The Asymmetric Effects of Deep Preferential Trade Agreements on Bilateral GVC Participation Levels

ABSTRACT

Although deepening of preferential trade agreements (PTAs) can promote the global value chain (GVC) participation of member countries, we argue that this effect is asymmetric for countries, which are at different levels of economic development. Specifically, deepening of PTAs has a stronger positive impact on GVC participation level of developing member countries than that of developed member countries. Furthermore, compared to the commodity provisions of PTAs, deepening of the factor provisions is more likely to generate a significant asymmetric effect. Empirical analysis using panel data, which covers 43 countries (including some emerging economies) over the 2000–2014 period, supports our predictions. Our findings are also confirmed by a series of robustness tests including the consideration for potential endogeneity issue.

KEYWORDS:

• Preferential trade agreements (PTAs)
• global value chain (GVC) participation
• asymmetric effects
• panel data analysis

JEL:

• F02
• F13
• F15

1. Introduction

Recent decades have witnessed rapid growth in global value chains (GVCs). By fragmenting the production process and spreading it across countries, GVCs have profoundly re-shaped international trade flows (Baldwin and Yan 2021). For developing countries, participation in GVCs has led to improvement in productivity, product quality as well as increase in the scale of international trade (Collier and Venables 2007; Criscuolo and Timmis 2017; Ndubuisi and Owusu 2021; Pahl and Timmer 2019). The rise of some emerging economies (such as China and India) is widely attributed to effective engagement with GVCs. Growth in GVCs has coincided with increase in preferential trade agreements (PTAs) aimed at increasing regional market access.¹ The number of PTAs in force increased from 70 to 573 over the 1990–2022 period. Modern PTAs involve policy cooperation far beyond the traditional tariff reductions, and include additional provisions such as investments, procurements, intellectual property rights, and competition policies (Hofmann, Osnago, and Ruta 2017). As a result, the effect of deep PTAs on GVC participation has received considerable attention in recent years.

Most existing studies showed that as an important way to increase regional cooperation and integration, the establishment of PTAs has a positive impact on GVC participation level of member countries, and compared to shallow PTAs, deep PTAs can lead to a larger increase in GVC participation level (e.g., Antràs and Staiger 2012; Boffa, Jansen, and Solleder 2019; Hur, Park, and Alba 2008; Johnson and Noguera 2017; Lee 2019; Limão 2016; Ornelas and Turner 2008; Zhang, Zhao, and Zhao 2021). Some recent studies also paid attention to the heterogeneous effects of deep PTAs on GVC participation level of member countries and found that the effect of deep PTAs on GVC participation level can be influenced by the structure of their members and provisions (Dahi and Demir 2017; Hiraide, Shen, and Silva 2020; Lee and Kim 2021).

Inspired by the fact that GVC governance systems vary across countries included in the same deep PTA, in this paper, we focus on the issue of whether the effect of deep PTAs on bilateral GVC participation level is asymmetric across member countries, which are at different levels of economic development.² Furthermore, we also consider whether the asymmetric effect is influenced by provisions of deep PTAs. The theoretical rationale behind the research questions is that deepening of PTAs generates both intensive and extensive-chains effect on bilateral GVC participation, where the former applies to both developing and developed members, but the latter is associated with the flows of know-how and capital from developed to developing countries (and hence it is mainly observed in developing countries). Furthermore, the extensive-chains trade involves re-organization of GVCs, which may also involve mobility of factors of production. Thus, factor mobility provisions of deep PTAs are more likely to generate the extensive-chains effect.

For empirical evaluation, we use a dataset which covers all sectors of 43 countries included in our sample over the 2000–2014 period.0 Analysis of the data, which includes emerging economies, reveals that deepening of PTAs has a stronger impact on GVC participation level of developing member countries than that of developed member countries. Moreover, compared to PTA deepening good and services-related provisions, the asymmetric effect is more significant when PTA deepening provisions involve mobility of factors of production. A series of robustness tests (as well as accounting for potential endogeneity) confirm the reliability of our findings.

This paper makes two important contributions to the existing literature that deals with the heterogeneous effect of deep PTAs on GVC participation. First, by investigating whether the effect of PTAs on GVC participation level is asymmetric across member countries which are at different levels of economic development, we contribute to the literature that examines how the effect of deep PTAs on GVC participation level is affected by the characteristics of member countries. Second, we also investigate the asymmetric effect of provisions of deep PTAs on GVC participation, which contributes to the literature on whether the effect of deep PTAs on GVC participation varies across PTA provisions.

The remainder of this paper is organized as follows. The theoretical foundation of our work and hypotheses tested are explained in Section 2. The empirical strategy is presented in Section 3. Data sources and summary statistics of the data are provided in Section 4. Estimation results are presented and discussed in Section 5. Some robustness checks are presented in Section 6. The regression equation is re-estimated after accounting for potential endogeneity in Section 7. The main conclusions and policy implications are presented in Section 8.

2. Theoretical Foundation and Hypothesis Development

2.1. Mechanism Through Which Deep PTAs Affect Bilateral GVC Participation Level

Deepening of PTAs can generally promote bilateral GVC participation level of member counties for three reasons. First, deep PTAs are primarily designed to secure or increase market access within the member countries (Limão 2016). In the presence of intermediates, which can be sourced internationally, deepening of PTAs can facilitate vertical integration by reducing the trade barriers for intermediates, and hence encouraging their members to increase engagement in GVC trade (Caliendo and Parro 2015; Johnson and Noguera 2017; Limão 2016; Ornelas and Turner 2008; WTO 2011; Yi 2003).

Second, deep PTAs go beyond the currently applied trade barriers, and cover some trade policy provisions in the future (Limão 2016). In this sense, deep PTAs can reduce trade policy uncertainty that may be caused by either internal or external shocks, such as the domestic political turbulence or trade wars (Amador and Bagwell 2013; Limão and Maggi 2015; Prusa and Teh 2010), thereby providing a predictable and stable environment for intermediates trade. This implies that deep PTAs can further promote the current bilateral GVC participation level by securing future market access within the member countries.

Finally, the production of intermediates is typically customized, and hence international intermediates trade often involves considerable relationship-specific investments (Feenstra and Hanson 2005; Nunn et al. 2009). Moreover, international contracts are difficult to enforce because the contractual safeguards are not sufficiently strong. This situation contributes to holdups in cross-border trade of intermediates, leading to GVC participation levels which are well below the optimal (Antràs and Staiger 2012; Grossman 2016). As a substitute for strong contractual safeguards that usually apply to domestic transactions, deep PTAs can encourage intermediates trade and improve bilateral GVC participation level by changing the ex-post bargaining conditions between the buyers and suppliers in the intermediates market.

2.2. The Asymmetric Effects of Deep PTAs on Bilateral GVC Participation Level

Among other things, GVCs facilitate the flows of know-how and capital from developed to developing countries (Baldwin and Yan 2021). To take advantage of the lower cost of factors of production in developing countries, developed country multinational corporations tend to fragment the production process and complete some stages of production in developing countries through foreign direct investment (FDI) or outsourcing. This gives rise to GVC-induced trade between developed and developing countries (Baldwin and Lopez-Gonzalez 2015). At the same time, with increased international policy coordination, the member structure of PTAs has become more diversified. Most PTAs include both developing and developed countries.³ Although deepening of such PTAs is expected to promote GVC participation level of member countries, it is not entirely clear whether this effect is asymmetric across developed and developed countries.

The overall impact of deep PTAs on bilateral GVC participation level of member countries can also be divided into “intensive-chains effect” and “extensive-chains effect.” The former refers to the fact that deepening of the PTAs can affect GVC participation level associated with original intermediate goods traded between member countries, while the latter refers to the effect which arises because deepening of PTAs can influence GVC participation level that is related to new intermediate goods traded by the member countries. Deepening of PTAs can help both developing and developed member countries to increase their exports of the original intermediate goods traded with each other. Thus, the intensive-chains effect of deep PTAs applies to both developing and developed member countries.

However, due to the dominant role of developed countries in GVC governance system, the extensive-chains effect of deep PTAs on bilateral GVC participation may be mainly observed for developing countries. With reductions in trade barriers, policy uncertainty and transaction costs, deepening of PTAs also induces developed member country multinational corporations to further fragment the production process and shift additional production stages to developing member countries through FDI and outsourcing activities (Antràs and Helpman 2004; Arnold 2000; Lacity and Willcocks 1995; Prahalad and Hamel 1997). Increased fragmentation of the production process generates the extensive-chains effect and thus leads to further increase in GVC exports of developing member countries. Therefore, compared to the developed countries, developing countries may experience a higher marginal effect on GVC participation from deepening of PTAs. Based on the above discussion, we have a hypothesis as follows.

Hypothesis 1:

Deepening of PTAs has a stronger positive impact on GVC participation level of developing member countries than that of developed member countries.

2.3. The Asymmetric Effects of Different Provisions on Bilateral GVC Participation Level

PTAs also differ in terms of their provisions (Antràs and Staiger 2012). A deep PTA typically involves two main categories of provisions: (i) provisions concerning trade in goods and services (commodity provisions) and (ii) provisions concerning the cross-border movement of factors of production (factor provisions). Specifically, commodity provisions cover import tariffs, non-tariff barriers, and the General Agreement on Trade in Services (GATS), while factor provisions cover policies on international technology transfer, foreign investment, and cross-border labor movement.

By generating the extensive-chains effect, deepening of both commodity and factor provisions contributes to the asymmetric effect on GVC participation across developing and developed member countries. However, the contribution of these factors to the asymmetric effect may not be equally statistically significant. GVC-trade is organized mainly by multinational corporations using FDI and outsourcing. This involves mobility of factors of production such as capital and exchange of technical know-how. The extensive-chains trade tends to involve re-organization of GVCs and corresponding flows of factors of production. Thus, deepening of factor provisions is more likely to generate the extensive-chains effect.

Furthermore, due to multilateral negotiations on trade liberalization under World Trade Organization (WTO), most of the barriers to goods and services trade have been largely reduced or even removed, and hence the contribution of commodity provisions to the depth of PTAs is limited.⁴ By contrast, as a politically sensitive issue, not much progress has been made on removing restrictions on cross-border movement of factors of production (especially labor) under multilateral negotiation framework of WTO. However, deep PTAs tend to deal with cross-border factor mobility issue. Accordingly, factor provisions can make a stronger contribution to the depth of PTAs compared to commodity provisions. Based on above discussion, we have a hypothesis as follows.

Hypothesis 2:

Compared to PTA commodity provisions, deepening of PTA factor provisions is more likely to have an asymmetric effect on GVC participation of developing and developed member countries.

3. Methodology

This section explains the methodology used to test the hypotheses concerning the effects of deep PTAs on bilateral GVC participation level. First, it is worthwhile to discuss how we measure GVC participation level. Based on the GVC-trade decomposition developed by Wang, Wei, and Zhu (2013), we disaggregate bilateral GVC participation level into forward and backward. The former measures domestic value added embodied in a country’s exports, whereas the latter measures foreign value added embodied in the country’s exports. Specifically, forward and backward GVC participation levels are calculated as follows.

(1) $FGVC{P}_{ijt}=\sum _{k=1}^n(DVA\mathrm{\_}IN{T}_{ijkt}+DVA\mathrm{\_}INTre{x}_{ijkt}+RD{V}_{ijkt}+DVA\mathrm{\_}FI{N}_{ijkt})$(1)

(2) $BGVC{P}_{ijt}=\sum _{k=1}^n(FVA\mathrm{\_}IN{T}_{ijkt}+FVA\mathrm{\_}FI{N}_{ijkt})$(2)

where t represents the year, i and j denote the exporting and importing countries, respectively. FGVCP_ijt and BGVCP_ijt, respectively, are country i’s forward and backward GVC participation level in country j. DVA_INT_ijkt is the domestic value added in country i’s intermediate good k exports used by direct importer j to produce local final products. DVA_INTrex_ijkt is the domestic value added in country i’s intermediate good k exports used to produce intermediate goods that are reexported to third countries for production of local final products. RDV_ijkt is the domestic value added in country i’s intermediate k first exported to country j but eventually returns and consumed in country i. DVA_FIN_ijkt is the domestic value added in country i’s final product k exported to country j. FVA_INT_ijkt is the foreign value added in country i’s intermediate good k exported to country j. FVA_FIN_ijkt is foreign value added in country i’s final product k exported to country j.

As to the depth of PTAs, following Hofmann, Osnago, and Ruta (2017), we measure the depth of PTAs by summing up the clauses that are associated with free trade policies. Specifically, if a PTA covers a certain clause of free trade, the clause takes the value of 1, 0 otherwise. The cumulative value of all clauses is defined as the depth of a PTA. The depth of the commodity and factor dimensions of PTAs is also measured in the same way. The classification of the provisions on trade in commodities and the movement of factors of production is shown in Table 1.

Table 1. Classification of commodity and factor provisions of PTAs.

Commodity provisions		Factor provisions
Import tariffs	FTA industrial	Technology	TRIPS
	FTA agriculture		IPR
Non-tariff barriers	Customs		Innovation policies
	Export taxes		Economic policy dialogue
	AD		Information society
	CVM		Research and technology
	SPS	Investment	TRIMS
	TBT		Investment
Post-border policies	STE		Movement of capital
	Competition policy	Labor	Visa and asylum
	State aid
	Anti-corruption
Services	GATS

Source: https://datacatalog.worldbank.org/dataset/wits-global-preferential-trade-agreement-database. AD denotes the retention of antidumping rights and obligations under the WTO Agreement. CVM represents the retention of countervailing measure rights and obligations under the WTO Agreement. SPS is the affirmation of rights and obligations under the WTO Agreement on SPS (Agreement on the Application of Sanitary and Phytosanitary Measures). TBT refers to the affirmation of rights and obligations under WTO Agreement on TBT (Agreement on Technical Barriers to Trade). STE denotes the establishment or maintenance of a state enterprise in accordance with and affirming provisions of GATT. GATS represents the liberalization of trade in services. TRIPS is the international treaties referenced in TRIPS (Agreement on Trade-Related Aspects of Intellectual Property Rights). IPR denotes the accession to international treaties not referenced in the TRIPs Agreement. TRIMS is the provisions concerning requirements for local content and export performance on FDI.

To investigate the asymmetric effects of deepening of PTAs on bilateral GVC participation level between member countries, which are at different levels of economic development, following the World Bank’s standards, we first select 20 developed countries,⁵ and classify PTAs into four groups: North-North (N-N), South-North (S-N), North-South (N-S), and South-South (S-S).⁶ We then estimate the impact of deepening of the PTAs on GVC participation level across the four PTA groups (Laget et al. 2020; Lee 2019). Moreover, we include several dummy variables to control the fixed effects that capture observed and unobserved heterogeneity across countries and country pairs. The resulting regression equations are as follows.

(3) $\begin{array}{rl}& GVC{P}_{ijt}=exp[{\beta }_0+{\beta }_1(Dept{h}_{ijt}\ast N{N}_{ij})+{\beta }_2(Dept{h}_{ijt}\ast N{S}_{ij})+{\beta }_3(Dept{h}_{ijt}\ast S{N}_{ij})\\ & +{\beta }_4(Dept{h}_{ijt}\ast S{S}_{ij})+{\delta }_{ij}+{\lambda }_{it}+{\vartheta }_{jt}]+{\epsilon }_{ijt}\end{array}$(3)

(4) $\begin{array}{rl}& GVC{P}_{ijt}=exp[{\beta }_0+{\beta }_1(CDept{h}_{ijt}\ast N{N}_{ij})+{\beta }_2(CDept{h}_{ijt}\ast N{S}_{ij})+{\beta }_3(CDept{h}_{ijt}\ast S{N}_{ij})\\ & +{\beta }_4(CDept{h}_{ijt}\ast S{S}_{ij})+{\beta }_{5}FDept{h}_{ijt}+{\delta }_{ij}+{\lambda }_{it}+{\vartheta }_{jt}]+{\epsilon }_{ijt}\end{array}$(4)

(5) $\begin{array}{rl}& GVC{P}_{i}jt=exp\left[{\beta }_0+{\beta }_1\left(FDept{h}_{i}jt\ast N{N}_{i}j\right)+{\beta }_2\left(FDept{h}_{i}jt\ast N{S}_{i}j\right)+\right.{\beta }_3\left(FDept{h}_{i}jt\ast S{N}_{i}j\right)\\ & \left.+{\beta }_4\left(FDept{h}_{i}jt\ast S{S}_{i}j\right)+{\beta }_{5}CDept{h}_{i}jt+{\delta }_{i}j+{\lambda }_{i}t+{\vartheta }_{j}t\right]+{\epsilon }_{i}jt\end{array}$(5)

where GVCP_ijt is country i’s GVC participation level in country j, which represents both FGVCP_ijt and BGVCP_ijt. Depth_ijt is the total depth of the trade agreement between country i and country j. CDepth_ijt is the commodity provision depth of the trade agreement between country i and country j. FDepth_ijt is the factor provision depth of the trade agreement between country i and country j. NN_ij, NS_ij, SN_ij and SS_ij are dummy variables that are used to identify the PTA group involving country i and j. ${\delta }_{ij}$ is the country pair fixed effects that are used to control time-invariant heterogeneity across country pairs. ${\lambda }_{it}$ and ${\vartheta }_{jt}$ are country-year fixed effects that capture time-variant heterogeneity across exporters and importers, respectively. ${\epsilon }_{ijt}$ is the usual error term.

4. Data

The data used to calculate the bilateral GVC participation level is sourced from the latest version of the World Input-Output Database (WIOD), released in 2016. This database contains Inter-Country Input-Output tables covering 43 countries and 56 sectors over the 2000–2014 period. The data on PTAs is sourced from Global Preferential Trade Agreements database of the World Bank (PTADB 2021). Most existing studies on preferential trade agreements use this database. The database provides information on PTAs of 189 countries from 1958 to 2015. After matching the data on PTAs and bilateral GVC participation level, we have a dataset which includes 43 countries and regions over the 2000–2014 period.

Table 2 shows the descriptive statistics of the main variables included in the regression equation. First, the average value of both GVC participation level and PTA depth is close to the median, indicating that the data dispersion of two variables in the sample is low, and there are no apparent outliers. Second, the average value of both forward and backward GVC participation level in the N-N group is the highest, while the average value of both forward and backward GVC participation level in the S-S group is the lowest. Finally, the average depth of PTAs in the N-N group is the highest, while the average depth of PTAs in the S-S group is the lowest.

Table 2. Descriptive statistics.

	Variables	Mean	median	SD	Min	Max	Observations
N-N	FGVCP	7.71	7.693	1.699	1.308	12.46	5700
	BGVCP	6.272	6.317	1.727	0.084	11.3	5700
	Depth	0.204	0.498	0.984	−0.855	1.42	5700
	CDepth	0.217	0.334	0.988	−0.869	1.536	5700
	FDepth	0.18	0.431	0.978	−0.814	1.364	5700
N-S	FGVCP	5.982	5.95	2.006	0.456	11.89	6900
	BGVCP	4.714	4.715	1.954	0.11	10.8	6900
	Depth	−0.019	−0.855	0.995	−0.855	1.666	6900
	CDepth	−0.02	−0.869	0.994	−0.869	1.536	6900
	FDepth	−0.016	−0.814	0.998	−0.814	1.831	6900
S-N	FGVCP	5.75	5.729	2.176	0.071	12.56	6900
	BGVCP	4.443	4.457	2.06	0.009	11.34	6900
	Depth	−0.015	−0.855	1	−0.855	1.666	6900
	CDepth	−0.014	−0.869	1	−0.869	1.536	6900
	FDepth	−0.015	−0.814	1	−0.814	1.831	6900
S-S	FGVCP	4.684	4.605	2.326	0.02	11.11	7590
	BGVCP	3.544	3.443	2.094	0.005	10.48	7590
	Depth	−0.123	−0.855	0.993	−0.855	1.666	7590
	CDepth	−0.132	−0.869	0.988	−0.869	1.536	7590
	FDepth	−0.107	−0.814	1.001	−0.814	1.831	7590

Note. Forward and backward GVC participation levels are in logarithms. The depth of PTAs is standardized.

5. Empirical Results and Analysis

In this section, we start by presenting and discussing the estimation results of Equation (3)(3) $\begin{array}{rl}& GVC{P}_{ijt}=exp[{\beta }_0+{\beta }_1(Dept{h}_{ijt}\ast N{N}_{ij})+{\beta }_2(Dept{h}_{ijt}\ast N{S}_{ij})+{\beta }_3(Dept{h}_{ijt}\ast S{N}_{ij})\\ & +{\beta }_4(Dept{h}_{ijt}\ast S{S}_{ij})+{\delta }_{ij}+{\lambda }_{it}+{\vartheta }_{jt}]+{\epsilon }_{ijt}\end{array}$(3) , which allows one to investigate the asymmetric effect of deepening of PTAs on bilateral GVC participation level of member countries. In Table 3, forward and backward GVC participation regression results, respectively, are presented in columns (1) and (2). As shown in Table 3, the effect of deepening of PTAs on forward and backward GVC participation level in both N-S and S-N groups is positive and statistically significant at the 1% level. However, magnitudes of the estimated effects vary considerably across the two groups. Specifically, one standard deviation increase in the depth of PTAs promotes forward and backward GVC participation level of developed countries by 5.21% and 6.33%, respectively. By contrast, one standard deviation increase in the depth of PTAs promotes forward and backward GVC participation level of developing countries by 9.94% and 9.57%, respectively. We use the F-test to investigate the significance of the differences in the estimated effect magnitudes. Based on the F-test results, we find that the impact of deepening of PTAs on forward (backward) GVC participation level of developing countries is greater than that of developed countries by 0.0473 (0.032) and this difference is statistically significant at the 1% (5%) level. This result is consistent with our first hypothesis in that the deepening of PTAs has a stronger positive impact on GVC participation level of developing member countries than that of developed member countries.

Table 3. The asymmetric effects of the depth of PTAs on bilateral GVC participation level.

Variables	(1)	(2)
	FGVCP	BGVCP
Depth*N-N	−0.0273**	−0.00655
	(0.0119)	(0.0116)
Depth*N-S	0.0521***	0.0633***
	(0.00926)	(0.00901)
Depth*S-N	0.0994***	0.0957***
	(0.00933)	(0.00908)
Depth*S-S	0.0435***	0.0341***
	(0.00952)	(0.00927)
Constant	5.926***	4.647***
	(0.00207)	(0.00202)
Export-time FE	YES	YES
Import-time FE	YES	YES
Country-pair FE	YES	YES
Observations	27,090	27,090
R^2	0.983	0.981

Note. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively. Values in parentheses underneath the estimated coefficients are the corresponding clustered robust standard errors.

We now examine whether the asymmetric effect of deepening of PTAs on GVC participation level varies significantly between the commodity and factor provisions. This task is completed by estimating Equations (4)(4) $\begin{array}{rl}& GVC{P}_{ijt}=exp[{\beta }_0+{\beta }_1(CDept{h}_{ijt}\ast N{N}_{ij})+{\beta }_2(CDept{h}_{ijt}\ast N{S}_{ij})+{\beta }_3(CDept{h}_{ijt}\ast S{N}_{ij})\\ & +{\beta }_4(CDept{h}_{ijt}\ast S{S}_{ij})+{\beta }_{5}FDept{h}_{ijt}+{\delta }_{ij}+{\lambda }_{it}+{\vartheta }_{jt}]+{\epsilon }_{ijt}\end{array}$(4) and (5(5) $\begin{array}{rl}& GVC{P}_{i}jt=exp\left[{\beta }_0+{\beta }_1\left(FDept{h}_{i}jt\ast N{N}_{i}j\right)+{\beta }_2\left(FDept{h}_{i}jt\ast N{S}_{i}j\right)+\right.{\beta }_3\left(FDept{h}_{i}jt\ast S{N}_{i}j\right)\\ & \left.+{\beta }_4\left(FDept{h}_{i}jt\ast S{S}_{i}j\right)+{\beta }_{5}CDept{h}_{i}jt+{\delta }_{i}j+{\lambda }_{i}t+{\vartheta }_{j}t\right]+{\epsilon }_{i}jt\end{array}$(5) ). The estimation results are shown in Table 4, where columns (1) and (2) show the impact of the commodity provisions of PTAs on GVC participation level, while columns (3) and (4) show the impact of the factors provisions of PTAs on GVC participation level. The results presented in Table 4 show that deepening of commodity provisions has an asymmetric effect on GVC participation level between developing and developed countries. Specifically, one standard deviation increase in the commodity provisions promotes forward and backward GVC participation level of developed countries by 5.53% and 8.87%, respectively. By contrast, one standard deviation increase in the commodity provisions promotes forward and backward GVC participation level of developing countries by 10.3% and 12.6%, respectively. The F-test results confirm that the impact of deepening of commodity provisions of PTAs on forward (backward) GVC participation level of developing countries is higher than that of developed countries by 0.0477 (0.0373) and this difference is statistically significant at the 1% (1%) level.

Table 4. The asymmetric effects of commodity and factor provisions on GVC participation level.

Variables	(1)	(2)	(3)	(4)
	FGVCP	BGVCP	FGVCP	BGVCP
	commodity provisions		factor provisions
Depth*N-N	−0.0506**	0.0115	−0.0561***	−0.0608***
	(0.0197)	(0.0192)	(0.0209)	(0.0160)
Depth*N-S	0.0553***	0.0887***	0.0357	0.0176
	(0.0168)	(0.0164)	(0.0269)	(0.0157)
Depth*S-N	0.103***	0.126***	0.0668**	0.0337**
	(0.0168)	(0.0163)	(0.0286)	(0.0157)
Depth*S-S	0.0564***	0.0739***	0.0303	−0.00712
	(0.0171)	(0.0166)	(0.0288)	(0.0154)
FDepth	0.0139	−0.00894
	(0.0138)	(0.0134)
CDepth			0.0371	0.0705***
			(0.0276)	(0.0150)
Constant	5.929***	4.649***	5.927***	4.647***
	(0.00217)	(0.00211)	(0.000936)	(0.00202)
Export-time FE	YES	YES	YES	YES
Import-time FE	YES	YES	YES	YES
Country-pair FE	YES	YES	YES	YES
Observations	27,090	27,090	27,090	27,090
R^2	0.983	0.982	0.983	0.982

Note. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively. Values in parentheses underneath the estimated coefficients are the corresponding clustered robust standard errors.

Moreover, as shown in columns 3 and 4 of Table 4, deepening of factor provisions on GVC participation level between developing and developed countries is also asymmetric, but the asymmetric effects are more significant compared to the case of commodity provisions. Specifically, one standard deviation increase in factor provisions promotes forward and backward GVC participation level of developing countries by 6.680% and 3.37%, respectively, while its effects on forward and backward GVC participation level of developed countries are statistically insignificant at the 10% level. This result is consistent with our second hypothesis, which suggests that compared to commodity provisions of PTA deepening, factors of production provisions are more likely to have a stronger effect on GVC participation of developing countries than that of developed countries.

6. Robustness Checks

In this Section, we present the results of robustness tests. First, although high-dimensional fixed-effects are controlled in Section 5, we examine whether our results are affected by omitted variables bias. Previous studies report that bilateral investment agreements (BIT) have a significant effect on the GVC participation level of member countries (Baldwin and Venables 2013; Egger and Wamser 2013). Therefore, we re-estimate the asymmetric impact of deepening of PTAs on bilateral GVC participation level by including BIT as an additional explanatory variable. The results are presented in Table 5. As can be seen in columns 1 and 2 where the depth of PTAs is measured by all provisions, the estimated coefficients of the interaction term are significantly positive in the case of both N-S and S-N, but they are higher in the case of S-N than them in the case of N-S, implying that deepening of PTAs has asymmetric effects on GVC participation level between developing and develop. In columns 3 to 6, deepening of both commodity and factor provisions have asymmetric effects on GVC participation level, but the asymmetric effects of factor provisions are more significant. These results are qualitatively similar to those presented in Section 5, implying the absence of omitted variable bias and thus robustness of our earlier results.

Table 5. Robustness test: Controlling for bilateral investment agreements.

Variables	(1)	(2)	(3)	(4)	(5)	(6)
	FGVCP	BGVCP	FGVCP	BGVCP	FGVCP	BGVCP
	all provisions		commodity provisions		factor provisions
Depth*N-N	−0.0255**	−0.00467	−0.0439**	0.0182	−0.0570***	−0.0610***
	(0.0119)	(0.0116)	(0.0198)	(0.0193)	(0.0209)	(0.0160)
Depth*N-S	0.0514***	0.0628***	0.0568***	0.0900***	0.0318	0.0144
	(0.00927)	(0.00902)	(0.0168)	(0.0164)	(0.0268)	(0.0157)
Depth*S-N	0.0998***	0.0956***	0.106***	0.128***	0.0641**	0.0307*
	(0.00933)	(0.00908)	(0.0168)	(0.0164)	(0.0286)	(0.0157)
Depth*S-S	0.0441***	0.0349***	0.0593***	0.0762***	0.0279	−0.00911
	(0.00952)	(0.00926)	(0.0171)	(0.0166)	(0.0287)	(0.0154)
BIT*N-N	−0.119***	−0.103***	−0.110***	−0.104***	−0.125***	−0.113***
	(0.0324)	(0.0315)	(0.0325)	(0.0316)	(0.0371)	(0.0315)
BIT*N-S	0.0235	0.0221	0.0235	0.0220	0.0251	0.0242
	(0.0259)	(0.0252)	(0.0259)	(0.0252)	(0.0357)	(0.0252)
BIT*S-N	0.0780***	0.0299	0.0784***	0.0309	0.0773**	0.0294
	(0.0248)	(0.0242)	(0.0248)	(0.0242)	(0.0338)	(0.0242)
BIT*S-S	0.0436**	−0.0501***	0.0450**	−0.0491***	0.0444	−0.0492***
	(0.0186)	(0.0181)	(0.0186)	(0.0181)	(0.0341)	(0.0181)
FDepth			0.0117	−0.0108
			(0.0138)	(0.0134)
CDepth					0.0409	0.0737***
					(0.0276)	(0.0151)
Constant	5.915***	4.637***	5.918***	4.639***	5.915***	4.637***
	(0.00330)	(0.00321)	(0.00340)	(0.00331)	(0.00323)	(0.00321)
Export-time FE	YES	YES	YES	YES	YES	YES
Import-time FE	YES	YES	YES	YES	YES	YES
Country-pair FE	YES	YES	YES	YES	YES	YES
Observations	27,090	27,090	27,090	27,090	27,090	27,090
R^2	0.983	0.982	0.983	0.982	0.983	0.982

Note. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively. Values in parentheses underneath the estimated coefficients are the corresponding clustered robust standard errors.

Second, geographical distance between countries plays a vital role in bilateral trade flows. In Section 5, a country-pair dummy variable is used to capture this effect. However, if the impact of geographical distance changes over time, it may affect the reliability of the benchmark estimation results presented in Section 5 (Bergstrand, Larch, and Yotov 2015). To control for the time-varying geographical distance cost effect, the interaction of distance (Dist) with years is included as an additional explanatory variable. The regression results are shown in Table 6, where the results in all cases are qualitatively similar to those presented in Tables 3 and 4. This implies that the asymmetric effect of deepening PTA on GVC participation level is not affected by the time-varying geographical distance effect between member countries.

Table 6. Robustness test: Including time-varying geographical distance effect.

Variables	(1)	(2)	(3)	(4)	(5)	(6)
	FGVCP	BGVCP	FGVCP	BGVCP	FGVCP	BGVCP
	all provisions		commodity provisions		factor provisions
Depth*N-N	−0.0266**	−0.00491	−0.0497**	0.0162	−0.0538**	−0.0568***
	(0.0119)	(0.0116)	(0.0198)	(0.0193)	(0.0209)	(0.0160)
Depth*N-S	0.0488***	0.0575***	0.0526***	0.0841***	0.0350	0.0150
	(0.00933)	(0.00907)	(0.0169)	(0.0164)	(0.0271)	(0.0157)
Depth*S-N	0.0960***	0.0898***	0.101***	0.122***	0.0661**	0.0311**
	(0.00940)	(0.00914)	(0.0169)	(0.0164)	(0.0289)	(0.0157)
Depth*S-S	0.0385***	0.0251***	0.0519***	0.0650***	0.0280	−0.0130
	(0.00968)	(0.00941)	(0.0172)	(0.0168)	(0.0290)	(0.0154)
FDepth			0.0138	−0.0105
			(0.0138)	(0.0135)
CDepth					0.0348	0.0671***
					(0.0276)	(0.0151)
Constant	5.926***	4.647***	5.929***	4.648***	5.926***	4.647***
	(0.00208)	(0.00202)	(0.00217)	(0.00211)	(0.000940)	(0.00202)
Dist*Year	YES	YES	YES	YES	YES	YES
Export-time FE	YES	YES	YES	YES	YES	YES
Import-time FE	YES	YES	YES	YES	YES	YES
Country-pair FE	YES	YES	YES	YES	YES	YES
Observations	27,090	27,090	27,090	27,090	27,090	27,090
R^2	0.983	0.982	0.983	0.982	0.983	0.982

Note. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively. Values in parentheses underneath the estimated coefficients are the corresponding clustered robust standard errors.

Third, the depth of PTAs can be re-measured by considering the legal enforceability of provisions. The depth of PTAs is measured by summing up the number of free trade provisions in the previous estimations. However, the provisions of PTAs can be different in legal enforceability, which may influence the impact of deep PTAs on GVC participation level. Thus, we re-calculate the depth of PTAs by taking the legal enforceability into consideration. Specifically, if a certain provision is in legal force, it takes the value of 2, 1 otherwise. The regression results are shown in Table 7. As can be seen, the estimation results are very similar to those presented in Tables 3 and 4, implying that the asymmetric effect of deep PTAs on GVC participation level is not influenced by considering the legal enforceability of provisions into the measurement of PTA depth.

Table 7. Robustness test: Re-measuring the depth of PTAs.

Variables	(1)	(2)	(3)	(4)	(5)	(6)
	FGVCP	BGVCP	FGVCP	BGVCP	FGVCP	BGVCP
	all provisions		commodity provisions		factor provisions
Depth*N-N	−0.0241**	−0.00693	−0.0543***	0.00897	−0.0475***	−0.0478***
	(0.0117)	(0.0113)	(0.0202)	(0.0196)	(0.0131)	(0.0116)
Depth*N-S	0.0692***	0.0821***	0.0653***	0.0922***	0.0254	0.0154
	(0.00904)	(0.00880)	(0.0145)	(0.0141)	(0.0208)	(0.0130)
Depth*S-N	0.112***	0.109***	0.113***	0.131***	0.0550**	0.0306**
	(0.00910)	(0.00885)	(0.0145)	(0.0141)	(0.0219)	(0.0130)
Depth*S-S	0.0896***	0.0864***	0.0746***	0.0854***	0.0341	0.00636
	(0.0102)	(0.00989)	(0.0147)	(0.0143)	(0.0239)	(0.0136)
FDepth			0.0106	−0.00427
			(0.0104)	(0.0101)
CDepth					0.0516**	0.0793***
					(0.0236)	(0.0130)
Constant	5.930***	4.651***	5.931***	4.650***	5.927***	4.648***
	(0.00217)	(0.00211)	(0.00226)	(0.00220)	(0.00109)	(0.00205)
Export-time FE	YES	YES	YES	YES	YES	YES
Import-time FE	YES	YES	YES	YES	YES	YES
Country-pair FE	YES	YES	YES	YES	YES	YES
Observations	27,090	27,090	27,090	27,090	27,090	27,090
R^2	0.983	0.982	0.983	0.982	0.983	0.982

Note. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively. Values in parentheses underneath the estimated coefficients are the corresponding clustered robust standard errors.

Forth, while the descriptive statistics presented in Table 2 do not show any obvious outliers in our sample, to ensure that our results are not driven by the presence of outliers, the asymmetric effect of deep PTAs on bilateral GVC participation level is re-estimated after excluding all dependent variable observations with residuals greater than two standard deviations (SD) from the sample. The estimation results, as shown in Table 8, remain highly consistent with those presented in Section 5. Specifically, the asymmetric effect of deep PTAs on bilateral GVC participation level is statistically significant, and this effect is more significant in the case of PTA factor provisions. Thus, it can be argued that our main findings are not affected by the presence of outliers in the sample.

Table 8. Robustness test: Removing outliers.

Variables	(1)	(2)	(3)	(4)	(5)	(6)
	FGVCP	BGVCP	FGVCP	BGVCP	FGVCP	BGVCP
	all provisions		commodity provisions		factor provisions
Depth*N-N	−0.0276**	−0.00792	−0.0520***	0.00654	−0.0510**	−0.0590***
	(0.0115)	(0.0117)	(0.0193)	(0.0195)	(0.0207)	(0.0163)
Depth*N-S	0.0595***	0.0640***	0.0600***	0.0854***	0.0483*	0.0226
	(0.00908)	(0.00911)	(0.0165)	(0.0166)	(0.0267)	(0.0159)
Depth*S-N	0.0898***	0.0915***	0.0887***	0.118***	0.0624**	0.0337**
	(0.00941)	(0.00930)	(0.0167)	(0.0167)	(0.0283)	(0.0159)
Depth*S-S	0.0351***	0.0363***	0.0388**	0.0695***	0.0230	−0.00252
	(0.00974)	(0.00955)	(0.0169)	(0.0169)	(0.0275)	(0.0156)
FDepth			0.0152	−0.00567
			(0.0135)	(0.0136)
CDepth					0.0290	0.0654***
					(0.0276)	(0.0153)
Constant	5.989***	4.605***	5.991***	4.606***	5.989***	4.605***
	(0.00202)	(0.00203)	(0.00210)	(0.00211)	(0.000751)	(0.00203)
Export-time FE	YES	YES	YES	YES	YES	YES
Import-time FE	YES	YES	YES	YES	YES	YES
Country-pair FE	YES	YES	YES	YES	YES	YES
Observations	25,971	26,247	25,971	26,247	25,971	26,247
R^2	0.981	0.979	0.981	0.979	0.981	0.979

Note. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively. Values in parentheses underneath the estimated coefficients are the corresponding clustered robust standard errors.

Finally, the dataset used in our study includes the 2008 global financial crisis and the simultaneous great trade collapse period (Baldwin 2009). Such events can have a significant impact on both GVC participation levels and the likelihood of the deepening PTAs. To ensure that our main findings are not affected by the GFC and world the trade collapse, following Lee (2019) and others, the empirical model was re-estimated after removing the data for 2008 and 2009. Estimation results, presented in Table 9, are qualitatively similar to the benchmark results presented in Tables 3 and 4. Thus, it can be argued that our findings are not affected by the global financial crisis and the great trade collapse.

Table 9. Accounting for global financial crisis and collapse of world trade in 2008 and 2009.

Variables	(1)	(2)	(3)	(4)	(5)	(6)
	FGVCP	BGVCP	FGVCP	BGVCP	FGVCP	BGVCP
	all provisions		commodity provisions		factor provisions
Depth*N-N	−0.0336***	−0.00991	−0.0639***	0.00261	−0.0594**	−0.0623***
	(0.0130)	(0.0126)	(0.0215)	(0.0209)	(0.0231)	(0.0175)
Depth*N-S	0.0555***	0.0672***	0.0566***	0.0907***	0.0428	0.0242
	(0.00993)	(0.00963)	(0.0183)	(0.0178)	(0.0297)	(0.0172)
Depth*S-N	0.101***	0.0967***	0.0995***	0.123***	0.0708**	0.0370**
	(0.00998)	(0.00967)	(0.0183)	(0.0177)	(0.0309)	(0.0172)
Depth*S-S	0.0498***	0.0391***	0.0588***	0.0755***	0.0378	−0.00117
	(0.0102)	(0.00985)	(0.0186)	(0.0180)	(0.0307)	(0.0168)
FDepth			0.0169	−0.00611
			(0.0153)	(0.0148)
CDepth					0.0332	0.0675***
					(0.0296)	(0.0164)
Constant	5.880***	4.602***	5.883***	4.605***	5.881***	4.603***
	(0.00231)	(0.00224)	(0.00243)	(0.00235)	(0.00136)	(0.00224)
Export-time FE	YES	YES	YES	YES	YES	YES
Import-time FE	YES	YES	YES	YES	YES	YES
Country-pair FE	YES	YES	YES	YES	YES	YES
Observations	23,478	23,478	23,478	23,478	23,478	23,478
R^2	0.983	0.981	0.983	0.981	0.983	0.981

Note. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively. Values in parentheses underneath the estimated coefficients are the corresponding clustered robust standard errors. The estimated results presented in this table are based on the sample which does not include the data for 2008 and 2009.

7. Accounting for Potential Endogeneity

In our previous estimation results, a big concern left is the potential endogeneity issue, which might arise from two main sources. First, the depth of PTAs may be affected by bilateral GVC participation level of member countries. Specifically, a country pair with high bilateral GVC participation level is more likely to establish a deep PTA. Second, countries decide to build a deep PTA exactly because they expect it to have a significant and positive effect, so the members of deep PTAs may be not randomly assigned.

To mitigate the impact of such a potential endogeneity issue caused by the reverse causality and selection bias, we employ the two-stage least squares (2SLS) technique to investigate the impact of deep PTAs on bilateral GVC participation level. Following Laget et al. (2020), we constructed an instrumental variable for the depth of PTAs, $IV\mathrm{\_}Dept{h}_{ijt}$, as follows.

(6) $IV\mathrm{\_}Dept{h}_{ijt}=\sum _{k\ne i}{W}_{ikt}Dept{h}_{ikt}+\sum _{j\ne k}{W}_{kjt}Dept{h}_{kjt}$(6)

where W_ikt (W_kjt) is the ratio of foreign value added to total exports from country i (k) to country k (j). Depth_ijt is the depth of the PTA between country i and j, and Depth_ikt (Depth_kjt) is the depth of the PTA between country i (k) and k (j).

This instrumental variable for the depth of PTAs has a straightforward intuition. In a world where production is fragmented across international boundaries, the PTA depth between two countries could affect the depth of their PTAs signed with third countries along the supply chain. Intuitively, a deeper trade agreement signed with a third country can lower the trade costs along the entire supply chain (Laget et al. 2020), thus encouraging two member counties belonging to a deep PTA to sign deep trade agreements with third countries. Therefore, the weighted PTA depth is associated with the PTA depth between two countries, but it is less likely to be affected by the bilateral GVC participation level of the country pair.

The 2SLS estimation results are shown in Table 10. After introducing the instrumental variable into the regression of Equations (3)(3) $\begin{array}{rl}& GVC{P}_{ijt}=exp[{\beta }_0+{\beta }_1(Dept{h}_{ijt}\ast N{N}_{ij})+{\beta }_2(Dept{h}_{ijt}\ast N{S}_{ij})+{\beta }_3(Dept{h}_{ijt}\ast S{N}_{ij})\\ & +{\beta }_4(Dept{h}_{ijt}\ast S{S}_{ij})+{\delta }_{ij}+{\lambda }_{it}+{\vartheta }_{jt}]+{\epsilon }_{ijt}\end{array}$(3) –(5(5) $\begin{array}{rl}& GVC{P}_{i}jt=exp\left[{\beta }_0+{\beta }_1\left(FDept{h}_{i}jt\ast N{N}_{i}j\right)+{\beta }_2\left(FDept{h}_{i}jt\ast N{S}_{i}j\right)+\right.{\beta }_3\left(FDept{h}_{i}jt\ast S{N}_{i}j\right)\\ & \left.+{\beta }_4\left(FDept{h}_{i}jt\ast S{S}_{i}j\right)+{\beta }_{5}CDept{h}_{i}jt+{\delta }_{i}j+{\lambda }_{i}t+{\vartheta }_{j}t\right]+{\epsilon }_{i}jt\end{array}$(5) ), the estimation results in all columns are qualitatively similar to those presented in Tables 3 and 4 of Section 5. This implies that the asymmetric effect of deep PTAs on GVC participation level between developing and developed member countries is not affected by the potential endogeneity issue. In addition, the results of both Kleibergen-P LM and Kleibergen-P Wald-F imply the absence of insufficient identification of instrumental variables and weak instrumental variables.

Table 10. The asymmetric effects of PTAs (using the 2SLS to estimate).

Variables	(1)	(2)	(3)	(4)	(5)	(6)
	FGVCP	BGVCP	FGVCP	BGVCP	FGVCP	BGVCP
	all provisions		commodity provisions		factor provisions
Depth*N-N	−0.284***	0.205	−0.121	0.180	−0.139	0.133**
	(0.0704)	(0.164)	(0.129)	(0.139)	(0.120)	(0.0562)
Depth*N-S	0.0435**	0.112**	0.00181	0.126*	0.0318	0.0466
	(0.0211)	(0.0511)	(0.0587)	(0.0697)	(0.0821)	(0.0422)
Depth*S-N	0.197***	0.259***	0.160***	0.303***	0.206***	0.200***
	(0.0191)	(0.0428)	(0.0565)	(0.0653)	(0.0663)	(0.0431)
Depth*S-S	0.0759***	0.137**	0.0263	0.172**	0.0637	0.0651
	(0.0283)	(0.0664)	(0.0604)	(0.0730)	(0.0814)	(0.0415)
FDepth			0.0340	0.000768
			(0.0378)	(0.0483)
CDepth					0.00506	0.0859**
					(0.0603)	(0.0432)
Export-time FE	YES	YES	YES	YES	YES	YES
Import-time FE	YES	YES	YES	YES	YES	YES
Country-pair FE	YES	YES	YES	YES	YES	YES
Observations	27,090	27,090	27,090	27,090	27,090	27,090
K-P LM	114.636	114.636	74.674	74.674	104.084	104.084
	[0.0000]	[0.0000]	[0.0000]	[0.0000]	[0.0000]	[0.0000]
K-P Wald F	30.521	30.521	21.164	21.164	25.440	25.440
	{NA}	{NA}	{NA}	{NA}	{NA}	{NA}

Note. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively. Values in parentheses underneath the estimated coefficients are the corresponding clustered robust standard errors.

8. Conclusion and Policy Implications

Rapid growth in preferential trade agreements (PTAs) and global value chain (GVC) trade has captured the attention of researcher in recent years. However, little attention has been paid to the impact of PTAs on GVC participation levels across groups of countries. In this paper, we argue that the impact of deep PTAs on GVC participation is asymmetric across member countries, which are at different levels of economic development. Using data on 43 countries (including some emerging economies) over the 2000–2014 period and employing panel estimation methodology, we investigate whether the impact of deep PTAs on bilateral GVC participation level is significantly asymmetric across developing and developed member countries. We also examine whether the asymmetric effect varies across the provisions of deep PTAs.

Empirical results presented in this paper show that deepening of PTAs can promote bilateral GVC participation level by reducing trade barriers, policy uncertainty and transaction costs, but such a positive effect is asymmetric between developed and developing member countries. Specifically, deepening of PTAs has a stronger impact on GVC participation level of developing member countries than that of developed member countries. Moreover, compared to the commodity provisions of PTAs, Deepening of PTA factor provisions is more likely to cause such an asymmetric effect on GVC participation level between developing and developed members. The findings are confirmed by a series of robustness tests including consideration for potential endogeneity.

Our findings have some important policy implications concerning the formation of PTAs and GVC trade participation. First, our empirical results suggest that through deepening of PTAs both developed and developing member countries can increase GVC production and trade. Thus, deepening of PTAs can also be used as a strategy to further improve living standards in emerging economies. Second, developing countries should try to establish deep PTAs with developed countries because, through creating both intensive-chains and extensive-chains effects, such agreements are more likely to promote GVC participation levels of developing countries. Finally, while establishing deep PTAs with developed countries, developing countries should endeavor to include additional provisions concerning the mobility of factors of production as such provisions can generate extensive-chains effects.

Acknowledgments

This paper has greatly benefitted from very helpful comments and suggestions received from three anonymous reviewers. All remaining errors are our own.

Disclosure Statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

The authors gratefully acknowledge financial support from the National Social Sciences Fund of China (Grant No.19BJY192) and the Fundamental Research Funds for the Central Universities of China (Grant No.19JNLH06).

Notes

1. Following Limão (2016), we define PTAs as international treaties with restrictive membership and including any articles that (i) apply only to member countries and (ii) aim to secure or increase their respective market access.

2. Participation level is measured using a value-added approach.

3. For example, Association of Southeast Asian Nations (ASEAN) includes Singapore, which is a newly industrialized economy. North American Free Trade Agreement (NAFTA) includes Mexico, which is a developing country.

4. The number of provisions included determines the PTA depth (Ray 1981).

5. The developed countries in the sample include Australia, Austria, Belgium, Canada, Switzerland, Germany, Denmark, Spain, Finland, France, United Kingdom, Ireland, Japan, Korea, Luxembourg, Netherlands, Norway, Portugal, Sweden, and USA.

6. “North” denotes developed countries, whereas “South” represents developing counters. N-N means that developed countries are both exporters and importers, S-N means that developing countries are the exporters and developed countries are importers, N-S means that developed countries are the exporters and developing countries are importers, and S-S means that both exporters and importers are developing countries.

Appendix

The empirical results presented in this paper are based on aggregate data in the sense that the sample includes various PTAs, especially a very large PTA (i.e., the EU) which involves many developed countries. Therefore, based on the suggestion received from an anonymous reviewer, four important non-EU countries (i.e., the US, the UK, Japan and China) were selected as GVC-trade exporters. We then investigated whether the asymmetric effect of deep PTAs applies to GVC exports of the selected countries to their trade partners. First, as shown in columns 1 and 2 of Table A1, the effect of deepening of PTAs on forward and backward GVC participation level is positive and statistically significant only for the S-N group, and the magnitude of this estimated effect is also relatively higher for the same group. Second, as can be seen in columns 3 and 4 of Table A1, the estimated effect of the deepening of the commodity provisions on forward and backward GVC participation level is statistically insignificant for both S-N and N-S groups, but the magnitude of the estimated effect is relatively higher for the S-N group. Finally, as indicated in columns 5 and 6 of Table A1, the estimated effect of the deepening of the factor provisions on forward and backward GVC participation level is statistically significant for the S-N group but insignificant for the N-S group, and the magnitude of the estimated effects is also relatively higher for the S-N group. The results presented in Table A1 are consistent with our (aggregate) full sample results.

Table A1. Estimation results using selected countries (i.e., the US, the UK, Japan and China).

Variables	(1)	(2)	(3)	(4)	(5)	(6)
	FGVCP	BGVCP	FGVCP	BGVCP	FGVCP	BGVCP
	all provisions		commodity provisions		factor provisions
Depth*N-N	−0.0415	−0.0304	−0.0972	−0.110*	0.0126	0.0338
	(0.0313)	(0.0300)	(0.0619)	(0.0594)	(0.0460)	(0.0506)
Depth*N-S	0.0285	0.0393	−0.0133	−0.0309	0.0914	0.110*
	(0.0252)	(0.0241)	(0.0578)	(0.0554)	(0.0751)	(0.0570)
Depth*S-N	0.107***	0.140***	0.0494	0.0557	0.149*	0.193***
	(0.0248)	(0.0237)	(0.0574)	(0.0550)	(0.0825)	(0.0572)
Depth*S-S	−0.0616	−0.185	−0.0721	−0.165	0.0301	−0.0465
	(0.274)	(0.262)	(0.170)	(0.163)	(0.272)	(0.212)
FDepth			0.0507	0.0752
			(0.0520)	(0.0499)
CDepth					−0.0534	−0.0655
					(0.0633)	(0.0545)
Constant	7.623***	6.046***	7.626***	6.053***	5.989***	4.605***
	(0.0318)	(0.0304)	(0.0208)	(0.0199)	(0.000751)	(0.00203)
Export-time FE	YES	YES	YES	YES	YES	YES
Import-time FE	YES	YES	YES	YES	YES	YES
Country-pair FE	YES	YES	YES	YES	YES	YES
Observations	4,905	4,905	4,905	4,905	25,971	26,247
R^2	0.990	0.991	0.990	0.991	0.981	0.979

Note. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively. Values in parentheses underneath the estimated coefficients are the corresponding clustered robust standard errors.