Análise da capacidade preditiva de modelos de regularização aplicado ao IPCA

Abstract

This study analyzes the short-term predictive power of the Shrinkage methods RIDGE and LASSO applied to the IPCA. With increasingly larger and more complex databases available, the use of Machine Learning methodology can be a great solution in situations in which it is not previously known which independent variables should be added to the regression model. The dissertation shows that the result of the Mean Square Error of both methods significantly surpasses the traditional method of Ordinary Least Squares (OLS). In the period analyzed, the LASSO regression obtained even a better result than the benchmarking in the Focus report, released by the Central Bank.