context. The HTM is increasingly applied in gravity models of trade in goods, but this is the first time it has been applied to services trade.21
The variables chosen as exogenous (the same variables are used in the model for total services and four disaggregated service categories) are the year dummies, which are the only time variant exogenous variables, distance, common language and adjacency. These variables are used to instrument the remaining variables, which are assumed to be endogenous. In addition, time invariant variables can be estimated using the HTM, giving it an advantage over the FEM.
To test the appropriateness of the HTM, the Hausman-Taylor over-identification test is applied to the FEM and HTM specifications. The test statistic of 1.39 is less than the critical chi-squared value with six degrees of freedom at 1 percent significance, so the null hypothesis that the unobserved effects are correlated with other regressors is not rejected: HTM is more efficient. Testing of the different specifications appears to confirm the findings of Egger (2002, 2005), that the HTM is the most appropriate estimator for the gravity equation of trade in goods, also holds for trade in services. This result holds for total services and the four sub-sectors. Consequently, the discussion of the results for these four sectors will focus on those obtained using the HTM estimator.
Prior to the interpretation of the results, it is useful to examine the changes in estimated coefficients between the pooled OLS and HTM. The differences are due to the bias in the OLS estimator (Egger, 2005). The OLS estimate of importer GDP per capita is likely to be biased downward by unobserved effects, such as restrictions on trade, that have been omitted from the model but are negatively correlated with imports. The OLS estimator overstates the importance of exporter GDP per capita due the correlation of this variable with omitted variables such as the technology and capital stock of the country. Similarly, the significance of EU membership in the OLS model is likely due to the high levels of factor endowments in EU members. While distance is negatively correlated with imports in the OLS model, using the HTM to control for unobserved characteristics shows it to be statistically insignificant. Egger (2005) suggests poor institutional standards of many peripheral economies as a possible source of the negative correlation.
Total Service Imports
As shown in table 3, when the gravity model is estimated using pooled OLS, all variables are significant and their coefficients take the signs that would be expected from the standard gravity literature. GDP per capita of the importer and exporter, common language, adjacency and EU membership all positively influence trade, as do the populations of the respective countries indicating that larger countries produce more services commodities for export and demand more services imports. Distance is the only variable that has a negative influence on trade.
In the HTM estimation (IV in table 3), the only variables that remain statistically significant are GDP per capita of both the importing and the exporting countries and the common language dummy. As discussed above, the HTM is the most appropriate choice
Egger and Pfaffermayr (2004) apply a HTM to FDI flows.