relative to the average barrier of the two regions with all trade partners. If a country has a relatively high average trade barrier, it will trade more with a country with which it has a low bilateral barrier. Anderson and van Wincoop argue that multilateral resistance cannot be measured using remoteness variables based on measures of distance as this does not capture border effects, rather the gravity model must be solved by taking into account the impact of barriers on prices. 10
The importance of Anderson and van Wincoops’ (2003) contribution is acknowledged in the literature. However, as Feenstra (2004) and others note, it has not been widely adopted in empirical research given the difficulties in its implementation (a customised programme is needed as the endogenous nature of the price terms requires a non-linear solution). Feenstra (2004) shows that the inclusion of country specific fixed effects generates the same results as Anderson and van Winccop (2003) with little loss of efficiency.11 A limitation of this approach is that it does not allow for the multilateral resistance (price) effects to be calculated explicitly. An alternative solution proposed by Baier and Bergstrand (2005a) is to use a Taylor series expansion to approximate for the price effect terms. The results are consistent with Anderson and van Wincoop (2003) and, unlike the fixed effects approach, this method allows for the multilateral resistance terms to be solved explicitly.
Recognising the nature of trading flows between countries as relationships that develop and change over time has resulted in an increasing use of panel (longitudinal) data approaches to the estimation of gravity models. This method is chosen in this paper. Anderson and van Wincoop (2004, p29) note that “improved econometric techniques based on careful consideration of the error structure are likely to pay off.” The use of different panel data methods, such as random or fixed (within) effects estimators, allows for various assumptions regarding trade flows to be analysed and tested. In particular, as is discussed in a later section, in panel data analysis of gravity models possible heterogeneity and endogeneity issues can be examined by isolating the effects of country pair effects (factors that influence trade between two countries).
3.2 Existing Studies Applying the Gravity Model to Services Trade
The existing literature on the application of the gravity model to services trade is quite limited. One of the first papers on the subject is Francois (2001), with the methodology further developed in Francois et al. (2003). Francois models the demand for imports of services as a function of the recipient country’s GDP per capita and population. Data on services trade flows are taken from the Global Trade Analysis Project (GTAP) database.12 The gravity equation is estimated using OLS and the resulting levels of predicted trade between countries are compared to actual trade flows to calculate tariff equivalents of the barrier to services using a constant elasticity import demand function. Francois’s estimated tariff equivalents have been widely employed in other studies.
10 Ferrantino (2006, p25) refers to this as taking into account the endogenous nature of prices in a general equilibrium context.
11 The inclusion of country dummy variables for importers and exporters is now widely employed, including research by van Wincoop (e.g., Rose and van Wincoop, 2001). Another option discussed but rejected by Feenstra (2004) is to include price indexes in the gravity equation to model the multilateral resistance terms.
This data is based on IMF Balance of Payments statistics.