Aarhus University Seal

Penalized Time Series Regression

Research output: Contribution to book/anthology/report/proceedingBook chapterResearchpeer-review

  • Anders Bredahl Kock, University of Oxford
  • ,
  • Marcelo Medeiros, Pontifical Catholic University of Rio de Janeiro
  • ,
  • Gabriel Vasconcelos, Pontifical Catholic University of Rio de Janeiro

This chapter covers penalized regression in the framework of linear time series models and reviews the most commonly used penalized estimators in applied work, namely Ridge Regression, the Least Absolute Shrinkage and Selection Operator (Lasso), the Elastic Net, the adaptive versions of the Lasso as well as Elastic Net and the group Lasso. Other penalties are briefly presented. We discuss theoretical properties such as consistent variable selection, the oracle property, and oracle inequalities and list time series models in which penalized estimators have been shown to possess these. Potentially problematic aspects of (some of) these properties are also discussed. Practical issues, such as the selection of the penalty parameters and available computer implementations, are also covered. A Monte Carlo simulation is presented in order to compare different penalties in terms of estimation precision, model selection capability, and forecasting performance. Finally, an application to forecasting US monthly inflation is presented.

Original languageEnglish
Title of host publicationMacroeconomic Forecasting in the Era of Big Data
EditorsP. Fuleky
Number of pages36
Place of publicationCham
Publication year2020
ISBN (print)978-3-030-31149-0
ISBN (Electronic)978-3-030-31150-6
Publication statusPublished - 2020
SeriesAdvanced Studies in Theoretical and Applied Econometrics

See relations at Aarhus University Citationformats

ID: 187170273