***    (0.023)  (0.023)    (0.018)  (0.018)  Employment (>150 and ≤250)    0.215***  0.202***    0.175***  0.136***    (0.025)  (0.025)    (0.022)  (0.021)  Employment (>250 and <500)    0.177***  0.163***    0.153***  0.100***    (0.025)  (0.025)    (0.020)  (0.020)  Employment (≥500)    0.209***  0.186***    0.251***  0.156***    (0.030)  (0.031)    (0.026)  (0.026)  Firm age (6–20 years)    −0.030  −0.027    −0.010  −0.010    (0.018)  (0.018)    (0.011)  (0.011)  Firm age (>20 years)    −0.041**  −0.039**    −0.016  −0.020*    (0.018)  (0.018)    (0.011)  (0.011)  Family member as CEO    0.023**  0.022**    0.003  0.004    (0.010)  (0.010)    (0.006)  (0.006)  Growth of domestic sector    0.033  0.052    0.040  0.059    (0.071)  (0.070)    (0.048)  (0.048)  R&D intensity domestic sector    0.004**  0.004**    −0.000  −0.000    (0.002)  (0.002)    (0.001)  (0.001)  Share of fixed-term contracts    −0.000  −0.000    −0.000  −0.000    (0.000)  (0.000)    (0.000)  (0.000)  ICT access    0.034**  0.031**    0.021**  0.016**    (0.015)  (0.015)    (0.008)  (0.008)  Domestic affiliates      0.019      0.022**      (0.013)      (0.010)  Foreign affiliates      0.006      0.119***      (0.020)      (0.017)  Foreign direct investments      0.020      0.060***      (0.023)      (0.021)  Arms’ length foreign production      −0.019      0.070***      (0.022)      (0.018)  Sales to order share (1–30%)      −0.013      0.014      (0.017)      (0.011)  Sales to order share (30%–70%)      −0.018      −0.000      (0.019)      (0.012)  Sales to order share (>70%)      0.031**      −0.020**      (0.014)      (0.008)  Domestic competitors      0.031**      −0.032***      (0.013)      (0.009)  Competitors in the United States      0.045***      0.062***      (0.015)      (0.011)  Competitors in EU      0.056***      0.011*      (0.010)      (0.007)  Competitors other geo areas      0.024**      0.001      (0.011)      (0.007)  Constant  0.071  −0.004  −0.030  −0.054  −0.173*  −0.130  (0.168)  (0.205)  (0.214)  (0.037)  (0.096)  (0.096)  Observations  12,905  12,905  12,905  12,905  12,905  12,905  R-squared  0.037  0.073  0.080  0.073  0.138  0.164    (1)  (2)  (3)  (4)  (5)  (6)  Process innovation  Brand new product innovation  Export  0.096***  0.052***  0.037***  0.093***  0.054***  0.034***  (0.009)  (0.010)  (0.010)  (0.006)  (0.006)  (0.006)  Share of R&D    0.908***  0.884***    0.742***  0.683***    (0.069)  (0.069)    (0.056)  (0.056)  Skilled labor force    0.048***  0.043***    0.035***  0.028***    (0.010)  (0.010)    (0.006)  (0.006)  National group    0.006  0.002    0.011  0.001    (0.014)  (0.014)    (0.009)  (0.009)  Foreign group    0.006  0.002    0.029**  0.017    (0.018)  (0.018)    (0.014)  (0.014)  Employment (>25 and ≤50)    0.076***  0.072***    0.027***  0.022***    (0.010)  (0.010)    (0.006)  (0.006)  Employment (>50 and ≤100)    0.125***  0.117***    0.074***  0.056***    (0.015)  (0.015)    (0.010)  (0.010)  Employment (>100 and ≤150)    0.156***  0.143***    0.134***  0.104***    (0.023)  (0.023)    (0.018)  (0.018)  Employment (>150 and ≤250)    0.215***  0.202***    0.175***  0.136***    (0.025)  (0.025)    (0.022)  (0.021)  Employment (>250 and <500)    0.177***  0.163***    0.153***  0.100***    (0.025)  (0.025)    (0.020)  (0.020)  Employment (≥500)    0.209***  0.186***    0.251***  0.156***    (0.030)  (0.031)    (0.026)  (0.026)  Firm age (6–20 years)    −0.030  −0.027    −0.010  −0.010    (0.018)  (0.018)    (0.011)  (0.011)  Firm age (>20 years)    −0.041**  −0.039**    −0.016  −0.020*    (0.018)  (0.018)    (0.011)  (0.011)  Family member as CEO    0.023**  0.022**    0.003  0.004    (0.010)  (0.010)    (0.006)  (0.006)  Growth of domestic sector    0.033  0.052    0.040  0.059    (0.071)  (0.070)    (0.048)  (0.048)  R&D intensity domestic sector    0.004**  0.004**    −0.000  −0.000    (0.002)  (0.002)    (0.001)  (0.001)  Share of fixed-term contracts    −0.000  −0.000    −0.000  −0.000    (0.000)  (0.000)    (0.000)  (0.000)  ICT access    0.034**  0.031**    0.021**  0.016**    (0.015)  (0.015)    (0.008)  (0.008)  Domestic affiliates      0.019      0.022**      (0.013)      (0.010)  Foreign affiliates      0.006      0.119***      (0.020)      (0.017)  Foreign direct investments      0.020      0.060***      (0.023)      (0.021)  Arms’ length foreign production      −0.019      0.070***      (0.022)      (0.018)  Sales to order share (1–30%)      −0.013      0.014      (0.017)      (0.011)  Sales to order share (30%–70%)      −0.018      −0.000      (0.019)      (0.012)  Sales to order share (>70%)      0.031**      −0.020**      (0.014)      (0.008)  Domestic competitors      0.031**      −0.032***      (0.013)      (0.009)  Competitors in the United States      0.045***      0.062***      (0.015)      (0.011)  Competitors in EU      0.056***      0.011*      (0.010)      (0.007)  Competitors other geo areas      0.024**      0.001      (0.011)      (0.007)  Constant  0.071  −0.004  −0.030  −0.054  −0.173*  −0.130  (0.168)  (0.205)  (0.214)  (0.037)  (0.096)  (0.096)  Observations  12,905  12,905  12,905  12,905  12,905  12,905  R-squared  0.037  0.073  0.080  0.073  0.138  0.164  Note: All models are estimated with OLS estimator. All models include country, sector, and region fixed effects. The reference category for firms’ size is less than 25 employees. The reference category for firms’ age is less than 6 years. The reference category for sales to order share is zero. Standard errors in parentheses are clustered at the firm level *** P < 0.01, ** P < 0.05, * P < 0.1. Once acknowledged that also in this sample, exporting activity is positively correlated with innovation, and that this effect is equal for the two innovative strategies identified, I investigate more specifically whether the technological learning and demand effects have a differentiated impact on the two innovative outputs. Indeed it might be that the positive coefficient found for the export dummy in Table 2 is sometimes due to the technological learning effect and sometimes to the foreign demand effect, according to the specific type of innovation considered. In this way I will be able to test the hypotheses spelled out in Section 2.1. In Columns (1)–(4) of Table 3, I substitute the export dummy with the two effects in the process innovation specification. In Column (1) I present the OLS results introducing only the structural controls such as size, belonging to a group and those related with innovation capacity (R&D; skilled labor force; adoption of Information and Communication Technologies – ICT): I find that only the demand index has a positive and significant effect, while the technological learning index is not significantly different from zero. The results do not change when in Column (2) I include the additional controls for internationalization activity and vertical integration: the coefficients decrease slightly, but only the demand effect is positive and significant. As already stated in Section 3.2, the procedure used to build the two indices assumes that firms are affected by the level of foreign demand and foreign technological spillovers of the sectors in which they are active. This might not be an appropriate strategy for large multinational operating in very different market segments: in that case considering only one of the various economic activities in which these firms are active would be highly misleading. To overcome this problem in Column (3), I restrict the sample to the firms that have at least 60% of their sales in one specific economic activity, and in Column (4), instead, I only consider firms with less than 500 employees. The OLS results however do not change: the only positive and significant coefficient is found for the foreign demand effect. Overall the results confirm Hypothesis 1a and reject Hypothesis 1b: the effect of exporting activities on process innovation is only due to the foreign demand effect, while the technological learning effect has no role. Table 3. The effect of technological learning and foreign demand, baseline results   (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  Process innovation   Brand new product innovation   All firms  All firms  No multiproduct firms  No large firms  All firms  All firms  No multiproduct firms  No large firms  Technological learning effect  0.141  0.106  0.123  0.111  0.358***  0.294***  0.267***  0.292***  (0.114)  (0.114)  (0.118)  (0.118)  (0.096)  (0.095)  (0.096)  (0.097)  Demand effect  0.265***  0.181***  0.154**  0.164**  0.282***  0.176***  0.184***  0.167***  (0.067)  (0.068)  (0.070)  (0.069)  (0.046)  (0.047)  (0.048)  (0.046)  Innovative capacity controls  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Internationalization controls  No  Yes  Yes  Yes  No  Yes  Yes  Yes  Observations  12,905  12,905  12,195  12,564  12,905  12,905  12,195  12,564  R-squared  0.073  0.080  0.081  0.080  0.140  0.165  0.166  0.150    (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  Process innovation   Brand new product innovation   All firms  All firms  No multiproduct firms  No large firms  All firms  All firms  No multiproduct firms  No large firms  Technological learning effect  0.141  0.106  0.123  0.111  0.358***  0.294***  0.267***  0.292***  (0.114)  (0.114)  (0.118)  (0.118)  (0.096)  (0.095)  (0.096)  (0.097)  Demand effect  0.265***  0.181***  0.154**  0.164**  0.282***  0.176***  0.184***  0.167***  (0.067)  (0.068)  (0.070)  (0.069)  (0.046)  (0.047)  (0.048)  (0.046)  Innovative capacity controls  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Internationalization controls  No  Yes  Yes  Yes  No  Yes  Yes  Yes  Observations  12,905  12,905  12,195  12,564  12,905  12,905  12,195  12,564  R-squared  0.073  0.080  0.081  0.080  0.140  0.165  0.166  0.150  Note: The LPMs in the table are estimated with OLS estimator and include country, sector, and region fixed effects. The technological learning effect is calculated as the 2003–2007 average level of sectoral R&D of the three export destinations. The demand effect is calculated as the 2003–2007 average sectoral import growth of the three export destinations. In Columns (3) and (7), only firms with at least 60% of their sales in a specific business are included, and in Columns (4) and (8), only firms with less than 500 employees are included. All specifications include the controls displayed in Table 2. Columns (1) and (5) do not include internationalization controls. Standard errors in parentheses are clustered at the firm level *** P < 0.01, ** P < 0.05, * P < 0.1. In Columns (5)–(8) of Table 3, I test the role of the technological learning and demand effect in the brand new innovation specification. In Column (1) I only include structural and innovation-capacity controls, in Column (2) I include all controls, and in Column (3) I select only firms with the majority of sales in one single type of economic activity. Also in this case in Column (4) I exclude firms with more than 500 employees: indeed it could be the case that brand new product innovations are introduced especially by very large firms; therefore, the results might be driven by this specific subset of firms. However across the different specifications, the results are very stable. In this case both effects are positive and significantly different from zero, even if the coefficient of the technological learning effect is larger than the foreign demand effect. The results confirm both Hypothesis 2a and Hypothesis 2b: the effect of exporting activities on brand new product innovation is due to both the foreign demand and the technological learning effect. 4.1. Identification 4.1.1. Selection into exporting: the role of productivity An important threat to identification is represented by the respective levels of productivity of firms. As stressed by the literature that highlights the role of self-selection into exports, more productive firms are much more likely to export, as they can bear the costs associated to exporting activity (Bernard and Jensen, 1999, Bernard and Wagner, 1997). Productivity is likely to be positively correlated both with export and innovation activity, representing a possible confounding factor in the specification of equation (4). Moreover productivity is likely to influence also the specific destination of exports: as shown by previous empirical research (Serti and Tomasi, 2012; Crinò and Epifani, 2012), more productive firms might be able to export to markets with higher levels of R&D intensity (or higher demand growth), leading to a positive correlation with the index of technological learning (or foreign demand effect) and with innovative activities. Based on these considerations I introduce, respectively, labor productivity and total factor productivity (TFP) as further controls in the estimation of equation (4).5 Both measures are provided by Bruegel together with the EFIGE data set: more specifically TFP is computed following the methodology suggested by Levinsohn and Petrin (2003) to estimate output elasticities. To decrease as much as possible problems related to reverse causality in both cases, I use the average productivity for the years prior to the introduction of innovations, i.e. before 2007. More specifically I use the average for the years 2001–2007.6 I introduce the two measures of productivity in Tables 4 and 5 to verify their impact on the two measures of innovation. In Table 4 I check how their inclusion affects the coefficient of the simple export dummy, while in Table 5 I check whether their inclusion influences the coefficients of the technological learning and foreign demand effects. The results in Table 4 show two interesting patterns: while labor productivity is only significant in the process innovation equation, TFP is instead significant only in the brand new product innovation specification. This confirms the fact that the two types of innovation chosen are indeed very different: while process innovations are innovative strategies aimed at increasing the efficiency of the productive processes, which might be associated with investments in machinery that increase capital intensity and labor productivity, brand new product innovations are instead a much more rare innovative strategy that can only be implemented by firms closer to the technological frontier and which display also higher levels of TFP. Table 4. The role of productivity   (1)  (2)  (3)  (4)  Process innovation   Brand new product innovation   All firms  All firms  All firms  All firms  Export  0.035***  0.036***  0.036***  0.036***  (0.010)  (0.010)  (0.006)  (0.006)  Labor productivity  0.036***    0.006    (0.011)    (0.008)    TFP    0.017    0.018**    (0.014)    (0.009)  All controls  Yes  Yes  Yes  Yes  Observations  12,905  12,905  12,905  12,905  R-squared  0.081  0.080  0.164  0.164    (1)  (2)  (3)  (4)  Process innovation   Brand new product innovation   All firms  All firms  All firms  All firms  Export  0.035***  0.036***  0.036***  0.036***  (0.010)  (0.010)  (0.006)  (0.006)  Labor productivity  0.036***    0.006    (0.011)    (0.008)    TFP    0.017    0.018**    (0.014)    (0.009)  All controls  Yes  Yes  Yes  Yes  Observations  12,905  12,905  12,905  12,905  R-squared  0.081  0.080  0.164  0.164  Note: The LPMs in the table are estimated with OLS estimator and include country, sector, and region fixed effects. All specifications include the controls displayed in Table 2. Standard errors in parentheses are clustered at the firm level *** P < 0.01, ** P < 0.05, * P < 0.1. Table 5. The effect of technological learning and foreign demand, controlling for productivity   (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  Process innovation   Brand new product innovation   All firms  All firms  No multiproduct firms  No large firms  All firms  All firms  No multiproduct firms  No large firms  Technological learning effect  0.140  0.106  0.123  0.112  0.353***  0.290***  0.264***  0.290***  (0.114)  (0.114)  (0.118)  (0.118)  (0.096)  (0.095)  (0.096)  (0.097)  Demand effect  0.252***  0.172**  0.145**  0.154**  0.279***  0.182***  0.182***  0.166***  (0.067)  (0.068)  (0.070)  (0.069)  (0.046)  (0.047)  (0.048)  (0.047)  Labor productivity  0.039***  0.037***  0.033***  0.037***  –  –  –  –  (0.011)  (0.011)  (0.012)  (0.012)  –  –  –  –  TFP  –  –  –  –  0.026***  0.018**  0.020**  0.013  –  –  –  –  (0.009)  (0.009)  (0.009)  (0.009)  Innovative capacity controls  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Internationalization controls  No  Yes  Yes  Yes  No  Yes  Yes  Yes  Observations  12,905  12,905  12,195  12,564  12,905  12,905  12,195  12,564  R-squared  0.074  0.081  0.082  0.081  0.141  0.164  0.166  0.150    (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  Process innovation   Brand new product innovation   All firms  All firms  No multiproduct firms  No large firms  All firms  All firms  No multiproduct firms  No large firms  Technological learning effect  0.140  0.106  0.123  0.112  0.353***  0.290***  0.264***  0.290***  (0.114)  (0.114)  (0.118)  (0.118)  (0.096)  (0.095)  (0.096)  (0.097)  Demand effect  0.252***  0.172**  0.145**  0.154**  0.279***  0.182***  0.182***  0.166***  (0.067)  (0.068)  (0.070)  (0.069)  (0.046)  (0.047)  (0.048)  (0.047)  Labor productivity  0.039***  0.037***  0.033***  0.037***  –  –  –  –  (0.011)  (0.011)  (0.012)  (0.012)  –  –  –  –  TFP  –  –  –  –  0.026***  0.018**  0.020**  0.013  –  –  –  –  (0.009)  (0.009)  (0.009)  (0.009)  Innovative capacity controls  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Internationalization controls  No  Yes  Yes  Yes  No  Yes  Yes  Yes  Observations  12,905  12,905  12,195  12,564  12,905  12,905  12,195  12,564  R-squared  0.074  0.081  0.082  0.081  0.141  0.164  0.166  0.150  Note: The LPMs in the table are estimated with OLS estimator and include country, sector, and region fixed effects. The technological learning effect is calculated as the average level of sectoral R&D of the three export destinations. The demand effect is calculated as the average sectoral import growth of the three export destinations. In Columns (3) and (7), only firms with at least 60% of their sales in a specific business are included, and in Columns (4) and (8), only firms with less than 500 employees are included. All specifications include the controls displayed in Table 2. Columns (1) and (5) do not include internationalization controls. Standard errors in parentheses are clustered at the firm level *** P < 0.01, ** P < 0.05, * P < 0.1. After this preliminary check of the relative impact of the two productivity measures in Table 4, in Table 5, I only include labor productivity in the process innovation equation and only TFP in the brand new product innovation equation. The results of both tables show that the introduction of the productivity measures does not significantly alter the size and significance of the coefficients of the technological learning and foreign demand effects. 4.1.2. Instrumental variable strategy Even after controlling for the levels of productivity, the results obtained with the OLS specification cannot be considered free from endogeneity issues. Di{t−1} and Li{t−1}, i.e. the indices of imports growth and of technological development of the main countries of export destination for each firm, might be endogenous because the choice of a firm to export in a specific country d is not random: firms chose strategically the destination of their exports. The specification chosen, which considers only innovation activities performed between 2007 and 2009 and export destinations in which firms were present before (from 2003 to 2007), allows to diminish to a certain extent the impact of reverse causality bias; however it does not allow to rule it out completely. Past innovative activities can still have an effect on firms’ export destinations, even if there are no clear expectations ex ante about the direction of such effect. Moreover, as also shown by the literature that explores the selection into different types of export markets (Blanes-Cristóbal et al. 2008; Serti and Tomasi, 2012), the fact that a firm exported to a specific country (with a specific demand growth and technological development level) in 2003–2007 might be related to the existence of unobservables, such as managerial ability or past international experience, that I might not be able to control for, since the sample is a cross-section. I try to mitigate these endogeneity problems through an IVs strategy, which builds on similar methodologies already used in the literature (Bratti and Felice, 2012), but which also introduces some novelties. The identification strategy relies on the average propensity of firms in a certain national sector to export toward specific destinations. For each national sector, I create a fictional “representative” exporter, which exports to the most common export destinations among the firms in that sector, and I build two new indexes of technological learning and foreign demand effects that are created using the R&D intensity and import growth of the most common export destinations among the firms in a sector-country. These can be considered as the indexes of a “representative” exporter in the same sector and country of the focal firm. The rationale behind this strategy is that in each country a firm will be more likely to export to the market destinations that are common among the other firms active in the same national sector, so the values of the indexes of technological learning and foreign demand of the “representative” exporter should be correlated with those of the focal firm: if the majority of exports in the German electronics industry is toward, say, the United States, France, and Italy, it is also likely that an exporting firm in the German electronics sector will export to these countries. The exclusion restriction instead is that this average measure of R&D intensity (or import growth) for the whole sector-country will not be correlated with the focal firm’s idiosyncratic innovativeness.7 Taking advantage of OECD trade data (STAN Bilateral Trade in Goods by Industry and End-use), I retrieved for each national sector the aggregate flow of exports to each country and selected the 25 most common destinations in the period 2003–2007. On the basis of these data, I built the two new indexes of technological learning and foreign demand as an average of import growth and of R&D intensity for the 25 most common export market destinations (weighted by their relative importance) of each national sector. A possible threat to identification could arise if in a specific country-sector all exporters were equally very productive and innovative: in this case the technological learning and foreign demand indexes calculated for the “representative” firm might still be correlated with the innovativeness of the focal firm. However, calculating the representative exporter at the (two-digit) sector-country aggregation level should allow for a substantial degree of intra-sectoral heterogeneity among firms. Indeed, as suggested by Dosi, Lechevalier, and Secchi (2010), within a sector, the heterogeneity of firms’ productivity levels is always very high. Also Melitz (2003) stresses that even within the same industry, there are substantial differences in productivity and export performances among firms. Therefore it seems unlikely that the level of R&D intensity and import growth, calculated using the most common export destinations among all the firms of a national sector, can be correlated with the idiosyncratic error term of the focal firm. Since the relationship between this instrument and the actual behavior of firms is likely to be not linear, I introduce some further factors that are supposed to determine heterogeneous responses by firms to the instrument. The first factor is the regional propensity to export: the probability that a firm exports in the same market destinations of the average firm in its own national sector also depends on the general propensity to export of the firm’s region, since this propensity varies quite a lot among regions in the same countries. Another factor that is likely to diminish the ability of the instrument to explain firms’ export choices is the size of firms: very small firms will have in general a lower ability to export, regardless of the sectoral averages, since they face relevant obstacles to access foreign markets, represented by sunk and information costs. On the basis of these preliminary considerations, I build the following instrument:   T^L=∑d=125wjLjdmr.Ljd is the level of technological development proxied by the R&D intensity of the 25 most-common d country-destinations of exports for the sector j in the specific European country in which the firm is active (France, Germany, Italy, Spain, or UK). wj is the share of export to each of the 25 most common destinations of exports over the total exports of national sector j in the period 2003–2007. mr is the share of exporters in each region. Finally to account for firm-size effects, an additional instrument will be added in which T^Li is multiplied by a dummy (0/1) equal to 1 if a firm’s number of employees is equal or lower than 25. The same procedure is used to instrument the foreign demand effect index TD:   T^D=∑d=125wjDjdmr. In this case Djd is the growth of imports between 2003 and 2007 of the 25 most-common c country-destinations of exports for the sector j in the specific country in which the firm is active. Also in this case T^D is multiplied by a dummy (0/1) equal to 1 if a firm employment is equal or lower than 25 employees. In Table 6 I present the results obtained with this IV strategy, using a two-stage least squares (2SLS) estimator. In Columns (1)–(4), I present the process innovation specification in which both the foreign demand and technological learning indices are instrumented by the instruments built with the national sectoral propensities. The results in Columns (1) and (2) in which I use the whole sample and progressively include internationalization controls show that only the foreign demand effect is positive and significant. Moreover the coefficient becomes larger than the one found in the OLS estimates, pointing to a downward bias in those estimates. The results also hold when I only consider firms with a dominant business activity and firms with less than 500 employees in Columns (3) and (4). The first-stage F-statistics of the two instrumented variables, reported in the lower part of Table 6, are always greater than 10, i.e. above the usual threshold identified by the weak instruments literature (Bound et al., 1995, see also Table A1 for first-stage regressions). Moreover the Hansen test on overidentifying restrictions shows that the instruments are exogenous to the error term and correctly excluded from the main regression. Table 6. Instrumental variables   (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  Process innovation   Brand new product innovation   All firms  All firms  No multiproduct firms  No large firms  All firms  All firms  No multiproduct firms  No large firms  Technological learning effect  0.583  0.603  0.617  0.304  1.860***  1.819***  1.825***  1.697***  (0.682)  (0.707)  (0.764)  (0.729)  (0.533)  (0.545)  (0.600)  (0.555)  Demand effect  2.072**  2.151**  2.107**  1.748*  0.739  0.722  0.962  0.328  (0.871)  (0.972)  (0.950)  (0.989)  (0.544)  (0.602)  (0.604)  (0.602)  Labor productivity  0.018  0.019  0.014  0.022  –  –  –  –  (0.015)  (0.015)  (0.015)  (0.015)  –  –  –  –  TFP  –  –  –  –  0.012  0.009  0.009  0.008  –  –  –  –  (0.011)  (0.010)  (0.011)  (0.010)  Innovative capacity controls  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Internationalization controls  No  Yes  Yes  Yes  No  Yes  Yes  Yes                    IV first stage                  F-statistics                  Technological learning effect  13.29  12.61  10.63  11.46  13.29  12.61  10.63  11.46  Demand effect  13.23  11.88  12.45  11.07  13.23  11.88  12.45  11.07  Number of instruments  4  4  4  4  4  4  4  4  Hansen J-statistics  0.014  0.015  0.077  0.003  1.111  1.259  3.370  1.197  P-value  0.993  0.992  0.962  0.998  0.573  0.532  0.185  0.549  Observations  12,905  12,905  12,195  12,564  12,905  12,905  12,195  12,564    (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  Process innovation   Brand new product innovation   All firms  All firms  No multiproduct firms  No large firms  All firms  All firms  No multiproduct firms  No large firms  Technological learning effect  0.583  0.603  0.617  0.304  1.860***  1.819***  1.825***  1.697***  (0.682)  (0.707)  (0.764)  (0.729)  (0.533)  (0.545)  (0.600)  (0.555)  Demand effect  2.072**  2.151**  2.107**  1.748*  0.739  0.722  0.962  0.328  (0.871)  (0.972)  (0.950)  (0.989)  (0.544)  (0.602)  (0.604)  (0.602)  Labor productivity  0.018  0.019  0.014  0.022  –  –  –  –  (0.015)  (0.015)  (0.015)  (0.015)  –  –  –  –  TFP  –  –  –  –  0.012  0.009  0.009  0.008  –  –  –  –  (0.011)  (0.010)  (0.011)  (0.010)  Innovative capacity controls  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Internationalization controls  No  Yes  Yes  Yes  No  Yes  Yes  Yes                    IV first stage                  F-statistics                  Technological learning effect  13.29  12.61  10.63  11.46  13.29  12.61  10.63  11.46  Demand effect  13.23  11.88  12.45  11.07  13.23  11.88  12.45  11.07  Number of instruments  4  4  4  4  4  4  4  4  Hansen J-statistics  0.014  0.015  0.077  0.003  1.111  1.259  3.370  1.197  P-value  0.993  0.992  0.962  0.998  0.573  0.532  0.185  0.549  Observations  12,905  12,905  12,195  12,564  12,905  12,905  12,195  12,564  Note: The LPMs in the table are estimated with a 2SLS estimator and include country, sector, and region fixed effects. The technological learning effect is calculated as the average level of sectoral R&D intensity of the three main export destinations. The demand effect is calculated as the average sectoral import growth of the three main export destinations. In Columns (3) and (7), only firms with at least 60% of their sales in a specific business are included, and in Columns (4) and (8), only firms with less than 500 employees are included. All specifications include the controls displayed in Table 2. Columns (1) and (5) do not include internationalization controls. Standard errors in parentheses are clustered at the firm level *** P < 0.01, ** P < 0.05, * P < 0.1 In Columns (5)–(8) of Table 6, I implement the same IV strategy in the brand new product innovation equation. In this case, I find that in all specifications only the technological learning index is positive and significant, while the foreign demand index becomes not significantly different from zero. These results suggest that, while there certainly is a positive correlation between foreign demand growth and brand new product innovation, such a relationship is probably not a causal one. Summing up the IV results confirm the initial OLS results with one important exception: the foreign demand effect is not anymore found to foster brand new product innovations. Hence only Hypothesis 1a and Hypothesis 2a are supported when the IV procedure is implemented. 4.1.3. Alternative measurements of the technological learning and foreign demand indexes The results presented so far could also be sensitive to the procedure chosen for the measurement of the technological learning and foreign demand effect. For this reason in this section, I introduce alternative measures for the two indexes of technological learning and foreign demand. First of all, it could be the case that the innovative strategies put in place by exporting firms in the period 2007–2009 are only affected by what happened in firms’ export markets in more recent years, for example in the previous 3 years (2005–2007). For example, only recent increases in the import growth of a specific destination might matter for the strategic innovative decisions of firms exporting to that market. If that is the case measures which take into account R&D intensity and import growth between 2003 and 2007 may contain a lot of information that is not relevant for firms’ strategies and therefore introduce noise. To account for this I replicate the two formulas introduced, respectively, in equations (5) and (6), but now I use the average sectoral R&D intensity and the average growth of imports of the three main markets in the period 2005–2007.8 Another potential shortcoming of the technological learning index is that using the average value of R&D across country destinations might not be very informative to make comparisons across exporting firms. Indeed it is likely that knowledge spillovers and opportunities to learn will proceed only from the firm’s most sophisticated market: using an average value means that if firm A exports to only one advanced market and firm B exports to an equally advanced market and a less advanced market, the average value of the technological learning effect would be lower for firm B. This might not be a legitimate choice, since both companies have the same opportunity to learn from the most advanced market in which they export. Therefore I build an alternative measure of technological learning using only the highest value of R&D intensity in sector j among the three main countries of destinations d, conditional on the fact that the firm was already exporting in that market in 2003.   Li{t−1}=max⁡(R&Ddj{t−1}). (7) A potential flaw of the foreign demand index instead is that all foreign markets are counted as equal, regardless of whether they are very important or marginal for a firm. In other words firms might only react to changes that occur in their most important export market, where most of their foreign sales come from. For this reason I build a second index of foreign demand in which I only consider the sectoral import growth between 2003 and 2007 in the most important market for the exporting firm, conditional on the fact that the firm already exported there in 2003.   Di{t−1}=impjd12007−impjd12003. (8) In equation (8)d1 is the country destination (among the three possible), where the firm exports the highest share of its export. Another way of considering the relative share of each export destination is by creating an index of technological learning and foreign demand effects in which I weight the three main export destinations by their relative share in the company’s overall exports.9 I compute the technological learning index as follows:   Li{t−1}=∑d=13avgR&D03−07d*ωd. (9) Where ωd is the share of export for the country of destination d. The same weights are also applied for the foreign demand effect index:   Di{t−1}=∑d=13(impjd2007−impjd2003)*ωd. (10) In Table 7 I test whether these three alternative ways of calculating the technological learning and foreign demand effects change results obtained through my IV methodology in Table 6 (see Table A2 for the first-stage statistics). The results show that also when I use these alternative measures, the results do not change dramatically. The foreign demand effect is still only positive and significant in the process innovation equation, while the technological learning effect only significantly affects the introduction of brand new product innovations. For what concerns the first two alternative indexes (only 2005–2007 period, largest market and highest level of R&D), the size of the coefficients in Columns (1), (2), (4), and (5) is not significantly different with respect to the previous estimates. In the case of the export-share-weighted measures, in Columns (3) and (6), instead the magnitude of the coefficients increases, but the overall statistical significance and direction of the effects do not change. Overall this suggests that the specific methodology chosen to measure the two indexes does not influence the overall results of the analysis. Table 7. Instrumental variables with alternative indexes   (1)  (2)  (3)  (4)  (5)  (6)  Process innovation   Brand new product innovation   Period 2005–2007  Largest market highest R&D  Weighted export shares  Period 2005–2007  Largest market highest R&D  Weighted export shares  Technological learning effect  1.012  0.375  1.044  1.641***  1.310***  3.140***  (0.775)  (0.509)  (1.292)  (0.535)  (0.390)  (1.073)  Demand effect  2.635***  2.027**  3.618**  −0.141  0.674  0.926  (0.991)  (0.937)  (1.749)  (0.594)  (0.587)  (1.098)  Labor productivity  0.010  0.022  0.020  –  –  –  (0.016)  (0.014)  (0.015)  –  –  –  TFP  –  –  –  0.015  0.010  0.011  –  –  –  (0.010)  (0.010)  (0.010)  All controls  Yes  Yes  Yes  Yes  Yes  Yes  IV first stage              F-statistics              Technological learning effect  12.14  14.35  6.85  12.21  14.43  6.87  Demand effect  12.20  11.14  6.81  11.86  10.71  6.54  Number of instruments  4  4  4  4  4  4  Hansen J-statistics  0.744  0.002  0.198  1.584  1.330  1.749  P-value  0.689  0.998  0.905  0.452  0.514  0.417  Observations  12,905  12,849  12,905  12,905  12,849  12,905    (1)  (2)  (3)  (4)  (5)  (6)  Process innovation   Brand new product innovation   Period 2005–2007  Largest market highest R&D  Weighted export shares  Period 2005–2007  Largest market highest R&D  Weighted export shares  Technological learning effect  1.012  0.375  1.044  1.641***  1.310***  3.140***  (0.775)  (0.509)  (1.292)  (0.535)  (0.390)  (1.073)  Demand effect  2.635***  2.027**  3.618**  −0.141  0.674  0.926  (0.991)  (0.937)  (1.749)  (0.594)  (0.587)  (1.098)  Labor productivity  0.010  0.022  0.020  –  –  –  (0.016)  (0.014)  (0.015)  –  –  –  TFP  –  –  –  0.015  0.010  0.011  –  –  –  (0.010)  (0.010)  (0.010)  All controls  Yes  Yes  Yes  Yes  Yes  Yes  IV first stage              F-statistics              Technological learning effect  12.14  14.35  6.85  12.21  14.43  6.87  Demand effect  12.20  11.14  6.81  11.86  10.71  6.54  Number of instruments  4  4  4  4  4  4  Hansen J-statistics  0.744  0.002  0.198  1.584  1.330  1.749  P-value  0.689  0.998  0.905  0.452  0.514  0.417  Observations  12,905  12,849  12,905  12,905  12,849  12,905  Note: The LPMs in the table are estimated with a 2SLS estimator and include country, sector, and region fixed effects. In Columns (1) and (4), the technological learning and foreign demand effects are calculated, respectively, as the average level of sectoral R&D in the period 2005–2007 and as the average sectoral import growth in the period 2005–2007 of the three main export destinations. In Columns (2) and (4), the technological learning effect is calculated as the highest level of sectoral R&D among the three main export destinations, while the demand effect is calculated as the sectoral import growth of the most important market (for the firm) among the three main export destinations. In Columns (3) and (6), the technological learning and foreign demand effects are calculated, respectively, as the average level of sectoral R&D in the period 2003–2007 and as the average sectoral import growth in the period 2003–2007 weighted for the firms’ share of export to each of the three main export destinations. Standard errors in parentheses are clustered at the firm level *** P < 0.01, ** P < 0.05, * P < 0.1. 5. Conclusions This article shows that the positive effect of export activity on innovation among European firms is strictly dependent on the specific export destinations of firms. While export destinations might differ along many dimensions, the article considers two of them in particular: the availability of foreign technological spillovers (the technological learning effect) and the access to foreign expanding markets (the foreign demand effect). The technological learning effect affects firms’ innovation strategies because it provides knowledge spillovers from foreign customers or competitors in very technologically advanced markets: this is important especially because it reduces the relevant internal research costs needed to develop brand new product innovations. On the contrary the foreign demand effect of exporting activities affects firms’ innovation strategies by increasing the potential output of firms, a factor that is often associated with process innovation. In the article I build two indices that are able to proxy these two effects through the use of R&D intensity data at the sectoral level of the destination countries (the technological learning effect) and the growth of sectoral imports of the destination countries (the foreign demand effect). I introduce these two indices in a LPM that explains the adoption of, respectively, brand new product and process innovations by European firms included in the EFIGE data set. The econometric strategy controls for different measures of productivity among the control variables and introduces an IV approach to address possible endogeneity issues. Moreover it also implements a number of robustness checks to control whether the way the two indexes are built affects the results. The results show that indeed the technological learning effect has a positive effect on the introduction of brand new product innovations, while the demand effect of exporting activity mainly induces process innovations. From a theoretical point of view, the results show that export destinations—a factor which has been so far mostly neglected in the existing literature—are instead an important moderator factor able to explain more precisely the relationship between exporting activity and innovation outcomes. Indeed not all export destinations exert the same effect on innovation: acknowledging this evidence could help future work in understanding better when and why exporting firms are able to increase their innovativeness. From a managerial perspective, the results of this article suggest that firms might also choose strategically the destination of their export, not only to increase their level of sales but also to upgrade their competences and capabilities. Moreover managers should be aware that exporting per se might not be sufficient to improve firms’ capabilities: on the contrary some export destinations might be much more effective than others to foster specific types of innovation outputs. Firms might choose export destinations also on the basis of their specific needs in terms of innovation competences. From a policy perspective, it seems important to acknowledge that the positive effect of exporting activity also depends on the specific export destinations. If in a specific country firms export mainly to expanding markets with little levels of technological development, they might be induced to put less efforts in developing truly innovative products. Since in advanced economies brand new product innovations—able to actually shift the world technological frontier—are those with the highest economic impact, exporting only to expanding markets might hinder the ability of firms to develop their future innovative capacities. Finally, while this article provides an interesting perspective on the role of export destinations for firms’ innovation strategies, it must be stressed that it only analyzes two possible dimensions along which export destinations might differ: however other dimensions, such as geographical or cultural distance, institutional settings, specific (environmental) regulations might also play a role in the causal relationship going from export to innovation. Future research should address these interesting additional perspectives. Funding The author has benefitted from the access to the EU-EFIGE/Bruegel-UniCredit database, managed by Bruegel and funded by the European Union’s Seventh Framework Programme ([FP7/2007-2013] under grant agreement number 225551), as well as by UniCredit. Footnotes 1 The data collection has been performed in early 2010 through a questionnaire submitted to the firms. The data set includes also information on Austrian and Hungarian firm which are not used in this analysis, since firms in small and open economies might show different dynamics with respect to firms in large European countries. 2 A non-exporting firm which only knows its domestic market might consider a new product with little innovative content as new to the market. On the contrary a highly competitive and internationalized firm operating in different foreign markets might consider a very innovative product as not new to the market because in some other markets, it might have been already introduced by another leading competitor. 3 An alternative would be the number of patent applications by national firms in each specific sector. However this approach is not straightforward because it is necessary to match firms’ sectorial classifications with the technological classes of patents. 4 The results obtained with probit and logit estimations are perfectly in line with the LPM results presented here both in terms of significance and of magnitude of the coefficients. 5 Both measures have advantages and disadvantages: labor productivity is a more straightforward index of efficiency, which does not rely on any assumption about firms’ production functions; however, in the absence of a measure of the stock of capital (due to the nonavailability of this measure for EFIGE data), it might also proxy the capital intensity of each firm. On the contrary TFP allows to account for all the inputs used by firms—and hence might be a better proxy for the technological sophistication of a firm—however, it also requires to estimate the output elasticities of firms on the basis of some assumptions about the functional form of firms’ production functions. 6 The results are robust to the use of productivity measures also for more recent years (such as 2008). The measures of productivity in the EFIGE data provided by Bruegel are not available for about 25% of the firms, with slightly higher shares in Germany and the UK, due to the nonavailability of ORBIS balance sheet data for such firms. For these firms I imputed their labor productivity and TFP with the following methodology: I estimated through OLS the determinants of both productivity measures in two separate regressions, using the rich set of independent variables included in equation (4) and including all the firms (Rubin, 1987). Then I calculated the predicted values for both productivity measures and substituted the predicted values only for the firms that had missing values for labor productivity (3823 firms) and TFP (3300 firms). A careful analysis of the distribution of imputed and real productivity measures (for the firms for which productivity was instead available) did not show any significant differences in terms of mean and skewness of the variables. Moreover results obtained without including the firms with missing productivity measures did not show significant differences. While this imputation strategy is not free of limitations, as it computes productivity on the basis of the average contribution of a set of other variables, it allows to maintain the representativeness of the sample for the European countries analyzed in the study. 7 This methodology borrows from Bratti and Felice (2012) the idea of using a specific feature of a foreign destination market and calculate an average across destinations, using as weights the export shares to each destination of all the firms active in the sector of the focal firm. However, differently from Bratti and Felice (2012), the instrument includes R&D intensity and imports growth rather than GDP per capita, and it is computed at the national sector level, and not at the sector-province level: the larger level of aggregation adds to the probability that the instrument is exogenous to the focal firms’ innovativeness. 8 To instrument these two new variables, I use the same IV methodology as in the previous specifications in Table (6), with the only difference that also the instruments will be computed for the period 2005–2007. 9 This information is available in EFIGE, since for each of three main export destinations, every firm was also asked what was the share of exports accounted for by each destination. Acknowledgments The author is grateful to Marcello Messori, Cristiano Antonelli and Davide Castellani for useful comments and suggestions. The article has benefited from comments received during the following seminars and conference presentations: Annual meeting of the Academy of International Business, New Orleans, July 2016; DRUID Conference, Rome, June 2015; EMAEE Conference, Maastricht, June 2015; Explaining Economic Change Workshop, Rome, November 2014. The author is also grateful to Bruegel for the access to the quasi full set EFIGE database at their premises. References Accetturo A., Bugamelli M., Lamorgese A. R.. ( 2013), ‘ Skill upgrading and exports,’ Economics Letters , 121( 3), 417– 420. Google Scholar CrossRef Search ADS   Altomonte C., Aquilante T.. ( 2012), ‘The EU-EFIGE/Bruegel-Unicredit dataset,’ Bruegel Working Paper, October 2012: Bruegel, Brussels. Andersson M., Lööf H.. ( 2009), ‘ Learning-by-exporting revisited: the role of intensity and persistence,’ Scandinavian Journal of Economics , 111( 4), 893– 916. Google Scholar CrossRef Search ADS   Angrist J. ( 2001), ‘ Estimation of limited dependent variable models with dummy endogenous regressors: simple strategies for empirical practice,’ Journal of Business and Economic Statistics , 19( 1), 2– 16. Google Scholar CrossRef Search ADS   Basile R.( 2001), ‘ Export behavior of Italian manufacturing firms over the Nineties: the role of innovation,’ Research Policy , 30 ( 8), 1185– 1201. Google Scholar CrossRef Search ADS   Bernard, A. B. and J. Wagner (1997), ‘Exports and Success in German Manufacturing’, Weltwirtschaftliches Archiv/Review of World Economics, 133(1), 134–57. Bernard A. B., Jensen B. J.. ( 1999), ‘ Exceptional exporter performance: cause, effect, or both?,’ Journal of International Economics , 47( 1), 1– 25. Google Scholar CrossRef Search ADS   Blanes-Cristóbal J. V., Dovis M., Milgram-Baleix J., Moro-Egido A. I.. ( 2008), ‘ Do sunk exporting costs differ among markets? Evidence from Spanish manufacturing firms,’ Economics Letters , 101, 110– 112. Google Scholar CrossRef Search ADS   Bound J., Jaeger D., Baker R.. ( 1995), ‘ Problems with instrumental variables estimation when the correlation between instruments and the endogenous explanatory variables is weak,’ Journal of American Statistical Association , 90( 430), 443– 450. Bratti M., Felice G.. ( 2012), ‘ Are exporters more likely to introduce product innovations?,’ The World Economy , 35( 11), 1559– 1598. Google Scholar CrossRef Search ADS   Brouwer E., Kleinknecht A..( 1999), ‘ Keynes-Plus? Effective demand and changes in firm-level R&D: an empirical note,’ Cambridge Journal of Economics , 23( 3), 385– 391. Google Scholar CrossRef Search ADS   Bustos P. ( 2011), ‘ Trade liberalization, exports and technology upgrading: evidence on the impact of MERCOSUR on argentinean firms,’ American Economic Review , 101( 1), 304– 340. Google Scholar CrossRef Search ADS   Cassiman B., Golovko E., Martínez-Ros E..( 2010), ‘ Innovation, exports and productivity,’ International Journal of Industrial Organization , 28( 4), 372– 376. Google Scholar CrossRef Search ADS   Cohen W. M., Klepper S.. ( 1996), ‘ Firm size and the nature of innovation within industries: the case of process and product R&D,’ The Review of Economics and Statistics , 78( 2), 232– 243. Google Scholar CrossRef Search ADS   Crespi G., Criscuolo C., Haskel J.. ( 2008), ‘ Productivity, exporting, and the learning-by-exporting hypothesis: direct evidence from UK firms,’ Canadian Journal of Economics , 41( 2), 619– 638. Google Scholar CrossRef Search ADS   Crinò, R. and P. Epifani, (2012), ‘Productivity, Quality and Export Behaviour,’ Economic Journal, 122(565), 1206–1243. Damijan J. P., Kostevc C., Polanec S.. ( 2010), ‘ From innovation to exporting or vice versa?,’ The World Economy , 33( 3), 374– 398. Google Scholar CrossRef Search ADS   Desmet K., Parente S. L.. ( 2010), ‘ Bigger is better: market size, demand elasticity, and innovation,’ International Economic Review , 51( 2), 319– 333. Google Scholar CrossRef Search ADS   Dosi, G., S. Lechevalier and A. Secchi. (2010), ‘Interfirm heterogeneity: nature, sources and consequences for industrial dynamics. An introduction’, Industrial and Corporate Change, 19, 1867. Dosi G., Grazzi M., Moschella D.. ( 2015), ‘ Technology and costs in international competitiveness: From countries and sectors to firms,’ Research Policy , 44( 10), 1795– 1814. Google Scholar CrossRef Search ADS   Fafchamps M., El Hamineb S., Zeufackc A.. ( 2008), ‘ Learning to export: evidence from moroccan manufacturing,’ Journal of African Economies , 17( 2), 305– 355. Google Scholar CrossRef Search ADS   Fagerberg J. ( 1988), ‘ International competitiveness,’ The Economic Journal , 98 ( 391), 355– 374. Google Scholar CrossRef Search ADS   Fitjar R. D., Rodriguez-Pose A.. ( 2012), ‘ Firm collaboration and modes of innovation in Norway,’ Research Policy , 42( 1), 128– 138. Google Scholar CrossRef Search ADS   Garcia-Quevedo J., Pellegrini G., Savona M.. ( 2017), ‘Reviving demand-pull perspectives: the effect of demand uncertainty and stagnancy on R&D strategy,' Cambridge Journal of Economics, 41(4), 1087–1122. Geroski P. A., Walters C. F.. ( 1995), ‘ Innovation activity over the business cycle,’ The Economic Journal , 105, 916– 928. Google Scholar CrossRef Search ADS   Golovko E., Valentini G.. ( 2014), ‘ Selective learning-by-exporting: firm size and product versus process innovation,’ Global Strategy Journal , 4, 161– 180. Google Scholar CrossRef Search ADS   Harirchi G., Chaminade C.. ( 2014), ‘ Exploring the relation between the degree of novelty of innovations and user-producer interaction across different income regions,’ World Development , 57, 19– 31. Google Scholar CrossRef Search ADS   Hall B., Mairesse J., Branstetter L., Crepon B.. ( 1999), ‘Does cash flow cause investment and R&D? An exploration using panel data for French, Japanese and United States scientific firms,’ in Audretsch D., Thurik R. (eds), Innovation, Industry Evolution and Employment . Cambridge University Press: Cambridge. Laursen K. ( 2011), ‘ User–producer interaction as a driver of innovation: costs and advantages in an open innovation model,’ Science and Public Policy , 38( 9), 713– 723. Google Scholar CrossRef Search ADS   Levinsohn J., Petrin A..( 2003), ‘ Estimating production functions using inputs to control for unobservables,’ The Review of Economic Studies , 70 ( 2), 317– 341. Google Scholar CrossRef Search ADS   Lileeva A., Trefler D.. ( 2010), ‘ Improved access to foreign markets raises plant-level productivity … for some plants,’ The Quarterly Journal of Economics , 125( 3), 1051– 1099. Google Scholar CrossRef Search ADS   Liu X., Buck T.. ( 2007), ‘ Innovation performance and channels for international technology spillovers: evidence from Chinese high-tech industries,’ Research Policy , 36( 3), 355– 366. Google Scholar CrossRef Search ADS   Malerba F., Nelson R. R., Orsenigo L., Winter S. G.. ( 2007), ‘ Demand, innovation and the dynamics of market structure: the role of experimental users and diverse preferences,’ Journal of Evolutionary Economics , 17( 4), 371– 399. Google Scholar CrossRef Search ADS   Melitz M. ( 2003), ‘ The impact of trade in intra-industry reallocations and aggregate industry productivity,’ Econometrica , 71( 6), 1695– 1725. Google Scholar CrossRef Search ADS   Piva M., Vivarelli M.. ( 2007), ‘ Is demand-pulled innovation equally important in different groups of firms?,’ Cambridge Journal of Economics , 31, 691– 710. Google Scholar CrossRef Search ADS   Posner M. V. ( 1961), ‘ International trade and technical change,’ Oxford Economic Papers , 13( 3), 323– 341. Google Scholar CrossRef Search ADS   Rubin D. B.( 1987), Multiple Imputation for Non Response in Surveys . Wiley: New York. Google Scholar CrossRef Search ADS   Salomon R. M. ( 2006), ‘ Spillovers to foreign market participants: assessing the impact of export strategies on innovative productivity,’ Strategic Organization , 4( 2), 135– 164. Google Scholar CrossRef Search ADS   Salomon R. M., Shaver J. M.. ( 2005), ‘ Learning by exporting: new insights from examining firm innovation,’ Journal of Economics and Management Strategy , 14( 2), 431– 460. Google Scholar CrossRef Search ADS   Scherer F. ( 1991), ‘Changing perspectives on the firms size problem,’ in Acs Z. J., Audretsch D. B. (eds), Innovation and Technological Change. An International Comparison . Harvester Wheatsheaf: New York. Schmookler J. ( 1954), ‘ The level of inventive activity,’ The Review of Economics and Statistics , 36, 183– 190. Google Scholar CrossRef Search ADS   Schmookler J. ( 1966), Invention and Economic Growth . Harvard University Press: Cambridge. Google Scholar CrossRef Search ADS   Serti F., Tomasi C..( 2012), ‘ Self selection among different export markets,’ Economics Letters , 117( 1), 102– 105. Google Scholar CrossRef Search ADS   Trofimenko N. ( 2008), ‘ Learning by Exporting: does it matter where one learns? Evidence from Colombian manufacturing firms,’ Economic Development and Cultural Change , 56( 4), 871– 894. Google Scholar CrossRef Search ADS   Van Biesebroeck J. ( 2005), ‘ Exporting raises productivity in sub-Saharan African manufacturing firms,’ Journal of International Economics , 67( 2), 373– 391. Google Scholar CrossRef Search ADS   Verhoogen E. ( 2008), ‘ Trade, quality upgrading and wage inequality in the Mexican manufacturing sector,’ Quarterly Journal of Economics , 123( 2), 489– 530. Google Scholar CrossRef Search ADS   von Hippel E. ( 1986), ‘ Lead users: a source of novel product concepts,’ Management Science , 32, 791– 805. Google Scholar CrossRef Search ADS   von Hippel E. ( 2005), Democratizing Innovation . MIT Press: Cambridge, MA. Wakelin K. ( 1998), ‘ Innovation and export behaviour at the firm level,’ Research Policy , 26( 7), 829– 841. Google Scholar CrossRef Search ADS   Woerter M., Roper S.. ( 2010), ‘ Openness and innovation—home and export demand effects on manufacturing innovation: panel data evidence for Ireland and Switzerland,’ Research Policy , 39, 155– 164. Google Scholar CrossRef Search ADS   Appendix: first stage statistics Table A1. First-stage statistics for Table 6 First stage  (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  Technological learning effect  Foreign demand effect  TL  1.295***  1.267***  1.244***  1.260***  −0.224*  −0.286**  −0.287**  −0.243*  (0.208)  (0.206)  (0.214)  (0.209)  (0.135)  (0.134)  (0.139)  (0.136)  TL*emp≤25  −0.504***  −0.505***  −0.445***  −0.434***  0.206***  0.201***  0.216***  0.206***  (0.120)  (0.119)  (0.125)  (0.120)  (0.070)  (0.069)  (0.073)  (0.070)  TD  0.185  0.143  0.122  0.149  1.310***  1.178***  1.273***  1.127***  (0.128)  (0.127)  (0.132)  (0.131)  (0.256)  (0.247)  (0.253)  (0.253)  TD*emp≤25  0.050  0.063  0.050  0.044  −0.519***  −0.475***  −0.477***  −0.463***  (0.050)  (0.050)  (0.051)  (0.050)  (0.095)  (0.092)  (0.094)  (0.094)  Productivity measures  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Structural and innovation controls  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Internationalization controls  No  Yes  Yes  Yes  No  Yes  Yes  Yes  F-statistics  13.29  12.61  10.63  11.46  13.23  11.88  12.45  11.07  Observations  12,905  12,905  12,195  12,564  12,905  12,905  12,195  12,564  First stage  (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  Technological learning effect  Foreign demand effect  TL  1.295***  1.267***  1.244***  1.260***  −0.224*  −0.286**  −0.287**  −0.243*  (0.208)  (0.206)  (0.214)  (0.209)  (0.135)  (0.134)  (0.139)  (0.136)  TL*emp≤25  −0.504***  −0.505***  −0.445***  −0.434***  0.206***  0.201***  0.216***  0.206***  (0.120)  (0.119)  (0.125)  (0.120)  (0.070)  (0.069)  (0.073)  (0.070)  TD  0.185  0.143  0.122  0.149  1.310***  1.178***  1.273***  1.127***  (0.128)  (0.127)  (0.132)  (0.131)  (0.256)  (0.247)  (0.253)  (0.253)  TD*emp≤25  0.050  0.063  0.050  0.044  −0.519***  −0.475***  −0.477***  −0.463***  (0.050)  (0.050)  (0.051)  (0.050)  (0.095)  (0.092)  (0.094)  (0.094)  Productivity measures  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Structural and innovation controls  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Internationalization controls  No  Yes  Yes  Yes  No  Yes  Yes  Yes  F-statistics  13.29  12.61  10.63  11.46  13.23  11.88  12.45  11.07  Observations  12,905  12,905  12,195  12,564  12,905  12,905  12,195  12,564  Note: This table reports the first stage statistics for the instruments used in the IV two stages least squares estimator in Table 6 (see Section 4.1 for details). In Columns (3) and (7), only firms with at least 60% of their sales in a specific business are included, and in Columns (4) and (8), only firms with less than 500 employees are included. Table A2. First-stage statistics for Table 7 First stage  (1)  (2)  (3)  (4)  (5)  (6)  Technological learning effect   Foreign demand effect   Period 2005–2007  Largest market highest R&D  Weighted export shares  Period 2005–2007  Largest market highest R&D  Weighted export shares  TL  1.273***  1.729***  0.722***  −0.310**  −0.237*  −0.158*  (0.204)  (0.265)  (0.163)  (0.133)  (0.141)  (0.093)  TL*emp≤25  −0.496***  −0.684***  −0.267***  0.210***  0.212***  0.127***  (0.116)  (0.159)  (0.080)  (0.073)  (0.075)  (0.048)  TD  0.058  0.252  0.121  0.844***  1.235***  0.778***  (0.106)  (0.169)  (0.096)  (0.190)  (0.258)  (0.187)  TD*emp≤25  0.041  0.055  0.031  −0.471***  −0.471***  −0.202***  (0.056)  (0.068)  (0.034)  (0.096)  (0.096)  (0.071)  Productivity measures  Yes  Yes  Yes  Yes  Yes  Yes  Structural and innovation controls  Yes  Yes  Yes  Yes  Yes  Yes  Internationalization controls  Yes  Yes  Yes  Yes  Yes  Yes  F-statistics  12.14  14.35  6.85  12.2  11.14  6.81  Observations  12,905  12,849  12,905  12.905  12,849  12.905  First stage  (1)  (2)  (3)  (4)  (5)  (6)  Technological learning effect   Foreign demand effect   Period 2005–2007  Largest market highest R&D  Weighted export shares  Period 2005–2007  Largest market highest R&D  Weighted export shares  TL  1.273***  1.729***  0.722***  −0.310**  −0.237*  −0.158*  (0.204)  (0.265)  (0.163)  (0.133)  (0.141)  (0.093)  TL*emp≤25  −0.496***  −0.684***  −0.267***  0.210***  0.212***  0.127***  (0.116)  (0.159)  (0.080)  (0.073)  (0.075)  (0.048)  TD  0.058  0.252  0.121  0.844***  1.235***  0.778***  (0.106)  (0.169)  (0.096)  (0.190)  (0.258)  (0.187)  TD*emp≤25  0.041  0.055  0.031  −0.471***  −0.471***  −0.202***  (0.056)  (0.068)  (0.034)  (0.096)  (0.096)  (0.071)  Productivity measures  Yes  Yes  Yes  Yes  Yes  Yes  Structural and innovation controls  Yes  Yes  Yes  Yes  Yes  Yes  Internationalization controls  Yes  Yes  Yes  Yes  Yes  Yes  F-statistics  12.14  14.35  6.85  12.2  11.14  6.81  Observations  12,905  12,849  12,905  12.905  12,849  12.905  Note: This table reports the first stage statistics for the instruments used in the IV 2SLS estimator in Table 7 (see Section 4.1.3. for details). In Columns (1) and (4), the instruments are calculated using only data for the period 2005–2007. © The Author 2017. Published by Oxford University Press on behalf of Associazione ICC. All rights reserved. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Industrial and Corporate Change Oxford University Press

Export-led innovation: the role of export destinations

Loading next page...
 
/lp/ou_press/export-led-innovation-the-role-of-export-destinations-loJRex3Dej
Publisher
Oxford University Press
Copyright
© The Author 2017. Published by Oxford University Press on behalf of Associazione ICC. All rights reserved.
ISSN
0960-6491
eISSN
1464-3650
D.O.I.
10.1093/icc/dtx028
Publisher site
See Article on Publisher Site

Abstract

Abstract This article analyzes the effect of exporting activity on the innovative performances of firms in France, Germany, Italy, Spain, and UK. It argues that the positive effect of exporting on innovation usually found in the literature varies according to the specific destinations of exports, and it identifies two dimensions along which export destinations might differ: the level of foreign technological spillovers available to exporting firms (the technological learning effect) and the type of foreign demand that exporting firms are able to access (the foreign demand effect). The empirical analysis, which takes advantage of firm-level information about the export destinations of exporters, shows that while the technological learning effect increases mainly the incentives to introduce brand new product innovations, the foreign demand effect fosters the adoption of process innovations. 1. Introduction Empirical contributions in both the trade literature and the technology gap tradition have provided extensive evidence that innovative capabilities are important factors able to explain the export participation and the export success of firms (Posner, 1961; Fagerberg, 1988; Wakelin, 1998). Indeed innovative activities, even more than cost advantages, are able to increase the overall competitiveness of firms and allow them to compete in international markets (Basile, 2001; Cassiman et al., 2010; Dosi et al., 2015). While the fact that innovation is an important boost for export activities is generally accepted as an established fact, recent empirical evidence has suggested that the relationship between innovation and exports can also work in the opposite direction: exporters might be induced to innovate precisely because they are active in export markets. The intuition is that by operating on international markets, firms are induced to implement new organizational and technological routines that eventually allow them to increase productivity levels. A growing literature has analyzed the causal link existing between exporting activity and innovation (Liu and Buck, 2007; Andersson and Lööf, 2009; Lileeva and Trefler, 2010; Bustos, 2011), finding, in most cases, that exporting activity indeed increases the probability to introduce innovations at the firm level. While these studies agree on the existence of a generalized positive effect of exporting activity on innovation performances, there is instead no unanimity concerning which specific types of innovations are mostly affected. The results differ according to the specific national setting of the study: positive effects have been found in different empirical studies on, respectively, product innovation (Bratti and Felice, 2012), process innovation (Damijan et al., 2010), or patent applications (Salomon and Shaver, 2005). A possible explanation of the heterogeneity of these results is that in most of the cases the existing literature has considered exporting as an undifferentiated activity, regardless of the specific export destinations of firms. Quite on the contrary export destinations can differ a lot along several dimensions (level of competition; geographical and cultural distance; institutional setting), and each of them might induce different types of innovation outcomes. Studies that do not acknowledge this factor are likely to find different results, according to the specific destination of exports of the firms analyzed. In this article I focus on two important dimensions along which export destinations can differ and which might affect innovation outcomes: one is the level of foreign technological spillovers available to exporting firms, and the other is the type of foreign demand that exporting firms can have access to. Firms might benefit from exporting either because they are active in very technologically advanced foreign markets, where they learn to innovate from foreign clients and competitors, or because they export to foreign markets with a strong demand growth, which increases the demand for the firms’ products and also the profitability of introducing new technologies. Throughout the article I will refer to the first effect as the technological learning effect and to the second as the foreign demand effect. I argue that each of these two effects, which directly depend on export destinations, will exert different incentives on the introduction of innovations: the technological learning effect mainly reduces the costs of developing innovations, while the foreign demand effect increases the firms’ output and hence the profitability of introducing innovations. To identify the differentiated impact of each of these two effects on innovation outcomes, I examine two very different types of innovations—brand new product innovations and process innovations—and explore how the technological learning and foreign demand effect influence the introduction of these two innovation outputs. I take advantage of the EU-EFIGE/Bruegel-UniCredit data set (European firms in a Global Economy), which includes detailed firm-level information about export activity and innovation performances of a large number of firms active in the five largest European economies (Germany, France, Italy, Spain and UK) in the period 2007–2009. The main advantage of the data set is that it includes information on the most important export destinations of firms. I use this information to build two indexes that measure how much an exporting firm is exposed to each of the two effects. I proxy the technological learning effect with the R&D intensity of the countries in which a firm exports, and the foreign demand effect with the growth of imports of the markets in which an exporting firm is active. Since I expect firms to be mainly affected by the features and dynamics of their sector of affiliation, rather than by country-level ones, in both cases I use the R&D intensity and import growth of the sector in which exporting firms are active. I include these two indices in a model that explains the introduction of, respectively, brand new product innovations and process innovations. I test the robustness of the results through the inclusion of different productivity measures, through the use of instrumental variable (IV) estimations and through different methods of building the indexes of the technological learning and foreign demand effect. The results of the empirical analyses show that indeed the technological learning and foreign demand effects induce different innovation outcomes. The technological learning effect only affects brand new product innovations. The foreign demand effect is positively correlated with both process innovation and brand new product innovation, but the IV analysis points to a causal effect only on process innovations. The article contributes to the literature in a number of ways: first, in line with other recent studies (Golovko and Valentini, 2014), the effect of exporting activity on innovation is not anymore assessed on a generic measure of innovation, but rather on specific types of innovation outcomes. Second, the article shows that export destinations are an important moderator factor, able to drive the effect of exporting activity on different types of innovations. In doing so, it also provides a more precise measurement of the role of export destinations: indeed so far existing studies that took into consideration export destinations, only distinguished between the effect of exporting to high-income or Organisation for Economic Co-operation and Development (OECD) countries with respect to low- or middle-income countries (Salomon, 2006; Trofimenko, 2008). The present analysis instead not only takes into account the features of the specific country of destination but also of the sector of activity of each firm. Finally, the article introduces two different dimensions along which export destinations might differ: the availability of technological spillovers (the technological learning effect) and the access to foreign expanding markets (the foreign demand effect). While the former is often acknowledged in the existing literature, the latter has been largely overlooked: however, both seems relevant to understand the effect of exports on innovation. The article is organized as follows: Section 2 presents the relevant literature and introduces the hypotheses concerning the role of the technological learning and foreign demand effect on different types of innovations; Section 3 presents the empirical strategy and the data used; Section 4 presents the results, while Section 5 draws some conclusions and implications for policy and management. 2. The effect of exports on innovation: technological learning and foreign demand effects The evidence that exporters are generally more productive than non-exporters is usually explained by two main hypotheses: either by the fact that more productive firms self-select into export or, alternatively, by the fact that being active in international markets allows firms to upgrade their technologies. The self-selection hypothesis explains the productivity differentials among exporters and non-exporters with the existence of additional costs related to the sale of goods in foreign countries that act as entry barriers: only highly productive firms will be able to overcome such costs and hence decide to export (Bernard and Jensen, 1999; Melitz, 2003). The alternative hypothesis instead suggests that exporters become more productive precisely because they are in international markets: an increasing number of studies report results highlighting an ex post positive effect of exporting both on the productivity levels (Lileeva and Trefler, 2010; Bustos, 2011) and on the innovation outcomes (Damijan et al., 2010) of exporting firms. These results are often explained by the fact that, through interactions with foreign customers, exporters benefit from international technological spillovers in the forms of know-how and new technological capabilities: hence, higher productivity is a consequence of exporting, rather than a prerequisite. In this article I refer to this as the “technological learning effect.” In the literature that studies the mechanism by which firm self-select into exporting, the role of export destinations has been explored in several theoretical and empirical contributions, which showed that the level of productivity required for firms to start to export strongly depends on destination-specific factors, such as distance, the level of income of export destinations, the size of the market, and the exporters’ familiarity with institutions (Blanes-Cristóbal et al. 2008; Serti and Tomasi, 2012). On the contrary the literature focused on the effect of exporting on innovation and productivity has largely overlooked the role that different export destinations may exert on exporters. Nevertheless, the specific characteristics of destination markets are likely to be extremely important also to understand the impact of exporting activities on innovation and productivity. Indeed the learning by exporting mechanism is implicitly related with the specific destination markets of exporters, since to learn firms should be active in markets that are more advanced than the domestic one. In line with this, most of the existing studies find a positive effect of exporting on innovation and productivity among firms in middle-income and emerging countries, which export to high-income countries. Verhoogen (2008) finds that increased access to foreign markets induces Mexican exporters to upgrade their quality. Similar findings are provided by Liu and Buck (2007) for Chinese firms, by Van Biesebroeck (2005) for Sub-Saharan African companies, and by Fafchamps et al. (2008) for Moroccan firms exporting to Europe. However only Trofimenko (2008) explicitly measures the effect of different export destinations, although at a very aggregate level: in her analysis on Colombian firms, she finds that only firms which export to OECD countries truly benefit from exporting activity. The positive effect of exports on innovation and productivity is also found among firms in high-income economies. Salomon and Shaver (2005) report a positive effect of exporting on patenting and product innovation among Spanish firms; Crespi et al. (2008) find that UK firms able to access relevant information from foreign buyers are also able to increase their productivity. Damijan et al. (2010) and Bratti and Felice (2012) find a positive effect of exporting activity on innovation among, respectively, Slovenian and Italian firms: however while the former report a positive effect on process innovation, the latter show that exporting increases product innovation. Also in these studies the specific destinations of exports are usually not considered, implicitly assuming that technological spillovers occur, regardless of the specific foreign markets in which firms are active. The only exception is Salomon (2006), who finds a positive effect of exports to OECD countries on patents among Spanish firms. Overall the literature that explains the positive effect of exports on innovation through the access to foreign technological spillovers (the “technological learning effect”) has largely overlooked the role of export destinations. Moreover when export destinations have been introduced, this was limited to the broad distinction between high- and low-income countries. However it seems likely that export destinations should indeed matter for learning processes, since these only occur when foreign buyers are more sophisticated than domestic ones. Moreover the literature is also not unanimous about which types of innovation are affected by the technological learning effect, whether patents, product, or process innovations. A second explanation of the positive effect of exporting activity on innovation and productivity that is found in the literature is related to the role of foreign demand: in this article I will refer to it as “the foreign demand effect.” In the tradition of Schmookler (1966), demand growth has often been considered as an important determinant of innovation. The demand pull literature has largely emphasized the positive effect of the growth of markets on the innovative efforts put in place by firms (Geroski and Walter, 1995; Brouwer and Kleinknecht, 1999). Therefore also the growth of foreign markets is likely to induce exporting firms to invest in new technologies, since the expected profits related to the introduction of innovations will increase with the level of firms’ sales. The evidence on the effect of foreign demand on exporters’ innovative activities is quite limited and not clear-cut. Piva and Vivarelli (2007) measure the impact of demand on the investments in R&D of Italian firms and find that the positive effect of demand is stronger among firms with higher export intensity. Woerter and Rope (2010) show that foreign demand does not have any effect on the innovation outputs of Irish firms, while it has a positive, but quite limited, effect on product and process innovation among Swiss firms. Both studies however adopt aggregate measures of foreign demand growth which do not distinguish between firms exporting to markets with high or low growth of demand; also in this case, the role of export destinations is not accounted for. However it seems plausible that the foreign demand effect should have a great importance in fostering innovation strategies only for firms exposed to a high growth of foreign sales. Moreover also in this case it is likely that the foreign demand effect might foster some specific types of innovations, those especially affected by the growth of demand and sales, rather than others. Summing up existing studies suggest that the positive effect of exporting activities on innovation might be due to (at least) two different underlying mechanisms: a technological learning effect that allows firms to benefit from foreign technological spillovers and a foreign demand effect that creates incentives for firms to introduce innovations due to the increase of their foreign markets. However both effects will strongly depend on the specific export destinations of each firm. Moreover each of these two effects might also induce different types of innovations. So far the existing literature has rather overlooked these last two points: in the next section, I will specifically investigate whether the technological and foreign demand effects, which depend on export destinations, also lead to different innovative strategies. 2.1 The effect of exporting activities on different innovation outcomes The technological learning and foreign demand effects are two factors able to explain the positive effect of exporting activity on innovation, and their impact depending on the specific export destinations of each exporting firm. However their impact on innovation is also likely to differ according to the specific innovative output considered. Indeed the technological learning effect allows firms to benefit from foreign knowledge spillovers which decrease the internal research costs necessary to develop new innovations. The foreign demand effect instead increases firms’ potential output and the number of units sold. Therefore different types of innovations will be affected, according to how much these factors affect the profitability of their introduction. In this article I will focus on the impact of the technological learning and foreign demand effects on two specific types of innovation outcomes: process innovations and brand new product innovations. The reason is that these two types of innovations are sufficiently different to allow checking if the technological learning and foreign demand effects affect different innovation types. Indeed while process innovations are a quite common type of innovation, available also to firms with little technological capabilities, brand new product innovations are instead introduced only by leading firms able to actually shift the technological frontier. If the two effects influence different types of innovation outcomes, this might explain why in the existing literature there is no unanimity on which innovation outputs are more affected by exporting activity: according to the specific destinations of exports (and to the main effect at stake), different types of innovations will be affected. 2.1.1 Process innovations Process innovations increase the efficiency of the productive processes, that is, they decrease the cost of inputs, given a certain quantity of output, typically leading to increase in productivity levels. Firms might be induced to introduce process innovation both by the foreign demand and by the technological learning effect. The foreign demand effect might foster process innovation: in line with Schmookler tradition (1954), the incentives to introduce process innovations should increase with the quantity of output produced, since the efficiency gains on each unit produced will be multiplied by a larger number of units. Scherer (1991) and Cohen and Klepper (1996) provide theoretical and empirical evidence that the increase in the number of units sold induces firms to dedicate more research efforts toward process innovations. Desmet and Parente (2010) develop a theoretical model to show that an increase of international sales for an exporting firm will mainly foster process innovations. Firms exporting to markets with a growing demand should hence have a strong incentive to introduce process innovations and increase the efficiency of their expanding production. It is hence possible to spell out Hypothesis 1a: Hypothesis 1a: The foreign demand effect of export activity increases the incentives for firms to introduce process innovations. At the same time process innovations might also be induced by the technological learning effect, i.e. firms in very advanced markets might benefit from spillovers that eventually lead them to introduce process innovations. Indeed Damijan et al. (2010) find that Slovenian firms exporting to advanced European markets tend to introduce more process innovations. Interacting with foreign customers, which require higher-quality standards, might induce exporters to upgrade their productive processes and introduce process innovations. Also the need to comply with specific regulations in advanced markets concerning production processes might induce exporters to introduce process innovations. Accordingly it is possible that also the technological learning effect might induce process innovations, as stated in Hypothesis 1b: Hypothesis 1b: The technological learning effect of export activity increases the incentives for firms to introduce process innovations. 2.1.2 Brand new product innovations The introduction of brand new product innovations able to shift the technological frontier allows firms to earn a temporary monopolistic profit on the products sold. However, it also entails very high research costs and long development processes, which not all firms are able to undertake. Both the technological learning effect and the foreign demand effect might increase the incentives for exporters to introduce brand new product innovations. The technological learning effect allows firms to decrease research costs through knowledge spillovers stemming from foreign users or foreign competitors. Since research costs are of crucial importance for brand new innovative strategies, the technological learning effect, which decreases such costs, is likely to have a positive impact on this innovative strategy. The literature that focuses on the role of users in the innovative process (von Hippel, 1986, 2005; Malerba et al., 2007) shows that interactions with users that can increase the firms’ competences typically lead to brand new product innovations. Moreover, a growing literature has found that interactions with international customers have a high probability to increase the novelty of a firm’s innovation output and eventually lead to truly new product innovations (Laursen, 2011; Fitjar and Rodriguez-Pose, 2012; Harirchi and Chaminade, 2014). Accordingly Hypothesis 2a is as follows: Hypothesis 2a: The technological learning effect of export activity increases the incentives for firms to introduce brand new product innovations. At the same time, also the foreign demand effect might increase the incentives for firms to introduce brand new product innovations, since this type of innovations allows for temporary monopolistic rents which are especially profitable when a market is growing. Firms active in foreign expanding markets might have much larger incentives to invest resources and introduce truly innovative products, since the costs associated with the development of the new products will be more than compensated by the markups typically associated to these products. Indeed in most cases truly new products are the outcome of the investments in R&D activities and, as shown by Hall et al. (1999), the elasticity of R&D expenditures to sales is quite high. Garcia-Quevedo et al. (2017) also find evidence that demand dynamics strongly influence R&D investments. Moreover Piva and Vivarelli (2007) show that the positive effect of sales on R&D expenditures is especially strong for exporting firms. Based on these considerations, it is possible to expect also a positive effect of foreign demand on brand new product innovation and hence Hypothesis 2b goes as follows: Hypothesis 2b: The foreign demand effect of export activity increases the incentives for firms to introduce brand new product innovations. 3. The empirical strategy 3.1. A simple model To test the hypotheses about the effect of exports on firms’ innovative outputs, I introduce the following model: the probability to introduce any of the different innovation strategies y of firm i is a linear function of the firm’s past exporting activity.   yis=c+β EXPOi{t−1}+δXi+μj+νr+ρc+ui, (1) where s= process innovation, brand new product innovation yi indicates whether firm i implemented an innovation s. Therefore I introduce an equation for each of the two possible innovation outputs. EXPOi{t−1} is a dummy equal to 1 if a firm exported in the previous period t−1 and equal to zero if the firm did not export in time t−1. Xi includes a set of firm-level control variables, while μj, νr, and ρc control, respectively, for sector, regional, and country effects. The idiosyncratic error term is denoted by uit. While the literature so far has only focused on the size and sign of the β coefficient of being an exporter, here the hypothesis is that for each firm i the marginal effect of exporting on innovation activities is a linear function of the technological learning effect L and the foreign demand effect D of exporting, which on their turn depend on the specific export destinations of each firm. Accordingly it is possible to write:   βi=γ1Li{t−1}+γ2Di{t−1}. (2) Where for each firm the coefficient of the export dummy depends on the specific impact of the two identified effects. Substituting (2) into (1) I obtain the following specification:   yis=c+γ1(Li{t−1}* EXPOi{t−1})+γ2(Di{t−1}* EXPOi{t−1})+δXi+μj+νr+ρc+ui. (3) To ease the notation, the interaction terms will be simply denoted as TL and TD, suppressing the time indicators, as follows:   yis=c+γ1TL+γ2TD+δ Xi+μj+νr+ρc+ui. (4) Hence the two variables of interest are TL=0 if a firm did not export in t−1 and TL=  Li{t−1} if the firm exported in t−1. Also TD will be equal to zero if a firm did not export in time t−1 and TD=  Di{t−1} if the firm exported in the previous time period. According to the hypotheses spelled out in Section 2.1, the two coefficients γ1 and γ2 are likely to differ according to the type of innovation output considered. 3.2. Data The data used are the EU-EFIGE/Bruegel-UniCredit data set, a unique firm-level database collected within the EFIGE project (European Firms in a Global Economy), coordinated by Bruegel, which includes detailed firm-level information about the destinations of exports and the innovation performances of representative samples of manufacturing firms (with a lower threshold of 10 employees) in seven European countries in the period 2007–2009: the survey includes around 3000 firms for France, Germany, Italy, and Spain, and 2000 for the UK.1 The survey followed a proper stratification strategy of the sample to ensure representativeness of the collected data for each country, on the basis of industries, regions, and size classes; however in the stratification strategy, large firms have been slightly oversampled (Altomonte and Aquilante, 2012). The EFIGE data set is an extremely rich data set with harmonized information across the different countries about firms’ structural information (size; group affiliation; ownership structure), as well as information about the labor force, the innovative investments, and the internationalization strategies. The great advantage of the EFIGE data set is that it has detailed information on both the innovation strategies adopted by firms and on the specific destinations of their exports. It is hence possible to know what type of innovation strategies were implemented by each firm and, for the firms who exported, the main markets of destination of their exports. The main limit of the data set is that it is a cross-section, which makes it more difficult to address causality issues: however even if there is only one observation per firm, the questions concerning exporting activity cover also past years, so they allow to introduce suitable lags in the empirical analysis. 3.2.1 Dependent variables To identify the possible types of innovation strategies, two dependent variables will be used: each of them indicating a specific innovation type, as outlined in Section 2.1. Process innovation is proxied by a dummy that is equal to 1 if a firm introduced a process innovation in the period 2007–2009 and 0 otherwise. Brand new product innovation is proxied by a dummy variable that is equal to 1 if in the period 2007–2009, a firm introduced a product innovation that is new to the market and it also applied for a patent. This specific combination assures that the firm which introduces the product innovation is also able to introduce truly novel technologies. While this definition is very restrictive and cannot be compared with the usual definition of product innovation, it is useful to identify product innovations introduced by leading firms actually able to shift the technological frontier, as proxied by the application for a patent (which is by definition associated with a technological novelty). Accordingly, only a restricted number of highly competitive and technologically advanced firms will be able to introduce this type of product innovations, due to the high research costs. This strategy seems especially appropriate to identify brand new product innovations when domestic and exporting firms are compared, since the simple notion of “product new to the market” does not clearly indicate which markets firms are referring to.2 3.2.2 Independent variables The EFIGE survey asks firms if they export. To the exporters it also asks to indicate their three main export destinations in 2008 and to specify if they were already active in those countries in 2003. To decrease as much as possible the problems of simultaneity, I only consider export destinations in which the firm was already active in the 5 years from 2003 up to 2007: in this way I introduce a relevant time lag between exporting activity (2003–2007) and the period considered for the introduction of innovations (2007–2009). This also allows me to identify long-term export destinations, which have a high degree of persistence for the firms, since usually the positive effect of export on innovation is found for persistent exporters (Andersson and Lööf, 2009) (Figure 1). Figure 1. View largeDownload slide The time lag between exporting and innovation. Figure 1. View largeDownload slide The time lag between exporting and innovation. Combining this information with the sectoral affiliation of each firm, I can build the two main indices that measure the technological learning effect and the foreign demand effect according to the specific export destinations of each firm: to each exporting firm, I associate the level of foreign market growth and technological advancement of the countries in which the firm exports and specifically in the sector in which the firm is active (see Figure 2). The main assumption behind this approach is that the possibility to learn through exporting activity (technological learning) and to benefit from the increase of the foreign markets does not depend on the features and dynamics of the overall economy of the countries of destination, but only by the characteristics of the same sector in which the firm is active. As a matter of example this implies that for a German firm active in the electronic industry which exports to the United States, the technological learning and foreign demand effects will depend on the characteristics and the dynamics of the electronic industry in the United States and not on the overall dynamics of the US economy. This approach seems legitimate, since firms, especially small- and medium-sized firms, are often working in a specific market niche; therefore, the features of the economy at the aggregate level may have little or no influence at all on their economic decisions. Figure 2. View largeDownload slide The construction of the technological learning and foreign demand indices. Figure 2. View largeDownload slide The construction of the technological learning and foreign demand indices. However the advantages of this sectoral strategy increase only up to a certain threshold: if the sectoral disaggregation is too thin, there is the risk to miss important inter-sectoral effects. Indeed a firm necessarily sells outputs to other firms that perform slightly different economic activities along the vertical supply chains. Restricting the sectoral focus too much may result in losing these interactions occurring with foreign buyers. To take into account both these effects, I use the two-digit (ISIC. Rev. 3) sectoral aggregation: this classification distinguishes between manufacturing firms that do completely different economic activities (such as the pharmaceutical industry and the automotive sector), but at the same time, it aggregates across similar economic activities (such as the production of basic chemicals and the production of plastic products). Technological learning effect index. The technological learning effect can be proxied by the level of technological sophistication of the country in which a firm is exporting, in the specific two-digit sector in which the firm is active. The higher is the level of technological advancement of the markets/sectors of destination, the higher will be the possibility for the exporting firm to acquire new knowledge and new useful routines to be eventually incorporated in new products or new processes. The share of Research and Development (R&D) expenditures over the total value added of a sector can be considered a reliable proxy of the general level of technological advancement of a sector in a country.3 For each national sector indicated as a long-term export destination by the firms in the EFIGE sample, I calculate the level of business R&D intensity over value added using data from the OECD-STAN, integrating it with data from the UNIDO and the World Bank: for each country-sector, I use the average value of R&D intensity for the years between 2003–2007. In this way the technological intensity of export destinations corresponds to the period to which firms refer when they indicate their export markets. As shown in Figure 2, the technological learning effect L hence corresponds to the average level of R&D intensity in sector j among the three main countries of destinations d indicated by firm i, conditional on the fact that the firm was already exporting in those markets in 2003.   Li{t−1}=∑d=13(avg R&D  03−07jd)/3, (5) where d = 1, …, 3. Foreign demand effect index. Contrary to the technological learning effect in the literature, there are already some attempts to measure the effect of foreign demand on the innovative performances of exporting firms: Bratti and Felice (2012) use the level of gross domestic product (GDP) per capita of export destinations weighted by the relative distance. Accetturo et al. (2013) instead use import growth as a proxy of the growth of demand. Here I follow the second strategy and build an index that is equal to the average rate of growth of imports in the period 2003–2007 in each specific two-digit sector in the three export destinations of each exporting firm. The data come from COMTRADE and are calculated in US dollars. Since I am only considering long-term export destinations in which firms were already active in 2003 and were still active in 2007, I can be sure that from 2003 to 2007, these firms have been continuously exporting to that specific country c which experienced that rate of growth of imports in sector j. As also shown in Figure 2, the foreign demand effect is:   Di{t−1}=∑d=13(impjd2007−impjd2003)/3, (6) where d = 1, …,3. And imp is the log of imports from country d and sector j in time t. This measure is able to capture the extent to which the markets in which the firm was exporting have grown in the period before the decision to adopt any of the innovative output identified above. Also here I adopt a lag specification to restrict the focus on the sectoral import growth for the period 2003–2007 of the markets in which firms were already operating in 2003. National R&D and market growth. The same level of R&D intensity in a foreign market might have different effects for a firm in a highly advanced country as compared to a firm in a less advanced one. Exporting to the United States might substantially increase knowledge spillovers for an Italian firm active in a low competitive domestic market, but not necessarily for a German firm operating in a very competitive national market. For this reason, I also include the level of R&D intensity in the national two-digit sector of affiliation of each firm (OECD-STANBERD data). For the same reason, I also include as a further control a proxy for the growth of the internal markets, measured by the growth of value added in the national two-digit sector of affiliation of each firm (OECD-STAN data). Structural variables. The model includes controls for structural characteristics of the firms such as employment size, age of the firm, group affiliation, and the type of ownership control (family versus non-family business). Innovative capacity. The innovative capacity of the firm is measured by the share of R&D expenses over turnover in 2007–2009. The level of human capital is controlled by a dummy equal to 1 if the firm has a higher share of graduate employees with respect to the national average (Altomonte, Aquilante, 2012) and a variable that measures the share of employees with a fixed-term contract, assuming that fixed-term contracts are associated to a lower quality of the employees. Internationalization activity. The model also controls if the firm runs part of its production activity in another country through direct investments or through contracts and arms’ length agreements and whether it has foreign affiliates. I also introduce a set of dummies indicating the geographic localization of the main competitors and the level of vertical integration of firms, since the possibility to learn from foreign customers will change a lot if the firm sells directly to final consumers or to other firms. The model also controls for country effects, two-digit sector effects, and regional effects at the nuts-2 level. 3.2.3 Descriptive statistics Table 1 presents the aggregate descriptive statistics of the main variables in the whole sample that includes French, German, Italian, Spanish, and UK firms. As expected the most common innovation output is process innovation, which is adopted by more than 40% of firms, while brand new product innovations, which entail high research costs, are implemented only by 11% of firms. In total, 7% of firms implement both innovative strategies. Firms with up to 50 employees represent the large majority of the overall sample (75%). Only a small share of firms belongs to national or foreign groups, respectively, 13% and 8%. The variables related with internationalization strategies show that only a limited fraction of firms (5%) has foreign direct investments abroad. The majority of firms considers domestic competitors as the most important, followed by European competitors (43%) and competitors in other geographical areas (27%). Table 1. Descriptive statistics Variable  Mean  Standard deviation  Minimum  Maximum  Dependent variables           Process innovation  0.438  0.496  0  1   Brand new product innovation  0.114  0.317  0  1   Process innovation and brand new product innovation  0.070  0.256  0  1  Independent variables           Export activities            Export in 2003  0.411  0.492  0  1    Foreign demand effect  0.058  0.077  −0.164  0.610    Technological learning effect  0.019  0.049  0  0.735   Alternative indexes            Foreign demand effect (2005–2007)  0.063  0.089  −0.844  0.609    Technological learning effect (2005–2007)  0.019  0.049  0  0.732    Foreign demand effect (largest market)  0.057  0.079  −0.307  0.666    Technological learning effect (highest R&D intensity)  0.026  0.066  0  0.735    Foreign demand effect (weighted by export shares)  0.033  0.053  −0.073  0.602    Technological learning effect (weighted by export shares)  0.010  0.030  0  0.735  Structural variables           Labour productivity            TFP            Employment (≤25)  0.470  0.499  0  1    Employment (>25 and ≤50)  0.283  0.450  0  1    Employment (>50 and ≤100)  0.111  0.314  0  1    Employment (>100 and ≤150)  0.041  0.197  0  1    Employment (>150 and ≤250)  0.033  0.180  0  1    Employment (>250 and <500)  0.037  0.189  0  1    Employment (≥500)  0.026  0.160  0  1    Share of fixed-term contracts  26.773  38.902  0  100    Firm age (<6 years)  0.338  0.473  0  1    Firm age (6–20 years)  0.338  0.473  0  1    Firm age (>20 years)  0.594  0.491  0  1    National group  0.137  0.344  0  1    Foreign group  0.081  0.273  0  1    Family member as CEO  0.631  0.482  0  1   Innovative capacities           Share of R&D  0.037  0.076  0  1   Skilled labor force  0.281  0.449  0  1   ICT access  0.914  0.280  0  1  Internationalization variables           Foreign direct investments  0.049  0.216  0  1   Arms’ length foreign production  0.040  0.197  0  1   Domestic affiliates  0.133  0.339  0  1   Foreign affiliates  0.075  0.263  0  1   Domestic competitors  0.855  0.352  0  1   Competitors in EU  0.431  0.495  0  1   Competitors in the United States  0.126  0.332  0  1   Competitors other geo areas  0.273  0.445  0  1   Vertical integration            Sales to order share (1–30%)  0.120  0.325  0  1    Sales to order share (30%–70%)  0.088  0.284  0  1    Sales to order share (>70%)  0.662  0.473  0  1   Domestic effects           Growth of domestic sector  0.116  0.126  −0.646  0.673   R&D intensity domestic sector (%)  3.392  5.996  0.106  51.061  National composition  Number of firms  (%)       France  2723  21.1       Germany  2827  21.91       Italy  2950  22.86       Spain  2728  21.14       UK  1677  12.99      Total number of observations  12,905  100      Variable  Mean  Standard deviation  Minimum  Maximum  Dependent variables           Process innovation  0.438  0.496  0  1   Brand new product innovation  0.114  0.317  0  1   Process innovation and brand new product innovation  0.070  0.256  0  1  Independent variables           Export activities            Export in 2003  0.411  0.492  0  1    Foreign demand effect  0.058  0.077  −0.164  0.610    Technological learning effect  0.019  0.049  0  0.735   Alternative indexes            Foreign demand effect (2005–2007)  0.063  0.089  −0.844  0.609    Technological learning effect (2005–2007)  0.019  0.049  0  0.732    Foreign demand effect (largest market)  0.057  0.079  −0.307  0.666    Technological learning effect (highest R&D intensity)  0.026  0.066  0  0.735    Foreign demand effect (weighted by export shares)  0.033  0.053  −0.073  0.602    Technological learning effect (weighted by export shares)  0.010  0.030  0  0.735  Structural variables           Labour productivity            TFP            Employment (≤25)  0.470  0.499  0  1    Employment (>25 and ≤50)  0.283  0.450  0  1    Employment (>50 and ≤100)  0.111  0.314  0  1    Employment (>100 and ≤150)  0.041  0.197  0  1    Employment (>150 and ≤250)  0.033  0.180  0  1    Employment (>250 and <500)  0.037  0.189  0  1    Employment (≥500)  0.026  0.160  0  1    Share of fixed-term contracts  26.773  38.902  0  100    Firm age (<6 years)  0.338  0.473  0  1    Firm age (6–20 years)  0.338  0.473  0  1    Firm age (>20 years)  0.594  0.491  0  1    National group  0.137  0.344  0  1    Foreign group  0.081  0.273  0  1    Family member as CEO  0.631  0.482  0  1   Innovative capacities           Share of R&D  0.037  0.076  0  1   Skilled labor force  0.281  0.449  0  1   ICT access  0.914  0.280  0  1  Internationalization variables           Foreign direct investments  0.049  0.216  0  1   Arms’ length foreign production  0.040  0.197  0  1   Domestic affiliates  0.133  0.339  0  1   Foreign affiliates  0.075  0.263  0  1   Domestic competitors  0.855  0.352  0  1   Competitors in EU  0.431  0.495  0  1   Competitors in the United States  0.126  0.332  0  1   Competitors other geo areas  0.273  0.445  0  1   Vertical integration            Sales to order share (1–30%)  0.120  0.325  0  1    Sales to order share (30%–70%)  0.088  0.284  0  1    Sales to order share (>70%)  0.662  0.473  0  1   Domestic effects           Growth of domestic sector  0.116  0.126  −0.646  0.673   R&D intensity domestic sector (%)  3.392  5.996  0.106  51.061  National composition  Number of firms  (%)       France  2723  21.1       Germany  2827  21.91       Italy  2950  22.86       Spain  2728  21.14       UK  1677  12.99      Total number of observations  12,905  100      About 40% of firms were already exporting in 2003: for each of them, it was possible to calculate their respective index of technological learning and foreign demand effects—as proxied by the intensity of R&D expenditures and import growth of the sectors and markets in which they were exporting in 2003. 4. Results As in some of the existing literature on export and innovation (Lileeva and Trefler, 2007; Bustos, 2011; Bratti and Felice, 2012), I estimate the equations presented in Section 3.1 with a linear probability model (LPM) instead of nonlinear estimators such as probit or logit. The main reason is that LPM is especially recommended when IVs strategies are implemented (Angrist, 2001), and since in this section I will also introduce IV estimates obtained with LPM, I use this estimator throughout the article to ease the comparisons of the results across different specifications. An additional advantage of LPM with respect to nonlinear estimators is that it yields unbiased and consistent estimates with no assumptions on the distribution of the error term and is therefore suggested if one is only interested in measuring average treatment effects. Of course this choice would not be appropriate were I interested in the specific distribution of outcomes.4 Before estimating the impact of the technological learning and demand effect on firms’ different innovative strategies, I start with the ordinary least squares (OLS) estimation of the LPMs that explain the implementation of the two possible innovation outputs, using the fact of being an exporter in 2003 as the main independent variable. This is a useful benchmark with respect to the previous literature. The results in Table 2 show that indeed exporting activity has always a positive effect on innovation and that the size of such effect is broadly equal for process and brand new product innovations. The results also show that the inclusion of further controls in the model decreases by roughly two-third the coefficient of export activity in both types of innovation. Table 2. The effect of exports on innovation strategies   (1)  (2)  (3)  (4)  (5)  (6)  Process innovation  Brand new product innovation  Export  0.096***  0.052***  0.037***  0.093***  0.054***  0.034***  (0.009)  (0.010)  (0.010)  (0.006)  (0.006)  (0.006)  Share of R&D    0.908***  0.884***    0.742***  0.683***    (0.069)  (0.069)    (0.056)  (0.056)  Skilled labor force    0.048***  0.043***    0.035***  0.028***    (0.010)  (0.010)    (0.006)  (0.006)  National group    0.006  0.002    0.011  0.001    (0.014)  (0.014)    (0.009)  (0.009)  Foreign group    0.006  0.002    0.029**  0.017    (0.018)  (0.018)    (0.014)  (0.014)  Employment (>25 and ≤50)    0.076***  0.072***    0.027***  0.022***    (0.010)  (0.010)    (0.006)  (0.006)  Employment (>50 and ≤100)    0.125***  0.117***    0.074***  0.056***    (0.015)  (0.015)    (0.010)  (0.010)  Employment (>100 and ≤150)    0.156***  0.143***    0.134***  0.104***    (0.023)  (0.023)    (0.018)  (0.018)  Employment (>150 and ≤250)    0.215***  0.202***    0.175***  0.136***    (0.025)  (0.025)    (0.022)  (0.021)  Employment (>250 and <500)    0.177***  0.163***    0.153***  0.100***    (0.025)  (0.025)    (0.020)  (0.020)  Employment (≥500)    0.209***  0.186***    0.251***  0.156***    (0.030)  (0.031)    (0.026)  (0.026)  Firm age (6–20 years)    −0.030  −0.027    −0.010  −0.010    (0.018)  (0.018)    (0.011)  (0.011)  Firm age (>20 years)    −0.041**  −0.039**    −0.016  −0.020*    (0.018)  (0.018)    (0.011)  (0.011)  Family member as CEO    0.023**  0.022**    0.003  0.004    (0.010)  (0.010)    (0.006)  (0.006)  Growth of domestic sector    0.033  0.052    0.040  0.059    (0.071)  (0.070)    (0.048)  (0.048)  R&D intensity domestic sector    0.004**  0.004**    −0.000  −0.000    (0.002)  (0.002)    (0.001)  (0.001)  Share of fixed-term contracts    −0.000  −0.000    −0.000  −0.000    (0.000)  (0.000)    (0.000)  (0.000)  ICT access    0.034**  0.031**    0.021**  0.016**    (0.015)  (0.015)    (0.008)  (0.008)  Domestic affiliates      0.019      0.022**      (0.013)      (0.010)  Foreign affiliates      0.006      0.119***      (0.020)      (0.017)  Foreign direct investments      0.020      0.060***      (0.023)      (0.021)  Arms’ length foreign production      −0.019      0.070***      (0.022)      (0.018)  Sales to order share (1–30%)      −0.013      0.014      (0.017)      (0.011)  Sales to order share (30%–70%)      −0.018      −0.000      (0.019)      (0.012)  Sales to order share (>70%)      0.031**      −0.020**      (0.014)      (0.008)  Domestic competitors      0.031**      −0.032***      (0.013)      (0.009)  Competitors in the United States      0.045***      0.062***      (0.015)      (0.011)  Competitors in EU      0.056***      0.011*      (0.010)      (0.007)  Competitors other geo areas      0.024**      0.001      (0.011)      (0.007)  Constant  0.071  −0.004  −0.030  −0.054  −0.173*  −0.130  (0.168)  (0.205)  (0.214)  (0.037)  (0.096)  (0.096)  Observations  12,905  12,905  12,905  12,905  12,905  12,905  R-squared  0.037  0.073  0.080  0.073  0.138  0.164    (1)  (2)  (3)  (4)  (5)  (6)  Process innovation  Brand new product innovation  Export  0.096***  0.052***  0.037***  0.093***  0.054***  0.034***  (0.009)  (0.010)  (0.010)  (0.006)  (0.006)  (0.006)  Share of R&D    0.908***  0.884***    0.742***  0.683***    (0.069)  (0.069)    (0.056)  (0.056)  Skilled labor force    0.048***  0.043***    0.035***  0.028***    (0.010)  (0.010)    (0.006)  (0.006)  National group    0.006  0.002    0.011  0.001    (0.014)  (0.014)    (0.009)  (0.009)  Foreign group    0.006  0.002    0.029**  0.017    (0.018)  (0.018)    (0.014)  (0.014)  Employment (>25 and ≤50)    0.076***  0.072***    0.027***  0.022***    (0.010)  (0.010)    (0.006)  (0.006)  Employment (>50 and ≤100)    0.125***  0.117***    0.074***  0.056***    (0.015)  (0.015)    (0.010)  (0.010)  Employment (>100 and ≤150)    0.156***  0.143***    0.134***  0.104***    (0.023)  (0.023)    (0.018)  (0.018)  Employment (>150 and ≤250)    0.215***  0.202***    0.175***  0.136***    (0.025)  (0.025)    (0.022)  (0.021)  Employment (>250 and <500)    0.177***  0.163***    0.153***  0.100***    (0.025)  (0.025)    (0.020)  (0.020)  Employment (≥500)    0.209***  0.186***    0.251***  0.156***    (0.030)  (0.031)    (0.026)  (0.026)  Firm age (6–20 years)    −0.030  −0.027    −0.010  −0.010    (0.018)  (0.018)    (0.011)  (0.011)  Firm age (>20 years)    −0.041**  −0.039**    −0.016  −0.020*    (0.018)  (0.018)    (0.011)  (0.011)  Family member as CEO    0.023**  0.022**    0.003  0.004    (0.010)  (0.010)    (0.006)  (0.006)  Growth of domestic sector    0.033  0.052    0.040  0.059    (0.071)  (0.070)    (0.048)  (0.048)  R&D intensity domestic sector    0.004**  0.004**    −0.000  −0.000    (0.002)  (0.002)    (0.001)  (0.001)  Share of fixed-term contracts    −0.000  −0.000    −0.000  −0.000    (0.000)  (0.000)    (0.000)  (0.000)  ICT access    0.034**  0.031**    0.021**  0.016**    (0.015)  (0.015)    (0.008)  (0.008)  Domestic affiliates      0.019      0.022**      (0.013)      (0.010)  Foreign affiliates      0.006      0.119***      (0.020)      (0.017)  Foreign direct investments      0.020      0.060***      (0.023)      (0.021)  Arms’ length foreign production      −0.019      0.070***      (0.022)      (0.018)  Sales to order share (1–30%)      −0.013      0.014      (0.017)      (0.011)  Sales to order share (30%–70%)      −0.018      −0.000      (0.019)      (0.012)  Sales to order share (>70%)      0.031**      −0.020**      (0.014)      (0.008)  Domestic competitors      0.031**      −0.032***      (0.013)      (0.009)  Competitors in the United States      0.045***      0.062***      (0.015)      (0.011)  Competitors in EU      0.056***      0.011*      (0.010)      (0.007)  Competitors other geo areas      0.024**      0.001      (0.011)      (0.007)  Constant  0.071  −0.004  −0.030  −0.054  −0.173*  −0.130  (0.168)  (0.205)  (0.214)  (0.037)  (0.096)  (0.096)  Observations  12,905  12,905  12,905  12,905  12,905  12,905  R-squared  0.037  0.073  0.080  0.073  0.138  0.164  Note: All models are estimated with OLS estimator. All models include country, sector, and region fixed effects. The reference category for firms’ size is less than 25 employees. The reference category for firms’ age is less than 6 years. The reference category for sales to order share is zero. Standard errors in parentheses are clustered at the firm level *** P < 0.01, ** P < 0.05, * P < 0.1. Once acknowledged that also in this sample, exporting activity is positively correlated with innovation, and that this effect is equal for the two innovative strategies identified, I investigate more specifically whether the technological learning and demand effects have a differentiated impact on the two innovative outputs. Indeed it might be that the positive coefficient found for the export dummy in Table 2 is sometimes due to the technological learning effect and sometimes to the foreign demand effect, according to the specific type of innovation considered. In this way I will be able to test the hypotheses spelled out in Section 2.1. In Columns (1)–(4) of Table 3, I substitute the export dummy with the two effects in the process innovation specification. In Column (1) I present the OLS results introducing only the structural controls such as size, belonging to a group and those related with innovation capacity (R&D; skilled labor force; adoption of Information and Communication Technologies – ICT): I find that only the demand index has a positive and significant effect, while the technological learning index is not significantly different from zero. The results do not change when in Column (2) I include the additional controls for internationalization activity and vertical integration: the coefficients decrease slightly, but only the demand effect is positive and significant. As already stated in Section 3.2, the procedure used to build the two indices assumes that firms are affected by the level of foreign demand and foreign technological spillovers of the sectors in which they are active. This might not be an appropriate strategy for large multinational operating in very different market segments: in that case considering only one of the various economic activities in which these firms are active would be highly misleading. To overcome this problem in Column (3), I restrict the sample to the firms that have at least 60% of their sales in one specific economic activity, and in Column (4), instead, I only consider firms with less than 500 employees. The OLS results however do not change: the only positive and significant coefficient is found for the foreign demand effect. Overall the results confirm Hypothesis 1a and reject Hypothesis 1b: the effect of exporting activities on process innovation is only due to the foreign demand effect, while the technological learning effect has no role. Table 3. The effect of technological learning and foreign demand, baseline results   (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  Process innovation   Brand new product innovation   All firms  All firms  No multiproduct firms  No large firms  All firms  All firms  No multiproduct firms  No large firms  Technological learning effect  0.141  0.106  0.123  0.111  0.358***  0.294***  0.267***  0.292***  (0.114)  (0.114)  (0.118)  (0.118)  (0.096)  (0.095)  (0.096)  (0.097)  Demand effect  0.265***  0.181***  0.154**  0.164**  0.282***  0.176***  0.184***  0.167***  (0.067)  (0.068)  (0.070)  (0.069)  (0.046)  (0.047)  (0.048)  (0.046)  Innovative capacity controls  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Internationalization controls  No  Yes  Yes  Yes  No  Yes  Yes  Yes  Observations  12,905  12,905  12,195  12,564  12,905  12,905  12,195  12,564  R-squared  0.073  0.080  0.081  0.080  0.140  0.165  0.166  0.150    (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  Process innovation   Brand new product innovation   All firms  All firms  No multiproduct firms  No large firms  All firms  All firms  No multiproduct firms  No large firms  Technological learning effect  0.141  0.106  0.123  0.111  0.358***  0.294***  0.267***  0.292***  (0.114)  (0.114)  (0.118)  (0.118)  (0.096)  (0.095)  (0.096)  (0.097)  Demand effect  0.265***  0.181***  0.154**  0.164**  0.282***  0.176***  0.184***  0.167***  (0.067)  (0.068)  (0.070)  (0.069)  (0.046)  (0.047)  (0.048)  (0.046)  Innovative capacity controls  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Internationalization controls  No  Yes  Yes  Yes  No  Yes  Yes  Yes  Observations  12,905  12,905  12,195  12,564  12,905  12,905  12,195  12,564  R-squared  0.073  0.080  0.081  0.080  0.140  0.165  0.166  0.150  Note: The LPMs in the table are estimated with OLS estimator and include country, sector, and region fixed effects. The technological learning effect is calculated as the 2003–2007 average level of sectoral R&D of the three export destinations. The demand effect is calculated as the 2003–2007 average sectoral import growth of the three export destinations. In Columns (3) and (7), only firms with at least 60% of their sales in a specific business are included, and in Columns (4) and (8), only firms with less than 500 employees are included. All specifications include the controls displayed in Table 2. Columns (1) and (5) do not include internationalization controls. Standard errors in parentheses are clustered at the firm level *** P < 0.01, ** P < 0.05, * P < 0.1. In Columns (5)–(8) of Table 3, I test the role of the technological learning and demand effect in the brand new innovation specification. In Column (1) I only include structural and innovation-capacity controls, in Column (2) I include all controls, and in Column (3) I select only firms with the majority of sales in one single type of economic activity. Also in this case in Column (4) I exclude firms with more than 500 employees: indeed it could be the case that brand new product innovations are introduced especially by very large firms; therefore, the results might be driven by this specific subset of firms. However across the different specifications, the results are very stable. In this case both effects are positive and significantly different from zero, even if the coefficient of the technological learning effect is larger than the foreign demand effect. The results confirm both Hypothesis 2a and Hypothesis 2b: the effect of exporting activities on brand new product innovation is due to both the foreign demand and the technological learning effect. 4.1. Identification 4.1.1. Selection into exporting: the role of productivity An important threat to identification is represented by the respective levels of productivity of firms. As stressed by the literature that highlights the role of self-selection into exports, more productive firms are much more likely to export, as they can bear the costs associated to exporting activity (Bernard and Jensen, 1999, Bernard and Wagner, 1997). Productivity is likely to be positively correlated both with export and innovation activity, representing a possible confounding factor in the specification of equation (4). Moreover productivity is likely to influence also the specific destination of exports: as shown by previous empirical research (Serti and Tomasi, 2012; Crinò and Epifani, 2012), more productive firms might be able to export to markets with higher levels of R&D intensity (or higher demand growth), leading to a positive correlation with the index of technological learning (or foreign demand effect) and with innovative activities. Based on these considerations I introduce, respectively, labor productivity and total factor productivity (TFP) as further controls in the estimation of equation (4).5 Both measures are provided by Bruegel together with the EFIGE data set: more specifically TFP is computed following the methodology suggested by Levinsohn and Petrin (2003) to estimate output elasticities. To decrease as much as possible problems related to reverse causality in both cases, I use the average productivity for the years prior to the introduction of innovations, i.e. before 2007. More specifically I use the average for the years 2001–2007.6 I introduce the two measures of productivity in Tables 4 and 5 to verify their impact on the two measures of innovation. In Table 4 I check how their inclusion affects the coefficient of the simple export dummy, while in Table 5 I check whether their inclusion influences the coefficients of the technological learning and foreign demand effects. The results in Table 4 show two interesting patterns: while labor productivity is only significant in the process innovation equation, TFP is instead significant only in the brand new product innovation specification. This confirms the fact that the two types of innovation chosen are indeed very different: while process innovations are innovative strategies aimed at increasing the efficiency of the productive processes, which might be associated with investments in machinery that increase capital intensity and labor productivity, brand new product innovations are instead a much more rare innovative strategy that can only be implemented by firms closer to the technological frontier and which display also higher levels of TFP. Table 4. The role of productivity   (1)  (2)  (3)  (4)  Process innovation   Brand new product innovation   All firms  All firms  All firms  All firms  Export  0.035***  0.036***  0.036***  0.036***  (0.010)  (0.010)  (0.006)  (0.006)  Labor productivity  0.036***    0.006    (0.011)    (0.008)    TFP    0.017    0.018**    (0.014)    (0.009)  All controls  Yes  Yes  Yes  Yes  Observations  12,905  12,905  12,905  12,905  R-squared  0.081  0.080  0.164  0.164    (1)  (2)  (3)  (4)  Process innovation   Brand new product innovation   All firms  All firms  All firms  All firms  Export  0.035***  0.036***  0.036***  0.036***  (0.010)  (0.010)  (0.006)  (0.006)  Labor productivity  0.036***    0.006    (0.011)    (0.008)    TFP    0.017    0.018**    (0.014)    (0.009)  All controls  Yes  Yes  Yes  Yes  Observations  12,905  12,905  12,905  12,905  R-squared  0.081  0.080  0.164  0.164  Note: The LPMs in the table are estimated with OLS estimator and include country, sector, and region fixed effects. All specifications include the controls displayed in Table 2. Standard errors in parentheses are clustered at the firm level *** P < 0.01, ** P < 0.05, * P < 0.1. Table 5. The effect of technological learning and foreign demand, controlling for productivity   (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  Process innovation   Brand new product innovation   All firms  All firms  No multiproduct firms  No large firms  All firms  All firms  No multiproduct firms  No large firms  Technological learning effect  0.140  0.106  0.123  0.112  0.353***  0.290***  0.264***  0.290***  (0.114)  (0.114)  (0.118)  (0.118)  (0.096)  (0.095)  (0.096)  (0.097)  Demand effect  0.252***  0.172**  0.145**  0.154**  0.279***  0.182***  0.182***  0.166***  (0.067)  (0.068)  (0.070)  (0.069)  (0.046)  (0.047)  (0.048)  (0.047)  Labor productivity  0.039***  0.037***  0.033***  0.037***  –  –  –  –  (0.011)  (0.011)  (0.012)  (0.012)  –  –  –  –  TFP  –  –  –  –  0.026***  0.018**  0.020**  0.013  –  –  –  –  (0.009)  (0.009)  (0.009)  (0.009)  Innovative capacity controls  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Internationalization controls  No  Yes  Yes  Yes  No  Yes  Yes  Yes  Observations  12,905  12,905  12,195  12,564  12,905  12,905  12,195  12,564  R-squared  0.074  0.081  0.082  0.081  0.141  0.164  0.166  0.150    (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  Process innovation   Brand new product innovation   All firms  All firms  No multiproduct firms  No large firms  All firms  All firms  No multiproduct firms  No large firms  Technological learning effect  0.140  0.106  0.123  0.112  0.353***  0.290***  0.264***  0.290***  (0.114)  (0.114)  (0.118)  (0.118)  (0.096)  (0.095)  (0.096)  (0.097)  Demand effect  0.252***  0.172**  0.145**  0.154**  0.279***  0.182***  0.182***  0.166***  (0.067)  (0.068)  (0.070)  (0.069)  (0.046)  (0.047)  (0.048)  (0.047)  Labor productivity  0.039***  0.037***  0.033***  0.037***  –  –  –  –  (0.011)  (0.011)  (0.012)  (0.012)  –  –  –  –  TFP  –  –  –  –  0.026***  0.018**  0.020**  0.013  –  –  –  –  (0.009)  (0.009)  (0.009)  (0.009)  Innovative capacity controls  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Internationalization controls  No  Yes  Yes  Yes  No  Yes  Yes  Yes  Observations  12,905  12,905  12,195  12,564  12,905  12,905  12,195  12,564  R-squared  0.074  0.081  0.082  0.081  0.141  0.164  0.166  0.150  Note: The LPMs in the table are estimated with OLS estimator and include country, sector, and region fixed effects. The technological learning effect is calculated as the average level of sectoral R&D of the three export destinations. The demand effect is calculated as the average sectoral import growth of the three export destinations. In Columns (3) and (7), only firms with at least 60% of their sales in a specific business are included, and in Columns (4) and (8), only firms with less than 500 employees are included. All specifications include the controls displayed in Table 2. Columns (1) and (5) do not include internationalization controls. Standard errors in parentheses are clustered at the firm level *** P < 0.01, ** P < 0.05, * P < 0.1. After this preliminary check of the relative impact of the two productivity measures in Table 4, in Table 5, I only include labor productivity in the process innovation equation and only TFP in the brand new product innovation equation. The results of both tables show that the introduction of the productivity measures does not significantly alter the size and significance of the coefficients of the technological learning and foreign demand effects. 4.1.2. Instrumental variable strategy Even after controlling for the levels of productivity, the results obtained with the OLS specification cannot be considered free from endogeneity issues. Di{t−1} and Li{t−1}, i.e. the indices of imports growth and of technological development of the main countries of export destination for each firm, might be endogenous because the choice of a firm to export in a specific country d is not random: firms chose strategically the destination of their exports. The specification chosen, which considers only innovation activities performed between 2007 and 2009 and export destinations in which firms were present before (from 2003 to 2007), allows to diminish to a certain extent the impact of reverse causality bias; however it does not allow to rule it out completely. Past innovative activities can still have an effect on firms’ export destinations, even if there are no clear expectations ex ante about the direction of such effect. Moreover, as also shown by the literature that explores the selection into different types of export markets (Blanes-Cristóbal et al. 2008; Serti and Tomasi, 2012), the fact that a firm exported to a specific country (with a specific demand growth and technological development level) in 2003–2007 might be related to the existence of unobservables, such as managerial ability or past international experience, that I might not be able to control for, since the sample is a cross-section. I try to mitigate these endogeneity problems through an IVs strategy, which builds on similar methodologies already used in the literature (Bratti and Felice, 2012), but which also introduces some novelties. The identification strategy relies on the average propensity of firms in a certain national sector to export toward specific destinations. For each national sector, I create a fictional “representative” exporter, which exports to the most common export destinations among the firms in that sector, and I build two new indexes of technological learning and foreign demand effects that are created using the R&D intensity and import growth of the most common export destinations among the firms in a sector-country. These can be considered as the indexes of a “representative” exporter in the same sector and country of the focal firm. The rationale behind this strategy is that in each country a firm will be more likely to export to the market destinations that are common among the other firms active in the same national sector, so the values of the indexes of technological learning and foreign demand of the “representative” exporter should be correlated with those of the focal firm: if the majority of exports in the German electronics industry is toward, say, the United States, France, and Italy, it is also likely that an exporting firm in the German electronics sector will export to these countries. The exclusion restriction instead is that this average measure of R&D intensity (or import growth) for the whole sector-country will not be correlated with the focal firm’s idiosyncratic innovativeness.7 Taking advantage of OECD trade data (STAN Bilateral Trade in Goods by Industry and End-use), I retrieved for each national sector the aggregate flow of exports to each country and selected the 25 most common destinations in the period 2003–2007. On the basis of these data, I built the two new indexes of technological learning and foreign demand as an average of import growth and of R&D intensity for the 25 most common export market destinations (weighted by their relative importance) of each national sector. A possible threat to identification could arise if in a specific country-sector all exporters were equally very productive and innovative: in this case the technological learning and foreign demand indexes calculated for the “representative” firm might still be correlated with the innovativeness of the focal firm. However, calculating the representative exporter at the (two-digit) sector-country aggregation level should allow for a substantial degree of intra-sectoral heterogeneity among firms. Indeed, as suggested by Dosi, Lechevalier, and Secchi (2010), within a sector, the heterogeneity of firms’ productivity levels is always very high. Also Melitz (2003) stresses that even within the same industry, there are substantial differences in productivity and export performances among firms. Therefore it seems unlikely that the level of R&D intensity and import growth, calculated using the most common export destinations among all the firms of a national sector, can be correlated with the idiosyncratic error term of the focal firm. Since the relationship between this instrument and the actual behavior of firms is likely to be not linear, I introduce some further factors that are supposed to determine heterogeneous responses by firms to the instrument. The first factor is the regional propensity to export: the probability that a firm exports in the same market destinations of the average firm in its own national sector also depends on the general propensity to export of the firm’s region, since this propensity varies quite a lot among regions in the same countries. Another factor that is likely to diminish the ability of the instrument to explain firms’ export choices is the size of firms: very small firms will have in general a lower ability to export, regardless of the sectoral averages, since they face relevant obstacles to access foreign markets, represented by sunk and information costs. On the basis of these preliminary considerations, I build the following instrument:   T^L=∑d=125wjLjdmr.Ljd is the level of technological development proxied by the R&D intensity of the 25 most-common d country-destinations of exports for the sector j in the specific European country in which the firm is active (France, Germany, Italy, Spain, or UK). wj is the share of export to each of the 25 most common destinations of exports over the total exports of national sector j in the period 2003–2007. mr is the share of exporters in each region. Finally to account for firm-size effects, an additional instrument will be added in which T^Li is multiplied by a dummy (0/1) equal to 1 if a firm’s number of employees is equal or lower than 25. The same procedure is used to instrument the foreign demand effect index TD:   T^D=∑d=125wjDjdmr. In this case Djd is the growth of imports between 2003 and 2007 of the 25 most-common c country-destinations of exports for the sector j in the specific country in which the firm is active. Also in this case T^D is multiplied by a dummy (0/1) equal to 1 if a firm employment is equal or lower than 25 employees. In Table 6 I present the results obtained with this IV strategy, using a two-stage least squares (2SLS) estimator. In Columns (1)–(4), I present the process innovation specification in which both the foreign demand and technological learning indices are instrumented by the instruments built with the national sectoral propensities. The results in Columns (1) and (2) in which I use the whole sample and progressively include internationalization controls show that only the foreign demand effect is positive and significant. Moreover the coefficient becomes larger than the one found in the OLS estimates, pointing to a downward bias in those estimates. The results also hold when I only consider firms with a dominant business activity and firms with less than 500 employees in Columns (3) and (4). The first-stage F-statistics of the two instrumented variables, reported in the lower part of Table 6, are always greater than 10, i.e. above the usual threshold identified by the weak instruments literature (Bound et al., 1995, see also Table A1 for first-stage regressions). Moreover the Hansen test on overidentifying restrictions shows that the instruments are exogenous to the error term and correctly excluded from the main regression. Table 6. Instrumental variables   (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  Process innovation   Brand new product innovation   All firms  All firms  No multiproduct firms  No large firms  All firms  All firms  No multiproduct firms  No large firms  Technological learning effect  0.583  0.603  0.617  0.304  1.860***  1.819***  1.825***  1.697***  (0.682)  (0.707)  (0.764)  (0.729)  (0.533)  (0.545)  (0.600)  (0.555)  Demand effect  2.072**  2.151**  2.107**  1.748*  0.739  0.722  0.962  0.328  (0.871)  (0.972)  (0.950)  (0.989)  (0.544)  (0.602)  (0.604)  (0.602)  Labor productivity  0.018  0.019  0.014  0.022  –  –  –  –  (0.015)  (0.015)  (0.015)  (0.015)  –  –  –  –  TFP  –  –  –  –  0.012  0.009  0.009  0.008  –  –  –  –  (0.011)  (0.010)  (0.011)  (0.010)  Innovative capacity controls  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Internationalization controls  No  Yes  Yes  Yes  No  Yes  Yes  Yes                    IV first stage                  F-statistics                  Technological learning effect  13.29  12.61  10.63  11.46  13.29  12.61  10.63  11.46  Demand effect  13.23  11.88  12.45  11.07  13.23  11.88  12.45  11.07  Number of instruments  4  4  4  4  4  4  4  4  Hansen J-statistics  0.014  0.015  0.077  0.003  1.111  1.259  3.370  1.197  P-value  0.993  0.992  0.962  0.998  0.573  0.532  0.185  0.549  Observations  12,905  12,905  12,195  12,564  12,905  12,905  12,195  12,564    (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  Process innovation   Brand new product innovation   All firms  All firms  No multiproduct firms  No large firms  All firms  All firms  No multiproduct firms  No large firms  Technological learning effect  0.583  0.603  0.617  0.304  1.860***  1.819***  1.825***  1.697***  (0.682)  (0.707)  (0.764)  (0.729)  (0.533)  (0.545)  (0.600)  (0.555)  Demand effect  2.072**  2.151**  2.107**  1.748*  0.739  0.722  0.962  0.328  (0.871)  (0.972)  (0.950)  (0.989)  (0.544)  (0.602)  (0.604)  (0.602)  Labor productivity  0.018  0.019  0.014  0.022  –  –  –  –  (0.015)  (0.015)  (0.015)  (0.015)  –  –  –  –  TFP  –  –  –  –  0.012  0.009  0.009  0.008  –  –  –  –  (0.011)  (0.010)  (0.011)  (0.010)  Innovative capacity controls  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Internationalization controls  No  Yes  Yes  Yes  No  Yes  Yes  Yes                    IV first stage                  F-statistics                  Technological learning effect  13.29  12.61  10.63  11.46  13.29  12.61  10.63  11.46  Demand effect  13.23  11.88  12.45  11.07  13.23  11.88  12.45  11.07  Number of instruments  4  4  4  4  4  4  4  4  Hansen J-statistics  0.014  0.015  0.077  0.003  1.111  1.259  3.370  1.197  P-value  0.993  0.992  0.962  0.998  0.573  0.532  0.185  0.549  Observations  12,905  12,905  12,195  12,564  12,905  12,905  12,195  12,564  Note: The LPMs in the table are estimated with a 2SLS estimator and include country, sector, and region fixed effects. The technological learning effect is calculated as the average level of sectoral R&D intensity of the three main export destinations. The demand effect is calculated as the average sectoral import growth of the three main export destinations. In Columns (3) and (7), only firms with at least 60% of their sales in a specific business are included, and in Columns (4) and (8), only firms with less than 500 employees are included. All specifications include the controls displayed in Table 2. Columns (1) and (5) do not include internationalization controls. Standard errors in parentheses are clustered at the firm level *** P < 0.01, ** P < 0.05, * P < 0.1 In Columns (5)–(8) of Table 6, I implement the same IV strategy in the brand new product innovation equation. In this case, I find that in all specifications only the technological learning index is positive and significant, while the foreign demand index becomes not significantly different from zero. These results suggest that, while there certainly is a positive correlation between foreign demand growth and brand new product innovation, such a relationship is probably not a causal one. Summing up the IV results confirm the initial OLS results with one important exception: the foreign demand effect is not anymore found to foster brand new product innovations. Hence only Hypothesis 1a and Hypothesis 2a are supported when the IV procedure is implemented. 4.1.3. Alternative measurements of the technological learning and foreign demand indexes The results presented so far could also be sensitive to the procedure chosen for the measurement of the technological learning and foreign demand effect. For this reason in this section, I introduce alternative measures for the two indexes of technological learning and foreign demand. First of all, it could be the case that the innovative strategies put in place by exporting firms in the period 2007–2009 are only affected by what happened in firms’ export markets in more recent years, for example in the previous 3 years (2005–2007). For example, only recent increases in the import growth of a specific destination might matter for the strategic innovative decisions of firms exporting to that market. If that is the case measures which take into account R&D intensity and import growth between 2003 and 2007 may contain a lot of information that is not relevant for firms’ strategies and therefore introduce noise. To account for this I replicate the two formulas introduced, respectively, in equations (5) and (6), but now I use the average sectoral R&D intensity and the average growth of imports of the three main markets in the period 2005–2007.8 Another potential shortcoming of the technological learning index is that using the average value of R&D across country destinations might not be very informative to make comparisons across exporting firms. Indeed it is likely that knowledge spillovers and opportunities to learn will proceed only from the firm’s most sophisticated market: using an average value means that if firm A exports to only one advanced market and firm B exports to an equally advanced market and a less advanced market, the average value of the technological learning effect would be lower for firm B. This might not be a legitimate choice, since both companies have the same opportunity to learn from the most advanced market in which they export. Therefore I build an alternative measure of technological learning using only the highest value of R&D intensity in sector j among the three main countries of destinations d, conditional on the fact that the firm was already exporting in that market in 2003.   Li{t−1}=max⁡(R&Ddj{t−1}). (7) A potential flaw of the foreign demand index instead is that all foreign markets are counted as equal, regardless of whether they are very important or marginal for a firm. In other words firms might only react to changes that occur in their most important export market, where most of their foreign sales come from. For this reason I build a second index of foreign demand in which I only consider the sectoral import growth between 2003 and 2007 in the most important market for the exporting firm, conditional on the fact that the firm already exported there in 2003.   Di{t−1}=impjd12007−impjd12003. (8) In equation (8)d1 is the country destination (among the three possible), where the firm exports the highest share of its export. Another way of considering the relative share of each export destination is by creating an index of technological learning and foreign demand effects in which I weight the three main export destinations by their relative share in the company’s overall exports.9 I compute the technological learning index as follows:   Li{t−1}=∑d=13avgR&D03−07d*ωd. (9) Where ωd is the share of export for the country of destination d. The same weights are also applied for the foreign demand effect index:   Di{t−1}=∑d=13(impjd2007−impjd2003)*ωd. (10) In Table 7 I test whether these three alternative ways of calculating the technological learning and foreign demand effects change results obtained through my IV methodology in Table 6 (see Table A2 for the first-stage statistics). The results show that also when I use these alternative measures, the results do not change dramatically. The foreign demand effect is still only positive and significant in the process innovation equation, while the technological learning effect only significantly affects the introduction of brand new product innovations. For what concerns the first two alternative indexes (only 2005–2007 period, largest market and highest level of R&D), the size of the coefficients in Columns (1), (2), (4), and (5) is not significantly different with respect to the previous estimates. In the case of the export-share-weighted measures, in Columns (3) and (6), instead the magnitude of the coefficients increases, but the overall statistical significance and direction of the effects do not change. Overall this suggests that the specific methodology chosen to measure the two indexes does not influence the overall results of the analysis. Table 7. Instrumental variables with alternative indexes   (1)  (2)  (3)  (4)  (5)  (6)  Process innovation   Brand new product innovation   Period 2005–2007  Largest market highest R&D  Weighted export shares  Period 2005–2007  Largest market highest R&D  Weighted export shares  Technological learning effect  1.012  0.375  1.044  1.641***  1.310***  3.140***  (0.775)  (0.509)  (1.292)  (0.535)  (0.390)  (1.073)  Demand effect  2.635***  2.027**  3.618**  −0.141  0.674  0.926  (0.991)  (0.937)  (1.749)  (0.594)  (0.587)  (1.098)  Labor productivity  0.010  0.022  0.020  –  –  –  (0.016)  (0.014)  (0.015)  –  –  –  TFP  –  –  –  0.015  0.010  0.011  –  –  –  (0.010)  (0.010)  (0.010)  All controls  Yes  Yes  Yes  Yes  Yes  Yes  IV first stage              F-statistics              Technological learning effect  12.14  14.35  6.85  12.21  14.43  6.87  Demand effect  12.20  11.14  6.81  11.86  10.71  6.54  Number of instruments  4  4  4  4  4  4  Hansen J-statistics  0.744  0.002  0.198  1.584  1.330  1.749  P-value  0.689  0.998  0.905  0.452  0.514  0.417  Observations  12,905  12,849  12,905  12,905  12,849  12,905    (1)  (2)  (3)  (4)  (5)  (6)  Process innovation   Brand new product innovation   Period 2005–2007  Largest market highest R&D  Weighted export shares  Period 2005–2007  Largest market highest R&D  Weighted export shares  Technological learning effect  1.012  0.375  1.044  1.641***  1.310***  3.140***  (0.775)  (0.509)  (1.292)  (0.535)  (0.390)  (1.073)  Demand effect  2.635***  2.027**  3.618**  −0.141  0.674  0.926  (0.991)  (0.937)  (1.749)  (0.594)  (0.587)  (1.098)  Labor productivity  0.010  0.022  0.020  –  –  –  (0.016)  (0.014)  (0.015)  –  –  –  TFP  –  –  –  0.015  0.010  0.011  –  –  –  (0.010)  (0.010)  (0.010)  All controls  Yes  Yes  Yes  Yes  Yes  Yes  IV first stage              F-statistics              Technological learning effect  12.14  14.35  6.85  12.21  14.43  6.87  Demand effect  12.20  11.14  6.81  11.86  10.71  6.54  Number of instruments  4  4  4  4  4  4  Hansen J-statistics  0.744  0.002  0.198  1.584  1.330  1.749  P-value  0.689  0.998  0.905  0.452  0.514  0.417  Observations  12,905  12,849  12,905  12,905  12,849  12,905  Note: The LPMs in the table are estimated with a 2SLS estimator and include country, sector, and region fixed effects. In Columns (1) and (4), the technological learning and foreign demand effects are calculated, respectively, as the average level of sectoral R&D in the period 2005–2007 and as the average sectoral import growth in the period 2005–2007 of the three main export destinations. In Columns (2) and (4), the technological learning effect is calculated as the highest level of sectoral R&D among the three main export destinations, while the demand effect is calculated as the sectoral import growth of the most important market (for the firm) among the three main export destinations. In Columns (3) and (6), the technological learning and foreign demand effects are calculated, respectively, as the average level of sectoral R&D in the period 2003–2007 and as the average sectoral import growth in the period 2003–2007 weighted for the firms’ share of export to each of the three main export destinations. Standard errors in parentheses are clustered at the firm level *** P < 0.01, ** P < 0.05, * P < 0.1. 5. Conclusions This article shows that the positive effect of export activity on innovation among European firms is strictly dependent on the specific export destinations of firms. While export destinations might differ along many dimensions, the article considers two of them in particular: the availability of foreign technological spillovers (the technological learning effect) and the access to foreign expanding markets (the foreign demand effect). The technological learning effect affects firms’ innovation strategies because it provides knowledge spillovers from foreign customers or competitors in very technologically advanced markets: this is important especially because it reduces the relevant internal research costs needed to develop brand new product innovations. On the contrary the foreign demand effect of exporting activities affects firms’ innovation strategies by increasing the potential output of firms, a factor that is often associated with process innovation. In the article I build two indices that are able to proxy these two effects through the use of R&D intensity data at the sectoral level of the destination countries (the technological learning effect) and the growth of sectoral imports of the destination countries (the foreign demand effect). I introduce these two indices in a LPM that explains the adoption of, respectively, brand new product and process innovations by European firms included in the EFIGE data set. The econometric strategy controls for different measures of productivity among the control variables and introduces an IV approach to address possible endogeneity issues. Moreover it also implements a number of robustness checks to control whether the way the two indexes are built affects the results. The results show that indeed the technological learning effect has a positive effect on the introduction of brand new product innovations, while the demand effect of exporting activity mainly induces process innovations. From a theoretical point of view, the results show that export destinations—a factor which has been so far mostly neglected in the existing literature—are instead an important moderator factor able to explain more precisely the relationship between exporting activity and innovation outcomes. Indeed not all export destinations exert the same effect on innovation: acknowledging this evidence could help future work in understanding better when and why exporting firms are able to increase their innovativeness. From a managerial perspective, the results of this article suggest that firms might also choose strategically the destination of their export, not only to increase their level of sales but also to upgrade their competences and capabilities. Moreover managers should be aware that exporting per se might not be sufficient to improve firms’ capabilities: on the contrary some export destinations might be much more effective than others to foster specific types of innovation outputs. Firms might choose export destinations also on the basis of their specific needs in terms of innovation competences. From a policy perspective, it seems important to acknowledge that the positive effect of exporting activity also depends on the specific export destinations. If in a specific country firms export mainly to expanding markets with little levels of technological development, they might be induced to put less efforts in developing truly innovative products. Since in advanced economies brand new product innovations—able to actually shift the world technological frontier—are those with the highest economic impact, exporting only to expanding markets might hinder the ability of firms to develop their future innovative capacities. Finally, while this article provides an interesting perspective on the role of export destinations for firms’ innovation strategies, it must be stressed that it only analyzes two possible dimensions along which export destinations might differ: however other dimensions, such as geographical or cultural distance, institutional settings, specific (environmental) regulations might also play a role in the causal relationship going from export to innovation. Future research should address these interesting additional perspectives. Funding The author has benefitted from the access to the EU-EFIGE/Bruegel-UniCredit database, managed by Bruegel and funded by the European Union’s Seventh Framework Programme ([FP7/2007-2013] under grant agreement number 225551), as well as by UniCredit. Footnotes 1 The data collection has been performed in early 2010 through a questionnaire submitted to the firms. The data set includes also information on Austrian and Hungarian firm which are not used in this analysis, since firms in small and open economies might show different dynamics with respect to firms in large European countries. 2 A non-exporting firm which only knows its domestic market might consider a new product with little innovative content as new to the market. On the contrary a highly competitive and internationalized firm operating in different foreign markets might consider a very innovative product as not new to the market because in some other markets, it might have been already introduced by another leading competitor. 3 An alternative would be the number of patent applications by national firms in each specific sector. However this approach is not straightforward because it is necessary to match firms’ sectorial classifications with the technological classes of patents. 4 The results obtained with probit and logit estimations are perfectly in line with the LPM results presented here both in terms of significance and of magnitude of the coefficients. 5 Both measures have advantages and disadvantages: labor productivity is a more straightforward index of efficiency, which does not rely on any assumption about firms’ production functions; however, in the absence of a measure of the stock of capital (due to the nonavailability of this measure for EFIGE data), it might also proxy the capital intensity of each firm. On the contrary TFP allows to account for all the inputs used by firms—and hence might be a better proxy for the technological sophistication of a firm—however, it also requires to estimate the output elasticities of firms on the basis of some assumptions about the functional form of firms’ production functions. 6 The results are robust to the use of productivity measures also for more recent years (such as 2008). The measures of productivity in the EFIGE data provided by Bruegel are not available for about 25% of the firms, with slightly higher shares in Germany and the UK, due to the nonavailability of ORBIS balance sheet data for such firms. For these firms I imputed their labor productivity and TFP with the following methodology: I estimated through OLS the determinants of both productivity measures in two separate regressions, using the rich set of independent variables included in equation (4) and including all the firms (Rubin, 1987). Then I calculated the predicted values for both productivity measures and substituted the predicted values only for the firms that had missing values for labor productivity (3823 firms) and TFP (3300 firms). A careful analysis of the distribution of imputed and real productivity measures (for the firms for which productivity was instead available) did not show any significant differences in terms of mean and skewness of the variables. Moreover results obtained without including the firms with missing productivity measures did not show significant differences. While this imputation strategy is not free of limitations, as it computes productivity on the basis of the average contribution of a set of other variables, it allows to maintain the representativeness of the sample for the European countries analyzed in the study. 7 This methodology borrows from Bratti and Felice (2012) the idea of using a specific feature of a foreign destination market and calculate an average across destinations, using as weights the export shares to each destination of all the firms active in the sector of the focal firm. However, differently from Bratti and Felice (2012), the instrument includes R&D intensity and imports growth rather than GDP per capita, and it is computed at the national sector level, and not at the sector-province level: the larger level of aggregation adds to the probability that the instrument is exogenous to the focal firms’ innovativeness. 8 To instrument these two new variables, I use the same IV methodology as in the previous specifications in Table (6), with the only difference that also the instruments will be computed for the period 2005–2007. 9 This information is available in EFIGE, since for each of three main export destinations, every firm was also asked what was the share of exports accounted for by each destination. Acknowledgments The author is grateful to Marcello Messori, Cristiano Antonelli and Davide Castellani for useful comments and suggestions. The article has benefited from comments received during the following seminars and conference presentations: Annual meeting of the Academy of International Business, New Orleans, July 2016; DRUID Conference, Rome, June 2015; EMAEE Conference, Maastricht, June 2015; Explaining Economic Change Workshop, Rome, November 2014. The author is also grateful to Bruegel for the access to the quasi full set EFIGE database at their premises. References Accetturo A., Bugamelli M., Lamorgese A. R.. ( 2013), ‘ Skill upgrading and exports,’ Economics Letters , 121( 3), 417– 420. Google Scholar CrossRef Search ADS   Altomonte C., Aquilante T.. ( 2012), ‘The EU-EFIGE/Bruegel-Unicredit dataset,’ Bruegel Working Paper, October 2012: Bruegel, Brussels. Andersson M., Lööf H.. ( 2009), ‘ Learning-by-exporting revisited: the role of intensity and persistence,’ Scandinavian Journal of Economics , 111( 4), 893– 916. Google Scholar CrossRef Search ADS   Angrist J. ( 2001), ‘ Estimation of limited dependent variable models with dummy endogenous regressors: simple strategies for empirical practice,’ Journal of Business and Economic Statistics , 19( 1), 2– 16. Google Scholar CrossRef Search ADS   Basile R.( 2001), ‘ Export behavior of Italian manufacturing firms over the Nineties: the role of innovation,’ Research Policy , 30 ( 8), 1185– 1201. Google Scholar CrossRef Search ADS   Bernard, A. B. and J. Wagner (1997), ‘Exports and Success in German Manufacturing’, Weltwirtschaftliches Archiv/Review of World Economics, 133(1), 134–57. Bernard A. B., Jensen B. J.. ( 1999), ‘ Exceptional exporter performance: cause, effect, or both?,’ Journal of International Economics , 47( 1), 1– 25. Google Scholar CrossRef Search ADS   Blanes-Cristóbal J. V., Dovis M., Milgram-Baleix J., Moro-Egido A. I.. ( 2008), ‘ Do sunk exporting costs differ among markets? Evidence from Spanish manufacturing firms,’ Economics Letters , 101, 110– 112. Google Scholar CrossRef Search ADS   Bound J., Jaeger D., Baker R.. ( 1995), ‘ Problems with instrumental variables estimation when the correlation between instruments and the endogenous explanatory variables is weak,’ Journal of American Statistical Association , 90( 430), 443– 450. Bratti M., Felice G.. ( 2012), ‘ Are exporters more likely to introduce product innovations?,’ The World Economy , 35( 11), 1559– 1598. Google Scholar CrossRef Search ADS   Brouwer E., Kleinknecht A..( 1999), ‘ Keynes-Plus? Effective demand and changes in firm-level R&D: an empirical note,’ Cambridge Journal of Economics , 23( 3), 385– 391. Google Scholar CrossRef Search ADS   Bustos P. ( 2011), ‘ Trade liberalization, exports and technology upgrading: evidence on the impact of MERCOSUR on argentinean firms,’ American Economic Review , 101( 1), 304– 340. Google Scholar CrossRef Search ADS   Cassiman B., Golovko E., Martínez-Ros E..( 2010), ‘ Innovation, exports and productivity,’ International Journal of Industrial Organization , 28( 4), 372– 376. Google Scholar CrossRef Search ADS   Cohen W. M., Klepper S.. ( 1996), ‘ Firm size and the nature of innovation within industries: the case of process and product R&D,’ The Review of Economics and Statistics , 78( 2), 232– 243. Google Scholar CrossRef Search ADS   Crespi G., Criscuolo C., Haskel J.. ( 2008), ‘ Productivity, exporting, and the learning-by-exporting hypothesis: direct evidence from UK firms,’ Canadian Journal of Economics , 41( 2), 619– 638. Google Scholar CrossRef Search ADS   Crinò, R. and P. Epifani, (2012), ‘Productivity, Quality and Export Behaviour,’ Economic Journal, 122(565), 1206–1243. Damijan J. P., Kostevc C., Polanec S.. ( 2010), ‘ From innovation to exporting or vice versa?,’ The World Economy , 33( 3), 374– 398. Google Scholar CrossRef Search ADS   Desmet K., Parente S. L.. ( 2010), ‘ Bigger is better: market size, demand elasticity, and innovation,’ International Economic Review , 51( 2), 319– 333. Google Scholar CrossRef Search ADS   Dosi, G., S. Lechevalier and A. Secchi. (2010), ‘Interfirm heterogeneity: nature, sources and consequences for industrial dynamics. An introduction’, Industrial and Corporate Change, 19, 1867. Dosi G., Grazzi M., Moschella D.. ( 2015), ‘ Technology and costs in international competitiveness: From countries and sectors to firms,’ Research Policy , 44( 10), 1795– 1814. Google Scholar CrossRef Search ADS   Fafchamps M., El Hamineb S., Zeufackc A.. ( 2008), ‘ Learning to export: evidence from moroccan manufacturing,’ Journal of African Economies , 17( 2), 305– 355. Google Scholar CrossRef Search ADS   Fagerberg J. ( 1988), ‘ International competitiveness,’ The Economic Journal , 98 ( 391), 355– 374. Google Scholar CrossRef Search ADS   Fitjar R. D., Rodriguez-Pose A.. ( 2012), ‘ Firm collaboration and modes of innovation in Norway,’ Research Policy , 42( 1), 128– 138. Google Scholar CrossRef Search ADS   Garcia-Quevedo J., Pellegrini G., Savona M.. ( 2017), ‘Reviving demand-pull perspectives: the effect of demand uncertainty and stagnancy on R&D strategy,' Cambridge Journal of Economics, 41(4), 1087–1122. Geroski P. A., Walters C. F.. ( 1995), ‘ Innovation activity over the business cycle,’ The Economic Journal , 105, 916– 928. Google Scholar CrossRef Search ADS   Golovko E., Valentini G.. ( 2014), ‘ Selective learning-by-exporting: firm size and product versus process innovation,’ Global Strategy Journal , 4, 161– 180. Google Scholar CrossRef Search ADS   Harirchi G., Chaminade C.. ( 2014), ‘ Exploring the relation between the degree of novelty of innovations and user-producer interaction across different income regions,’ World Development , 57, 19– 31. Google Scholar CrossRef Search ADS   Hall B., Mairesse J., Branstetter L., Crepon B.. ( 1999), ‘Does cash flow cause investment and R&D? An exploration using panel data for French, Japanese and United States scientific firms,’ in Audretsch D., Thurik R. (eds), Innovation, Industry Evolution and Employment . Cambridge University Press: Cambridge. Laursen K. ( 2011), ‘ User–producer interaction as a driver of innovation: costs and advantages in an open innovation model,’ Science and Public Policy , 38( 9), 713– 723. Google Scholar CrossRef Search ADS   Levinsohn J., Petrin A..( 2003), ‘ Estimating production functions using inputs to control for unobservables,’ The Review of Economic Studies , 70 ( 2), 317– 341. Google Scholar CrossRef Search ADS   Lileeva A., Trefler D.. ( 2010), ‘ Improved access to foreign markets raises plant-level productivity … for some plants,’ The Quarterly Journal of Economics , 125( 3), 1051– 1099. Google Scholar CrossRef Search ADS   Liu X., Buck T.. ( 2007), ‘ Innovation performance and channels for international technology spillovers: evidence from Chinese high-tech industries,’ Research Policy , 36( 3), 355– 366. Google Scholar CrossRef Search ADS   Malerba F., Nelson R. R., Orsenigo L., Winter S. G.. ( 2007), ‘ Demand, innovation and the dynamics of market structure: the role of experimental users and diverse preferences,’ Journal of Evolutionary Economics , 17( 4), 371– 399. Google Scholar CrossRef Search ADS   Melitz M. ( 2003), ‘ The impact of trade in intra-industry reallocations and aggregate industry productivity,’ Econometrica , 71( 6), 1695– 1725. Google Scholar CrossRef Search ADS   Piva M., Vivarelli M.. ( 2007), ‘ Is demand-pulled innovation equally important in different groups of firms?,’ Cambridge Journal of Economics , 31, 691– 710. Google Scholar CrossRef Search ADS   Posner M. V. ( 1961), ‘ International trade and technical change,’ Oxford Economic Papers , 13( 3), 323– 341. Google Scholar CrossRef Search ADS   Rubin D. B.( 1987), Multiple Imputation for Non Response in Surveys . Wiley: New York. Google Scholar CrossRef Search ADS   Salomon R. M. ( 2006), ‘ Spillovers to foreign market participants: assessing the impact of export strategies on innovative productivity,’ Strategic Organization , 4( 2), 135– 164. Google Scholar CrossRef Search ADS   Salomon R. M., Shaver J. M.. ( 2005), ‘ Learning by exporting: new insights from examining firm innovation,’ Journal of Economics and Management Strategy , 14( 2), 431– 460. Google Scholar CrossRef Search ADS   Scherer F. ( 1991), ‘Changing perspectives on the firms size problem,’ in Acs Z. J., Audretsch D. B. (eds), Innovation and Technological Change. An International Comparison . Harvester Wheatsheaf: New York. Schmookler J. ( 1954), ‘ The level of inventive activity,’ The Review of Economics and Statistics , 36, 183– 190. Google Scholar CrossRef Search ADS   Schmookler J. ( 1966), Invention and Economic Growth . Harvard University Press: Cambridge. Google Scholar CrossRef Search ADS   Serti F., Tomasi C..( 2012), ‘ Self selection among different export markets,’ Economics Letters , 117( 1), 102– 105. Google Scholar CrossRef Search ADS   Trofimenko N. ( 2008), ‘ Learning by Exporting: does it matter where one learns? Evidence from Colombian manufacturing firms,’ Economic Development and Cultural Change , 56( 4), 871– 894. Google Scholar CrossRef Search ADS   Van Biesebroeck J. ( 2005), ‘ Exporting raises productivity in sub-Saharan African manufacturing firms,’ Journal of International Economics , 67( 2), 373– 391. Google Scholar CrossRef Search ADS   Verhoogen E. ( 2008), ‘ Trade, quality upgrading and wage inequality in the Mexican manufacturing sector,’ Quarterly Journal of Economics , 123( 2), 489– 530. Google Scholar CrossRef Search ADS   von Hippel E. ( 1986), ‘ Lead users: a source of novel product concepts,’ Management Science , 32, 791– 805. Google Scholar CrossRef Search ADS   von Hippel E. ( 2005), Democratizing Innovation . MIT Press: Cambridge, MA. Wakelin K. ( 1998), ‘ Innovation and export behaviour at the firm level,’ Research Policy , 26( 7), 829– 841. Google Scholar CrossRef Search ADS   Woerter M., Roper S.. ( 2010), ‘ Openness and innovation—home and export demand effects on manufacturing innovation: panel data evidence for Ireland and Switzerland,’ Research Policy , 39, 155– 164. Google Scholar CrossRef Search ADS   Appendix: first stage statistics Table A1. First-stage statistics for Table 6 First stage  (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  Technological learning effect  Foreign demand effect  TL  1.295***  1.267***  1.244***  1.260***  −0.224*  −0.286**  −0.287**  −0.243*  (0.208)  (0.206)  (0.214)  (0.209)  (0.135)  (0.134)  (0.139)  (0.136)  TL*emp≤25  −0.504***  −0.505***  −0.445***  −0.434***  0.206***  0.201***  0.216***  0.206***  (0.120)  (0.119)  (0.125)  (0.120)  (0.070)  (0.069)  (0.073)  (0.070)  TD  0.185  0.143  0.122  0.149  1.310***  1.178***  1.273***  1.127***  (0.128)  (0.127)  (0.132)  (0.131)  (0.256)  (0.247)  (0.253)  (0.253)  TD*emp≤25  0.050  0.063  0.050  0.044  −0.519***  −0.475***  −0.477***  −0.463***  (0.050)  (0.050)  (0.051)  (0.050)  (0.095)  (0.092)  (0.094)  (0.094)  Productivity measures  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Structural and innovation controls  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Internationalization controls  No  Yes  Yes  Yes  No  Yes  Yes  Yes  F-statistics  13.29  12.61  10.63  11.46  13.23  11.88  12.45  11.07  Observations  12,905  12,905  12,195  12,564  12,905  12,905  12,195  12,564  First stage  (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  Technological learning effect  Foreign demand effect  TL  1.295***  1.267***  1.244***  1.260***  −0.224*  −0.286**  −0.287**  −0.243*  (0.208)  (0.206)  (0.214)  (0.209)  (0.135)  (0.134)  (0.139)  (0.136)  TL*emp≤25  −0.504***  −0.505***  −0.445***  −0.434***  0.206***  0.201***  0.216***  0.206***  (0.120)  (0.119)  (0.125)  (0.120)  (0.070)  (0.069)  (0.073)  (0.070)  TD  0.185  0.143  0.122  0.149  1.310***  1.178***  1.273***  1.127***  (0.128)  (0.127)  (0.132)  (0.131)  (0.256)  (0.247)  (0.253)  (0.253)  TD*emp≤25  0.050  0.063  0.050  0.044  −0.519***  −0.475***  −0.477***  −0.463***  (0.050)  (0.050)  (0.051)  (0.050)  (0.095)  (0.092)  (0.094)  (0.094)  Productivity measures  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Structural and innovation controls  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Yes  Internationalization controls  No  Yes  Yes  Yes  No  Yes  Yes  Yes  F-statistics  13.29  12.61  10.63  11.46  13.23  11.88  12.45  11.07  Observations  12,905  12,905  12,195  12,564  12,905  12,905  12,195  12,564  Note: This table reports the first stage statistics for the instruments used in the IV two stages least squares estimator in Table 6 (see Section 4.1 for details). In Columns (3) and (7), only firms with at least 60% of their sales in a specific business are included, and in Columns (4) and (8), only firms with less than 500 employees are included. Table A2. First-stage statistics for Table 7 First stage  (1)  (2)  (3)  (4)  (5)  (6)  Technological learning effect   Foreign demand effect   Period 2005–2007  Largest market highest R&D  Weighted export shares  Period 2005–2007  Largest market highest R&D  Weighted export shares  TL  1.273***  1.729***  0.722***  −0.310**  −0.237*  −0.158*  (0.204)  (0.265)  (0.163)  (0.133)  (0.141)  (0.093)  TL*emp≤25  −0.496***  −0.684***  −0.267***  0.210***  0.212***  0.127***  (0.116)  (0.159)  (0.080)  (0.073)  (0.075)  (0.048)  TD  0.058  0.252  0.121  0.844***  1.235***  0.778***  (0.106)  (0.169)  (0.096)  (0.190)  (0.258)  (0.187)  TD*emp≤25  0.041  0.055  0.031  −0.471***  −0.471***  −0.202***  (0.056)  (0.068)  (0.034)  (0.096)  (0.096)  (0.071)  Productivity measures  Yes  Yes  Yes  Yes  Yes  Yes  Structural and innovation controls  Yes  Yes  Yes  Yes  Yes  Yes  Internationalization controls  Yes  Yes  Yes  Yes  Yes  Yes  F-statistics  12.14  14.35  6.85  12.2  11.14  6.81  Observations  12,905  12,849  12,905  12.905  12,849  12.905  First stage  (1)  (2)  (3)  (4)  (5)  (6)  Technological learning effect   Foreign demand effect   Period 2005–2007  Largest market highest R&D  Weighted export shares  Period 2005–2007  Largest market highest R&D  Weighted export shares  TL  1.273***  1.729***  0.722***  −0.310**  −0.237*  −0.158*  (0.204)  (0.265)  (0.163)  (0.133)  (0.141)  (0.093)  TL*emp≤25  −0.496***  −0.684***  −0.267***  0.210***  0.212***  0.127***  (0.116)  (0.159)  (0.080)  (0.073)  (0.075)  (0.048)  TD  0.058  0.252  0.121  0.844***  1.235***  0.778***  (0.106)  (0.169)  (0.096)  (0.190)  (0.258)  (0.187)  TD*emp≤25  0.041  0.055  0.031  −0.471***  −0.471***  −0.202***  (0.056)  (0.068)  (0.034)  (0.096)  (0.096)  (0.071)  Productivity measures  Yes  Yes  Yes  Yes  Yes  Yes  Structural and innovation controls  Yes  Yes  Yes  Yes  Yes  Yes  Internationalization controls  Yes  Yes  Yes  Yes  Yes  Yes  F-statistics  12.14  14.35  6.85  12.2  11.14  6.81  Observations  12,905  12,849  12,905  12.905  12,849  12.905  Note: This table reports the first stage statistics for the instruments used in the IV 2SLS estimator in Table 7 (see Section 4.1.3. for details). In Columns (1) and (4), the instruments are calculated using only data for the period 2005–2007. © The Author 2017. Published by Oxford University Press on behalf of Associazione ICC. All rights reserved.

Journal

Industrial and Corporate ChangeOxford University Press

Published: Feb 1, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off