Decoding specialisation patterns in global value chains: the role of multinationals and domestic firms

The emergence of global value chains (GVCs) ^[1] significantly changed global trade, especially through the growth of intermediate trade in goods and services produced in different countries around the world (Pleticha, 2021). However, trade in GVCs has lost momentum in the last decade, and its growth is not expected to pick up again (World Bank, 2020). Given a decline in trade dynamics in GVCs, the important question is: how can countries achieve more benefits from participation in GVCs? Previous empirical research shows that such benefits can be visible through multiple channels, e.g. higher value added/income per capita (Ignatenko et al., 2019), increase in productivity (Ndubuisi and Owusu, 2023), and new job opportunities (Taglioni and Winkler, 2016). However, there is high heterogeneity and the benefits appear to be more significant for high- and middle-income countries (Kim et al., 2019; Ignatenko et al., 2019).

In recent literature, Timmer et al. (2019) propose a new approach to explain how the benefits of GVC participation in terms of value added are distributed. Based on the concept of the “smile curve” ^[2] (Shih, 1996), literature on global production networks (Ernst and Kim, 2002), and the governance of GVCs (Gereffi et al., 2005), they propose the “functional specialisation” concept (functional specialisation (FS)), which allows a country’s specialisation to be measured in GVC tasks/activities. Timmer et al. find that some countries (so-called headquarters economies) generate more value added due to their specialisation in activities at the beginning and end of the production process, while other countries (so-called factory economies) generate less value added due to their specialisation in pure fabrication activities related to the middle stage of production. This “functional dichotomy” among countries is confirmed by Stöllinger (2019), Kordalska and Olczyk (2023a) and Coveri and Zanfei (2023), showing that the role of headquarters economies is often played by more advanced countries, whereas less developed countries are factory economies.

The concept of FS opens up new research directions on the benefits of GVC participation and, as Pleticha (2021) notes, “any analysis of GVC participation is incomplete if it does not take FS into account”. To obtain larger benefits from GVC participation, a country can shift towards a different FS pattern. The research question in this study is: since countries are characterised by different patterns of FS, what are the determinants of FS?

The existing literature on FS determinants is limited. Stöllinger (2019), linking the FS concept to economic growth, identifies the impact of the level of gross domestic product (GDP) per capita on the pattern of business functions. Kordalska and Olczyk (2023a), in the analysis of functional upgrading ^[3] in Central and Eastern European (Central Eastern European (CEE)) countries, find a significant impact of GVC linkages and employee skills on FS patterns. Based on the GVC literature, we search for new potential FS determinants and focus on the heterogeneity of firms in terms of ownership for several reasons.

First, research on the role of multinational (MNE) enterprises (MNEs) in GVCs shows that they are an important driver of GVC growth (OECD, WTO and World Bank, 2014), increasingly act as networks within the international production systems of GVCs (Cadestin et al., 2018) and have a dominant presence in the manufacturing and service sectors (Gao et al., 2023). Fetzer and Strassner (2015) as well as Piacentini and Fortanier (2015) find that value added, trade, and imported inputs vary between different types of firms. Fortanier et al. (2020) show, using input-output models, that firm heterogeneity in ownership matters in terms of the contribution to value added embodied in exports. This leads to the assumption that firm heterogeneity in terms of ownership could influence country/sector specialisation in different tasks in GVCs.

This paper enriches the existing research in three ways. First, our paper is the first to examine the role of MNEs vis-à-vis domestic firms in building FS patterns. The previously mentioned studies did not take this into account, although the relationship between firm heterogeneity and trade activities is well recognised in the empirical literature (Bernard et al., 2012; Melitz and Redding, 2014). Second, another novelty is that we take into account not only the heterogeneity of firms in terms of ownership, but also the export channels chosen by firms, i.e. direct (own) export or indirect export (via others). We conduct our analysis also for two groups of European Union (EU) countries (EU15 and CEE countries). Third, our analysis is based on a new extensive database containing indicators of GVC activities of MNE and domestic companies, calculated using data from the Organisation for Economic Co-operation and Development (OECD) Activity of Multinational Enterprises (AMNE) database (Cadestin et al., 2018; Cai et al., 2023), and FS indices calculated using the fDi Markets database.

The paper is organised as follows. The next section provides a literature review on the relationship between the concepts of firm heterogeneity and GVC activities/FS. Other potential determinants of FS are also discussed. Section 3 introduces data, the methodology of the research, and presents the model. Section 4 begins the empirical part of the paper with descriptive results about indicators that are subsequently used in the main analysis. Section 5 includes the results assessing the impact of firms’ heterogeneity on changes in FS patterns. The last section provides our conclusions.

FS is a new and significant concept in the literature on GVCs. According to the World Bank (2020), specialisation in business functions is considered crucial not only for developing countries, but also for developed countries. GVCs lead to a finer international division of labour, which takes place at the task level and complements specialisation at the product level. It allows countries or regions to functionally specialise at the stages of the value chain where they have a comparative advantage and lack the capabilities to produce complete products.

In the literature on GVCs we can find very few papers related to FS and most of them are empirical. Kordalska et al. (2022) embed the concept of specialisation functions in economic theories. According to the authors, the FS concept has a multidimensional character and its roots lie in theories of trade, vertical specialisation, functional upgrading, and economic development. Due to the novelty of the FS concept, potential determinants of FS can only be found in the literature on GVCs, especially in empirical analyses related to GVC participation/activity.

To show how firm heterogeneity can potentially influence FS, we refer to the new new trade theories, especially to the concept of heterogeneity in international trade (Melitz, 2003; Bernard et al., 2007). According to these theories, not only are exporting firms very different from non-exporters, but high heterogeneity is seen between firms in each of these two groups. Ownership, size and age are the most important characteristics that make firms differ (Wagner, 2007, 2012; Bernard et al., 2007, 2012). Firm heterogeneity is shown in massive dispersion in firm outcomes, such as revenue, employment, exports and total factor productivity (TFP) (Bernard et al., 2022). Taking into account ownership heterogeneity, some stylised facts are observed, i.e. (1) MNEs are larger and often more productive than domestic firms and (2) the relative importance of MNEs in economic activity is higher in capital-intensive and R&D-intensive goods, and a significant share of two-way Foreign direct investment (FDI) flows is intra-industry in nature (Antràs and Yeaple, 2014).

Recent studies concerning the heterogeneity of firms involved in export activities indicate the export channels as a strong differentiating factor that distinguishes firms (Cieślik et al., 2023). Several empirical studies (e.g. Bernard and Fort, 2015; Taglioni and Winkler, 2016; Lu et al., 2017) confirm a strong relationship between firm productivity and the choice of export channel in the presence of intermediaries. The most productive firms both sell domestically and export directly abroad, less productive firms sell domestically and export both directly and through intermediaries, then follow the firms that sell domestically and export through intermediaries, and finally, the least productive firms sell only domestically. Firms are more likely to choose direct exporting when they are from countries where institutional hardship is high and uncertainty avoidance is low (Di Cintio, 2020). Additionally, direct exporters compared to indirect exporters are on average larger and perform better (Davies and Jeppesen, 2015), acquire more knowledge, and are more willing to introduce product innovations (Elango and Pangakar, 2020).

The above literature review shows that companies with different ownership structures (MNEs vs domestic companies) differ from each other in terms of the firm outcomes (such as export activities and productivity). We infer that companies with different ownership structures use different export channels, since they have different productivity. The aim of the paper is to decide whether ownership heterogeneity influences FS patterns, i.e. in which tasks and through which channels do MNE and domestic firms help EU countries to gain a competitive advantage in GVCs? Few empirical papers (e.g. Cadestin et al., 2018; De Backer et al., 2019; Fortanier et al., 2020) assess the role of heterogeneity in ownership in GVC activities based on inter-country input-output (IO) tables, which allow a comparison of the role of MNEs vs domestic companies in GVCs. De Backer et al. (2019) find that MNEs purchase relatively more intermediate goods from abroad than domestic firms, thereby integrating more foreign (and less domestic) value added into their production. Cadestin et al. (2021) show foreign subsidiaries of MNEs compared to domestic firms generate more income from intangible assets which are often related to high value-added activities at the beginning (R&D) and end of the value chain (sales). In turn, domestic firms concentrate on fabrication activities in GVCs. None of the above analyses, even if they point to the different role of MNE versus domestic firms in GVC activities, examine the role of these firms in creating a comparative advantage in GVC tasks. Our paper is the first to do so and to answer the above questions.

In our analysis, we also consider the effects of factors other than heterogeneity of ownership, which may be potential determinants of FS. These determinants will be used in our empirical analyses as control variables. For each factor, we briefly explain why it could be related to FS (to be tested empirically) and refer to the relevant GVC literature. With this approach, we can compare the effects of these factors with each other and also with those of the export intensity of MNE versus domestic firms regarding FS patterns. This is relevant for policy makers, since they can influence most of these factors.

First, as an explanatory variable, we use GDP per capita. Stöllinger (2019) finds that FS in fabrication tasks is associated with lower growth rates. Kordalska and Olczyk (2023b) confirm this result. They find that GDP per capita positively influences FS in R&D activities, but negatively in fabrication activities. We expect that higher income per capita will make a country too expensive (high wages) to have a comparative advantage concerning fabrication tasks and countries will specialise in R&D activities.

Second, we consider workers’ skills and human capital as important factors of FS. The majority of studies show that a higher level of education or employees’ skills support GVC activities because the high value-added activities in GVCs require special skills and knowledge (Ignatenko et al., 2019; Kersan-Škabić, 2019; Banerjee and Zeman, 2022). Additionally, Kordalska and Olczyk (2023b) find that the increase in human capital and the growth in higher-skilled workers over lower-skilled ones, lead to specialisation in R&D activities, but negatively influence specialisation in fabrication activities. So we assume that as skill levels and human capital increase, R&D will be easier to achieve than before, which could lead to specialisation in R&D, away from fabrication tasks.

Third, we also include GVC linkages as potential FS determinants in our analysis. In general, backward linkages refer to imports that are used for exports, and forward linkages refer to the exports that are subsequently used by the importing country to produce for exports. In the empirical literature, we find ambiguous results on the impact of linkages on the type of GVC activities (Tian et al., 2019; Ignatenko et al., 2019). Kordalska and Olczyk (2023b) find that backward (forward) participation in GVCs has a positive (negative) effect on specialisation in fabrication (R&D) ^[4].

In our empirical research, we use generalised structural equation models (GSEM) in which we explain (1) how the export activities of domestic and foreign firms affect FS in GVCs, and (2) which factors are deemed to be important for export intensity in GVCs. Our search for the determinants of export intensity of both domestic and foreign firms is based on the empirical GVC literature related to the activities of such firms in GVCs. We selected a set of factors that, according to empirical findings, are important for domestic and foreign companies’ activities in GVCs. This approach also makes it possible to identify differences in the effects of the same factors on the export intensity of domestic versus foreign companies.

First, we focus on the factors of production, i.e. capital and labour, whose abundance can create economies of scale, as suggested by trade theories, and can support the export intensity of domestic and foreign firms. More capital in an industry might replace labour in fabrication tasks (FS in fabrication activities will diminish), but also lead to activity in more sophisticated tasks (FS in R&D might increase). The empirical literature on GVCs does not clearly show the impact of employment and capital on the intensity of GVC integration. Hollweg’s (2019) analysis shows that higher employment within sectors and firms is associated with higher GVC integration. However, analyses looking at the impact of capital on GVC participation give mixed results: either they are not statistically significant (Van der Marel, 2015), or they have a negative impact on domestic value added (Olczyk and Kordalska, 2017), or a positive relationship between capital intensity and GVC is observed (Banerjee and Zeman, 2022).

Second, we use distance as a determinant of trade intensity in GVCs. Despite the diffusion of information and communication technology and the decline in transportation costs over the last 3 decades, distance still plays an important role in determining export intensity. In recent literature, distance is measured as the distance to the GVC hub (e.g. to Germany in Europe). The World Bank (2020) and Fernandes et al. (2022) confirm that geographic distance to GVC hubs is an important determinant of trade in GVCs.

Finally, we consider FDI to be a factor affecting the export intensity of domestic and foreign firms. This is because the motives for FDI, i.e. asset exploiting motivations, and in particular, market seeking FDI, influence export intensity the most (Franco, 2013). Stöllinger (2016) shows that FDI inflows exacerbate the fragmentation of cross-border production between countries and research. Additionally, Head and Mayer (2017) and Martínez-Galan and Fontoura (2019) confirm a strong positive statistical relationship between the growth of FDI stock in a country and a country’s participation in GVCs.

This section first explains the measurement of FS. Subsequently, we explain how we derive direct and indirect export intensities. Next, the empirical model and the estimation strategy are explained.

We employ the fDi Markets cross-border investment monitor database and follow the methodology proposed by Stöllinger (2021). The fDi Markets database provides data on the number of investment projects, capital investment, and job creation at the level of a single greenfield FDI project. We concentrate on job creation resulting from these types of projects. In the database, each greenfield project is assigned to one out of five groups reflecting activities located along GVCs, i.e. (1) headquarter services, (2) R&D, (3) fabrication, (4) sales and distribution services, and (5) technical support services and training. We aggregate these data to a country-industry level along with the methodology described in detail in Stöllinger (2021). In this study, we consider manufacturing industries only (Table A.I in the online supplementary materials).

Next, we specify how particular countries and industries specialise. Following Stöllinger (2021), we adopt the Balassa relative comparative advantage index to calculate relative functional specialisation (RFS):

where $Ji,jf$ denotes the number of jobs created by greenfield FDI projects in value chain function f in manufacturing industry i and EU country j. An RFS value between 0 and 1 indicates that a particular country-industry does not specialise in function f, RFS values that are greater than 1 mean that a country-industry achieves relative comparative advantages in serving function f.

It is suggested to transform this asymmetric RFS measure to a symmetric one when using specialisation measures in econometric applications (Laursen, 2015) ^[5]. That is why we define normalised RFS measures in the following way:

Such a measure is symmetric around 0 and ranges between −1 and 1. A country-industry reveals a comparative advantage when its normRFS measure is greater than 0.

In the empirical part of this paper, we focus on two out of five business functions mentioned at the beginning of this subsection, i.e. fabrication function and R&D function.

We use an early version of the OECD Analytical AMNE database (Cadestin et al., 2018; Cai et al., 2023) for constructing variables that reflect the direct and indirect trading activities of domestically and foreign-owned firms. These data consist of IO tables split according to ownership, i.e. they show the link between the foreign-owned Dutch metal industry and the domestically owned German car industry. Here a foreign-owned MNE is a subsidiary based in the country but ultimately controlled from abroad. A domestically owned firm is ultimately controlled from the country where it is located. The AMNE tables make it possible to trace the activities of foreign affiliates across countries from a value-added perspective.

We construct indicators that reflect how much firms export directly (their own exports) and indirectly (due to supplies in the supply chain of exporters). We do this separately for domestically owned and foreign-owned firms and by industry by country, since we are interested in the differences between domestically and foreign-owned firms.

To derive the indicators, we first collapse the AMNE Inter-Country Input-Output (ICIO) tables into single country IO tables. For each country, all intermediate imports (int) are combined as total imports M at the level of type of firm by industry, e.g. the domestically owned car industry. In other words:

where c, i and o respectively denote country, industry and ownership dimension in the home country. For the foreign countries they import from, this is denoted by d, j and w, respectively.

Intermediate exports and exports for final use (X) are combined at the level of the type of firm by industry as well. In other words:

The resulting single country IO table contains 41 industries, each split into 2 types of firms: domestically owned and foreign-owned.

Now we calculate the different channels that a type of firm uses to export. First, we set:

where EXGR_DVA_direct is the 82 × 82 ^[6] matrix of domestic value added embodied in the gross exports of a type of firm in an industry that is due to its own direct exports. The off-diagonal elements of this matrix are zero. EXGR_DVA_indirect is an 82 × 82 matrix that contains domestic value added embodied in the indirect exports of a type of firm, where it supplies in the value chain of exporters. VA is the value added of each industry, GO is its gross output, L is the 82 × 82 Leontief inverse and I is the unity matrix with 1 on the diagonal and 0 elsewhere of the same dimension.

To calculate direct and indirect export intensity by industry by type of firm, we set:

The estimation approach is based on the work of Rabe-Hesketh et al. (2004). They incorporate multiple equations and latent variables. A GSEM provides several advantages. First, the estimations method enables us to consistently estimate parameters in a mixed-process simultaneous system using full-information maximum likelihood (FIML) by including common unobserved firm heterogeneity components that are captured by $λ$ (Drukker, 2014). Secondly, in GSEM, these unobserved components can correct for omitted variable bias, which makes it more robust. As such, in contrast to the classical linear Two-Stage Least Squares (2SLS) approach, the FIML tends to be more efficient than 2SLS because the full covariance structure is taken into account. Finally, the GSEM requires fewer assumptions than the classical 2SLS for identification (e.g. rank condition and identification issues). In a system of equations defined by (9-11), we specify the following equations:

where the suffix i denotes a country, j refers to sectors, and t refers to time. $δi$, $δj$ and $δt$ are country, industry and time fixed effects, λ includes the common unobserved firm heterogeneity components, and $ϵijt$ is a random disturbance term. In Equation (9), f denotes two value chain functions – fabrication and R&D, considered separately.

In this equation, (9), normRFS refers to normalised relative FS as described in (2). In Equation (9), we include Dom and For explained by (10) and (11). We also include RGDPpc, which reflects real GDP per capita, LMEmp, which describes the share of hours worked by low and medium-skilled workers in the total number of hours worked, and HC is the human capital in the industry. BWpart and FWpart illustrate GVC backward and forward participation, respectively.

A detailed description of all explanatory variables along with their sources is presented in the online supplementary materials, Table A.II. A correlation matrix for explanatory variables is included in the online supplementary materials, Table A.III.

In the domestic (10) and foreign (11) equations, Dom and For refer to the export intensity measures of, respectively, the domestically and foreign-owned firms. In the empirical section, we consider (1) the direct (Dom-Dir/For-Dir in the estimation results), (2) indirect (Dom-Indir/For-Indir), and (3) total (Dom-Tot/For-Tot) export intensity of both types of firms, in separate model specifications. In equations (10)-(11), K refers to the capital income of the industry, Emp illustrates the employment in this industry, Dist is the geographical distance to Germany as the main GVC hub for Factory Europe, and FDIratio is inward FDI divided by outward FDI. Similarly to equation (9), the description of explanatory variables is included in the online supplementary materials, Tables A.II. and A.III.

The variance-covariance matrices for the error terms for each of the three equations, for simplification denoted by $Σε$⁠, take the following form:

where $m$ = {1, 2, 3} and $n$ = {1, 2, 3}. We assume that the correlation among Dom, For and FS is due to the common observed and unobserved factors which enter in all equations. In addition, we require the unobserved individual heterogeneity term $λ$ to be independent of all explanatory variables in the system, as well as independent of errors $εijt1$⁠, $εijt2$ and $εijt3$⁠:

where $D(·)$ denotes the distribution of “.”.

Using the estimation approach described above, we estimate (9–11) by adding blocks of explanatory variables into Equation (11). First, we estimate models with Dom and For variables only. In the next step, Dom and For are supplemented by real GDP per capita. Next, we add factors that reflect human capital resources (LMEmp and HC), and finally, GVC participation measures (BW_part and FW_part, respectively, backward and forward participation) are taken into account. This allows us to evaluate the robustness of coefficient estimates.

In this article we consider a database which comprises twenty-five EU countries ^[7] divided into two subgroups – EU15 countries and CEE countries, and ten manufacturing NACE ^[8] rev.2 industries (Table A.I in the online supplementary materials.). These country-industry data are observed over the period 2003–2018 ^[9].

As a background for further analyses, we show the development of real GDP per capita for the CEE countries and the EU15 during 2003–2018 (Figure 1). Both groups show an upward trend, with a temporary decline during the financial crisis in 2008–2009. The real GDP per capita grew from 6 thousand euro to 15 thousand euro in the CEE countries, an increase of 147%. The real GDP per capita grew from 27 thousand euro to 39 thousand euro in the EU15, an increase of 43%. Although the CEE countries have relatively larger growth than the EU15, the absolute difference in real GDP per capita remains at a similar level.

FS is very different in the two country groups and the differences seem to be persistent (Figure 2). The CEE countries are, compared to the EU15, specialised in fabrication. Yet the EU15 countries are, compared to the CEE countries, specialised in R&D. FS in fabrication remains stable for both groups, whereas FS in R&D may fluctuate over time. This results from the character of the fDi Markets data. New FDI projects related to fabrication are considerably more frequent than those related to R&D. Therefore, a new R&D project or its absence in a particular year can make quite a difference for R&D specialisation.

Because of the heterogeneity between countries, Figure 3 zooms in on the individual countries. Generally, specialisation in fabrication and specialisation in R&D are negatively correlated. Furthermore, the countries that specialise in R&D are relatively more often in the EU15 (whereas fabrication – in the CEE countries), confirming the macro results of Figure 2. Comparing 2003 and 2018 shows large differences for individual countries. For example, for FS in R&D, Greece was third in 2003, but last in 2018. For Portugal, it was the other way around. The results for FS in fabrication seem more stable over time. This might be related to the type of data, there are more new FDI projects with fabrication than with R&D.

There is heterogeneity in direct export intensity (Figure 4). It varies from 0.16 (Romania) to 0.65 (Ireland) for domestically owned firms in 2018 and from 0.42 (United Kingdom) to 0.87 (Ireland) for foreign-owned firms. Note that for almost all 25 countries, the direct export intensity is greater for foreign-owned firms than for domestically owned firms, both in 2003 and in 2018. The only exception is Germany in 2003. The direct export intensity is higher in 2018 than in 2003, for both types of firms, with a few exceptions. This reflects the rise of integration in GVCs during 2003–2018. Larger economies, such as Germany, the UK and France, are less export-oriented than smaller economies such as Hungary, Slovakia and Slovenia. This is well-known from the literature; after specialisation, the home market is not large enough for smaller economies to sell all their products.

As expected, the (average) indirect export intensity is far lower than the direct export intensity (Figure 5). In 2018, it varied from 0.06 (Croatia) to 0.20 (Czech Republic) for domestically owned firms and from 0.01 (Greece) to 0.11 (Italy) for foreign-owned firms. With the exception of Germany, the indirect export intensity of domestically owned firms was larger than that of foreign-owned firms. The first group has a lower direct export intensity than the second group, hence it has more space to accommodate indirect exports. In the majority of the countries, the indirect export intensity of domestically owned firms was higher in 2003. This might be related to the fact that these firms started to export more themselves. Conversely, in the majority of the countries, the indirect export intensities of foreign-owned firms were higher in 2003. This could be caused by better integration in the domestic value chain, but it could also indicate a shift towards more indirectly exporting industries.

This section presents the estimation results of the econometric system of Equations (9-11) introduced in Section 3.3. First we provide the main results for FS in fabrication, then for FS in R&D. Models for the fabrication function and R&D function are always separate models. The results are presented for the whole group of EU countries, but bearing in mind their FS pattern, we also zoom in on the country group dimension by distinguishing between EU15 and CEE countries. Finally, we discuss the results for the export intensity of domestically owned and foreign-owned firms.

Regardless of the model specifications, and whether we analyse a total, direct, or indirect export intensity, the export intensity of domestically owned enterprises is positively related to FS in fabrication (Table 1). Comparing direct and indirect exports, indirect exports have more than twice as strong an effect on the increase of specialisation in fabrication as that of direct exports, i.e. a 1% point increase in direct export intensity leads to a growth of FS in fabrication by 0.0014 on average (specification (8)), while in the case of indirect export intensity, the growth of FS is about 0.0035 (specification (12)). In contrast, the indirect export intensity of foreign-owned enterprises is linked negatively.

Taking a closer look at control variables, we find that real GDP per capita is negatively related to specialisation in fabrication. 1% of GDP per capita growth is followed by a decrease of FS in fabrication by 0.0015 on average in the total export intensity model (specification (4)) and by 0.0014 in the direct export intensity model (specification (8)). Since fabrication is of lower value-added per worker than other activities, relatively more fabrication and a lower GDP are aligned.)

We now zoom in on the regional dimension of our models, by discerning between the EU15 and the CEE countries. In terms of the impact of domestically owned firms (and their way of exporting) on FS in fabrication, both groups of countries exhibit a similar picture (Table 2) ^[10]. FS in fabrication is positively related to the total and direct export intensities of domestically owned firms. However, the impact of the indirect export linkages of these firms is generally not significant. Considering foreign-owned firms and their exporting paths, we observe a difference between EU15 and CEE countries. The total export intensity of these firms is negatively and significantly related to FS in fabrication. An increase in total export intensity by 1% will lead to a drop in normalised RFS by 0.0012 in EU15 countries and 0.0017 in CEE countries. Direct export intensity is negatively related to FS in fabrication in CEE countries. Indirect export intensity is negatively related to this function in EU15 countries, but positively in CEE countries.

This might reflect the reasons of the MNEs for the choice of location of their subsidiaries. Fabrication in the EU15 for export will generally be more expensive than in the CEE countries, hence there has to be a good reason to do so. Having a plant in the EU15 anyway could be to aim for the domestic market and not for exports. However, for fabrication in the CEE countries, it is the opposite. It is less expensive to produce here than in the EU15, hence such a country will export relatively more.

Again, we see that a higher real GDP per capita in CEE countries limits specialisation in fabrication. The higher the GDP, the higher the wages and the more expensive it is to produce there. Regarding the remaining control variables, the growth of low- and medium-skilled workers in comparison to highly skilled ones in EU15 countries would result in an increase of FS in fabrication. The results for backward and forward GVC integration for EU15 countries are similar to those for the full sample models (Table 1). For these countries, FS in fabrication is positively affected by both backward and forward GVC integration intensity. Furthermore, for CEE countries, FS in fabrication decreases along with the growth of forward GVC integration.

The estimation results that assess the impact of firm heterogeneity on FS in R&D reveal different patterns from those for FS in fabrication. We see that advancements in the total export intensity of foreign-owned firms are accompanied by a growth of FS in R&D (Table 3). This is because of their direct exports. In turn, domestically owned firms can increase their RFS in R&D through strengthening their indirect exporting channel. Regardless of the model specifications, these results remain robust. Additionally, when domestic firms increase their exports through the indirect channel mentioned above (specifications (9)–(12)), this affects the R&D function twice as much (growth of RFS by 0.006) as when foreign-owned firms export directly (growth by about 0.003, specifications (5)–(8)).

Real GDP per capita in all models but one is positively related to FS in R&D. A higher living standard is reflected in higher wages, countries withdraw from specialising in fabrication, and specialisation in other functions occurs. More human capital goes hand in hand with the growth of comparative advantages in R&D. This is not surprising, since this variable measures years of schooling in combination with rates of return on education. The last factors reflecting integration in GVCs do not play a part; neither backward nor forward participation significantly affect FS in R&D.

In the next step, we divide the full sample into two sub-samples containing EU15 and CEE countries separately, analogously to the analysis on FS in fabrication (Table 4).

The results for FS in R&D in the models with direct export intensity are different from those for FS in fabrication. Only the direct export intensity of foreign-owned firms in EU15 countries has a positive relation with this type of specialisation in the full model (column (6)). In the models with indirect export intensity ((9)–(12)), we observe a positive relation between R&D specialisation and the indirect export intensity of domestically owned firms in CEE countries.

In these models, estimated separately for EU15 and CEE countries, the importance of real GDP per capita disappears. Both factors describing human capital, LMEmp and HC, in CEE countries affect specialisation in the R&D function as expected. An economy that employs more educated/highly-skilled workers has more FS in R&D. Yet in the EU15 we do not see those effects. Backward and forward GVC integration are positively related to FS in the EU15 countries. Similarly to the results for FS in fabrication, the impact of forward participation is stronger than the impact of backward participation. In CEE countries, we observe that the growth of GVC backward integration leads to a decrease of specialisation in R&D.

The estimation results in Tables 1-4 correspond to Equation (9) of the system of Equations (9)-(11). The estimation results for the total/direct/indirect export intensity of domestically owned firms (Equations (10)), and the estimation results for the total/direct/indirect export intensity of foreign-owned firms (Equations (11)) are presented in Table B.I in the online supplementary materials for the full sample, and Table B.II in the online supplementary materials. for EU15 and CEE countries, respectively. These results differ from each other when we consider the full sample of countries, the EU15 group, or the group of CEE countries. They vary from each other when we take into account total/direct/indirect export intensity. Due to the construction of the model, the results are identical when we consider Equation (9) either for the fabrication function or for the R&D function. Note that the results for Equations (10) and (11) would also be identical if we had other specifications of Equation (9).

As for the variables in the models explaining total export intensity, some general patterns emerge. Capital income and employment are often positively related to direct export intensity. Both of these variables are related to the size of the industry. The larger an industry, the more difficult it is to sell on the domestic market, which will become saturated, thus the more it needs to export. The distance to the main trade partner is negatively related to the total export intensity of domestically owned firms, but not to that of foreign-owned firms. The further away the main trade partner, the higher the costs such as transport and trade partner finding, thus the less likely a domestically owned firm can overcome these barriers. It is well-known from the literature that foreign-owned firms can do this more easily since they can rely on the (network of the) parent firm. The FDIratio is also only (positively) related to the total export intensity of domestically owned firms. The results for direct export intensity are very similar, with one exception. Capital income is not related to the total export intensity of foreign-owned firms, but it is positively related to their direct export intensity.

The patterns in the models explaining indirect export intensity are slightly different from those in the models regarding total (and direct) export intensity. First of all, employment and capital income (measures of the size of the economy) are no longer, or only in a much smaller way, positively related to indirect export intensity. Distance to Germany is positively related to the indirect export intensity of domestically owned firms, but negatively related to that of foreign-owned firms. For total and indirect export intensity, this factor was negative and neutral, respectively.

In the study, we analyse the impact of firm heterogeneity in ownership, measured by the direct and indirect export intensities of MNEs and domestic firms, on FS patterns. We find that in EU economies the direct export intensity of domestic firms is positively related to FS in fabrication, while the direct export intensity of MNE firms supports specialisation in R&D. Our research is consistent with the analyses of Van Assche (2020) and Cadestin et al. (2021), which show that MNEs are often engaged in high value-added activities at the beginning of the value chain (R&D), as these tasks rely on intangible assets and allow MNEs to achieve higher value added compared to domestic firms.

We find that the indirect export intensity of domestically owned firms is positively related to EU specialisation in both business functions (fabrication and R&D). Our results can be explained by the findings of Criscuolo and Timmis (2017), who revealed that GVCs are not primarily global in nature but are centred on regional production clusters, and that services and MNE enterprises (MNEs) play a key role in these networks. This is why some domestic companies export their products through intermediaries (often through MNEs).

Our more detailed analyses show that the types of firms and the export channels which support the most desirable specialisation profile in R&D activities differ between the two groups of EU countries (EU15 and CEE). For EU15 countries, they are direct exports of MNEs, and for CEE countries, indirect exports of domestic firms. The importance of indirect exporting channels for CEE firms can be explained by Chiacchio et al. (2018) and Ito and Saito (2021). The former identify a decline in TFP growth and, in particular, a drop in R&D investment in CEE countries since 2011, while the latter find that domestic firms which are not productive enough to participate directly in GVCs tend to export their products through intermediaries. Our results challenges the conventional wisdom that direct exporting is the only pathway for firms to integrate into high-value GVC activities and suggests that intermediary relationships within domestic value chains are crucial for upgrading FS patterns.

Our results contribute to the existing literature in several ways. First, we extend the literature on firm heterogeneity in GVC analyses by identifying the different roles of MNE versus domestic firms in building FS patterns, especially in headquarters and factory economies. Second, by decomposing export intensity into direct and indirect channels, we provide a nuanced understanding of how firms contribute to national comparative advantages in fabrication and research and development (R&D) functions. Particularly, we extend the knowledge on the role of MNEs in domestic value chains and the spillover effects of MNEs are expanded. We discover that MNEs play a new role in host countries: they are not only buyers of locally produced inputs or suppliers of final/intermediate products, but they are also the gateway linking the domestic economy to the most desirable activities of GVCs in terms of value added (R&D). The spillover effects of MNEs are not limited to upgrading exported products (e.g. Bajgar and Javorcik, 2020) or absorbing knowledge when buying inputs from MNEs (e.g. Di Ubaldo and Siedschlag, 2022). MNEs also help host countries to gain comparative advantages in R&D activities in GVCs. Additionally, our findings contribute to the existing literature on domestic linkages and the role of domestic firms in GVCs, especially in CEE countries. According to Pellényi (2020), the initial comparative advantage of low wages in CEE countries leads the region to remain specialised in fabrication tasks, which limits the domestic value-added content of exports. Our results suggest that a new channel for changing this specialisation patterns by domestic firms in CEE is the development of indirect exports. To sum up, by bridging firm-level trade dynamics with macroeconomic FS patterns, our study provides a more comprehensive framework for understanding the role of ownership structure and export channels in shaping comparative advantages within GVCs. This contribution lays the groundwork for future research exploring the interplay between domestic firms, MNEs, and FS across different institutional and industrial contexts.

Our findings have policy implications for EU countries. So far, EU countries have tried to attract FDI to certain sectors, e.g. technology or knowledge-intensive sectors. They could also consider a new system of incentives that are not targeted at a specific sector, but a system that is more flexible and offers customised incentives for specific tasks, e.g. for R&D tasks. This new system could be further supported by the development of specialised methods for evaluating and screening potential investment projects. For Central and Eastern Europe, where we find that in addition to the direct export of MNE companies, the indirect export of domestic companies also supports specialisation in R&D, it could be considered to implement programmes to promote the indirect export of domestic companies. This could facilitate the matchmaking of indirect exporters with exporters at home rather than with business partners abroad. A joint report by the Statistics Denmark and OECD (2017) states that promoting integration through domestic channels can stimulate companies that export indirectly.

Our analyses have some limitations. Our study shows relations between variables, but it is not a cause-effect study. For example, does inward FDI in R&D lead to strong domestically owned firms, or do strong domestically owned firms attract inward FDI in R&D? There are not yet data available about the types of tasks performed at foreign-owned and domestically owned firms in a given industry. Currently, there is only a total at the industry level. The basis for our analysis is the OECD AMNE, which is not in constant prices, hence different price changes in different sectors may play a part. Further analyses are needed. It would be advisable to conduct in-depth analyses from the sectoral perspective, based on more detailed data about the specific occupations at the sector level, such as in Kruse et al. (2023). Ideally, such data would be available at ownership by industry by country level. That would also yield new insights into the reasons for differences in the performances (such as productivity) of the two types of firms. Such data can be compiled by matching the employees in the labour force survey to their employers, matching these firms with the business register and aggregating the data to ownership by industry using appropriate weights.

We hope to have shown that a combination of micro (FS) and macro (AMNE) statistics allowed for more detailed explanations of the drivers and consequences of international production fragmentation. Additionally, we posit that this approach is useful in better understanding the relation between economic development and integration into GVCs for different types of countries.

The authors acknowledge the financial support from Gdańsk University of Technology by the DEC-4/2022/IDUB/II.1.1/Cu Grant under the CUPRUM Supporting Research Team Building in emerging areas and the program “Excellence Initiative – Research University”. The authors acknowledge participants of the 24th AEEFI Conference on International Economics (15-16.06.2023, Alcalá de Henares, Spain), the 29th International Input-Output Association Conference (25-30.06.2023, Alghero, Italy), the 15th Input-Output-Workshop (29.02-1.03.2024, Osnabrück, Germany), and the 25th European Trade Study Group Conference (12-14.09.2024, Athens, Greece) for their valuable comments and suggestions.

The supplementary material for this article can be found online.