The relationship between the Chinese ‘going out’ strategy and international trade

This study is the first to estimate a system of simultaneous gravity equations for Chinese exports, imports and foreign direct investment (FDI) using a sample of 167 countries over the period 2003–2012. The main results indicate that trade and outward FDI are complementary. In particular, the authors show that outward Chinese FDI is related to higher exports and imports and that China trades more with countries hosting Chinese FDI. Results are also robust to the use of instrumental variables. Therefore, Chinese investment seems to foster trade. (Published in Special Issue FDI and multinational corporations) JEL F14 F21 F59


Introduction
The main results support both hypotheses. More specifically, we find that China exports more to destinations in which it is active in FDI and that higher FDI stocks are associated with increases in trade. In particular, an increase of 10 percent in FDI stocks increases exports (imports) by about 2.1 percent (1.1 percent).
The rest of the paper is structured as follows. Section 2 presents a review of the closely related literature. Section 3 describes the data and presents some stylized facts. Section 4 specifies the model, shows and discusses the main results and presents some robustness checks. Finally, Section 5 concludes.

Trade and FDI in the gravity model
Gravity models have been considered the workhorse of international trade in the recent decades and are a widely accepted empirical tool (Head and Mayer, 2014). These models have also been used to estimate the determinants of bilateral FDI and some authors estimate FDI and trade models simultaneously. In particular, Brouwer et al. (2008) estimated gravity models of trade and FDI separately for a sample of 28 European countries over the period 1990 to 2004 and find a positive and significant correlation between bilateral FDI and bilateral trade, when FDI is included as explanatory variable in the gravity model of trade. However, the authors do not tackle the problems related to missing data in FDI (around 50 percent), endogeneity of the FDI variable or reverse causality. In contrast to these authors, Egger (2001) estimated a system of simultaneous equations for trade and FDI using intra-EU bilateral flows from 1988 to 1996, allowing for the endogeneity of both exports and FDI variables in the system. He finds that, in line with the theoretical models of Helpman (1984) and Markusen and Maskus (1999), bilateral exports are an increasing function of outward FDI stocks. However, the effect is only statistically significant in the long run. Chen et al. (2012) analyzed the relationship between outward FDI and exports of 15 Taiwanese manufacturing industries over the period from 1991 to 2007. The main results, obtained using random and fixed effects estimators, show the existence of complementarity between FDI and exports. Most of the abovementioned studies use lagged FDI values to control for the endogeneity of FDI in the trade equation, whereas lagged exports are used in the FDI equation. The reverse causality issue is also considered in Cheung and Qian (2009) who analyze the effect of Chinese exports as a determinant of Chinese outward FDI, also using the lagged value of exports to mitigate the endogeneity problem. They find that this relationship is positive and gets stronger when the host economies are developing countries. Also focusing on China, Caporale et al. (2015) analyzed China's trade with North America, Asia and Europe and its relationship with inward FDI. They found a positive relationship, stronger for the period after China joined the WTO. Their main concern is the endogeneity due to time-invariant variables, but they fail to account for the reverse causality problem that could arise by the inclusion of FDI in this setting. We differ from this study given that we focus on outward FDI and how it correlates with exports, and imports, plus employing econometric methods that aim to consider the correlation of the determinants of the different variables, and simultaneity issues. Moreover, we include all available countries for which there is data, regardless of the continent they belong to. A second paper focused on China's trade is Yang and Martinez-Zarzoso (2014), which assessed the effect of the ASEAN-China trade agreement on sectoral trade. The authors found mainly net trade creation effects, but did not consider FDI as a control variable in their gravity model focusing exclusively on trade flows. Some recent studies use firm level data to investigate the relationship between FDI and trade in Africa. In particular, Broadman (2007) using firm level data of the World Bank Africa Asia Trade Investment (WBAATI) survey and the World Bank's newly developed business case studies of Chinese firms in Africa, find that there are positive links between FDI and trade among Chinese firms involved in Africa. In particular, the attraction of investment for infrastructure and related services development seems to create "spillovers" on the continent. Moreover, intangible assets, such as technology transfer and transfer of managerial skills, which usually accompany FDI, also act as vehicles stimulating trade. Similar evidence is shown in Chen and Tang (2014). Applying propensity score matching techniques to compare firms that have similar characteristics ex-ante, the authors show that Chinese firms engaged in outward FDI export 0.6 log points more than firms that do not invest abroad. These results show that horizontal FDI from China complements firms' trade, consistent with the idea that exporting entails high fixed costs and that FDI helps reduce those fixed costs.

Data and stylized facts
We use bilateral FDI data from UNCTAD (2015), trade data from COMTRADE (2015) and gravity variables, namely distance between the capital cities (lnDist), colonial relationship (Colony), 2 common legal origin (Comleg), 3 and common language that is spoken by at least 9% of the population (Comlang) 4 from CEPII. Gross Domestic Product (GDP) 5 and population are from the World Development Indicators (2015), while the regional trade agreement (RTA) dummy is from De Sousa (2012) 6 . The bilateral investment treaty dummy variable (BIT) is created with information obtained from UNCTAD (2015). We use BIT ratification instead of BIT signature since the relevant date is the one in which the agreement enters into force; the same applies for the RTA variable. The sample includes 167 partner countries (see Table A1 in the Appendix) and cover the years from 2003 to 2012. Summary statistics for all the variables included in the analysis are shown in Table 1. Graphical inspection of the data shows that Chinese exports are significantly higher in destinations where China is also engaged in FDI ( Figure 1) and Chinese outward FDI is positively correlated with Chinese exports (Figure 2), the same applies to imports (Figures 3 and  4).

Model specification
We estimate a system of seemingly-unrelated gravity equations in which FDI, exports and imports are the endogenous variables and enter with one lag as explanatory variables. The model is specified as follows: = 0 + 1 ln (max {1, −1 }) + 2 NFDI −1 + α 3 ln −1 + 4 ln + 5 ln + 6 ln + 7 + 8 + 9 + 10 + ∑ −1 =1 + ∑ −1 =1 + (1) = 0 + 1 ln (max {1, −1 }) + 2 NFDI −1 + β 3 ln −1 + 4 ln + 5 ln + 6 ln + 7 + 8 + 9 + 10 ln (max {1, }) = 0 + 1 −1 + 2 ln −1 + 3 ln + 4 ln + 5 ln + 6 + 7 + 8 + 9 where j denotes the partner country and t the year. , and are time dummies, while , and are regional dummies. Regional dummies account for multilateral resistance factors and the time dummies account for common trends in Chinese exports, imports and FDI. Given the existence of zeros in the FDI variable, 7 we follow Martinez-Zarzoso et al. (2017) and Wagner (2003) and create a dummy to account for the absence of FDI and another variable to measure the impact of the level of the observed FDI. The effect of FDI is then specified in the following way: The amount of zero values for outward FDI is 23% for exports and 24% for imports. Thus, 1 ( 1 ) measures the elasticity where FDI is positive and 2 ( 2 ) modifies the constant term when FDI is zero. FDI j,t-1 denotes the lagged value of outward Chinese FDI stock in country j and period t-1 and NFDI is a dummy variable that takes the value of one when the FDI stock is zero in country j and time t.

Main results
The main results using equations (1)-(3) are shown in Table 2. Column 1 reports the results for exports (eq.1), column 2 for imports (eq.2) and column 3 for FDI (eq. 3). We observe a positive and statistically significant effect of FDI on exports and imports. For instance, increasing FDI to a host country by 10 percent, increases Chinese exports by 2.14 percent and imports by 1.12 percent (column 1 and 2, Table 2). We can use this elasticity to calculate how much export (imports) should increase per dollar of FDI according to our results. Each dollar of additional FDI yields on average an additional USD 4.63 of exports (USD 2.09 of imports). 8 The results in column 3 indicate that increases in the volume of exports and imports also foster Chinese FDI outflows significantly.
Concerning the no-FDI dummy (NFDI), the coefficient, which is -0.214 in column 1 (-0.397 in column 2), should be interpreted as follows. Logged exports (imports) when FDI is positive exceed logged exports (imports) when FDI is zero by 0.214*lnFDI+0.214 (0.112*ln FDI+0.397). In Figure 5 and Figure 6 we can observe the "excess" of log exports or imports, for the amount invested, compared to a scenario of no investment. For smaller amounts of FDI, the presence of FDI generates higher "excess returns" for imports than for exports, but the situation is the opposite for investments above 6 millions of USD.
As regards the control variables, the coefficient of the GDP of China's trading partners is positive and statistically significant, in the export and import equations (columns 1 and 2) as the gravity model predicts. The population coefficient is positive for Chinese exports and FDI and negative for imports, indicating that the size of the destination market is associated to higher exports and more FDI but with a reduction of Chinese imports. Among the time-invariant gravity variables, colony, sharing a common legal origin and sharing a common language show the expected positive effect on imports and FDI, but the effect in not always statistically significant, e.g. in the export equation for the former two variables. It is interesting to notice that Chinese exports are explained mainly by market size variables (GDP and population) and common language of ethnic groups, while imports are more sensitive to historical and cultural links (colonial relationship and sharing a common legal origin) and regional trade agreements. Distance has an unexpected positive effect on imports, which indicates that imports are higher from far away destinations, perhaps indicating that the continental dummies do not fully capture multilateral resistance factors. 9 _________________________ 8 These figures are obtained using the sample means of exports, imports and FDI (i.e. (Mean exports/Mean FDI)*α 1 ).
9 In auxiliary regressions that include interactions of the distance variable with the continental dummies, results show that the positive distance effect obtained in the import regression is mainly driven by the interactions with the Africa and Latin America dummy variables.  Concerning the RTA dummy, it indicates that entering into trade agreements promotes Chinese imports in the period considered. Finally, the BIT dummy in equation (3) has a positive and significant coefficient indicating that China invests around 52 percent more in host countries with whom it has signed a BIT than in non-signatory countries.
In Table 3 we can see the correlation matrix of the residuals. Since the null hypothesis of no correlation among the residuals is rejected, the SUR methodology improves the estimation over the Ordinary Least Squares.
Summarizing, the main results show that both hypotheses are confirmed. China exports less to (import less from) destinations with zero FDI, and an increase of 10 percent in FDI stocks increases exports by 2.1 percent (imports by 1.1 percent).

Robustness
As a robustness test, we estimate independently each equation (outward FDI, exports and imports) and instrument the two variables that are potentially endogenous (as instruments in each specification we use the corresponding variable lagged two periods). The Hansen test statistics (see last row of  Finally, we estimated the model using a between estimator 11 and the main results are shown in Table 5. Some authors argue that the estimated elasticities could be interpreted as long-run _________________________ 11 We have estimated three independent regressions with country fixed effects. Results are included in the appendix (Table A2) and show that the coefficients are imprecisely estimated, most likely due to the lack of sufficient withincountry variation over time. For this reason, and also to be able to estimate the coefficients of the variables that vary by destination/origin, we refrain from using this as the main specification. The main limitation of using continental  (3) are the results of running a between estimator and columns (4) to (6) from running a SUR regressions but using time-averages of the variables of interest. The default for the continental dummies is Africa.

_________________________
fixed effects could be that we are not accounting for all the unobserved heterogeneity that is time invariant and country specific, which could be correlated with the error term in the estimated model. The main results are robust in Table A3, where we have estimated a feasible generalized least squares model with continental dummies and country RE, allowing for panel specific autocorrelation.
effects (Stern, 2010). Basically, the model is estimated for the time-averages of the variables and it does not make a priori assumptions concerning the nature of the time effects. Hence, in real world data situations, with this estimator we are likely to obtain estimates that are robust to misspecification of dynamics. The first part of Table 5 shows the results obtained when the three equations (for exports imports and FDI) are considered independent, whereas the second part shows the results when we allow for unrestricted correlation between the error terms of the three equations (BE-SUR). The main results indicate that higher Chinese FDI induces higher exports from China in the long-run (column 1, Table 5), and the same can be said with respect with higher exports inducing higher outward FDI (column 3, Table 5). However, the coefficient of lagged imports (exports) is not statistically significant in column 1 (column 2), but turns significant and positive when accounting for the correlation across error terms in column 4 (column 5). Results in column 6 also confirm that higher imports (exports) from China attract more FDI from the same country (column 4) in the long-run. The estimated coefficients are higher in magnitude when the BE-SUR estimator is used, indicating a downward bias in the estimations shown in the first part of Table 5.

Conclusions
In the 2000s, China has been actively investing abroad, becoming the third largest investor in the world. Many have challenged the benefits of the Chinese investments in the local economies. For instance, Adisu et al. (2010) find that Chinese investments have negatively impacted internal trade. However, other authors highlight also some benefits as for example increasing trade and investment in a continent that was systematically marginalized in the past from international flows of goods and capital (Zafar, 2007). In this paper, using a system of seemingly unrelated gravity equations for exports, imports and FDI we show that FDI appears to be complementary to Chinese exports and imports. These results are also robust to an instrumental variable approach. Chinese FDI -despite being correlated to higher imports from China -is also associated to higher exports to China.
The findings are of relevance to the wider FDI literature in that some evidence of the complementarity between FDI and trade is found for the Chinese case. This supports the general view that FDI and trade are mutually reinforcing channels and also sources of economic development and prosperity.
Future work entails the analysis of different product groups to investigate the potential heterogeneity of the relationships and to extend the analysis to trade in services.

Additional Material -Data sources
(retrieved in 2015, therefore some discrepancies might arise due to updates in the data if downloaded as of now) -Exports and imports were retrieved from https://comtrade.un.org/ (as reported by China).
Reporter country is China.
-Gross Domestic Product (GDP, in current US$) and population were compiled from the World Development Indicators, retrieved from http://datatopics.worldbank.org/world-developmentindicators/.
-Distance between capital cities, and dummies that take the value of 1 if the two countries where ever in colonial relationship, have a common legal origin or if a language is spoken by at least 9% of the population in both countries were retrieved from: http://www.cepii.fr/CEPII/en/bdd_modele/bdd_modele.asp, especially from the GeoDist dataset -Regional Trade Agreement's data was obtained from: http://jdesousa.univ.free.fr/data.htm#RegionalTradeAgreements -Bilateral Investment Treaties was obtained from: https://investmentpolicyhub.unctad.org/IIA/CountryBits/42#iiaInnerMenu Tables 2 and 3 Results in these tables were obtained using the Stata command "sureg", options "small dfk cor". Table 4 Results in this table were obtained using the (user written) Stata command "ivreg2", options "first robust". Table 5 Results in this table were obtained using the Stata command "xtreg", options "be vce(jack)" (first three columns) and using the Stata command "sureg", options "small dfk cor" for the last three columns.
Appendix Table A2 Results in this table were obtained using the Stata command "xtreg", options "fe clust(countries)". Here "countries" denote the partner countries of China.
Appendix Table A3 Results in this table were obtained using the Stata command "xtgls", options "panels(h) corr(psar1) force".