Análise sobre a relação do retorno acionário com variáveis macroeconômicas utilizando o LASSO

Abstract

In front of the evidences in the literature that, in general, macroeconomic variables can influence shareholder returns, many studies emerge trying to demonstrate this relationship by applying different methodologies. In this sense, the Least Absolute Shrinkage and Selection Operator (LASSO) presents itself as a differentiated methodology that selects relevant variables in a set of possible predictors, allowing the irrelevant predictors to have their coefficients zero (shrink toward 0), collaborating for more parsimonious and simple interpretation models. This work aims to apply the LASSO methodology of variables selection to relate the Brazilian stock returns with macroeconomic variables of the country. The database used includes the return of the Ibovespa index and a further six stock exchanges of the financial sector (ITUB4 and BBAS3), consumer sector (ABEV3 and BRFS3) and basic materials sector (PETR4 and VALE5). For possible predictors, 15 macroeconomic variables were chosen in the literature prior to this study. The variable with the highest probability of inclusion was "Prêmio de Risco", contained in all regressions. Regarding the prediction, the Average Quadratic Forecast Error was calculated for each of the equities and the result indicated a quality of adjustment of the regressions because they presented a lower value than other methodologies that used the same database, indicating a superiority of LASSO for forecasting.