Banca de DEFESA: Fabio Augusto Fujita

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
STUDENT : Fabio Augusto Fujita
DATE: 10/08/2022
TIME: 20:00
LOCAL: a definir
TITLE:

Forecasting Brazilian industrial goods inflation with machine learning methods

KEY WORDS:

 Forecasting, machine learning, industrial goods inflation.

PAGES: 53
BIG AREA: Ciências Sociais Aplicadas
AREA: Economia
SUMMARY:

There is great interest in improving inflation forecasts for better planning and decision making by households, the private sector, and policy makers. However, even outperforming univariate models can be a difficult task. We use machine learning methods and a large data set to forecast industrial goods inflation on Brazilian IPCA for horizons up to t+12, considering the time span between January 2007 and August 2021. We assess the forecasts of four regularized linear methods and two nonlinear tree based methods, with random walk and AR models as benchmarks, in a pseudo out-of-sample framework. We also assess the results without unemployment data as regressors, considering the discussions around the relevance of unemployment data on inflation forecasting. The nonlinear methods outperform the regularized linear methods and the benchmarks. We also find evidence that the variable selection mechanisms of random forest and gradient tree boosting perform better than on linear regularized models to forecast industrial goods inflation. Random forest stands out in terms of forecasting error and as the method that better controls the bias-variance trade-off. It also displays a more uniform performance than gradient tree boosting across the forecasting horizons.

BANKING MEMBERS:
Presidente - 1642911 - DANIEL OLIVEIRA CAJUEIRO
Externo à Instituição - FABIO AUGUSTO REIS GOMES - USP
Externo ao Programa - 1550794 - JOSE GUILHERME DE LARA RESENDE
Interna - 2262223 - MARINA DELMONDES DE CARVALHO ROSSI