Predição de risco soberano a partir de indicadores de desenvolvimento do Banco Mundial

Abstract

The sovereign classi cation is an indicator that seeks to express the risk to which foreign investors are subjected when acquiring securities of some country. Being issued by rating agencies, companies independent of governments or private companies. Risk agencies are being widely criticized for their lack of transparency in classi cation processes. In this context, the prediction of rating through machine learning models is shown as an option for the purpose of simulation of the grades assigned by the agencies. This paper proposes two studies, one involving the prediction of sovereign risk from macroeconomic fundamentals through the algorithm Random Forest, and another aiming to analyze some hypotheses through a regression model, more precisely to answer if the nancial crisis of 2008 resulted in a structural break in the sovereign risk assessments, and whether risk agencies have di erent ratings for countries with distinct degrees of economic development.The results show accuracy up to 98.28% in the prediction problem, but also through the "p-value"test, from a statistical point of view, there was a structural break in assessments after the 2008 nancial crisis, and that agencies have come to evaluate countries with developed economies in di erent ways. These results indicate that the inability of the agencies to predict the crisis has led to a change in the methodology of evaluations.