Modelagem Bayesiana dos níveis máximos do Índice de Preços ao Consumidor

Abstract

The Brazilian Consumer Price Index (CPI) is the index used by the Central Bank of Brazil in establishing its inflation targets. By serving as an inflation reference, the Brazilian CPI is closely monitored by foreign and Brazilian investors as well as by public managers. It is known that high and uncontrolled inflation causes distortions and economic losses in the country, so there is an interest on the part of managers and financial managers to predict maximum inflation for a certain period of time. Thus, the objective of the work was to model the maximum Brazilian CPI levels, which can occur in a four-month period. The choice of four-month periods aims to equate the analysis with the intervals between the presentations of the statements of compliance with the fiscal targets by the government. The Generalized Extreme Values (GEV) distribution was used for modeling. For the estimation of the parameters of the GEV distribution the maximum likelihood method and the Bayesian Inference were used. In the elicitation of information for the construction of the prior distributions, we used data from countries economically similar to Brazil: Russia, China and India, which belong to BRICS. In addition, different combinations of prior distribution were created, using information from these countries with different variance structures. In order to evaluate the best estimation methodology, the accuracy and precision of the estimates of the maximum inflation levels for certain return times were analyzed. The results showed that the Bayesian approach, which used as information the mean data of the BRICS countries for construction of the Normal Trivariate prior distribution, led to more accurate and accurate predictions.