1 Introduction

According to Feldstein and Krugman (1990), in an international system where countries charge VAT for imports (as do China), the non-distortionary policy is to also rebate VAT fully on exports, so that the effective VAT rates charged on domestically produced and imported goods are equalized within each country (Garred 2018). However, compared to most other countries with VAT system, China does not fully refund the VAT on exports. Instead, exporters may receive VAT rebates, which vary across products (Gourdon et al. 2017). In other words, in China, the system of incomplete VAT rebates for exporters is adopted and incomplete rebates constitute a tax on exports (Feldstein and Krugman 1990).

China's VAT export rebate system has been identified as a key component of its industrial policy, exerting influence on its international competitiveness (Gourdon et al. 2020). Consequently, the Chinese government has faced persistent allegations of granting unfair advantages to selected firms in global trade through the use of incomplete VAT export rebates (Evenett et al. 2012; Gourdon et al. 2017, 2020). While numerous studies demonstrate the positive impact of China's VAT export rebates on domestic exports (Chen et al. 2006; Evenett et al. 2012; Chandra and Long 2013; Gourdon et al. 2017, 2020), there is a noticeable absence of research addressing their effects on global trade and external exporters, specifically trade diversionFootnote 1 effects. This paper aims to address this gap in the literature by conducting a thorough empirical examination of the trade diversion effects resulting from Chinese VAT export rebates.

The study relates to a heated policy debate on the third-country effects of domestic export policies (see, e.g., Hirono 2019; Hoekman and Nelson 2020) presenting one of the early attempts to provide extensive empirical evidence in this field. The paper is also closely tied to a contentious policy debate regarding China’s utilization of VAT rebates for exporters as a potent tool in trade wars (see, e.g., Reuters 2018; Liu 2020).

Viewing VAT on exported goods as an export tax, its rebate can be considered an export subsidy that aligns effectively with the structural gravity framework employed as the theoretical foundation of empirical analysis in this paper. Chinese VAT export rebates further satisfy important requirements of empirical analysis that ultimately allows to get refined and reliable empirical evidence. First, China`s weight in the world trade is exceptionally large.Footnote 2 This per se implies that China`s trade policies can significantly affect the world trade. Second, China`s VAT rebates` policy has relatively long history. Hence, the time span is long enough for obtaining reliable results. Third, data for Chinese VAT export rebates is available at detailed industry level that enables to include refined industry dimension into the analysis. Finally, Chinese government changes the rebates rather frequentlyFootnote 3 that implies enough variation of rebates across industries and time for adequate panel data analysis.

This study utilizes comprehensive data on VAT rebates spanning a lengthy period from 2004 to 2018, distinguishing it from previous empirical studies on China's VAT rebate policies. The data is sourced from the Chinese tax refund website,Footnote 4 which provides records dating back to 2004. The data is presented in graphical form for 8 and 10-digit HS classification codes, with separate webpages dedicated to each code range. Unfortunately, it is not downloadable in bulk format. This limitation has led previous studies to compile export VAT rebate rates either based on official announcements released by the Chinese government (Eisenbarth 2017; Garred 2018; Bai and Liu 2019) or to rely on the Chinese Customs' Etax yearbooks, which are solely available in hardcopy format (Gourdon et al. 2016, 2020). To address this issue, an Excel VBA program has been developed to automate the extraction of data from the Chinese tax refund website into Excel.

For the baseline empirical analysis, a "two-stage" estimation procedure was employed. In the first stage, an explanatory variable that represents the level of China`s VAT export rebate in a 6-digit industry i and year t is introduced into a “four-way” fixed effects Poisson Pseudo-Maximum Likelihood (PPML) structural gravity empirical specification of bilateral trade flows of 75 countries over the period of 2004–2018 disaggregated at 6-digit industries. The estimation sample encompasses approximately half a billion observations. “Four-way” fixed effects are represented by exporter-importer-4-digit industry-year fixed effects. The usage of 4-digit industries in four-way fixed effects, on one hand, enables to include VAT rebate that varies across 6-digit industries and year as explanatory variable, and, on the other hand, allows to significantly eliminate omitted variables` bias and potential endogeneity in general: fixed effects control for all sources of reverse causality and omitted variables that operate at exporter-importer-4-digit industry-year levels.

Subsequently, country-specific estimates are obtained. This stage yields 75 country-specific estimates, which are subsequently employed as dependent variable in the second stage, examining the determinants of variance in these estimates.

Highly statistically significant and robust evidence reveals trade diversion effects resulting from Chinese VAT export rebates. The findings indicate that, all else being equal, a one-unit increase (represented by percent) in China's VAT rebate corresponds to approximately a 2.62 percent decrease in third-country exports. The results from the second-stage analysis are largely intuitive. Wealthier nations with greater levels of trade liberalization tend to experience fewer adverse effects from Chinese export rebate policies. The extent of these effects is influenced by a country's level of export diversification, particularly its intensive margin. Consequently, countries with a more evenly balanced mix of exports or trading partners demonstrate greater resilience to Chinese competition in export markets. Additionally, there is evidence suggesting that countries exporting higher quality goods are more vulnerable to the negative effects of Chinese competition. This underscores Chinese efforts to upgrade export quality, as evidenced by previous research (see, for example, Bas and Strauss-Kahn 2015; Anwar and Sun 2018), thereby enhancing Chinese exporters' ability to compete with high-quality foreign goods. This may also suggest that a primary objective of Chinese export policy is to foster improvements in the quality of Chinese goods.

The sector-specific analysis reveals that in bulk negative effects of China`s VAT export rebates for global trade largely concentrate in machinery and equipment sector (84–85 HS chapters). Comprising half of Chinese export and one fourth of world export, this sector has been experiencing second largest VAT export rebates in China after the sector of transport equipment (the latter one comprises only 3 percent of Chinese export and 11 percent of world export). Both qualitative and empirical analyses examining the factors contributing to the heterogeneity of country-specific effects within this sector suggest that the rebates for machinery and equipment goods are partly intended to outcompete China's neighboring countries in Asia.

We proceed to estimate the effects of rebates on both the extensive and intensive margins of trade. We consider the decomposition for the extensive partner margin as suggested in Helpman et al. (2008) and apply Hummels and Klenow (2005) approach to compute the margins. Our findings indicate that the negative impact of Chinese VAT rebates on global trade primarily stems from the intensive margin, while the effects on the extensive margin are observed to be positive. This suggests that foreign exporters tend to expand the extensive margin, i.e., increase the number of trading partners, to offset the adverse effects of Chinese VAT rebates at the intensive margin of export.

The paper is linked with several literatures. First, the paper naturally contributes to the literature on China`s VAT export rebate policy. Taxation of Chinese exports has received increasing attention from academic researchers in recent years. One direction of these studies has been focusing on the effects of VAT export rebates on China`s domestic export performance (Chen et al. 2006; Evenett et al. 2012; Chandra and Long 2013; Gourdon et al. 2017, 2020). The other direction investigates the consequences of China`s VAT rebate policy reforms (Liu and Lu 2015; Fan et al. 2020; Liu 2020). The third direction considers general patterns of the policy (Eisenbarth 2017; Garred 2018). This study adds to this literature by examining the policy effects for competitors in global trade.

Second, this study provides detailed and comprehensive empirical evidence regarding the trade diversion effects of the export policies of a major exporting nation. While there is a substantial body of literature on the effects of trade policies on non-targeted countries, the focus has primarily been on the trade diversion effects of Free Trade Agreements (FTAs) (Romalis 2007; Magee 2008; Freund 2010; Dai et al. 2014; Conconi et al. 2018), and the trade deflection effects of antidumping duties (Bown and Crowley 2006, 2007; Baylis and Perloff 2010; Chandra 2017; Sandkamp 2020). In stark contrast, there is a dearth of systematic empirical evidence on the third-country effects of export promotion policies, despite their significant policy implications. To bridge this gap, this paper investigates one of the world's largest exporters and underscores the noteworthy and highly heterogeneous third-country effects of its export policy.

Thirdly, the paper contributes to the ongoing discourse concerning China's role in global competition. A significant portion of this research focuses on the impact of Chinese imports on the domestic production activities of third countries (Acemoglu et al. 2016; Autor et al. 2013; Mion and Zhu 2013; Balsvik et al. 2015). Additionally, there are studies that examine China's competitive effects on third-country exporters. For instance, Jenkins (2014) and Iacovone et al. (2013) observed negative impacts of Chinese competition on Brazilian and Mexican exports, respectively. Conversely, Martin and Mejean (2014) identified a positive impact of Chinese competition on the quality content of French exports. This study investigates China's competitive effects through the lens of third-country effects resulting from Chinese governmental industrial policies that selectively promote the export of certain products and industries. By analyzing the heterogeneity of these effects across 75 countries, the paper provides, to the best of our knowledge, the first comprehensive evidence on the global patterns of China's export competition effects.

The rest of the paper is structured as follows. Next section briefly discusses China`s VAT rebate policy for exporters and VAT rebates` data. Second section outlines theoretical framework of the study. Third section presents empirical framework while fourth and fifth sections discuss estimation results and their robustness checking, respectively. Final section concludes.

2 Background

2.1 Export Value-Added Tax Rebate Policy in China

Despite its name, Value-Added Tax (VAT) is typically designed as a tax on consumption rather than solely on value added. It is charged at all stages of production, but with the provision of some mechanism enabling firms to offset the tax they have paid on their own purchases of goods and services against the tax they charge on their sales of goods and services (Ebrill et al. 2001). In theory, neutral VAT implies a zero rate on exported goods and a full refund of VAT paid by exporters on their inputs (Gourdon et al. 2020). However, in practice, refunds can be partial, as observed in China.

In 1994, China initiated a significant tax reform by replacing the old industrial and commercial standard tax with a new value-added tax (Cui 2003). The standard VAT rate in China remained at 17 percent until 2017. A reduced rate of 13 percent was applied to basic foodstuffs, utilities, newspapers, and inputs to agricultural production. In 2018, the standard VAT rate in China was reduced to 16 percent, followed by a further reduction to 13 percent in 2019. Additionally, China introduced a partial VAT refund system for inputs paid by exporters, with the refund rate varying by industry. Consequently, the modern Chinese VAT system imposes an additional tax burden on exporters whose goods receive a VAT refund rate lower than the applicable VAT rate.

Chinese VAT export rebates undergo frequent changes, primarily driven by objectives such as supporting sophisticated high-technology products, restricting the export of energy-intensive and polluting products, managing risks associated with trade disputes, and ensuring food security (Gourdon et al. 2016, 2020; Eisenbarth 2017).

2.2 VAT Rebates Data Patterns

In this study, we utilize product-level data on Chinese VAT rates and VAT rebate rates for exporters sourced from the Chinese tax refund websiteFootnote 5 where data has been available from 2004. The data is presented in graphical form on separate webpages for each 8–10-digit product. To efficiently extract the data, an Excel VBA program was developed to passively transfer the information from the website into Excel. For the purposes of analysis, we first arrange the raw data across years and 8–10-digit products, and then aggregate it to 6-digit HS codes (rebate rates within 6-digit codes are usually identical; see also Bai and Liu 2019) and convert it into HS2002.

As depicted in Fig. 1, there has been a downward trend in the mean VAT rebate until 2008, followed by an increase and subsequent fluctuation around 10 percent between 2010 and 2018. Notably, during the global economic downturn of 2008–2010, the Chinese government implemented a fiscal stimulus program. Additionally, from July 2008 to June 2009, all modifications to the Chinese VAT export regime entailed expansions of rebates on thousands of product lines. Unlike the fiscal stimulus, changes in VAT rebates were not reversed at the end of 2010 (Evenett et al. 2012). The standard deviation of VAT rebates gradually increased until 2010 (see Fig. 2). From 2010 to 2018, the standard deviation remained relatively stable at around 6.

Fig. 1
figure 1

Source: Chinese tax refund website and author`s calculations

Mean of VAT rebate rates, VAT rates and final VAT (rate minus rebate) for exporters in China in 2004–2018; based on 6-digit HS industry data.

Full size image
Fig. 2
figure 2

Source: Chinese tax refund website and author`s calculations

Standard deviation of VAT rebate rates, VAT rates and final VAT (rate minus rebate) for exporters in China in 2004–2018; based on 6-digit HS industry data.

Full size image

Figure 3 illustrates the distribution of mean VAT rebates across 6-digit industries from 2004 to 2018. It is evident from the histogram that the distribution of VAT rebates resembles a plateau-type distribution, with the most prominent peak occurring within the rebate rates ranging between 14.3 and 15.4, approximately 90% of the standard VAT rate during the study period (which was set at 17 from 2004 to 2017).

Fig. 3
figure 3

Source: Chinese tax refund website and author`s calculations

Histogram of China`s VAT export rebates (across 6-digit HS industries; as average over 2004–2018).

Full size image

Table 1 summarizes sectoral patterns of final export VATs (calculated as the difference between the VAT rate and the VAT rebate) spanning the years 2004 to 2018. Highest VAT rebates (relative to VAT rates) and, hence, lowest final VATs for exporters are observed in the sectors of vehicles, aircraft, vessels, and associated transport equipment (HS chapters 86–89) and of machinery and mechanical appliances, electrical equipment, etc. (HS chapters 84–85). While the former sector comprises around 3% of Chinese export, the latter one represents almost half of it. Other goods with high VAT rebates include textile, footwear, optical and the like instruments and apparatus. All these sectors also exhibit the lowest standard deviations of final export VATs both across goods and time that points to the consistency of export promotion component of Chinese VAT export rebate policy.

Table 1 Sectoral patterns of final export VATs (the difference between the VAT rate and the VAT rebate), 2004–2018
Full size table

The sectors of mineral products (classified under HS chapters 25–27), wood and paper (HS chapters 44–49), precious stones and metals (HS chapter 71), raw hides and skins, leather, furskins (HS chapters 41–43), and metallurgy (HS chapters 72–83) exhibit the lowest VAT rebates relative to VAT rates. Notably, within the sectors of metallurgy, mineral products, and precious stones and metals, there is a discernible trend of high standard deviations in final export VATs across various goods and over time. This variability suggests a nuanced approach within China's VAT export rebate policy, aiming to impose export restrictions on specific product categories.

In Online Appendix A we offer a comprehensive description of goods, delineated by their 6-digit product codes, that have either been subject to zero VAT rates (entailing full VAT rebates) or have received zero VAT rebates throughout the entire study period spanning from 2004 to 2018. Notably, cotton (comprising HS codes 520,100 and 520,300) and various items falling under HS chapters 84–90 (a total of 185 6-digit items) stand out for enjoying full VAT rebates across the entire duration of the study. Remarkably, these goods collectively account for 8.85% of Chinese exports during the period under examination. Interesting observation is that only one 6-digit good (out of 185 with full VAT rebates across the period) – 847,130, portable digital automatic data processing machines, weighing not more than 10 kg, consisting of a least a central processing unit, a keyboard, and a display – comprises 4.6% of Chinese export in 2004–2018.

The range of sectors encompassing products that have never received VAT rebates throughout the 2004–2018 period is notably broader, spanning items from HS chapters 5, 9, 12, 22–29, 38, 40, 44, 45, 47, 48, 63, 71, 72, 74, 76, 81, 87, and 93, totaling 244 6-digit products. However, their collective contribution to Chinese exports over the same period is marginal, accounting for less than 1% (0.81%) of the total export volume.

Our analysis of VAT rebate data aligns with previous research findings. Eisenbarth (2017) asserts that industries characterized by higher water pollution intensity, such as metallurgy, mining, and wood processing, tend to experience elevated levels of non-refunded VAT tax in China. This trend is particularly evident in sectors dealing with resources such as wood, mineral, and metal products, as well as precious stones. Similarly, Gourdon et al. (2016) observe that Chinese export VAT rebates largely aim to enhance export of industries with high technological intensity while damping the export of environmentally unfriendly industries.

3 Empirical Strategy

3.1 Data and Baseline Specification

The panel data utilized in this study comprises four dimensions: exporter, importer, 6-digit product, and year. Data pertaining to bilateral trade was sourced from the CEPII BACI database, covering the period from 2004 to 2018 and encompassing 6-digit HS-2002 industries across 75 countries, both as exporters and importers. The dataset was balanced across all four dimensions. The BACI database, derived from UN COMTRADE data, offers several advantages over raw COMTRADE data. Firstly, the CEPII developed a procedure that reconciles the declarations of the exporter and the importer, that may be different in the original data. Secondly, the differentiation between zero values and missing trade values is more straightforward in the BACI version. Specifically, BACI provides two zero trade flow dummies: the first dummy is assigned a value of one if either the exporter or importer has at least one nonzero trade flow during the year (imports or exports), and zero otherwise; the second dummy takes a value of one if the exporter-importer dyad has at least one nonzero bilateral trade flow during the year, and zero otherwise. In this study, observations are classified as missing if the sum of these two dummies equals zero, and as zero trade value otherwise.

The selection of countries for inclusion in the analysis is predicated on their aggregate export volumes from 2004 to 2018, with the first 75 largest exporters being incorporated. Notably, China is omitted from the final sample of 75 countries due to the theoretical premise that China's VAT rebates exert contrasting effects on domestic and foreign trade. The chosen 75 countries collectively represent 85% of global exports. In the remaining 15% China`s weight is 12%. The list of both included and excluded countries is detailed in the Online Appendix (Appendix B).

In tackling the prevalence of zero trade observations in the refined data sample (comprising 85% of the total observations in this study), the widely advocated approach is to employ the Poisson Pseudo Maximum Likelihood (PPML) model for estimating the structural gravity equation. This methodology is endorsed by various scholars, including Head and Mayer (2014) and Larch and Yotov (2016). Thus, the baseline specification for our analysis is formulated as follows:

$${X}_{abit}=\mathit{exp}[{{a}_{0}+\pi }_{abst}+{a}_{1}VAT{R}_{it,China}]+{\varepsilon }_{abit}$$
(1)

where Xabit is bilateral trade flows in US dollars from country a (exporter; 1 … 75) to country b (importer; 1 … 75) in year t (2004 … 2018) and industry i (6-digit HS industry; 1 … 5219).

\(VAT{R}_{it, China}\) represents China’s Value-Added Tax (VAT) export rebate, expressed as a percentage, corresponding to the 6-digit Harmonized System (HS) industry category i in year t. Each unit increase in \(VAT{R}_{it, China}\) corresponds to a one percentage point (pp) change in the rebate rate. The data pertaining to Chinese export VAT rebate rates at the product level is sourced from the Chinese tax refund website, with additional details regarding the data transfer process provided in the Introduction and Background section. This source furnishes a comprehensive compilation of export VAT rebate rates corresponding to 8–10-digit HS product codes, starting from the year 2004. Initially, we aggregate the rebate rates to 6-digit HS codes by computing arithmetic averages, given that rebate rates within 6-digit codes typically exhibit uniformity (as noted by Bai and Liu 2019). Subsequently, we convert the aggregated data into the HS-2002 classification system.

The term \({\pi }_{abst}\) represents the ensemble of four-way exporter-importer-four-digit sector-year fixed effects. This encompasses:

  1. 1.

    Time-varying four-digit sector-specific exporter (country a) dummies, denoted as \({d}_{ast}\), which consider sector-specific outward multilateral resistances, countries' output shares, and any other observable and non-observable exporter-specific factors influencing bilateral trade.

  2. 2.

    Time-varying four-digit sector-specific importer (country b) dummies, denoted as \({d}_{bst}\), that control for sector-specific inward multilateral resistances, total expenditure, and other observable and non-observable importer-specific characteristics.

  3. 3.

    Four-way fixed effects representing time-varying four-digit sector-specific country-pair fixed effects, denoted as \({d}_{abst}\), which assess all observable and non-observable sector-specific bilateral trade costs over time. These effects encapsulate variables such as free trade agreements, alterations in trade tariffs, and non-tariff restrictions specific to countries and sectors across time.

Additionally, the set comprises export-specific (\({d}_{a}\)), importer-specific (\({d}_{b}\)), four-digit sector-specific (\({d}_{s}\)), and year-specific (\({d}_{t}\)) dummies, along with all bilateral combinations of these dummies. ε_abit represents the error term in the model.

Acknowledging the limitations associated with employing multiple levels of fixed effects in regression, it is essential to emphasize that, according to Weidner and Zylkin (2019), while the Poisson Pseudo-Maximum Likelihood (PPML) estimator for “three-way” fixed effects gravity models remains consistent when the time dimension is fixed, the confidence intervals are not accurately centered, and the cluster-robust variance estimates exhibit bias. Although Weidner and Zylkin (2019) offer practical remedies for more reliable inferences concerning trade policies and network panel data in a "three-way" gravity model, they explicitly state that addressing the bias in the four-way gravity model is left for future work. Consequently, we are unable to apply their insights to rectify potential biases in our specific context.

It is important to note that in this study, we treat the VAT rebate variable as exogenous. Firstly, it is technically challenging to incorporate an endogeneity assumption into the multiple-level fixed effects PPML model in an accurate and straightforward manner. Secondly, there are several reasons to support the plausibility of the exogeneity assumption for the VAT rebate variable within the framework of this study. In all previous studies examining the effects of China's VAT rebates on domestic (Chinese) exports, VAT rebates are treated as endogenous. However, in this study, the dependent variable focuses on third-country exports, distinguishing it from the domestic context. Although adjustments to China's VAT rebates may align with global trade patterns to some extent, the majority of these patterns have been accounted for by the four-way fixed effects used in the model.

Any remaining endogeneity of VAT rebates is likely limited to the 4-digit sector-year dimension and is primarily attributed to omitted variable bias. In our model, the omitted variables are confined to the exporter-importer-4-digit sector-year dimension. The primary observable omitted variable in this context would be importer-specific time-varying import tariffs at the 6-digit industry level that is endogenous itself. Importer-specific time-varying import tariffs at the 6-digit industry level exhibit endogeneity in trade gravity regressions due to several factors. Firstly, policy endogeneity arises as import tariffs are often influenced by various economic factors such as domestic conditions, trade relationships, and international agreements, which are themselves affected by trade volumes. Secondly, reverse causality occurs as tariffs impact trade volumes by altering the attractiveness of imports, leading to adjustments in tariffs driven by changes in trade volumes. Lastly, trade policy responses further compound the endogeneity, as policymakers adjust tariffs in reaction to shifts in trade volumes or patterns, aiming to safeguard domestic industries or address trade imbalances.

However, addressing the endogeneity of this variable within the PPML model with multiple fixed effects is technically challenging. In addition, based on computations using the study's data settings, the correlation between China's VAT rebate variable and the importer-specific import tariff variable at the 6-digit industry level is very small, at 0.05. Therefore, omitting this control variable is unlikely to introduce significant omitted variable bias.

Fortunately, the specific focus of this study's research question allows for the utilization of a sophisticated framework that effectively addresses the endogeneity problem. To counteract trade policy endogeneity, recent approaches in empirical trade policy analysis advocate for incorporating multiple levels of fixed effects into the structural gravity specification. This approach compels identification to stem from the within dimension of the data, as discussed in studies by Head and Mayer (2014), Yotov et al. (2016), and Weidner and Zylkin (2019). However, it's essential to highlight that in the robustness section, we partially address the endogeneity issue by incorporating the one-year lagged VAT rebate variable.

3.2 Country-Specific Specification

The estimation framework can be readily adapted to derive estimates of China’s VAT rebate effects for each affected country. The model is expressed as follows:

$${X}_{abit}=\mathit{exp}[{a}_{0}+{\pi }_{abst}+{a}_{1}VAT{R}_{it,China}+{a}_{2}{D}_{Y,b}+{a}_{3}VAT{R}_{it,China}{\times D}_{Y,b}]+{\varepsilon }_{abit}$$
(2)

where \({D}_{Y,b}\) is a dummy variable that equals one if the export originates from a particular third country Y among countries a, and zero otherwise. The cumulative impact of rebates for a specific exporter Y is thus captured by the summation of coefficients \({a}_{1}\) and \({a}_{3}\). We estimate Eq. (2) for each country a separately.

To elucidate the determinants influencing the heterogeneity of rebates' trade diversion effects across third countries-exporters, we employ a regression analysis where country-specific effects, \({CSE}_{a}\), as sum of coefficients \({a}_{1}\) and \({a}_{3}\) from Eq. (2) are regressed on a set of explanatory variables:

$$\begin{array}{c}{CSE}_{a}={b}_{0}+{{b}_{1}GDPpc}_{a}+{b}_{2}{POP}_{a}+{b}_{3}{DIST}_{aChina}+{b}_{4}{RCA}_{a}+\\ +{b}_{5}{ImpTar}_{a}+{b}_{6}{ExpQ}_{a}+{b}_{7}{ExpDiv}_{a}+{\varepsilon }_{a}\end{array}$$
(3)

where a denotes specific country-exporter, \({b}_{n}\) are coefficients to be estimated and \({\varepsilon }_{a}\) is an error term. We begin by considering general factors, such as GDP per capita in USD (\({GDPpc}_{a}\)) and population (\({POP}_{a}\)) in country a computed as arithmetic averages over the period of 2004–2018, sourced from the World Bank. Our consideration of these general factors stems from their fundamental roles in shaping economic activity and trade dynamics. Higher GDP per capita levels often correlate with increased purchasing power and demand for imports, while larger population sizes may signify larger markets and greater consumption potential, thus exerting notable effects on trade patterns.

Subsequently, we incorporate more specific factors. The first indicator is the geodesic distance between country a and China (\({DIST}_{aChina}\)) derived from the CEPII database. We utilize the variant of CEPII distance computed based on geographical coordinates of the capital cities, following the methodology outlined by Mayer and Zignago (2011). The incorporation of this variable is driven by its pivotal role in shaping trade costs and patterns, acknowledging the substantial influence of geographic proximity on trade dynamics.

The second variable under consideration is the overall revealed comparative advantage (RCA) of country a (\({RCA}_{a}\)) sourced from the World Integrated Trade Solution (WITS). To compute this, revealed comparative advantage indices for country a in each year from 2004 to 2018 have been calculated using a weighted average formula:

$${RCA}_{at }=\frac{\sum_{s=1}^{s=15}{RCA}_{ast}\times {WESh}_{st}}{\sum_{s=1}^{s=15}{WESh}_{st}}$$
(4)

Here, \({RCA}_{ast}\) represents the revealed comparative advantage index of country a in year t within sector s. WITS provides data for 15 broad sectors, including animal, vegetables, food products, minerals, fuels, chemicals, plastic, or rubber, hides and skins, wood, textile and clothing, footwear, stone and glass, machinery/electrical, transportation, and miscellaneous. \({WESh}_{st}\) denotes the sector's share in world exports for the respective year. Subsequently, arithmetic averages of the weighted revealed comparative advantage indices over the period of 2004–2018 are computed for each country a. The consideration of country-specific revealed comparative advantage (RCA) indices serves to capture the relative strengths and weaknesses of country a in specific sectors vis-à-vis global competitors. As elucidated in Eq. (4), the RCA index reflects both the breadth and intensity of a country's export specialization, providing insights into its comparative advantage profile. This variable is instrumental in assessing how trade diversion effects may vary based on countries' comparative strengths across different sectors.

The third variable pertains to the level of import tariffs in country a (\({ImpTar}_{a})\) which is quantified as the arithmetic average of three import tariff indicators. These indicators include the Most Favored Nation (MFN) simple average tariff (%), effectively applied simple average tariff (%), and the weighted average of MFN tariffs by their corresponding trade value (%) over the period of 2004–2018. Data for this variable is sourced from WITS. Its inclusion serves to account for the impact of trade policy measures on trade diversion dynamics. Higher tariff barriers tend to hinder imports and may prompt increased reliance on alternative suppliers, thereby influencing the magnitude and direction of trade diversion effects.

Finally, we introduce export quality and diversification indicators to our analysis, recognizing the nuanced dimensions of countries' export characteristics. These variables offer insights into the quality and diversity of countries' export baskets, which can affect their vulnerability to trade diversion effects. First, we examine the export quality indicator in country a (\({ExpQ}_{a}\)) represented as the simple average over the period of 2004–2010. Data for this indicator is sourced from the diversification toolkit of the International Monetary Fund (IMF) via IMF DataMapper. The methodology utilized to compute the IMF export quality variable is elucidated in detail by Henn et al. (2013) and is based on unit values. Higher values for the quality indices signify elevated quality levels in exports.

Next, we incorporate three indicators of export diversification (\({ExpDiv}_{a}\)) sourced from the IMF's diversification toolkit. The primary export diversification index assesses the diversification of country a's exports, focusing on either product narrowly defined or trading partners. This index is further segmented into the extensive and intensive margins of diversification. Extensive diversification denotes an increase in the number of export products or trading partners, while intensive diversification considers the distribution of export volumes across active products or partners. Consequently, a country's export portfolio is deemed less diversified when export revenues are concentrated in a few sectors or with a limited number of trading partners, despite exporting a wide range of goods or to various trading partners (extensive margin). Conversely, countries with a more balanced mix of exports or trading partners exhibit a higher level of intensive diversification (IMF policy paper 2014). The computation of all three indices relies on the Theil index methodology outlined by Cadot et al. (2011). It is noteworthy that higher values for all three indices indicate lower levels of diversification. We calculate arithmetic averages of the respective indices over the period of 2004–2010 for each country a.

Given the potential for unobservable error in the first-stage estimates of trade diversion effects, it is crucial to address how this error affects the second stage. Lewis and Linzer (2005) propose that in regression models reliant on estimates, when the share of the regression residual attributed to sampling error in the dependent variable is substantial, weighted least squares (WLS) is preferable to ordinary least squares (OLS) with robust standard errors. In this study, notable sampling errors are anticipated for two reasons: firstly, only 75 countries out of 195 were included in the first-stage estimations, and secondly, due to data availability constraints, analysis was limited to a specific period. Consequently, standard WLS has been employed as the estimation method.

4 Results

4.1 Baseline Results

Table 2 displays the economy-wide outcomes derived from the baseline specification (1). Estimations were conducted utilizing the Poisson Pseudo Maximum Likelihood (PPML) model incorporating multiple levels of fixed effects.Footnote 6 Descriptive statistics for the variables are provided in the Online Appendix (Appendix C).

Table 2 Baseline structural gravity equation economy-wide results
Full size table

We observe that the coefficient estimate of the VAT variable exhibits a sign consistent with theoretical expectations and is statistically significant at a high level. Additionally, the pseudo R-square value is considerable. The interpretation of the coefficient \({a}_{1}\) is straightforward: all else being equal, a one-unit increase in China's VAT rebate in industry i (measured in percentage points) corresponds to an average decrease of 2.62%Footnote 7 in the export of that industry to third countries.

Table 3 provides an overview of the estimation outcomes for specification (2), which accounts for the country-specific impacts of Chinese VAT export rebates on third-country trade. A comprehensive summary detailing the effects, including estimates of coefficients \({a}_{1}\) and \({a}_{3}\), along with their standard errors and significance levels, is accessible in the Online Appendix (Appendix D). Figure 4 illustrates the distribution of country-specific estimates for the coefficient \({a}_{3}\), along with their 95% confidence intervals, offering crucial insights into the variability of \({a}_{3}\) across countries and aiding interpretation of the model results. Figures 5 and 6 represents a Pareto chart that plots the distribution of the final effects (as reported in Table 3) in descending order of frequency with a cumulative line on a secondary axis as a percentage of the total.

Table 3 Country-specific effects of China`s VAT export rebates for trade in monetary terms
Full size table
Fig. 4
figure 4

Distribution of country-specific estimates of the coefficient \({a}_{3}\) with 95% Cis

Full size image
Fig. 5
figure 5

Pareto chart of the final effects as reported in Table 3

Full size image
Fig. 6
figure 6

Machinery and equipment sector: Distribution of country-specific estimates of the coefficient \({a}_{3}\) with 95% Cis

Full size image

We can discern that the influence of the Chinese export rebate policy on its international competitors varies quite considerably. Notably, the coefficient \({a}_{3}\) was statistically insignificant for merely 18 countries, while 72% of countries encountered negative effects, aligning with expectations. Nonetheless, the effect was observed to be positive for 21 countries in the sample, accounting for 28% of the total. While the adverse effects on third countries resulting from Chinese VAT rebates are indicative of competitive consequences, the positive effects may arise from the heavy reliance of many countries on importing intermediate goods from China. Consequently, these countries benefit from more affordable Chinese inputs due to the VAT rebates. Illustratively, within the group of nations experiencing the most substantial positive effects of Chinese VAT export rebates, we observe several Asian countries, including the Philippines, Republic of Korea, Malaysia, Japan, Singapore, and Thailand and several Arabian countries, such as Saudi Arabia, Kuwait, and Iran, all of which are notably reliant on Chinese inputs (see, e.g., EconPol Europe 2022; Kavanagh and Wehrey 2023).

Next, we delve into the analysis of the factors contributing to the heterogeneity of the effects by estimating Eq. (3) as described earlier. The estimation results using WLS are showcased in Table 4. Descriptive statistics and the correlation matrix of the variables are available in the Online Appendix (Appendix E). Given the significant issue of multicollinearity in the data, where the RCA variable and all the export characteristics indicators exhibit strong correlations with each other, we systematically introduce these correlated variables one by one into our regression models. Additionally, GDP per capita was omitted from most regressions due to its strong correlation with the majority of the variables of interest. Generally, each regression (excluding Model 3a) incorporates only variables with pairwise correlations not exceeding 0.5. To facilitate a more straightforward interpretation of the results, all independent variables have been standardized.

Table 4 Country-specific determinants of trade diversion effects of China`s export VAT rebates: WLS estimation results
Full size table

Firstly, the results predictably indicate that wealthier countries tend to experience less adverse effects from Chinese export competition. Secondly, countries situated farther from China tend to be more adversely affected by Chinese rebates. This outcome is unsurprising given that China is regarded as a primary competitor for many European countries and the USA. Furthermore, as anticipated, we find that trade liberalization, reflected in lower import tariffs, mitigates the negative effects of Chinese competition.

Thirdly, empirical evidence indicates that countries specializing in higher quality goods are more prone to experiencing adverse effects from the Chinese export rebate policy, as indicated by Models 3 and 3a. Notably, in Model 3a, the effect is amplified when controlling for GDP per capita, which is closely correlated with the export quality variable. This reaffirms the presence of Chinese export quality enhancement strategies aimed at bolstering the competitiveness of Chinese exporters against high-quality foreign goods, as evidenced by studies such as Bas and Strauss-Kahn (2015) and Anwar and Sun (2018). Furthermore, this observation suggests that one of the primary objectives of the Chinese rebate policy is to facilitate quality improvements in Chinese exports.

Fourthly, the export diversification of third countries at both the extensive and intensive margins moderates the effects of rebates differently. Lower diversification at the intensive margin exacerbates the negative trade diversion effects of the rebates. Conversely, lower diversification at the extensive margin mitigates these negative effects. This implies that concentrating on a select few trade partners could be a prudent trade strategy, to some extent. Such a strategy entails cultivating close, robust, and enduring relationships with trade partners that are more resilient to external competitive pressures.

Lastly, it is surprising to discover that countries with a stronger revealed comparative advantage are more adversely affected by China's rebate policy. This finding implies that China's export policy is notably assertive, deliberately focusing on countries with a stronger position in global trade.

4.2 Results for Machinery and Equipment Sector

Based on the findings presented in Table 1 of Section 1, we deduced that the machinery and equipment sector, constituting half of Chinese exports and one-fourth of global exports, has been granted the second-largest VAT export rebates in China. Consequently, it is reasonable to expect that the adverse effects on third countries resulting from China's VAT rebates would predominantly concentrate in this sector. Thus, a more detailed examination of the sector-specific effects would enhance our analysis.

In this subsection, we conduct estimations for the machinery and equipment sector using a modified version of Eq. (1):

$${X}_{abit}=\mathit{exp}[{{a}_{0}+\pi }_{abst}+{\mu }_{abi}+{a}_{1}VAT{R}_{it,China}]+{\varepsilon }_{abit}$$
(5)

where the term \({\mu }_{abi}\) denotes three-way exporter-importer-6-digit industry fixed effects. Incorporating these fixed effects allows for a more nuanced consideration of time-invariant industry-specific bilateral trade costs compared to the baseline specification. This refinement is made feasible by a notable reduction in the number of observations in the sector-specific regression. The estimation results are outlined in Table 5.

Table 5 Machinery and equipment sector-specific effects of China`s VAT export rebates for trade in monetary terms
Full size table

It is important to highlight that our sector-specific estimations benefit from the inclusion of additional exporter-importer-6-digit industry fixed effects. This enhancement leads us to believe that our sector-specific estimates are more robust and less susceptible to omitted variable bias and general endogeneity compared to the baseline estimates. The findings indicate that, all else being equal, a one-unit increase (measured in percentage points) in China's VAT rebate for product i in the machinery and equipment sector results in an average decrease of 2.65% in third-country exports of that product. Notably, this result closely mirrors our baseline finding.

Next, we proceed to estimate the adapted form of Eq. (2), which considers the country-specific effects of Chinese VAT export rebates on third-country trade, focusing specifically on the machinery and equipment sector (HS chapters 84–85):

$${X}_{abit}=\mathit{exp}[{{a}_{0}+\pi }_{ab"\text{4di"}t}+{\mu }_{abi}+{a}_{1}VAT{R}_{it,China}+{a}_{2}{D}_{Y,b}+{a}_{3}VAT{R}_{it,China}{\times D}_{Y,b}]+{\varepsilon }_{abit}$$
(6)

Table 6 provides an overview of the country-specific effects in the machinery and equipment sector. For detailed information on the effects, including estimates of coefficients a1 and a3, their standard errors, and significance levels, please refer to the Online Appendix (Appendix F). Additionally, Figs. 5 and 6 illustrates the distribution of country-specific estimates of the coefficient \({a}_{3}\) along with their associated 95% confidence bounds. Furthermore, Fig. 7 depicts a Pareto chart showing the distribution of the final effects reported in Table 6, arranged in descending order of frequency, with a cumulative line displayed on a secondary axis as a percentage of the total.

Table 6 Machinery and equipment sector: Country-specific effects of China`s VAT export rebates for trade in monetary terms
Full size table
Fig. 7
figure 7

Machinery and equipment sector: Pareto chart of the final effects as reported in Table 6

Full size image

It can be inferred that the effects within the machinery and equipment sector exhibit less heterogeneity compared to the entire sample. Specifically, for 34 out of 75 countries, the coefficient \({a}_{3}\) is not statistically significant, indicating that the effects do not significantly deviate from the average effect within the sector's subset. Additionally, a discernible pattern emerges wherein China's Asian neighbors (including the Philippines, Vietnam, Thailand, Hong Kong, and South Korea) have consistently experienced significantly larger negative effects than the sector's average.

We proceed to empirically analyze the factors contributing to the heterogeneity of the effects by estimating Eq. (3). It should be noted that RCA and export quality variables have been measured at the machinery and equipment sector level in these estimations. The WLS estimation results are presented in Table 7. While there remains a multicollinearity issue in the data (albeit slightly smaller than in the entire sample; refer to Online Appendix G), each regression (except Model 3a) incorporates only variables with pairwise correlations not exceeding 0.5. To enhance the clarity of interpretation, all independent variables have been standardized.

Table 7 Country-specific determinants of trade diversion effects of China`s export VAT rebates: WLS estimation results; machinery and equipment sector
Full size table

Overall, the findings closely resemble the economy-wide results outlined in Table 4. It is evident that countries with higher wealth, lower import tariffs, lower revealed comparative advantage (in the machinery and equipment sector), higher export diversification at the intensive margin, and lower diversification at the extensive margin tend to experience lesser adverse effects from the Chinese export rebate policy.

However, there are two notable differences in the results. The first difference pertains to the export quality variable. In Model 3, the export quality variable (specific to the machinery and equipment sector) exhibits a positive association with the dependent variable, suggesting that countries exporting higher quality goods in the machinery and equipment category are more resilient to Chinese competition. This finding contrasts with the conclusion drawn from the analysis of the entire sample. However, in Model 3a, when the GDP per capita variable is added (which is highly correlated with the export quality variable, with a pairwise correlation coefficient of 0.94), the coefficient of the export quality variable changes sign to negative. Consequently, the evidence regarding the impact of export quality on the effects of Chinese competition is inconclusive.

The second distinction from the results of the entire sample pertains to the distance variable. In the whole sample analysis, the findings unequivocally indicate that countries farther away from China tend to be more adversely affected by the Chinese rebate policy. However, in Model 1 of Table 7, specifically for the machinery and equipment sector, the corresponding association is contrary, indicating that countries neighboring China tend to suffer more from the rebates. Although in other models the distance variable is not statistically significant (yet remains positive as in Model 1), these results collectively suggest that in the machinery and equipment sector, both neighboring and distant countries may be equally impacted by Chinese rebates.

4.3 Effects at Intensive and Extensive Margins

In the preceding analysis, we examined the trade diversion effects of China's VAT rebates for total trade. However, it is important to recognize that these effects may differ across the intensive and extensive trade margins. Notably, the unique aspect of studying China's VAT rebate effects at the extensive and intensive trade margins lies in the inability to apply the commonly used decomposition for the extensive product margin (as explained by Dutt et al. 2013) due to the variation of the explanatory variable of interest, VAT rebate, across products and years. Consequently, we adopt the decomposition for the extensive partner margin, as suggested by Helpman et al. (2008), which aligns well with our empirical framework.

The basic measure of the extensive margin is determined by the count of destinations (importers) j to which country i exports product k at time t:

$${EM}_{kit}^{simple}=\sum_{j\in J}I\left({T}_{kit}\left(j\right)>0\right)$$
(7)

where J is the set of destinations in world trade, I(x) is a logical function that returns a value of one if statement x is true and zero otherwise. \({T}_{kit}\left(j\right)\) is the export flows to destination j of product k from exporter i in year t. However, this measure does not consider the relative importance of destination j in product k world market. To correct this, we apply Hummels and Klenow (2005) approach to derive the weighted extensive margin measure:

$${EM}_{kit}^{weighted}=\sum_{j\in J}I\left({T}_{kit}\left(j\right)>0\right)\left(\frac{{T}_{kt}(j)}{{T}_{kt}}\right)$$
(8)

where \({T}_{kt}\left(j\right)={\sum }_{k\in K}{T}_{kit}\left(j\right)\) is product k import by destination j from the world, K is the set of k products. \({T}_{kt}={\sum }_{k\in K}{T}_{kit}={\sum }_{j\in J}{T}_{kt}(j)\) is k product imported to all destinations from the world.

The simple measure of the intensive margin of exports of product k from country i in year t is represented by the average volume of the export per destination j:

$${IM}_{kit}^{simple}=\frac{{T}_{kit}}{{\sum }_{j\in J}I({T}_{kit}\left(j\right)>0)}$$
(9)

where \({T}_{kit}={\sum }_{j\in J}{T}_{kit}(j)\) is total export of product k from country i in year t. Once more, this measure overlooks the significance of destination j in the global market for product k. To overcome this limitation, we adopt the approach proposed by Hummels and Klenow (2005), which entails weighting each destination j with the total imports of product k by destination j from the world:

$${IM}_{kit}^{weighted}=\frac{{T}_{kit}}{{\sum }_{j\in J}I({T}_{kit}\left(j\right)>0){T}_{kt}(j)}$$
(10)

The product of the weighted measures is the market share of exporter i in product k overall import market:

$${EM}_{kit}^{weighted}\bullet {IM}_{kit}^{weighted}=\sum_{j\in J}I\left({T}_{kit}\left(j\right)>0\right)\left(\frac{{T}_{kt}\left(j\right)}{{T}_{kt}}\right)\bullet \frac{{T}_{kit}}{{\sum }_{j\in J}I\left({T}_{kit}\left(j\right)>0\right){T}_{kt}\left(j\right)}=\frac{{T}_{kit}}{{T}_{kt}}$$
(11)

To evaluate the rebates` effects at extensive and intensive margins we estimate the following specification:

$${EM}_{kit}^{weighted}/{IM}_{kit}^{weighted}=exp\left[{{a}_{0}+\pi }_{a"\text{4di"}t}+{\mu }_{ai}+{a}_{1}{VATR}_{it, China}\right]+{\varepsilon }_{ait}$$
(12)

where \({\pi }_{a"\text{4di"}t}\) represents 3-way exporter-4-digit industry-year fixed effects and \({\mu }_{ai}\) represents 2-way exporter-6-digit industry fixed effects. Table 8 presents the results. PPML model with multiple fixed effects was used for estimations.

Table 8 Results at trade margins
Full size table

As observed, the negative trade diversion effects of Chinese export rebates on VAT are primarily influenced by the intensive margin, whereas positive effects are evident at the extensive margin. This implies that exporters offset their losses at the intensive margin by expanding the extensive margin, namely by increasing the number of trade partners or export destinations. At a superficial glance, these results may appear contradictory to our findings regarding the factors influencing the heterogeneity of effects (Tables 4 and 7). However, there is no contradiction, as countries with lower intensive margin and higher extensive margin have fewer opportunities to counteract negative effects at the intensive margin by expanding the extensive margin.

5 Robustness Check

5.1 Estimations with Lagged VAT Rebate Variable

In this subsection we estimate the following modification of Eq. (1):

$${X}_{abit}=\mathit{exp}[{{a}_{0}+\pi }_{ab"\text{4di"}t}+{a}_{1}VAT{R}_{i,t-1,China}]+{\varepsilon }_{abit}$$
(13)

where \(VAT{R}_{i,t-1,China}\) represents VAT rebate variable lagged one year. Despite the relatively frequent changes in VAT rebates in China, it is anticipated that a one-year lag would not be substantially influenced by these alterations. Moreover, the utilization of a lagged variable assists in mitigating endogeneity concerns to some extent. The results are presented in Table 9. PPML model with multiple fixed effects was used for estimations.

Table 9 Results with lagged VAT rebate variable
Full size table

As observed, the results are essentially identical to the baseline findings.

5.2 Estimations with Observations Accumulated across Periods

Next, we estimate our baseline Eq. (1) using trade flows accumulated over two, three, and four years. The VAT rebate variable has been averaged over two, three, and four years, respectively. The results are outlined in Table 10.

Table 10 Economy-wide results with observations accumulated across periods
Full size table

As evident, the results closely resemble the baseline, albeit with slightly larger effect magnitudes in absolute terms.

6 Conclusions

The series of analyses yields several key findings: Firstly, there is consistent evidence that Chinese export VAT rebates exert negative effects on external exporters. On one hand, these effects are found to be stronger for poorer countries with lower level of trade liberalization and lower level of export diversification at intensive margin, i.e., the characteristics that constitute a common portrait of a developing country. On the other hand, such common characteristics of the developed countries as exporting higher-quality goods and having stronger revealed comparative advantage in the world trade are also associated with larger negative trade diversion effects of Chinese rebates. Secondly, the adverse effects of China's rebates are predominantly concentrated in the machinery and equipment sector. Lastly, the negative impacts primarily stem from the intensive margin of trade, while exporters appear to use the extensive margin to offset losses incurred from rebates at the intensive margin.

These findings hold significant policy implications. Despite the longstanding debate surrounding the trade diversion effects of export policies, comprehensive empirical evidence has been lacking until now. This study contributes substantially by offering refined, systematic, and wide-ranging empirical insights into the third-country effects of export policies. Given China's prominent role in the global economy, the focus on China's export VAT rebates policy enhances the study's relevance for economic and business practices, as well as economic policy analysis.

Moreover, the study underscores the pivotal role of trade liberalization and export diversification at the intensive margin as robust factors for the sustainable development of trade.