9. International Linkages, Value-Added Trade, and Firm Productivity in Latin America and the Caribbean

Participation in international trade can be an important source of information, knowledge spillovers, technology transfers, technical assistance, competitive pressures, and other productivity advantages for firms, leading to significant performance improvements (Grossman and Helpman 1991; Clerides et al. 1998; Verhoogen 2007; Fafchamps et al. 2008; Bernard et al. 2003). At the same time, firms with FDI and/or multinational firms may generate a total cost reduction through low-priced production factors. All of these factors may generate a positive learning effect of global activities. This “learning-by-exporting” hypothesis has spurred a large number of empirical studies that seek to assess the causal effect of exporting at the firm level. ¹ However, there is no consensus among scholars on whether such a learning effect exists or what specific factors may be behind it. While a comprehensive survey by Wagner (2007) indicated that the evidence on this learning effect was mixed and unclear, a significant positive effect of the export experience on firm productivity has been found in several studies. ² The meta-analysis conducted by Martins and Yang (2009) indicated that the impact of exporting on productivity was higher for developing than for developed economies. Most importantly, the direction of causality between openness and firm performance is controversial (see Greenaway and Kneller 2007).

Firm productivity and sunk costs play important roles in how firms select international activities. These costs tend to discourage less productive firms from international linkages; therefore, firms generally self-select to participate in global markets. This selection mechanism according to the level of productivity is called the “selection effect” in exporting. Melitz (2003), who showed that exporting firms had relatively higher productivity, has provided the theoretical benchmark for the above selection mechanism, while the pioneering empirical work by Bernard and Jensen (1999) on US firms has been followed by many scholars. López (2005), Greenaway and Kneller (2007), and Wagner (2007) have done surveys on the topic. Most of the studies on the selection effect found that more productive producers self-select into the export market (Clerides et al. 1998; Álvarez and López 2005; Hayakawa et al. 2012).

Identifying the learning effects of FDI is also important. On the one hand, the performance of domestic firms may improve with FDI, particularly inward FDI in the form of cross-border mergers and acquisitions (M&A). Fostered by superior know-how, human capital, and organization of foreign firms, local firms with FDI could strengthen their local advantages (i.e. experience in the local market and knowledge of the local institutional environment) and enhance their productivity (see UNCTAD’s World Investment Reports from various years). On the other hand, Helpman et al. (2004) theoretically showed that investing firms have relatively high productivity. Several studies have empirically tested this proposition (see Greenaway and Kneller 2007, for a survey within this literature). ³ Studies do not necessarily detect a positive causal effect of investing on firm productivity. While Barba Navaretti and Castellani (2004) and Kimura and Kiyota (2006) found significant positive impacts, Aitken and Harrison (1999), Hijzen et al. (2007), and Ito (2007) detected a small or non-positive effect. Hijzen et al. (2006) and Barba Navaretti et al. (2006) further explored a possible qualitative difference in learning related to two types of FDI: horizontal (many plants doing the same activities in a number of countries to put production near consumers) and vertical (different stages of production in different countries). For French firms, they found positively significant enhancements in productivity from horizontal FDI but not from vertical FDI. Other papers have focused specifically on the impact of M&A on firm performance, and most have found a significant positive impact. ⁴

The increasing international fragmentation of production that has occurred in recent decades has challenged the conventional wisdom on how we look at and interpret trade. Traditional measures of trade record gross flows of goods and services each and every time they cross borders, leading to a multiple counting of trade, which may lead to misguided empirical analyses (Cattaneo et al. 2013; OECD–WTO 2012). Furthermore, since these days a large number of countries have developed comparative advantages in specific parts of the value chains and not necessarily on final goods, standard trade statistics are becoming much less informative.

Enterprise-level data offers crucial information to understand the drivers of productivity and competitiveness, as aggregate performance depends strongly on firm-level factors such as size, ownership, and technological capacity. For our empirical exercise, we use a subset of the WBES database specifically focused on firms in LAC countries. This subset provides information on the characteristics of firms across various dimensions, including size, ownership, trading status, and performance, and collects data for 14,657 firms and 31 LAC countries. ⁹ Table 9.6 in the Appendix presents information about the international linkages we analyze (exports, imports, and foreign-owned firms) for the whole LAC sample by country and survey year. In addition, in order to provide a richer picture of the phenomena we analyze (and to combine different levels of aggregation) and to map out sources and components of trade in value added, we use the OECD-WTO TiVA dataset by industry (see “Trade in Valued-Added and GVCs” above). We focus specifically on the following countries for which TiVA and WBES data are both available for the same fiscal year: Argentina, Brazil, Chile, and Mexico. ¹⁰ Looking at all of the data on firms and industries, we can draw a picture of the current international linkages of the four LAC countries as well as trade in value added components and GVC characteristics.

Table 9.1 presents a descriptive analysis of the characteristics of the firms’ international linkages (WBES) and GVC indicators (TiVA) for the four LAC countries. ¹¹ The WBES subsample includes 5120 firms across the four countries. The first five columns of Table 9.1 (WBES data) show that, overall, almost 15 % of these firms declare themselves to be exporters ¹² and their export intensity is on average more than 33 % of their total sales. ¹³ Only 8.5 % are foreign-owned firms, but on average foreign investors own a significant share (85.4 %). The level of firm internationalization is heterogeneous across these four countries. With regards to international trade, Argentina has the highest number of exporting firms (over 27 %), followed by Chile (17 %), Mexico (15 %), and Brazil (7 %). Chile shows the highest export intensity (42 %), followed by Mexico (36 %), Argentina (33 %), and Brazil (30 %). With regards to FDI, Chile and Argentina have the highest number of foreign-owned firms (both around 13 %), while Brazil has the lowest (4 %). The foreign ownership share of these firms is high, ranging from about 83 % in Mexico to nearly 90 % in Argentina.

Concerning trade in value added components, columns 6 to 9 in Table 9.1 (TiVA data) present the main components of the decomposition of the overall gross exports described in Fig. 9.1 by country (IVA and FVA). The last two columns provide some perspective on the role and position in GVCs by country, using the indicators of GVC participation and position illustrated in the “Trade in Valued-Added and GVCs” section above.

The reported decomposition components show some degree of heterogeneity among the countries. Specifically, Chile has the highest IVA value, suggesting it provides relatively higher domestic added value inputs to other countries’ exports. This is followed by Brazil and Argentina, with Mexico having the lowest value, which is in line with its relative specialization in processing foreign inputs. On the other hand, Argentina and Brazil show, on average, a lower level of FVA, suggesting they contribute to their gross exports mainly with domestic value added, relying less on imported inputs. This can be related to the fact that these countries are, on average, more involved in exporting goods in which the main source of inputs comes from the primary domestic sector (Blyde 2014). We note that Brazil shows the smallest share of FVA—only 9 % of the value added incorporated in the Brazilian exports comes from other countries. On the other hand, Chile and Mexico show a relatively higher level of FVA, suggesting a relevant presence of foreign inputs in their overall exports. It is worth noting that about 30 % of the value of Mexican processing exports comes from abroad. This heterogeneity is associated to some extent with the country dimension (Cattaneo et al. 2013), but also with differences in the patterns of specialization: a relative specialization in producing primary goods requires, on average, less imported inputs than manufacturing them.

Figure 9.2 presents the international comparison of the value added decomposition for selected industrialized, emerging, and developing/transition economies, ¹⁴ as well as the sample of LAC countries. Other than Mexico, the LAC countries show a relatively high level of IVA but generally lower FVA (i.e. a lower content of intermediate inputs coming from abroad). This confirms the relative specialization of LAC countries within GVCs in exports of primary goods. In Table 9.7 in the Appendix, we provide the details of the main value added components of gross exports for the countries in Fig. 9.2.

Figure 9.3 provides an international comparison of the GVC participation and GVC position indicators. The figure shows that Chile’s GVC participation is substantial, Mexico’s is somewhat lower, and Argentina’s and Brazil’s are below the selected world counterparts, except for South Africa. This heterogeneity may be attributed to a size effect (i.e. larger economies tend to have a relatively higher degree of self-sufficiency in producing inputs for exports: Brazil and Argentina) and/or patterns of specialization (relatively high specialization in manufacturing can justify a higher degree of global participation: Chile and Mexico). This empirical evidence is consistent with similar analyses of LAC integration into the global production network (see UNCTAD 2013; Blyde 2014).

As stated in the “Trade in Valued-Added and Global Value Chains” section above, the GVC position index reflects where countries fit in the value chain. A country can be upstream or downstream, depending on its specialization. Upstream countries produce inputs and/or raw materials used at the beginning of the production process and do not rely on foreign inputs for their exports. Downstream countries assemble products and provide relatively less intermediates to the exports of other countries. The higher the value of the index (higher than 1), the more upstream the country’s exporters are in GVCs. In general, the sample of LAC countries is upstream (i.e. away from the final customer) in GVCs more than their international counterparts; however, again, there is a degree of heterogeneity. Brazil is the most upstream and has the highest GVC position in our international comparison. This is consistent with the fact that Brazil, more specialized in natural resources, mainly provides inputs to other countries’ exports and does not rely much on other countries’ inputs. Thus, Brazil is positioned more at the beginning of the GVCs. On the opposite side is Mexico, which is located more at the end of GVCs and acts as a final producer, using inputs provided by upstream countries in the form of maquila (factory) processing operations (Contreras et al. 2012; De La Cruz et al. 2011; Dussel Peters 2003) and does not provide many intermediates to other countries’ exports.

Figure 9.4 presents the comparison of the GVC position indicator by industry for the four big LAC countries. We provide an international comparison with the selected counterparts in Table 9.8 in the Appendix. The indicator used for the industry analysis is obtained from TiVA data, as proposed by Fally (2012) and Antràs et al. (2012). ¹⁵ In line with the literature, industries such as mining and quarrying, wood, paper, paper products, printing and publishing, chemicals and non-metallic mineral products, and basic metals and fabricated metal products are at the highest level of upstreamness since they provide raw materials and inputs for the beginning of the value chain. Among those industries, the big LAC countries (except Mexico) show relatively high specialization, with a degree higher than 2. Brazil has the highest index of upstreamness in mining and quarrying, with a value higher than 3.5. Concerning services, Fig. 9.4 shows that the most upstream services are, on average, telecommunications and financial. Of the LAC countries, Chile is positioned more upstream in the value chain in all the services sectors, while Mexico, in line with the result at the aggregate level is, on average, the most downstream, with the relevant exception of financial intermediates.

The aim of our empirical exercise is to investigate whether LAC firms characterized by international linkages have higher productivity than other LAC firms. Specifically, we want a more in-depth assessment of whether there is a causal relationship between the degree and type of involvement in international production networks and firm performance in the LAC region.

We start by presenting static differences in firm productivity premia between exporters and non-exporters, and foreign-owned ¹⁶ and domestic enterprises. First we pool data for the entire sample of LAC countries included in the WBES. Productivity premia are measured as the coefficients for export and inward FDI dummies in a regression of the form:where θ _i is the log of firm labor productivity, ¹⁷ d _i is a set of dummies for exporting firms and firms characterized by foreign ownership (i.e. our proxy of inward FDI); η _c and η _j are dummies for country and industry, respectively, to control for bias due to unobserved factors; ε _i is the error term. Table 9.2 confirms the expected positive relationship between international linkages and firm productivity based on firm-level data. These findings are in line with the theoretical predictions that low-productivity firms stay in the domestic market, while firms with higher productivity export and/or engage in FDI (Helpman et al. 2004).

So far we have presented stylized facts, which cannot yet allow any causal interpretation. Taking advantage of the availability of the set of firm-level covariates provided by the WBES, we can test the above relationship by presenting a version of the standard constant returns to scale Cobb–Douglas production function with labor, capital, and knowledge expanded to international linkages as follows:Equation 9.2 adds the following explanatory variables (all variables are in logs) to equation 9.1: k _i for firm “capital intensity” and z _i for a bundle of firm-level observables (human capital, employment, firm size, and technological innovation). As in Farole and Winkler (2012), the latter variable is a dummy that controls whether firms use technology licensed from a foreign-owned company (excluding office software), their own internationally recognized quality certification (e.g. ISO), or use a firm website and/or email to communicate with clients and suppliers. To avoid bias due to unobservable factors, ¹⁸ we control for the geographical location and industry of the firms. A full description of the above variables is provided in Table 9.9 in the Appendix.

Table 9.3 shows the regression results of the base model. It is organized in ten columns. The first five columns report the estimates for equation 9.2. Columns 6 to 10 report the same estimates for the subsample of exporting firms by substituting the dummy variable for exports with a continuous variable (i.e. the value of sales exported directly). Also in this case, we use pooled data for the entire LAC dataset. The signs of the relationship between labor productivity and the set of firm-level explanatory variables are significant and consistent with the theory. A positive coefficient is estimated for the relationship between labor productivity, capital intensity, employment, and innovation, while a negative coefficient is estimated for unskilled workers (a proxy of human capital). Also in this case, on average and all else being equal, our findings are consistent with the view that exporter and/or foreign-owned firms (i.e. characterized by inward FDI) show higher productivity. To look more in depth at firm heterogeneity, we also carried out separate regressions by firm size (distinguishing micro, small, medium, and large firms). On average and all else being equal, the subsample of exporting firms (columns 6 to 10) confirms the positive relationship between the level of gross exports and productivity for all of the size categories.

Because of the lack of panel data, our base model cannot avoid further bias due to unobserved characteristics that are correlated with firm characteristics and productivity. To this end, we provide additional empirical estimates for the subsample of exporting firms located in the LAC region by controlling for endogeneity bias in the relationship between firm productivity and the value of their gross exports with excluded instruments. More specifically, from the WBES dataset we select some additional explanatory firm-level variables that are supposed to be correlated with gross exports but not with domestic productivity: average time to clear imports from customs (days), and days to obtain import license. We use these variables as proxies for international trade obstacles that are negatively correlated with export flows but do not depend on firm productivity. ¹⁹

Table 9.4 provides estimates for an instrumental variable (IV-2SLS) and a control function (CF) for the pooled data (the first stage estimates are not reported in the table). The IV results are robust and significant. Moreover, the Hansen’s J statistics of over-identifying restrictions—which is consistent in the presence of heteroskedasticity—does not reject the null hypothesis that our instruments are valid. However, the Angrist-Pischke (AP) F-statistic of weak identification is significant only at the 5 % level. Since the IV inconsistency increases with the number of instruments used, we opt for more parsimonious behavior by using only one instrument, the average time to clear imports from customs. Further, we apply a CF approach that controls for the endogeneity bias by directly adding the estimated residual of the first stage equation to the main regression providing an unbiased CF estimator that is generally more precise than the IV estimator (Wooldridge 2010). The significance of the CF estimates confirms the above evidence of a relationship between trade and firm-level productivity for the full sample and by firm size (with the exception of small firms) ²⁰ as well as the absence of reverse causality. ²¹

Finally, we provide a more detailed investigation of the linkages between firm-level exports and productivity and specifically address our second research question, which is related to the effect of firm involvement in GVCs (participation and position) on firm productivity. Thus, we present a further empirical test of equation 9.2 for the subsample of exporting firms by controlling for the decomposition of the value added embodied in national exports as well as the GVC indicators at the industry level. This further test assumes that firm performance in value added trade is heterogeneous across industries but homogeneous within them. We acknowledge this is a strong assumption. However, it is consistent with the high level of aggregation of TiVA industry data that supports the hypothesis of firm heterogeneity across industries. It is also consistent with detailed investigations at the industry level that show a very low degree of firm heterogeneity across sector functions in the LAC region (Gereffi et al. 2005; Pietrobelli and Rabellotti 2011). However, we empirically test this assumption by applying a Levine test (i.e. similar to the standard ANOVA test but less sensitive to the violation of normality assumption) to a set of firm characteristics. The outcomes of the Levine test confirm, on average, that we can reject the null hypothesis that the “within variances” of the set of firm-level characteristics across industries are equal (with a probability below 0.05). This strongly supports the assumption of intra-industry firm homogeneity across ISIC industries and, thus, the relative homogeneity in value added trade across industries.

Before presenting this further empirical test, it is worth recalling that FVA (foreign value-added embodied in total exports) and IVA (indirect domestic value-added embodied in intermediate exports used in other countries’ exports) are the key value-added components of total exports. Moreover, the ratio between these two components provides a measure of country/industry relative upstreamness/downstreamness (i.e. the GVC position index). Since the GVC participation index is a linear combination of IVA and FVA, the parameters associated with these components of gross exports are jointly considered indicators of GVC participation.

Table 9.5 presents the results of the value added and GVC estimates. Unfortunately, due to data constraints, we can run the latter test only for a restricted sample of exporting firms from the four LAC countries for which TiVA data are available (Argentina, Brazil, Chile, and Mexico). ²² The results are fully consistent with the theory and with the results of the previous empirical exercises (the coefficients of the base model are all significant and show the expected signs): firms’ international linkages are positively correlated with productivity. There appears to be an additional and heterogeneous impact on firm productivity in clustering firms by trade in value added (specifically in value added embodied in foreign intermediate imports) once the causal impact of gross exports is controlled for. These estimates confirm that international trade participation has a positive effect on productivity at the firm level and suggest it is not independent of the decomposition of the added value of gross exports by industry. Furthermore, the robust and positive relationship between firm-level productivity and the industry GVC position suggests that the position of the industry in the GVC matters as well: the higher the industry upstreamness in the GVC, the greater the impact of its international linkages on firm productivity. In other words, firms operating in the industries that get added value from exporting intermediates and primary goods used in other countries’ exports tend to be more productive than firms operating in industries whose value added comes primarily from imported inputs.

In this chapter, we have addressed two key research questions:

Our empirical analysis provides a rich picture of the relationship between firm performance and country/industry involvement in international production networks in the LAC region by combining the WBES firm-level data and the OECD-WTO TiVA data.

First, we estimated the productivity premiums associated with participation in trade and the presence of inward FDI, while controlling for firm heterogeneity by using dummies for country, sector, and survey waves. Second, we analyzed the relationship between firm international linkages and productivity by using a standard output function with constant returns to scale Cobb–Douglas technology with labor, capital, and knowledge, presenting both OLS, IV, and CF estimates. Third, we ran a final test of the same equation expanded to account for TIVA-based indicators of value-added trade and industry involvement in global production networks.

Our empirical analysis confirmed a positive causal relationship between international activities and firm performance in the LAC region. Furthermore, focusing on four big LAC countries (Argentina, Brazil, Chile, and Mexico), we showed that the level of involvement in GVCs matters as well. More specifically, our empirical analysis highlighted the key role of both trade in value added and GVC position, with a positive impact of upstreamness on firm performance. Firms operating in the industries exporting intermediates and primary goods used in other countries’ exports tend to be more productive than firms operating in industries whose value-added comes primarily from imported inputs.

We suggest that research into constraints preventing a country from fuller engagement in GVCs would be a natural next step to our research. We also believe it will be important to propose adequate criteria to prioritize different constraints depending on whether a country tries to go upstream or to integrate downstream, or to broaden the variety of its exports and opportunities to attract greater GVC participation. Research to assess feasible changes in the business or policy environment in relation to the above factors is alsonecessary.