Modelos Gamlss para fluxos comerciais: uma abordagem alternativa à gravitação clássica

Abstract

The gravity model is frequently used to describe international trade flows, but it faces challenges when dealing with data characterized by heteroskedasticity, asymmetry, and excess zeros. This study analyzes the effectiveness of Generalized Additive Models for Location, Scale, and Shape (GAMLSS) in modeling such flows, compared to conventional estimators such as the Poisson Pseudo-Maximum Likelihood (PPML) and the Gamma Pseudo-Maximum Likelihood (GPML). The empirical application uses data on Brazilian exports of carnauba wax—a product of economic and strategic relevance—considering both economic and institutional factors that influence trade. Models with continuous and zero-inflated distributions were estimated, highlighting the Zero Adjusted Gamma (ZAGA) distribution, which showed the best performance according to information criteria and residual diagnostics. The results indicate that although PPML and GPML are consistent in certain contexts, both displayed limitations when facing high variability and the presence of zeros. In contrast, the GAMLSS (ZAGA) model achieved superior performance, allowing for the simultaneous modeling of the mean, dispersion, and probability of zero trade flows. From a substantive standpoint, the main effects predicted by gravity theory were confirmed, including the positive impact of the importers’ GDP and population, as well as the negative effects of distance, exchange rate, and lack of coastline. It was also found that regional agreements, although not increasing the average export volume, contributed to reducing its variability and the frequency of zero-trade periods, acting as stabilizing instruments for trade relations. The main contribution of this research is to demonstrate how GAMLSS models are robust and flexible tools for analyzing trade flows, expanding the range of methodologies available for studies in international economics.