Heterogeneous spillover effects of outward FDI on global value chain participation

This study delves into the effects of outward foreign direct investment (FDI) on global value chain (GVC) participation from 2000 to 2014. The utilization of traditional panel models, the spatial Durbin model (SDM), and the threshold model provides a comprehensive understanding of the heterogeneous spillover effects of outward FDI. The results show that increased outward FDI not only facilitates the GVC participation of parent countries but also has a profound impact on that of other countries. The spillover effects of outward FDI play a vital role in the GVC participation of low total factor productivity (TFP) countries. However, for developed countries with high TFP levels, outward FDI has positive impacts on deep GVC participation while not influencing shallow participation. These findings serve as an extension to the relevant theories and suggest a way for developing countries to capture gains from outward FDI and participate further in GVCs.


Introduction
The accelerating globalization of production, coupled with capital account liberalization at the multilateral level, prompts firms to invest abroad to expand their operations. Despite the increasing attention paid to the complementary or substitutionary effects of outward foreign direct investment (FDI) on employment, investment, trade in exports, and so on, the relationship between outward FDI and global value chain (GVC) participation has been neglected by mainstream academics to some extent. Hence, our study intends to fill this gap and explores the potential mechanisms within the framework of GVCs and the "knowledge capital model" (KC model). In addition to traditional panel models, we estimate the spatial Durbin model (SDM) jointly with the threshold model to investigate the heterogeneous spillover effects of outward FDI on GVC participation. By doing so our study has sought to act as an extension to the existing theories on multinational enterprises (MNEs) and GVCs and lends support to the opening-up strategies implemented in many developing and transition economies.
The rest of this paper is arranged as follows: Section 2 explores the potential mechanisms and presents the main hypotheses based on the relevant literature; Section However, the preponderance of extant literature favors the complementary effects of outward FDI. On the one hand, through technology spillovers, marketing skills, newly developed products, and so on, outward FDI is conducive to the expansion of the industry size in host countries, thus inducing more exports of intermediate inputs from parent countries (Walid Hejazi and Peter Pauly 2003;Peter Egger 2007;Zitian V. Chen, Jing Li, and Daniel M. Shapiro 2012). Subsequent to the increasing value added embedded in the exported intermediate inputs, the GVC participation of the home country would show a rise, which may surpass the initial reduction. On the other hand, investing in low-cost host countries helps to cut down the production costs of MNEs, promote economies of scale, and improve their competitiveness (Jaan Masso, Urmas Varblane, and Priit Vahter 2007). Moreover, the repatriation of profits brought by such improvement to the parent firms allows the scaling up of production and boosts R&D investment, hence contributing to increasing the GVC participation of the parent country. Summarizing this line of argument produces the first hypothesis.
H1. An increase of outward FDI may have positive effects on the GVC participation of parent countries.
Meanwhile, outward FDI may benefit the GVC participation of the host countries when the outward investors relocate the production of intermediate inputs to these countries. The local producers may gain "spillover" effects from externalities generated by the existence of transnational firms, since the monopolistic advantages of the outward investors may not be completely internalized by the foreign affiliate firms and thus spill over to local firms in the host countries. As verified by some empirical evidence, the spillover effects of outward FDI facilitate the product innovation of local firms (Jaya P. Pradhan 2004; Pinelopi K. Goldberg et al. 2010). The improvement in innovation ability facilitates more GVC participation of the host countries. Besides, this can be extended a little further to other countries. A spatial perspective may enable a better understanding of the spillover effects of outward FDI in other countries. As Zoltan J. Acs, Luc Anselin, and Attila Varga (2002)  Furthermore, there are obvious distinctions of outward FDI between developed and developing countries. The monopolistic advantages of outward investors from developing countries are different with those from developed countries, due to different levels of productivity, infrastructure, industry, skills, and so on. More specifically, multinationals from developing economies show competition generally in price rather than sophisticated product diversification or cutting-edge technology (Pradhan 2004;Hsiu-Yun Lee, Kenneth S. Lin, and Hsiao-Chien Tsui 2009). Hence, less developed countries (LDCs) are more likely to act as platforms for exporting back to their parent countries (Stephen R. Yeaple 2003). In particular, the adverse spillovers of advanced technology in management, production, and marketing from the host countries could also promote the GVC participation of developing countries. By contrast, in many developed countries, like the United Kingdom, Korea, and the US, vertical outward FDI prevails, which follows a pattern in accordance with the comparative advantage (Maskus and Allan Webster 1995;Yeaple 2003). These distinctions in outward FDI may result in different effects on GVC participation. Accordingly, we posit that: H3. There may be a broad range of effects of outward FDI on GVC participation in developing and developed countries.
In a nutshell, the question on the effects of outward FDI on GVC participation remains very much unexplored, which may partially be attributed to the limited quantity of empirical studies conducted on this issue. Moreover, the relevant literature generally has not considered the spatial dependence, namely the correlation of GVC participation among neighboring countries. This omission motivates our study to utilize spatial econometric techniques to investigate the spillover effects of outward FDI on GVC participation. The consideration of the parent and other countries, together with developing and developed countries, allows for a more comprehensive view of the effects that outward FDI has on GVC participation.

Research Methodology
This section illustrates the details of the construction of the baseline model, the SDM, and the threshold model as well as the spatial autocorrelation test with Moran's I index.

Baseline Model
The specification of the baseline model is shown in formula (1): = 0 + 1 + 2 + + + (1) where and indicate the country and year, respectively, is the logarithm of the overall GVC participation, which consists of two parts, shallow participation and deep participation; denotes the logarithm of outward FDI; represents a series of control variables, including imports and exports of intermediate goods, foreign direct investment (FDI), the labor force, the capital density, human resources, research and development (R&D) expenditure, infrastructure level, and the Theil index; and are the country dummy and year dummy, respectively ; and is the error term.
The underlying assumption of the baseline model is that the investigated countries are spatially separated and independent of each other. However, on the one hand, the rapid development of GVCs leads to closer cooperation and dependence among countries; on the other hand, as mentioned previously, the GVC participation of one country is influenced not only by its outward FDI but also by other countries' outward FDI, the neglect of which may result in omitted-variable bias and endogeneity. Therefore, we construct the SDM with a spatially lagged dependent variable and the key independent variable to determine the real effects of outward FDI. Before that, the spatial autocorrelation test is applied to verify the existence of GVC participation correlation among countries.

Spatial Autocorrelation Test
The global Moran's I index is utilized to diagnose the spatial autocorrelation of GVC participation, as shown in formula (2): where n is the number of countries; and indicate the GVC participation of country and country , respectively; ̅̅̅̅̅̅ represents the sample mean of GVC; is the geographical distance-based weight matrix, the element of which is calculated using formula (3): is the geographical distance between and ; when = , equals 0. The value of Moran's I is between [-1,1]. Its main statistical test index is the Z value, as shown in formula (4): ( where E(Moran's I) denotes the mathematical expectation of Moran's I and VAR (Moran's I) is the variance. The test of spatial autocorrelation makes a difference only when the Z-statistic is significant at least at the 10% level. Then, if Moran's I index is higher than zero, the autocorrelation of GVC participation in different countries is positive, implying the spatial agglomeration of regions with similar higher levels of GVC participation and areas with lower GVC participation; if the index is lower than zero, the spatial autocorrelation is negative, indicating a significant difference between country and its neighboring countries in GVC participation; finally, if the index is close to zero, different regions' GVC participation is independent.

Spatial Durbin Model
The SDM is thought to be more efficient than the other spatial models, like the spatial lag model (SLM) and the spatial error model (SEM) (James P. LeSage and Kelley R. Pace 2008). It can estimate the direct and indirect effects of each variable without prior restrictions, while these effects are nearly the same in the SLM and the SEM (Paul J. Elhorst and Solmaria H. Vega 2013). The specification of the SDM is shown in formula (5): , , , and are the same as those in formula (1); is the geographical distance-based weight matrix; represents the parameter for the spillovers of GVC participation from neighboring countries; and is the parameter for outward FDI spillovers. We estimate formula (5) by utilizing maximum likelihood (ML) techniques rather than ordinary least squares (OLS) to obtain consistent and unbiased estimators.
Our study is more than a parameter estimation of or , because the existence of spatial correlation of GVC participation indicates that GVC participation in one country is influenced not only by its own outward FDI but also by the outward FDI from other countries. The approach developed by LeSage and Pace (2008) is applied to understand this point better. Rearranging formula (5) yields: includes , , and ; incorporates , , and ; and ( ) equals (1 − ) −1 . The extensive form of formula (6) is: where ( ) is the element of matrix ( ). According to LeSage and Pace (2008), the mean value of the diagonal elements in matrix ( ) is defined as the direct effects, denoting the average effects of outward FDI from parent countries on GVC participation, while the average of a row or column sum of non-diagonal elements is regarded as the indirect effects, which are produced by the variation of outward FDI from other regions.

Threshold Model
As mentioned previously, the obvious distinctions in outward FDI between developed and developing countries may cause heterogeneous effects on GVC participation. Hence, we employ the threshold model proposed by Bruce E. Hansen (1999) and set the total factor productivity (TFP) as the threshold variable to distinguish different levels of economic development. The fundamental form with a threshold is shown in formula (8): is the threshold variable, denotes the threshold value, and (•) represents the indicator function. Given the purpose of our research, we obtain formula (9): is the same as in formula (8), representing the different levels of development. Following Stephanie Kremer, Alexander Bick, and Dieter Nautz (2013), we first calculate the critical value of the TFP and confidence intervals with formula (9), then we divide the samples into sub-panels according to the critical value, and finally we estimate these sub-panels with spatial econometric techniques. Additionally, bootstrap methods are adopted to obtain the critic values, which are verified by the likelihood ratio.

Variable Definitions and Data Sources
Our panel data set ranges from 2000 to 2014 and covers 42 World Input-Output Database (WIOD) countries, which account for approximately 76% of exports and 95% of imports of intermediate goods and 80.85% of the global outward FDI in 2014. Our main data sources are the WIOD, World Development Indicators (WDI), Organization for Economic Cooperation and Development (OECD), Penn World Table (PWT), CEPII, World Bank (WB), and so on. Details of the calculation and data processing are provided in sequence.
As mentioned earlier, in Figure 1, the measurement of GVC participation is based on the decomposition of the value added created by domestic production. Only the value added crossing national borders for production purposes rather than consumption purposes is treated as GVC participation. Further, according to the times of crossing national borders, the overall GVC participation can be divided into a shallow part and a deep part, the former crossing a national border only once and the latter traversing national borders at least twice. Taking a chip firm in Japan as an example, if the chips are exported to China for toy manufacturing and ultimately consumed in China, then the value added embedded in the chips belongs to the shallow GVC participation of Japan. If those Chinese-made toys are exported to a third country, the US, for example, then the value added is treated as deep GVC participation. Mathematical expressions of the decomposition are shown in formula (10) 1 : 1 For simplification, we assume that there are G countries and N industries, X is a GN*1 vector of gross outputs, and Value-added is a 1*GN vector. The intermediate products Z plus the final outputs Y equal the gross outputs, that is, + = ; the direct input coefficient A can be defined as divided by on its diagonal, namely = −1 , then we have ( − ) −1 = . Similarly, we can obtain the domestic Leontief inverse matrix ( − ) −1 , where is the domestic input coefficient matrix. Dividing the total value added by the gross outputs, we can (10) where the superscripts and denote domestic and foreign, respectively; ̂ denotes the diagonal matrix with value-added coefficients ; and are the direct input coefficients and final outputs; and ( − ) −1 and ( − ) −1 are the global and domestic Leontief inverse matrix, respectively. The calculation of formula (10)  Infras is measured by a simple average of three infrastructure indicators: i) the air transport of passengers carried per capita; ii) the simple average of per capita fixed broadband, fixed telephone, and mobile cellular subscriptions; and iii) the per capita kilometers of railroad. Besides, the Theil index (TL) is defined by formula (11) where i represents three industries, including primary, secondary, and tertiary; indicates the total employment; and / implies the productivity per unit of labor. The closer the TL index is to 0, the more rational the industrial structure is.
As regards the threshold variable, the TFP is used to determine the heterogeneity of the spillover effects of outward FDI on GVC participation. The calculation of the TFP is shown in formula (12) where denotes the capital stocks. We utilize stochastic frontier analysis (SFA) together with Frontier 4.1 version to estimate formula (12).
obtain the value-added coefficient, that is, = /̂. For details of the decomposition, please refer to Wang et al. (2017).  Table  1.

Results and Discussion
In this section, both the descriptive and the empirical results are displayed to demonstrate the effects of outward FDI on GVC participation.

Descriptive Results
As shown in Figure 2, during the period 2000 to 2014, the network of overall GVC participation almost tripled, that of shallow GVC participation increased 2.88 times, and that of deep GVC increased 3.15 times, indicating the gradually closer relations among countries. In 2000, the GVC participation was mainly dominated by developed countries, as more than 45% of the total value added for GVC participation was generated by countries such as the US (15.39%, proportion of the total hereinafter), Germany (7.46%), Japan (7.57%), the UK (5.65%), France (4.43%), Canada (4.02%), and so on, while the proportion of developing countries, like China, Russia, and Brazil, was relatively small. In 2014, the share of developed countries dropped by various degrees, while that of developing countries, especially China, rose sharply to 10.13%, ranking only second to the US. Though the GVC participation network with developed countries at the core did not change, developing countries like China, Russia, and Brazil participated more and more in the GVCs and gradually moved from the periphery to the core.
The spatial distribution of outward FDI in 2000 and 2014 is also mapped with the geographical information system (GIS) (Figure 3). In 2000, outward FDI was generally distributed among developed countries, such as the US, the UK, France, and Canada. However, with the rapid development of globalization and production fragmentation, developing countries placed increasing emphasis on outward FDI, which was gradually oriented towards developed countries (Carlos Rodríguez and Ricardo Bustillo 2011). As a result, the outward FDI from developing countries saw dramatic changes during the fifteen-year period from 2000 to 2014. Furthermore, the surging outward FDI took on obvious characteristics of geographical agglomeration. For instance, around 80% of outward FDI from some developed countries and developing countries (such as Russia and Mexico) flowed to the US and European countries. By contrast, more than 80% of Chinese outward FDI flowed to developing areas.
The estimation of the TFP is reported in Figure 4. The TFP varies greatly from country to country and is positively correlated with the average GDP per capita. Additionally, for most developed countries, the TFP is generally more than 0.7, and some countries even hit the maximum value of 0.99, like the US and Luxembourg.

Notes:
The dots represent different countries; the thickness of each line represents the volume of overall GVC participation. The thicker the lines, the more GVC participation is. Situations of shallow and deep GVC participation remain roughly the same, so details are not repeated.
Source: The authors' calculations.  Meanwhile, for developing countries, such as Brazil, China, and Indonesia, the average TFP values are less than 0.5.

Baseline Model Regression
As reported in Table 2, we replace the dependent variable in formula (1) with overall, shallow, and deep GVC participation successively to obtain the first three models, then we add the intersection terms of outward FDI and TFP to gain the last three models, which are expected to uncover the potential pathways. The variance inflation factor (VIF) value of the baseline model is 6.03, less than 10, indicating that the multicollinearity problem among the explanatory variables can be ignored. The Hausman tests of each model turn out to be significantly positive, implying that fixed-effect models are better than random-effect ones. The coefficients of lnOFDI in the first three models are statistically significant at the 1% level and range from 0.047 to 0.072, indicating that the rise of outward FDI promotes GVC participation, especially deep participation. Outward FDI reduces production and trade costs and improves the economies of scale and resource productivity of MNEs, thus facilitating the GVC participation of parent countries. This evidence is broadly in line with hypothesis H1 (Section 2). In the latter three models, the estimators for OFDI are significantly positive, in contrast to the significant negative coefficients for the interaction terms of OFDI and TFP, indicating that the effects of outward FDI on GVC participation depend on the TFP level of the home country. Using model (4) as an example, the partial derivative of GVC_overall over OFDI equals 0.254-0.380*TFP; therefore, if the value of TFP is over the critical value 0.668, the partial derivative is negative, and otherwise it is positive. A possible reason may be that, compared with low-TFP countries, high-TFP countries are more likely to transfer their manufacturing industry to countries with cheap factors, as they have done since the middle of the last century, which causes a reduction in the GVC participation.    708 Notes: t statistics in the parentheses; *, **, and *** denote significant at the 10%, 5%, and 1% level, respectively; ln means the natural logarithm of the original values.
Source: The authors' calculations.
Regarding the results for the other control variables, the coefficients of lnEINT, lnEMP, lnR&D, Infras, and TFP are significantly positive under most circumstances, denoting that an increase in intermediate goods exports, employment, R&D expenditure, infrastructure level, and TFP contributes to GVC participation. Table 3 presents the results of the global Moran's I index from 2000 to 2014. For shallow GVC participation, the index is positive and significant at least at the 5% level, while, for overall and deep GVC participation, the index is not statistically significant initially. However, the results of the overall and deep GVC participation gradually become significant and show an increase since 2004 and 2008, respectively. These discoveries indicate that the GVC participation levels of different countries are positively correlated and the Matthew effect on GVC participation gradually comes into being. Notes: z statistics in parentheses; *, **, and *** denote significant at the 10%, 5%, and 1% level, respectively. Source: The authors' calculations.

Spatial Autocorrelation Analysis
The spatial dependence and interaction of GVC participation in different countries suggests that the traditional econometric models without the consideration of spatial factors are far from enough to identify the true effects of outward FDI on GVC participation. Therefore, the SDM model is applied in the following section.

SDM Estimates
We carry out the same strategy for formula (5) as for the baseline regression, the estimates of which are displayed in Table 4. The coefficients for lnOFDI are significantly positive in all six models; however, as mentioned in the third part, such estimates cannot be interpreted as the influence coefficients of outward FDI on GVC participation. In addition, likelihood ratio (LR) tests show that the models with interaction terms are more appropriate. Therefore, our attention in the following discussion is mainly focused on the decomposition of the spatial effects in the latter three models. In models (4) and (5), the spatial effects of lnOFDI are positive and significant at the 1% level, suggesting that the GVC participation of the parent country is not only influenced directly by its own outward FDI but also affected indirectly by the outward FDI from other countries. The other countries, the host countries in particular, may benefit from the outward FDI spillovers, such as production technology and management skills, which are conducive to their GVC participation. The estimation with spatial techniques can act as a robustness test of the baseline models, at the same time confirming the spillover effects of outward FDI on GVC participation. These results seem to be consistent with the second hypothesis, H2.
The spatial effects of the interaction term in models (4), (5), and (6) are significantly negative, verifying again that the effects of outward FDI on GVC participation change along with the TFP level of the parent countries. Hence, we will test the heterogeneous effects of outward FDI in high-TFP countries and low-TFP countries.

Heterogeneity Test
The results of the threshold tests are illustrated in Table 5. The estimated threshold values are 0.517, 0.711, and 0.787, respectively, but the LR test in Figure 5 implies that only the first threshold is significant at the 5% level, so we select 0.517 as the critical value to determine the threshold effects. Firstly, we divide the samples into two groups according to the critical value: one group with an average TFP over 0.517 and another below 0.517, which are termed the developed and developing group, respectively. Subsequently, we estimate the spatial effects of these two sub-panels separately with the SDM. Table 6 reports the results for the developed group. The direct effects of outward FDI on deep GVC participation are 0.052 and significant at the 1% level, while the indirect effects remain insignificant. Contrarily, the indirect effects on the overall and shallow GVC participation are significantly positive, while the direct effects remain insignificant. These results indicate that outward FDI from developed countries contributes to their deep GVC participation and facilitates the shallow GVC participation of neighboring countries at the same time. A probable explanation is that multinationals from developed countries generally dominate the design, R&D, and marketing of the GVCs and that the value added embedded in their exports may traverse national borders more than twice for processing and packaging, thus facilitating deep GVC participation. Nevertheless, local foreign firms in developed countries are more competitive in price, which may increase the intermediate goods supply, thus contributing to the shallow GVC participation.  However, a different picture emerges in the low-TFP countries. As displayed in Table 7, the parameter estimation of outward FDI in all three models turns out to be positive and significant at least at the 5% level, indicating that outward FDI in developing countries not only promotes the shallow and deep GVC participation of parent countries but also improves those of the other countries. For parent countries, outward FDI from developing countries is generally in the character of technology seeking and market seeking (Davide Castellani and Fabio Pieri 2016), which may accelerate the productivity and economies of scale, thus inducing more GVC participation. For other countries, many foreign affiliates of the multinationals from developing countries possess comparative advantages in price and cheap factors. These advantages may facilitate the value-added exports of the host countries. The different results of both sub-panels appear to be in line with the third hypothesis, H3. 473 Notes: t statistics in parentheses; *, **, and *** denote significant at the 10%, 5%, and 1% level, respectively.
Source: The authors' calculations

Conclusions
Based on the KC model and the GVC framework, our study sheds light on the heterogeneous spillover effects of outward FDI on GVC participation using a panel data model covering 42 WIOD countries from 2000 to 2014. The results show that the expansion of outward FDI not only contributes to the GVC participation of parent countries but also has a profound effect on the GVC participation of other countries. The spillover effects of outward FDI make sense under the spatial correlation of GVC participation among countries. In addition, heterogeneous effects of outward FDI on GVC participation appear in developing and developed countries. Specifically, for the developing sub-panel, the effects of outward FDI on both the shallow and the deep GVC participation of parent countries are dramatically positive. However, for the developed sub-panel, outward FDI has positive effects on deep GVC participation while not influencing shallow participation. The indirect effects of outward FDI on GVC participation in both sub-panels are significantly positive. The augmentation of R&D, infrastructure level, labor and capital input, and so on, can also accelerate the integration into globalization and GVCs. Actually, outward FDI brings more than cost reduction and market expansion. Cross-border investment among countries motivates the transnational flows of technology, labor, goods, even cultures, and so on; thus, the relocation of production activities also motivates the reallocation of resources, which are no longer confined by national borders and initial factor endowments. If the economic specialization and integration in global production facilitate the development of GVCs, then outward FDI offers both developed and developing countries an alternative way to specialize in the production stages in which they have comparative advantages and to integrate into specific regional production networks characterized by high entry barriers. Developing countries, in particular, can exploit their advantages of backwardness and encourage domestic enterprises to invest abroad to access the advanced and sophisticated technology in developed countries.
As an extension to the KC theoretical model, our study enables a comprehensive understanding of the fundamental role that multinationals play in GVC participation. Our findings suggest a way for developing countries to capture gains from outward FDI and participate further in GVCs. However, with regard to developed countries, some of them have transferred their manufacturing and producer service industries to developing countries, typically with a lower labor cost, through outward FDI during the past couple of years. As a result, the GVC participation of parent countries is declining and the problem of hollowing out of manufacturing industries has grown into a major concern in countries like the US, Japan, and South Korea.