2.4.1 Data
We use a worldwide database11 over the period 1948-2012 with more than 1.3 million observations. This database includes the traditional variables in a gravity model like GDP, GDP per capita, geographical distance, shared language, contiguity, former colonisers, trade agreements12. Our dependent variable is the value in million dol-lars of bilateral exports (Xijt) constructed by DOTS (IMF) and complemented by COMTRADE.
2.4.2 Empirical specifications and estimation issues
To investigate the effects of South-South through the BRIC countries on the geo-graphical diversification of intra-SSA exports, we apply a structural gravity model from three complementary approaches. First of all, the gravity equation shows the positive and proportional effect of economic size of trading partners on bilateral trade but also the reverse effect of trade costs for example geographical distance.
More specifically, the gravity model mainly admits two sets of explanatory vari-ables: (i) monadic or unilateral variables which vary or not in time for origin and destination countries; (ii) dyadic or bilateral variables constants or which vary in time. For several decades, there have been a many theoretical foundations about the gravity model. Indeed, the contribution of Tinbergen (1962) has been
progres-11See the website of Keith Head: http://strategy.sauder.ubc.ca/head/sup/
12For more details about the data origin see the appendix of Head et al. (2010): http://
strategy.sauder.ubc.ca/head/Papers/erosion.pdf.
sively improved by integrating more and more microeconomic elements and the new international trade theory framework13.
Equation 1 shows the economic intuition derived from Newton’s law with the pos-itive and proportional effect of economic size of trading partners (Yi,Ej) on bilateral trade (Xij), the reverse effect of trade costs (τij) like geographical distance and trade costs across the other export and import markets through relative price effects (Pi, Pj). Indeed, based on Armington hypotheses about specialisation, identical constant elasticity of substitution (σ) and the theoretical-consistent background developed by Anderson and van Wincoop (2003), we sum up this theoretical gravity equation:
Xij = From a structural gravity model, we account for multilateral resistance demon-strated by Anderson and van Wincoop (2003) through fixed effects14 (country-year and country-pair). This empirical method is the most used by the literature (Head and Mayer, 2014) and also solves the endogeneity issue. One of the recurrent issues in the gravity model is the presence of zero trade flows where the PPML estimator is the most appropriated compared to the other standard methods (Santos Silva and Tenreyro, 2006 ; Santos Silva and Tenreyro, 2011 ; Head and Mayer, 2014 ; Fally, 2015). We divide our approach into three parts with robust standard errors clustered by country-pair.
First, the dependent variable is based on the value of bilateral exports (Equations 2 and 3). We wish to empirically estimate whether the BRICs exports to SSA affect intra-SSA trade flows. The variable of interest equals 1 if i and j are SSA trading
13See Head and Mayer (2014) for have more details and references.
14In order to capture other trade costs across the other export and import markets through relative price effects.
partners for each year since Brazil, Russia, India and China export to i countries, 0 otherwise.
Second, we decide to estimate the marginal effect of South-South trade on geo-graphical extensive margin through a LPM. The dependent variable is a binary vari-able for strictly positive export flows which takes 1 in this case, 0 otherwise (Head et al., 2010), and with the same variable of interest described above (Equations 4 and 5). More precisely, we see the likelihood of realising positive trade flows with SSA countries when these are a destination of the BRIC exports. In other words, we rely and extend the economic intuition of Amighini and Sanfilippo (2014) where African imports from developing countries enhance the diversification of African bilateral trade flows. We use LPM, which is implemented by robust ordinary least squares (OLS) with fixed effects instead of probit and logit as advised by Angrist and Pischke (2009). An increase in the probability of strictly positive export flows signifies a rise of new trading partners applied to intra-SSA exports due to the BRIC exports to SSA.
Third, we assess the impact of South-South trade on the geographical extensive margin of trade by using PPML with country-pair fixed effects. The dependent variable is the total number of countries to which a given country has nonzero exports for each year (Shepherd, 2010 ; Lavall´ee and Lochard, 2015) and with the same variable of interest (Equation 6). This method introduced by Shepherd (2010) allows us to better investigate the geographical export diversification through the number of export markets and in our case for intra-SSA trade.
Accordingly, our different specifications of gravity equation are as follows:
lnXijt=β0+β1lnMi(j)t+β2Dijt+β2SSTijt+γij +λt+ijt (2.2)
lnXijt=β0+β1Dijt+β2Fij+β2SSTijt+γit+γjt+ijt (2.3) xijt =β0 +β1lnMi(j)t+β2Dijt+β2SSTijt+γij +λt+ijt (2.4) xijt =β0+β1Dijt+β2Fij +β2SSTijt+γit+γjt+ijt (2.5) nb xijt=β0+β1lnMi(j)t+β2Dijt+β2SSTijt+γij +λt+ijt (2.6) whereXijt is country i exports to j at yeart, xijt is a binary variable taking the value 1 for strictly positive export flows (Xijt>0), 0 otherwise. nb xijtis the number of countries to which the exporting country has strictly positive export flows (Xijt
> 0). Fij represents time-fixed dyadic variables like geographical distance, shared language, contiguity. Mi(j)t regroups time-varying monadic variables like GDP and population. Dijt are time-varying dyadic variables like FTAs, GATT-WTO. Our variable of interest is SSTijt as previously described. Equations 3 and 5 include country-year fixed-effects (γi(j)t). Equations 2, 4 and 6 have country-pair fixed effects (γij) and time dummies (λt) as advised by Baldwin and Taglioni (2006) and Head and Mayer (2014).