Evidence beyond specifications is unstable. The results of the loglinary version of the gravitational equation with OLS (including variable fixed effects of exporters and importers and fixed effects of country pairs at variable times) indicate that reciprocal and non-reciprocal trade agreements have had an economic and statistically significant impact, regardless of the direction of trade flows. However, this evaluator may give distorted results because he does not address the problems related to heterostitity and zeros. As soon as we discuss these issues with the PPML evaluator, the situation is very different. We find the apparently implausible result that no economic integration agreement has had an impact on export flows. Given these differences and according to other authors, we choose as our preferred specification a PPML estimate that uses export shares as a dependent variable and in which default errors are rewarded several times. According to this specification, only mutual agreements had a positive effect on trade flows between developed and developing countries for the entire sampling period, and only if the exporter is the developing country. For data up to 2008, the NRPTA had a positive impact on exports from beneficiary countries to industrialized countries that have disappeared in recent years. Finally, for exports from industrialized countries to beneficiary countries, our preferred specification does not provide any evidence of a significant effect.
In summary, the results of our preferred specification increasingly support the argument of critics of non-reciprocal preferential regimes that developing countries should abandon their dependence on unilateral trade preferences in favour of mutual agreements. In addition, it is important to address the economic problems posed by both heterosticity and the existence of zero values in bilateral trade flows. It allows us to avoid other potential sources of distortions in the estimate that could be present in the loglinary specification. This document benefits from the recent computer development implemented by  and which makes it possible to execute the PPML evaluator, including the three types of high-dimensional solid effects needed to obtain unbiased and theoretical estimates. Column 1 of Table 3 shows the estimates of severity Eq (2) with PPML. To save space, we report under each coefficient the two types of default errors: grouped by country pairs (in parentheses) and default errors grouped multi-page (in parentheses). Comparing these results with those in column 1 of Table 1, we can see that switching from OLS to PPML results in a loss of statistical significance of all variables. At this point, it should be noted that the loss of meaning of the estimated coefficients of the PTAs and GATT variables, when we move from OLS to PPML, corresponds to the result of . In addition, as before , these authors also present insignificant results with regard to the commercial effect of the euro on the use of PPML, but not in the use of OLS (including high-dimensional fixed effects in both cases). . . .