Alocação da Poupança das Famílias Brasileiras: Microdados da POF 2017–2018 e Script em R para Modelagem Composicional com Regressão Dirichlet

This technical note presents a proposal for analyzing the allocation of Brazilian households’ savings across different categories of assets, grouped by their degree of liquidity, based on microdata from the Household Budget Survey (POF 2017–2018). The proportions of savings allocated to financial assets, durable goods, and pension/real estate holdings are modeled using Dirichlet regression — a method well-suited for handling compositional data. Although the ultimate goal is the application of a statistical model, the study stands out for the detailed and careful construction of an analytical dataset composed of homogeneous families (couples with two children), as well as for the development of a rigorous and replicable methodology for identifying and quantifying the net variation in assets, based on the various datasets of the POF 2017–2018. The note documents all stages of data processing, along with the theoretical and technical decisions embedded in the R script, enabling the generation of results for different occupational profiles — with a particular focus on the savings behavior of private sector employees. The proposal aims to contribute both to the methodological advancement in measuring wealth formation and to the dissemination of best practices in the use of microdata and the R programming language.