Time-varying Lasso

George Kapetanios; Filip Zikes

doi:10.1016/j.econlet.2018.04.029

Time-varying Lasso

George Kapetanios, Filip Zikes

King's Business School

Research output: Contribution to journal › Article › peer-review

9 Citations (Scopus)

338 Downloads (Pure)

Abstract

This paper introduces a Lasso-type estimator for large linear models with time-varying parameters. The estimator is easy to implement in practice and standard algorithms developed for Lasso with fixed parameters can be readily used. We derive theoretical properties of the estimator, allowing for deterministic or stochastic smoothly varying parameter processes and discuss ways in which tuning parameters can be data dependent. Monte Carlo simulation and an application to forecasting inflation with macroeconomic variables illustrates the usefulness of our method.

Original language	English
Journal	ECONOMICS LETTERS
Early online date	3 May 2018
DOIs	https://doi.org/10.1016/j.econlet.2018.04.029
Publication status	E-pub ahead of print - 3 May 2018

Keywords

Large datasets
Structural change
Penalised regressions
Lasso

Access to Document

10.1016/j.econlet.2018.04.029

Time-varying Lasso_KAPETANIOS_Accepted24April2018_GREEN AAM (CC BY-NC-ND)
© <2018> This manuscript version is made available under the CC-BY-NC-ND 4.0 license https://creativecommons.org/licenses/by-nc-nd/4.0/legalcode
Accepted author manuscript, 492 KBLicence: CC BY-NC-ND

Cite this

@article{09ae7a4020cd44c688044323e5058cd3,

title = "Time-varying Lasso",

abstract = "This paper introduces a Lasso-type estimator for large linear models with time-varying parameters. The estimator is easy to implement in practice and standard algorithms developed for Lasso with fixed parameters can be readily used. We derive theoretical properties of the estimator, allowing for deterministic or stochastic smoothly varying parameter processes and discuss ways in which tuning parameters can be data dependent. Monte Carlo simulation and an application to forecasting inflation with macroeconomic variables illustrates the usefulness of our method.",

keywords = "Large datasets, Structural change, Penalised regressions, Lasso",

author = "George Kapetanios and Filip Zikes",

year = "2018",

month = may,

day = "3",

doi = "10.1016/j.econlet.2018.04.029",

language = "English",

journal = "ECONOMICS LETTERS",

issn = "0165-1765",

publisher = "Elsevier",

}

TY - JOUR

T1 - Time-varying Lasso

AU - Kapetanios, George

AU - Zikes, Filip

PY - 2018/5/3

Y1 - 2018/5/3

N2 - This paper introduces a Lasso-type estimator for large linear models with time-varying parameters. The estimator is easy to implement in practice and standard algorithms developed for Lasso with fixed parameters can be readily used. We derive theoretical properties of the estimator, allowing for deterministic or stochastic smoothly varying parameter processes and discuss ways in which tuning parameters can be data dependent. Monte Carlo simulation and an application to forecasting inflation with macroeconomic variables illustrates the usefulness of our method.

AB - This paper introduces a Lasso-type estimator for large linear models with time-varying parameters. The estimator is easy to implement in practice and standard algorithms developed for Lasso with fixed parameters can be readily used. We derive theoretical properties of the estimator, allowing for deterministic or stochastic smoothly varying parameter processes and discuss ways in which tuning parameters can be data dependent. Monte Carlo simulation and an application to forecasting inflation with macroeconomic variables illustrates the usefulness of our method.

KW - Large datasets

KW - Structural change

KW - Penalised regressions

KW - Lasso

U2 - 10.1016/j.econlet.2018.04.029

DO - 10.1016/j.econlet.2018.04.029

M3 - Article

SN - 0165-1765

JO - ECONOMICS LETTERS

JF - ECONOMICS LETTERS

ER -

Time-varying Lasso

Abstract

Keywords

Access to Document

Fingerprint

Cite this