Please use this identifier to cite or link to this item: https://hdl.handle.net/10419/284870 
Year of Publication: 
2022
Series/Report no.: 
Texto para Discussão No. 2814
Publisher: 
Instituto de Pesquisa Econômica Aplicada (IPEA), Brasília
Abstract (Translated): 
In this Discussion Paper, we test forecasting models for inflation and economic activity with macroeconomic data and economic surveys between January 2002 and October 2019 on a monthly basis. Due to the high dimension nature of the set of explanatory variables, we use machine learning (ML) models that offer different ways to deal with large datasets and we compare with benchmark models. We find that ML methods substantially improve inflation forecasts for shorter horizons (one and three months). While Least Absolute Shrinkage and Selection Operator (Lasso) is the model that best performs for the one-month horizon, a combination of ML models performs better for the three months horizon. However, for longer-term horizons (six and twelve months), individual ML methods and economic surveys do not perform well, despite the fact that a combination of ML models are better than benchmark models. Concerning GDP forecasts, the reverse is true. ML methods do not perform well for the one month horizon, but combinations of ML methods (three and twelve month) and complete subset regression (CSR) (six month) overcame traditional models.
Subjects: 
forecast
econometrics
macroeconomics
machine learning
JEL: 
C01
C53
Persistent Identifier of the first edition: 
Creative Commons License: 
cc-by Logo
Document Type: 
Working Paper

Files in This Item:
File
Size
1.78 MB





Items in EconStor are protected by copyright, with all rights reserved, unless otherwise indicated.