A Systems Perspective on the Embeddedness of Foreign-Invested Enterprises and Functional Upgrading in Manufacturing: The Threshold Effect of Industry Chain Centrality

Abstract

This study adopts a systems perspective to explain why the embeddedness of foreign-invested enterprises (FIEs) generates divergent effects across countries—promoting upgrading in some while inducing low-end lock-in in others. Based on complex network theory, we construct an industry chain centrality indicator and examine how the embeddedness of FIEs affects functional upgrading in manufacturing, as well as the threshold effect created by industry centrality. Using panel data on manufacturing sectors of 42 economies from 2003 to 2020, we employ a panel threshold model to analyse the nonlinear impact of FIEs embeddedness. The results show a significant single threshold. When industry centrality is low, the positive effect of FIE embeddedness on functional upgrading is not significant; once the threshold is crossed, the effect strengthens markedly. This pattern indicates that industries occupying hub positions in global production networks can better absorb and amplify knowledge and technology spillovers generated by FIEs, promoting upgrading of high value-added functions such as R&D, management, and marketing. Robustness checks confirm these findings, and heterogeneity analysis shows that different types of functional upgrading exhibit distinct threshold levels. Overall, the study highlights that the impact of FIEs depends critically on an industry’s structural position within the global network.

1. Introduction

A prominent feature of contemporary globalization is the rapid rise and continuous deepening of global value chains [1,2]. As production activities have been fragmented into a series of tradable tasks across borders, the global division of labour has gradually shifted from an inter-industry pattern to a functional pattern in which economies specialize in distinct stages such as research and development, design, manufacturing, and marketing [3]. At the same time, the embeddedness of foreign-invested enterprises (FIEs) in host-country industrial chains introduces capital, knowledge, and advanced technologies and combines these with local labour, natural resources, and infrastructure [4]. Together, these forces constitute a major driver of the organization and participation of host countries in the functional allocation of value chain activities [5]. Therefore, a key question that follows is whether the embeddedness of FIEs can facilitate the functional upgrading of host-country manufacturing—namely, the shift from low value-added processing and assembly activities towards high value-added functions such as research and development (R&D), management, and marketing.

Existing research addressing this question reveals a marked divergence of views. On the one hand, a substantial body of empirical evidence shows that foreign direct investment can enhance the productivity and innovation capacity of host-country industries through mechanisms such as technological spillovers, managerial know-how transfer, and expanded market access [6]. The entry of multinational enterprises not only brings advanced production processes and organizational practices but also stimulates local firms’ technological learning and efficiency improvements through competition and demonstration effects [7]. From this perspective, the embeddedness of FIEs should, in principle, provide momentum for functional upgrading. On the other hand, some scholars argue that under the unequal governance structure of global value chains, firms in developing economies often become trapped in ‘low-end lock-in’, remaining confined to low-technology, low-profit processing and assembly activities and thus struggling to achieve substantive upgrading [8,9]. This lock-in is largely attributable to the restrictions that multinational corporations impose on the transfer of critical technologies and knowledge in order to safeguard their core competencies. Such constraints cultivate a dependency relationship that entrenches local firms’ peripheral position in the global division of labour [10]. Hence, the effects of the embeddedness of FIEs do not naturally translate into functional upgrading.

These contradictory empirical findings suggest that the effect of the embeddedness of FIEs on functional upgrading may not be linear; rather, it may constitute a conditional outcome shaped by multiple contextual factors. Although existing studies have identified several moderating mechanisms at the firm, industry, and institutional levels [11], relatively few have examined the issue from a broader system-level perspective by considering the ‘structural position’ of industries within the global production system. As the networked evolution of GVCs deepens, the global production system has increasingly taken the form of a complex, multi-layered network in which industries occupy highly differentiated positions in terms of power and connectivity [12]. Some industries are located at the core of the network, possessing high levels of connectedness and informational dominance, whereas others remain at the periphery and are subject to the control of core nodes. Such differences in structural position may fundamentally determine whether, and under what conditions, the embeddedness of FIEs can facilitate functional upgrading.

Building upon this intuition to develop a system-level perspective, this paper introduces the concept of industrial chain centrality as a key indicator for measuring the structural power and influence of industries within the global value chain network. We argue that industrial chain centrality reflects the structural advantages of industries in terms of information access, resource integration, and bargaining power, and that it systematically affects the mechanisms through which the embeddedness of FIEs influences functional upgrading [13]. Specifically, this paper proposes a core hypothesis: that there is a non-linear relationship between the embeddedness of FIEs and functional upgrading in manufacturing, mediated by industrial chain centrality as a threshold. Once the centrality of an industry surpasses a certain threshold value, its dense industrial connections, mature ecosystem, and enhanced bargaining power will significantly amplify the potential spillover effects brought by foreign investment, thus activating its positive role in facilitating functional upgrading.

To test the above hypothesis, this paper constructs an indicator of industrial chain centrality based on cross-country panel data covering 14 manufacturing industries across 42 economies from 2003 to 2020. Using these data, the study employs a panel threshold regression model to empirically examine the non-linear impact of the embeddedness of FIEs on functional upgrading in manufacturing.

The contributions of this study are threefold. First, at the theoretical level, this paper incorporates complex network theory into GVC research and, from a systemic perspective, proposes industrial chain centrality as a key moderating variable shaping the spillover effects of foreign investment. This approach provides an integrative analytical framework for reconciling the contradictory findings in the existing literature and advances the paradigm shift in GVC studies from a ‘chain-based’ to a ‘network-oriented’ perspective. Second, at the empirical level, this study utilizes the latest cross-country input–output datasets and greenfield investment data to construct measures of functional upgrading and industrial chain centrality, applying a threshold regression model to conduct the first empirical test of the non-linear characteristics of foreign-invested enterprise embeddedness. The model generates new empirical evidence that sheds light on the complex interactions between foreign investment and industrial upgrading in host countries. Third, at the policy level, the findings offer important implications. They indicate that the focus of industrial policy in developing economies should shift from merely increasing the quantity of foreign investment towards strategically enhancing the industrial chain centrality of domestic industries. Only when industries occupy a favourable structural position within the global production network can foreign investment truly serve as an effective catalyst for functional upgrading, thereby helping countries avoid the trap of low-end lock-in.

2. Literature Review and Research Hypotheses

2.1. Literature Review

As functional specialization deepens and value-added activities become fragmented across geographical spaces, the global production system has gradually evolved into a complex and multilayered production structure. This structure not only integrates sequential processes and synchronized operations but also further develops a hybrid operating model. In such a system, where high levels of specialization and collaboration coexist, the connections between industries or countries are no longer confined to the traditional linear upstream–downstream relationships but have woven into a multi-layered, networked system of interaction [14]. Thus, it is evident that the form of global production networks has transcended the single chain structure, exhibiting significant network characteristics and hierarchical nesting [8]. The identification and characterization of these structural features require theoretical tools capable of systematically analysing complex interrelationships. Therefore, to more accurately reveal the strength of the relationships and the structural positions of various entities, there is a pressing need to introduce an analytical framework that can systematically depict these complex interconnections, reflecting the structural differences and functional characteristics of different economies and industrial segments within the global value chain.

In this context, complex network theory provides a powerful analytical tool [15]. In particular, centrality metrics are widely used to quantitatively measure the relative importance of nodes within a network and are key to understanding the role of nodes within a system’s hierarchy. However, earlier centrality metrics—such as degree centrality, closeness centrality, and betweenness centrality—while offering some explanatory power, primarily focus on first-order connections within the network and fail to capture the higher-order interactions and indirect effects that reflect systemic characteristics [16]. To address these limitations, some researchers have introduced centrality models capable of capturing indirect effects, such as eigenvector centrality and PageRank centrality [17]. The core idea behind these models is that a node’s importance is determined not only by the number of its direct connections but also by the importance of the nodes to which it is connected. In other words, being connected to more central nodes within the network enhances a node’s significance. Building upon this insight, this paper introduces the concept of ‘industry chain centrality’, aimed at capturing the higher-order connections and transmission effects within global production networks. This metric is calculated using production networks constructed from global or regional input–output tables and is intended to measure the pivotal position and structural influence of specific industries within the global production network.

Besides clarifying the conceptual meaning of industry chain centrality, it is also necessary to explicate the structural power it reflects. Within global production networks, industries with higher levels of industry chain centrality typically exhibit several structural characteristics: (1) they maintain extensive and tightly knit linkages with a large number and wide variety of upstream suppliers and downstream customers; (2) they perform a ‘hub’ function in cross-industry production and value transmission pathways, thereby substantially reducing the transaction costs of overall network connectivity; and (3) they form dense interactions with other high-impact core industries, collectively constituting clusters in which power is highly concentrated. This structurally central position endows these industries with pronounced forms of structural power [17,18]. More specifically, such industries possess stronger bargaining power in price negotiations, contract formation, and profit distribution; they can secure priority access to critical market information, advanced technologies, and scarce resources [19]; and they often assume leading roles in the formation and evolution of value-chain governance mechanisms—including standard-setting, certification systems, and compliance requirements. Consequently, the distribution of power within global production networks is markedly asymmetric. A limited number of countries or industries located at the core of the network, by virtue of their pivotal node positions, control a disproportionate share of structural power, thereby exerting profound influence on global value distribution patterns and pathways of functional upgrading [12,20].

2.2. Research Hypotheses

The threshold effect of industry chain centrality manifests in two ways. On the one hand, in industries with lower levels of industry chain centrality, the effect of the embeddedness of FIEs on functional upgrading is significantly constrained. First, at this stage, the industry is positioned at the periphery of the global value chain, with limited channels for knowledge spillover. Second, local firms possess weak bargaining power and high dependence on foreign firms, with the types of incoming foreign investment primarily being cost-driven. Consequently, the knowledge and technologies of these foreign firms struggle to diffuse within the local network, resulting in the limited conversion of foreign investment into substantial functional upgrading. On the other hand, when industry chain centrality is higher, the industry occupies a central position within the global production network, with dense knowledge transfer pathways and more balanced power dynamics. This position also, first, attracts higher-quality foreign investment oriented towards innovation and collaboration [17], and second, enables the advanced functions of foreign investment to be more easily absorbed and recreated by local firms. As a result, high-value-added activities such as R&D, management, and marketing are significantly promoted, leading to functional upgrading.

Building on this general distinction, the mechanisms underlying the threshold effect can be further elaborated. In industries with lower levels of industry chain centrality, the role of the embeddedness of FIEs in promoting functional upgrading is severely constrained. First, channels for knowledge spillover are limited. Low centrality indicates that industries are situated at the periphery or semi-periphery of the global production network, characterized by weak and sparse external connections and an underdeveloped local industry ecosystem. Under such conditions, new technologies and managerial knowledge introduced by foreign firms struggle to transfer and diffuse through the fragmented industry network [19], confining spillover effects to only a few local firms directly linked to foreign investors. Second, the lack of structural power and increased dependence exacerbate the situation. Local firms at the network periphery have very weak bargaining power when negotiating with powerful multinational corporations. They are more likely to accept unfavourable terms set by these corporations, passively taking on low-value-added tasks and thus falling into the trap of low-end lock-in [21]. Instead of fostering upgrading, foreign investment may reinforce their marginal position in the global division of labour [22]. Third, the ability to attract high-quality foreign investment is limited. While efficiency-seeking, “cost-driven” foreign investment may still enter these industries, “empowerment-driven” foreign investment, which seeks to leverage local innovation ecosystems and strategic partnerships, is unlikely to be attracted [23]. As a result, the types of foreign investment entering these industries are inherently less conducive to functional upgrading. Therefore, in the low-centrality range, the impact of foreign investment on functional upgrading may be weak, insignificant, or even negative.

In contrast, in industries with higher levels of industry chain centrality, the role of the embeddedness of FIEs in promoting functional upgrading is significantly amplified. First, the channels for knowledge spillover are dense and efficient. High centrality means that the industry benefits from a dense and mature industry ecosystem, comprising numerous local suppliers, customers, research institutions, and supporting service providers [24]. A more developed local network enables the rapid absorption, digestion, and diffusion of new technologies and management practices introduced by foreign firms. Knowledge and information can quickly spread throughout the entire industrial cluster via dense inter-firm connections, generating significant multiplier effects [25]. Second, structural power is enhanced, and win–win cooperation is fostered. Local firms and industry clusters located at the core of the network possess stronger attraction and bargaining power in relation to multinational corporations. Multinational firms, in order to enter or leverage this efficient local network, are more willing to establish equal partnerships with local firms, engaging in deeper technical cooperation and knowledge sharing, rather than merely exerting control and dominance [26]. This creates favourable conditions for local firms to achieve functional upgrading. Third, a virtuous cycle of attracting high-quality foreign investment is established. A high-centrality industrial cluster itself constitutes a valuable strategic resource. It attracts more empowerment-driven foreign investment, which seeks technological cooperation, market innovation, and knowledge creation. These high-quality foreign firms further strengthen the cluster’s central position and innovation capacity, creating a self-reinforcing positive cycle [27].

Based on the above analysis, industry chain centrality determines the nonlinear nature of the relationship between foreign-invested enterprise embeddedness and functional upgrading. Building on this theoretical framework, the paper proposes the following core research hypothesis: the impact of foreign-invested enterprise embeddedness on manufacturing functional upgrading follows a nonlinear relationship shaped by a threshold in industry chain centrality. Specifically, foreign-invested enterprise embeddedness can only have a significant and positive impact on functional upgrading when the industry’s level of industry chain centrality surpasses a certain threshold.

3. Materials and Methods

3.1. Model Construction

To capture the potential non-linear relationship between the embeddedness of foreign-invested enterprises and functional upgrading, this study employs the threshold regression model proposed by Hansen [28]. This method automatically partitions the sample into distinct regimes based on the value of the threshold variable and estimates separate coefficients for each regime. By determining the threshold endogenously, the model avoids the subjectivity associated with arbitrarily pre-setting cut-off points and enables a more accurate identification of segmented effects of the explanatory variable.

Building on this approach, the present study uses industrial chain centrality (Centrality) as the threshold variable to examine whether the embeddedness of FIEs exerts a non-linear influence on functional upgrading. The specified threshold model is expressed as follows:

$${Upgrade}_{ij,t}={\beta }_{1^{\prime }}{FIEs}_{ij,t}I({Centrality}_{ij,t}\le \gamma )+{\beta }_{2^{\prime }}{FIEs}_{ij,t}I({Centrality}_{ij,t}>\gamma )+\delta X_{ij,t}+u_{ij}+{\eta }_t+{\epsilon }_{ij,t}$$

(1)

In this model, i denotes the country, j represents the manufacturing sub-sector, and t indicates the year. Upgrade refers to functional upgrading, while FIEs represents the embeddedness of foreign-invested enterprises. Centrality serves as the threshold variable, referring to industrial chain centrality, and γ is the threshold value to be estimated. The indicator function I(⋅) equals 1 when the condition inside the parentheses is satisfied and 0 otherwise. $X_{it}$ represents a set of control variables, $u_{ij}$ captures country–industry fixed effects, ${\eta }_t$ denotes year fixed effects, and ${\epsilon }_{ij,t}$ is the random disturbance term. In the empirical procedure, we first test the statistical significance of the threshold effect and determine the number of thresholds. Based on the results of these tests, the appropriate threshold model will be estimated.

3.2. Selection of Variables

3.2.1. Explained Variable

The explained variable in this study is functional upgrading. Following the approach of Stöllinger [29], we use the project-level information recorded in the fDi Markets database regarding the primary activities performed by each greenfield investment project in the host country. These activities are matched to five distinct functional segments along the GVC, namely headquarters functions, research and design, manufacturing, logistics and retail services, and after-sales services. Subsequently, we calculate the share of greenfield investment projects undertaken by the manufacturing sector of the specific country that correspond to upstream and downstream functional activities, and compare this share with the corresponding global benchmark. Based on this comparison, we construct an index of functional specialization along the GVC, which reflects the extent to which manufacturing industries in the specific country have moved toward the two ends of the “smiling curve.” The index is computed as follows:

$${FS\_Total}_{it}={\displaystyle \frac{P_{cit}^{UD}/P_{cit}}{P_{wit}^{UD}/P_{wit}}}$$

(2)

where $P_{wit}^{UD}$ denotes the number of global greenfield investment projects in year t that fall within upstream functions (headquarters, research and design) and downstream functions (logistics, retail services, and after-sales services) in manufacturing industry i; $P_{wit}$ denotes the total number of global projects in industry i in year t. $P_{cit}^{UD}$ represents the number of greenfield projects in industry i and year t within the specific country that engage in upstream or downstream functional activities, while $P_{cit}$ denotes the total number of projects recorded for the same country, industry, and year. To mitigate potential sample loss arising from years in which no greenfield investment projects are recorded, the functional upgrading index is transformed by taking the natural logarithm after adding one.

3.2.2. Explanatory Variable

The core explanatory variable of this study is the embeddedness of FIEs. This indicator captures the depth of multinational enterprises’ participation in a given country–industry pair and the extent to which their activities are integrated into the host economy. We follow existing research and measure the embeddedness of FIEs using data from the OECD Activities of Multinational Enterprises (AMNE) database, calculating the share of value added generated by foreign-invested enterprises within an industry relative to the industry’s total value added. This measure provides a more accurate reflection of the substantive role played by foreign firms in sectoral production activities. Conceptually, the embeddedness of FIEs refers to the overall involvement of multinational enterprises in production, sourcing, and sales activities within the host country. Drawing on the value-added decomposition framework proposed by Wang et al. (2021) [30], we trace the value-added creation of multinational enterprises along the global value chain (GVC) to identify different layers of their embeddedness both within and across industries. Specifically, by decomposing the destination of value added generated by multinational enterprises’ affiliates located in the host country, we characterize the embeddedness of FIEs from two dimensions: forward linkages and backward linkages. Following the established literature, we adopt the average of forward embeddedness and backward embeddedness as indicators of the embeddedness of FIEs in this study.

3.2.3. Threshold Variable

The threshold variable used in this study is industrial chain centrality (Centrality). Following the literature [17], we construct a composite measure of centrality by taking the average of PageRank and CheiRank scores. Both measures are derived from the probabilistic transition properties of directed and weighted networks. Their underlying intuition is that the structural position and influence of each node within the industrial chain can be captured through its weighted directed connections.

We begin by constructing an $N\times N$ weighted and directed adjacency matrix W, which serves as the basis for computing centrality. The matrix element $W_{i,j}$ represents the weight of the directed edge from node i to node j. We further define the weighted out-degree as $d_j^{out}=\sum _i{W}_{j,i}$ and the weighted in-degree as $d_j^{out}=\sum _i{W}_{j,i}$.

Based on the above, two Markov transition matrices describing the random walk process over the network, S and $S^{\ast }$, are constructed as follows:

$$S_{i,j}=\left\{\begin{array}{ll}1/N,& \mathrm{if} d_j^{\mathrm{out}}=0\\ W_{j,i}/d_j^{\mathrm{o}\mathrm{u}\mathrm{t}},& \mathrm{if} d_j^{\mathrm{out}}\ne 0\end{array}\right.$$

(3)

$$S_{i,j}^{\ast }=\left\{\begin{array}{ll}1/N,& \mathrm{if} d_j^{in}=0\\ W_{i,j}/d_j^{in},& \mathrm{if} d_j^{in}\ne 0\end{array}\right.$$

(4)

the Google matrices G and $G^{\ast }$ are then constructed, and their dominant eigenvectors are obtained to compute PageRank centrality $C_{PR}\left(i\right)$ and CheiRank centrality $C_{CR}\left(i\right)$, respectively. PageRank reflects the importance of a node in terms of being “pointed to,” whereas CheiRank captures its importance in terms of “pointing to others.” Both metrics account not only for direct connections but also for high-order propagation effects embedded within complex networks. Their respective formulations are given by:

$$C_{PR}(i)=\alpha \sum _j{\displaystyle \frac{W_{j,i}}{d_j^{\mathrm{out}}}}{C}_{PR}(j)+{\displaystyle \frac{1-\alpha }{N}}$$

(5)

$$C_{CR}(i)=\alpha \sum _j{\displaystyle \frac{W_{j,i}}{d_j^{\mathrm{out}}}}{C}_{PR}(j)+{\displaystyle \frac{1-\alpha }{N}}$$

(6)

In matrix form, the PageRank solution can be expressed as:

$$C_{PR}={\displaystyle \frac{1-\alpha }{N}}(I-\alpha S{)}^{-1}e$$

(7)

$$C_{CR}={\displaystyle \frac{1-\alpha }{N}}(I-\alpha S{)}^{-1}e$$

(8)

where I is the N × N identity matrix and e is a vector of ones. Following Carvalho (2014), the damping factor α is set to 0.5 [31]. Because the World Input–Output Tables include both domestic and international production linkages, we follow Criscuolo and Timmis (2018) and decompose the inverse matrices $(I-\alpha S{)}^{-1}$ and $(I-\alpha S^{\ast }{)}^{-1}$ into domestic and foreign components [17]:

$$(I-\alpha S{)}^{-1}=(I-\alpha S{)}_{dom}^{-1}+(I-\alpha S^{*}{)}_{dom}^{-1}$$

(9)

$$(I-\alpha S^{\ast }{)}^{-1}=(I-\alpha S{)}_{fgn}^{-1}+(I-\alpha S^{*}{)}_{fgn}^{-1}$$

(10)

where $(I-\alpha S{)}_{fgn}^{-1}$ and $(I-\alpha S^{\ast }{)}_{fgn}^{-1}$ retain only the foreign-cycle (i.e., cross-border input–output) elements, with domestic-cycle elements set to zero. The resulting foreign PageRank $C_{PR}^{fgn}$ and foreign CheiRank $C_{CR}^{fgn}$ effectively strip out purely domestic input–output linkages, thereby providing a more accurate representation of a country or industry’s status within the global production network. We use the average of foreign PageRank and foreign CheiRank as the measure of a country or industry’s position in the global value chain. This composite indicator captures both the extent to which a node is dependent on others (controlledness) and the extent to which it exerts control over others (controlling power), thereby offering a dynamic and structure-sensitive perspective on global value chain positioning.

3.2.4. Control Variable

To mitigate potential omitted variable bias, we include a series of industry-level control variables in the model to capture other factors that may influence functional upgrading in the manufacturing sector [32]. Specifically, the following variables are incorporated. (1) Trade openness (Open): measured as the ratio of total exports to total output in each industry. Higher trade openness implies greater exposure to international competition as well as increased opportunities for learning and technology absorption, both of which may promote functional upgrading. (2) Economic development level (GDPpc): captured by the natural logarithm of industry-level GDP per capita, reflecting the overall development foundation of each industry. (3) Capital intensity (CI): measured by the total fixed asset formation of each manufacturing industry, indicating the extent of capital investment. In addition, we incorporate several country-level institutional variables to control for the influence of the macro-institutional environment on industrial upgrading. Rule of Law (ROL) reflects societal confidence in the legal environment, including the quality of contract enforcement, property-rights protection, police and judicial systems, and the likelihood of crime and violence. Political Stability and Absence of Violence/Terrorism (PSAVT) captures the probability of political instability and/or politically motivated violence, including terrorism. Voice and Accountability (VA) measures the perceived extent to which citizens in a country can take part in choosing their government, and the level of freedom of expression, association, and the press.

3.3. Data Sources and Descriptive Statistics

The data on the embeddedness of FIEs are obtained from the latest 2024 release of the OECD AMNE database. This database is an extension of the Inter-Country Input–Output (ICIO) tables along the ownership dimension, distinguishing between domestic and foreign enterprises, thereby ensuring consistency with the unit of analysis used in constructing the functional upgrading index. It is noteworthy that the 2024 edition introduces an upward revision to the criteria used to identify foreign-invested enterprises in China: the ownership threshold was raised from at least 25% to 50%, which helps avoid overstating the actual significance of foreign firms in the host economy. The functional upgrading indicator is sourced from the fDi Markets database. Trade openness and other industry-level control variables are drawn from the OECD Trade in Employment (TiM) database and the ICIO tables. Government governance variables, including Government Effectiveness and Rule of Law, are obtained from the Worldwide Governance Indicators (WGI) database. Combining these data sources, we construct an unbalanced panel dataset covering 42 economies and 14 industries over the period 2003–2020. The descriptive statistics of the variables are presented in

Table 1. Descriptive statistics.

Variable	N	Mean	Sd	Min	Max
Upgrade	6757	0.609	0.379	0	2.303
FIEs	6757	22.21	7.703	1.696	51.46
Centrality	6757	1.238	1.475	0.204	15.860
Open	6757	0.639	0.401	0.001	1.864
GDPpc	6757	5.035	1.111	0.680	9.522
CI	6757	7.305	1.700	1.668	12.650
ROL	6757	0.771	0.906	−0.874	2.125
PSAVT	6757	0.281	0.819	−2.376	1.687
VA	6757	0.674	0.853	−1.907	1.801

.

4. Results

4.1. Multiple Collinearity Test

The first step in the analysis was to perform a variance inflation factor (VIF) test on the panel data. The results of this test are presented in

Table 2. VIF multicollinearity test results.

Variable	VIF	VIF/1
ROL	4.690	0.213
PSAVT	3.520	0.284
VA	2.410	0.416
GDPpc	2.090	0.478
CI	1.220	0.822
Open	1.190	0.841
FIEs	1	0.999
Mean	VIF	2.300

. As shown in Table 2, the highest VIF value among all variables is 4.690, which is well below the threshold of 10. Therefore, we can conclude that multicollinearity is not a concern in this model.

4.2. Selection of Panel Data Model

Before proceeding with the parameter estimation, it is necessary to test the specification of the panel data model to ensure that the appropriate model is selected. This helps avoid significant bias in the parameter estimates due to incorrect model specification. To determine the appropriate panel data model, we perform both the F-test and Hausman test [33]. The results of these tests are shown in

Table 3. Model Specification Tests for Panel Data.

Test Type	F-Test	Hausman Test
Statistic type	F-statistic	Chi-Sq. Statistic
Statistic value	8.93	50.62
p-value	0.000	0.000
Model decision	Individual fixed-effects model	Fixed-effects model

. The p-values for both the F-test and the Hausman test are 0.000, indicating that a fixed-effects model should be selected.

4.3. Threshold Effect Significance Test

Before constructing the threshold regression model, it is essential to test for the existence of threshold effects in order to determine whether the model exhibits a nonlinear structure and to identify the number of threshold values. In this study, we employed the bootstrap method proposed by Hansen (1999) [28], performing 1000 resampling iterations to test for potential single, double, and triple threshold effects. The detailed results of these tests are presented in

Table 4. Significance test and confidence intervals of industrial chain centrality.

Threshold	F-Value	p-Value	10%	5%	1%	Threshold Estimates	95% CI
Single	185.890	0.026	135.478	158.713	220.242	1.941	[1.915, 2.000]
Double	80.120	0.285	115.941	137.280	205.925
Triple	93.480	0.833	276.670	321.136	395.536

.

The results in Table 4 first indicate the presence of a significant single-threshold effect. In the single-threshold test, the F-statistic reaches 185.890, with a corresponding p-value below 0.05, allowing us to reject the null hypothesis that “the embeddedness of foreign-invested enterprises has no threshold effect on functional upgrading.” This confirms that when industrial chain centrality is employed as the threshold variable, there is indeed at least one structural break in the relationship between the embeddedness of foreign-invested enterprises and functional upgrading, thereby suggesting a nonlinear effect. Building on this, we further examine whether a dual-threshold structure exists. The results report an F-statistic of 80.120 with a p-value of 0.285, which fails to reject the null hypothesis of “a single-threshold effect only.” A similar pattern is observed in the triple-threshold test, where the statistic again lacks significance. Therefore, the model is determined to contain only one statistically significant threshold and does not exhibit a multiple-threshold structure. Taken together, these findings indicate that the relationship between the embeddedness of foreign-invested enterprises and functional upgrading is not linear, but rather undergoes a structural shift at a critical point, illustrating a typical nonlinear effect driven by industrial chain centrality.

Table 4 also presents the estimated threshold value. The estimated value for the single threshold is 1.941, with a 95% confidence interval of [1.915, 2.000]. According to the basic principle of the threshold model, when the likelihood ratio (LR) statistic approaches 0, the corresponding γ represents the threshold value. Figure 1 illustrates the likelihood ratio function for the threshold estimate of 1.941 within the 95% confidence interval. In the figure, the lowest point of the LR curve corresponds to the true threshold estimate, and the dashed line indicates the critical value of 7.35. Since the LR curve is significantly below this critical value near the threshold point, it indicates that the threshold estimate has passed the significance test and is statistically valid and reliable.

4.4. Threshold Regression Model Estimation Results

Table 5. Threshold regression results.

	(1)	(2)
	Linear Regression	Threshold Regression
FIEs	0.001
	(0.114)
FIEs (centrality ≤ 1.941)		0.001
		(0.083)
FIEs (centrality > 1.941)		0.161 ***
		(3.239)
Open	0.008	0.006
	(0.235)	(0.187)
GDPpc	−0.006	−0.008
	(−0.319)	(−0.394)
CI	0.005	0.001
	(0.316)	(0.086)
ROL	−0.068 *	−0.069 *
	(−1.780)	(−1.813)
PSAVT	0.024	0.025
	(1.437)	(1.470)
VA	0.058 *	0.059 *
	(1.675)	(1.695)
Constant	0.571 ***	0.591 ***
	(5.010)	(5.183)
Country FE	YES	YES
Industry FE	YES	YES
Year FE	YES	YES
Observations	6757	6757
R2	0.013	0.015

Note: *, *** indicate significance at the 10%, and 1% levels, respectively.

presents the final estimation results of the threshold regression model based on the identified threshold value. For comparison, we also employed a fixed-effects model to test the model parameters. The results in Table 5 clearly show that the fixed-effects model (column 1) indicates that the effect of the embeddedness of foreign-invested enterprises on functional upgrading is not significant. In contrast, the threshold regression results (column 2) reveal that the regression coefficients for the embeddedness of foreign-invested enterprises differ significantly across different threshold ranges. This strongly supports the existence of a single threshold effect and affirms the core hypothesis of this study. Specifically, when the centrality of the industrial chain is at a low level (i.e., Centrality ≤ 1.941), the regression coefficient for the embeddedness of foreign-invested enterprises on functional upgrading is 0.001, but it is statistically insignificant. This suggests that for industries on the periphery of global production networks, with weak ties to core networks, simply increasing the embeddedness of foreign-invested enterprises does not effectively promote functional upgrading. In contrast, when an industry’s industrial chain centrality surpasses the threshold value of 1.941 and enters the high-level range, a fundamental change occurs. The regression coefficient for the embeddedness of foreign-invested enterprises on functional upgrading becomes 0.161, and it is highly significant at the 1% level. This means that for industries that have become hubs in global or regional production networks, the embeddedness of foreign-invested enterprises can significantly promote functional upgrading. In this range, for every 1% increase in the embeddedness of foreign-invested enterprises, the corresponding functional upgrading index increases by 0.161 percentage points. This result suggests that a high-centrality industrial ecosystem, with its dense channels for knowledge spillovers, strong bargaining power, and attraction of high-quality foreign investment, can effectively transform the potential of foreign investment into the driving force for functional upgrading. The results for the control variables are also largely consistent with expectations.

4.5. Robustness Tests

To further verify the robustness of the regression results, this section employs several additional approaches, including using a one-period lag of the key explanatory variable and clustering standard errors at the industry level.

4.5.1. One-Period Lag of Explanatory Variable

Considering that the effect of the embeddedness of foreign-invested enterprises on functional upgrading may exhibit time lags—because technology spillovers and managerial improvements often materialize gradually—we re-estimate the threshold regression model after lagging the explanatory variable by one period. Column (1) of

Table 6. Robustness Tests: Lagged Explanatory Variable and Industry-Clustered Errors.

	(1)	(2)
	Lagged Explanatory	Industry-Clustered SEs
FIEs (centrality ≤ threshold value)	0.002	0.001
	(0.299)	(0.109)
FIEs (centrality > threshold value)	0.162 ***	0.164 ***
	(3.246)	(3.268)
Control variables	YES	YES
Country FE	YES	YES
Industry FE	YES	YES
Year FE	YES	YES
Observations	5731	6757
R-squared	0.019	0.014

Note: *** indicate significance at the 1% levels.

presents the results. The direction and significance of the coefficients across all threshold intervals remain consistent with the baseline results, indicating strong robustness of the main findings.

4.5.2. Industry-Clustered Robustness Test

Given the potential technological heterogeneity across industries, which may induce within-industry correlation in the error terms and affect the accuracy of standard errors, we additionally apply industry-clustered robust standard errors. Following the industry technology-intensity classification published by OECD (2007), industries are grouped into low-, and high-intensity categories, and standard errors are clustered based on this grouping [34]. As shown in Column (2) of Table 6, after applying industry-clustered robust standard errors, the magnitude and statistical significance of the coefficients for the embeddedness of foreign-invested enterprises remain largely unchanged. This further confirms the robustness of the empirical results.

4.6. Heterogeneity Analysis

According to Zhong et al. (2021), functional specialization refers to the upgrading of functional comparative advantages that emerges as economies shift from manufacturing activities toward higher value-added functions such as R&D, management, and marketing, based on their specialization advantages [35]. Building on this definition, we examine the impact of the embeddedness of foreign-invested enterprises on functional upgrading across different dimensions and explore the heterogeneity of these effects. We use the updated occupational skills database developed by Kruse et al. (2024) and match it with the Asian Development Bank (ADB) input–output database to construct indicators of functional upgrading at the country–industry level [36]. Specifically, production functions are classified according to workers’ occupational types. Following the correspondence between ISCO-88 occupational categories and production functions provided in the appendix of Kordalska and Olczyk [37], we categorize production functions into four types: (1) R&D functions, involving knowledge-intensive occupations such as engineering, health services, and education; (2) management functions, including legislators, senior managers, and decision-making positions; (3) fabrication functions, corresponding to technical workers, machine operators, and transport equipment operators; and (4) marketing functions, encompassing clerical work, sales, and customer service.

Table 7. Heterogeneity analysis.

	(1)	(2)	(3)
	R&D	Management	Marketing
FIEs (centrality ≤ threshold value)	0.001	−0.008	−0.000
	(0.471)	(−0.331)	(−0.123)
FIEs (centrality > threshold value)	0.024 ***	0.493 ***	0.002 **
	(11.117)	(12.959)	(2.456)
Control variables	YES	YES	YES
Country FE	YES	YES	YES
Industry FE	YES	YES	YES
Year FE	YES	YES	YES
Observations	4496	4530	4592
R-squared	0.079	0.170	0.056

Notes: The threshold values corresponding to R&D function upgrading, management function upgrading, and marketing function upgrading are 1.945, 1.658, and 1.399, respectively. **, *** indicate significance at the 5%, and 1% levels, respectively.

reports the regression results for functional upgrading across the three dimensions. Columns (1)–(3), respectively, use R&D, management, and marketing specialization within manufacturing value chains as the dependent variables, with all models controlling for country, industry, and year fixed effects. As shown in Column (1), there is a significant threshold effect in the impact of the embeddedness of foreign-invested enterprises on R&D upgrading. When industrial-chain centrality is below 1.945, the coefficient is 0.001 and statistically insignificant. When centrality exceeds 1.945, the coefficient becomes positive (0.024) and statistically significant at the 1% level. This indicates that the promotion of R&D functions by foreign-invested enterprises emerges only when the industry reaches a relatively high level of centrality. Column (2) shows that management upgrading also exhibits a threshold pattern. When centrality is below 1.658, the effect is negative and insignificant; once the threshold is exceeded, the coefficient becomes 0.493 and significantly positive. This suggests that management upgrading requires a lower centrality threshold than R&D upgrading, but still depends on achieving a certain degree of integration within the production network. Column (3) indicates a similar threshold pattern for marketing functions. When centrality is below 1.399, the embeddedness of foreign-invested enterprises has no significant effect. Beyond this threshold, the coefficient increases to 0.002 and is significant at the 5% level. Taken together, these results reveal that the dependence of functional upgrading on industrial-chain centrality varies across functions: R&D upgrading requires the highest level of centrality, followed by marketing, while management has the lowest threshold.

The finding that the industry chain centrality threshold required for R&D function upgrading is higher than that for management and marketing upgrading is consistent with the differentiated knowledge conditions associated with these functions. Upgrading into R&D is widely recognized as the most demanding step in the functional hierarchy, as it requires the absorption and recombination of highly tacit, complex, and cumulative knowledge, which can only occur in industries embedded in dense and structurally central network positions [38]. Recent empirical studies also show that innovation spillovers are strongly conditioned by production-network centrality: Ito (2023) demonstrates that industries located in central “hub” positions within global production networks benefit disproportionately from international R&D spillovers [38], while Chen and Liu (2022) find that innovation in China diffuses primarily through supplier–customer linkages rather than geographic proximity, implying that only sufficiently central network structures can sustain meaningful technological upgrading [39]. In contrast, management upgrading requires lower levels of ecosystem sophistication, as managerial and organizational practices can diffuse through moderately dense inter-firm networks, while marketing upgrading is even less dependent on production-network centrality because it relies more on market access and demand-related information. Together, these differences in knowledge complexity and network dependence explain why the estimated thresholds follow a clear hierarchy—R&D > management > marketing—reflecting the increasingly stringent structural conditions necessary for foreign-invested enterprise embeddedness to translate into substantive functional upgrading. In addition, the case of Germany further illustrates this mechanism: as a highly central economy, Germany shows predominantly positive effects of FIE embeddedness on functional upgrading in industries such as Other transport equipment, Basic metals, Machinery and equipment, and Chemicals, where industry chain centrality is high. In contrast, in sectors with lower centrality—such as Food products, beverages and tobacco—the impact is not significant, demonstrating that the benefits of FIE embeddedness materialize only when the required centrality threshold is met.

5. Discussion

The core finding of this study is that the impact of the embeddedness of FIEs on the functional upgrading of the manufacturing sector is not linear, but instead exhibits a pronounced nonlinear pattern characterized by a threshold in industrial chain centrality. Only when an industry’s centrality within the production network surpasses a certain critical value can the embeddedness of FIEs effectively promote its transition into higher value-added functions. This result provides a new systems-level perspective for understanding the complex interaction between foreign capital and host-country upgrading, and it also carries important implications for understanding how developing countries can formulate effective industrial policies in the context of globalization.

The present analysis further shows that industrial chain centrality is the key factor for explaining the positive effects of the embeddedness of FIEs. This finding provides strong evidence for reconciling and interpreting the contradictory conclusions reported in the existing literature. For decades, scholars have debated whether FDI facilitates or inhibits industrial upgrading in host economies [40]. Proponents argue that FDI can significantly enhance the productivity and innovation capacity of domestic firms through technology spillovers, managerial knowledge transfer, and expanded market access [41,42]. Critics, however, contend that under the unequal governance structures of global value chains, multinational corporations may leverage their dominant positions to lock host-country firms into low-value-added assembly tasks, thereby hindering functional upgrading [1,22]. The threshold-effect model developed in this study indicates that these seemingly contradictory views are in fact manifestations of the same underlying mechanism operating under different structural conditions. Whether the embeddedness of FIEs promotes or suppresses industrial upgrading largely depends on the structural position of the host-country industry within global production networks.

Industrial chain centrality, as an indicator of structural power within an industry, plays a critical role in distinguishing between the two contrasting outcomes observed in the embeddedness of FIEs. When an industry is positioned at the periphery of the global production network (low centrality), its channels for knowledge absorption are often limited, bargaining power is weak, and it is more susceptible to domination by multinational corporations, which makes it difficult for the positive effects of foreign capital to materialize [43,44]. In contrast, when an industry is located at the core of the network (high centrality), its dense industrial linkages, mature ecosystems, and stronger bargaining power allow it to maximize the potential spillover effects of foreign capital, thereby activating its positive role in promoting functional upgrading [45]. Thus, by introducing ‘industrial chain centrality’ as a systemic variable, this study provides an integrated analytical framework to reconcile the debates in the existing literature.

Furthermore, this study contributes to the shift in GVC research from the traditional ‘chain-based’ paradigm to a more ‘networked’ perspective. Traditional GVC analysis often depicts the production process as a linear, value-adding chain extending from upstream to downstream. However, with the increasing fragmentation and complexity of global production, inter-industry links have evolved into intricate, interconnected networks [46]. By introducing the concept of centrality from complex network theory, this study aims to capture these network structural characteristics. Industrial chain centrality not only reflects the number of direct connections within an industry but, more importantly, it embodies the transmission effects and influence within higher-order networks—where the importance of an industry depends on the significance of the industries to which it is connected. This indicator provides a powerful analytical tool for quantifying and understanding the power structure and uneven development within GVCs, where related analyses often require sophisticated methods to handle large-scale datasets [47]. Through empirical testing of the threshold effect of centrality, this study reveals how an industry’s ‘structural position’ within the global production system fundamentally determines its upgrading path and its efficiency in utilizing foreign capital. This opens up new theoretical avenues for future GVC research, emphasizing the need for greater attention to systemic factors such as network topology, node positions, and power distribution in shaping industrial development.

The conclusions of this study have significant policy implications, especially for developing countries seeking to leverage foreign capital for industrial upgrading. First, the focus of industrial policy should shift from merely attracting FDI to enhancing the industrial chain centrality of domestic industries. Traditional policies often prioritize the total inflow of FDI, assuming that foreign capital will naturally bring about technology transfer and growth. However, the findings of this study caution that if domestic industries are positioned at the periphery of the global production network, simply attracting large amounts of foreign capital may not lead to the expected functional upgrading, and may even exacerbate dependency on multinational corporations. Therefore, policymakers need to adopt a more strategic perspective, placing the cultivation and enhancement of domestic industries’ central position within global networks as a prerequisite for effectively utilizing foreign capital. This approach entails substantial investments in infrastructure, reducing transaction costs, fostering mature supplier networks, and building dynamic industrial clusters [48]. Only when the domestic industrial ecosystem is strong enough to play an indispensable role as a hub in the global network can foreign capital become a catalyst for industrial upgrading.

Furthermore, foreign capital policy should be differentiated and dynamic, with targeted strategies based on the level of industrial chain centrality in different industries. For industries with low industrial chain centrality, the focus of policy should not be on blindly attracting foreign capital, but rather on strengthening the foundational aspects of the domestic industry. This includes supporting domestic firms in technological innovation, strengthening domestic industry linkages, and cultivating specialized talent to gradually enhance the firms’ connectivity and influence within the global network. At this stage, the government can selectively attract ‘empowering’ foreign capital to fill domestic technological gaps and drive local supply chain development, rather than merely pursuing ‘efficiency-oriented’ foreign capital focused on cost advantages. For industries that have already surpassed the centrality threshold, policy should shift towards attracting higher-quality FDI and promoting its deep integration with the domestic innovation system. For example, encouraging multinational corporations to establish R&D centers locally, collaborate with local universities and research institutions, and jointly develop industry standards [49]. This phased, differentiated policy framework can more effectively guide foreign capital to serve the strategic goals of long-term industrial development, avoiding the negative effects of a one-size-fits-all approach.

Despite the significant findings of this study, several limitations remain, which also point to fruitful directions for future research. First, functional upgrading and the embeddedness of FIEs are multidimensional ideas for variable measurement. The value-chain functional distribution index measures functional upgrading overall but cannot identify process, product, and chain upgrading [50]. The embeddedness of FIEs statistic does not distinguish between FDI from various nations or motivated by different reasons (e.g., market-seeking or efficiency-seeking), yet such heterogeneity surely affects spillover effects. Future research could use micro-level, firm-based datasets to show upgrading types and FDI features more nuancedly, revealing deeper mechanisms. Second, while this study employs a fixed-effects model to account for time-invariant industry features, potential endogeneity—especially reverse causality—remains a methodological concern. Industries with stronger upgrading potential and greater structural centrality may attract higher-quality foreign capital, thereby biasing coefficient estimates. To more rigorously identify causal linkages, future studies may adopt instrumental variables or quasi-experimental approaches such as natural experiments or regression discontinuity designs. Third, industry-level analysis hides firm-level heterogeneity in this study. Absorptive capacity, ownership structure, and firm size vary widely within an industry, which may affect the embeddedness of FIEs. Thus, studying the relationships between business characteristics, foreign capital links, and network positions at the firm level is a promising research topic. Future research could examine whether state-owned versus private firms upgrade differently when faced with foreign capital shocks and how local industrial cluster embedding affects firms’ ability to benefit from foreign capital spillovers [51].

6. Conclusions

In the context of deepening global economic integration, understanding how the embeddedness of FIEs can facilitate functional upgrading in the manufacturing sector has become an urgent and significant research question. The key to addressing this issue lies in identifying the structural conditions that shape the mechanisms through which the embeddedness of FIEs exerts its influence. Drawing on a complex network systems perspective, this paper challenges the traditional linear assumptions regarding the effects of FIE embedding and introduces industrial chain centrality as a critical threshold variable. In doing so, it systematically uncovers the nonlinear mechanism through which the embeddedness of FIEs affects the functional upgrading of manufacturing. Based on panel data for cross-economy manufacturing industries from 2003 to 2020, the empirical analysis yields several core conclusions. The most significant finding to emerge from this investigation is that the impact of the embeddedness of FIEs on manufacturing functional upgrading exhibits a clear threshold effect. Results from the threshold regression model demonstrate that only when industrial chain centrality—defined as an industry’s hub position and structural influence within the global production network—exceeds a specific critical value does the embeddedness of FIEs significantly promote functional upgrading. Conversely, when industrial chain centrality falls below this threshold, the positive effect of the embeddedness of FIEs becomes negligible and may even result in a ‘lock-in’ at the lower end of the value chain due to insufficient structural power. These findings remain robust when the explanatory variable is lagged by one period and under varying clustering of standard errors, thereby reinforcing the validity of the identified threshold effect.

In turn, this study yields several important theoretical and policy implications. On the theoretical side, the findings offer fresh insight into how the structural configuration of global production networks shapes the spillover effects of FIEs and influences upgrading trajectories within global value chains. This perspective advances the literature beyond a predominantly “chain-based” division of labour toward a more nuanced, “network-oriented’’ analytical framework. On the policy side, the results indicate that the effectiveness of industrial upgrading strategies hinges not only on the scale of inward FDI but also on the quality of the domestic industrial ecosystem and its centrality within global production networks. Policymakers should therefore incorporate industrial chain centrality into the design of FDI-related strategies. Only when domestic industries possess relatively high network density and structural advantages can the embeddedness of FIEs effectively promote functional upgrading rather than exacerbate lock-in risks.