Hierarchical Regularizers for Reverse Unrestricted Mixed Data Sampling Regressions

Alain Hecq; Marie Ternes; Ines Wilms

Hierarchical Regularizers for Reverse Unrestricted Mixed Data Sampling Regressions

^*Corresponding author for this work

Research output: Working paper / Preprint › Preprint

Abstract

Reverse Unrestricted MIxed DAta Sampling (RU-MIDAS) regressions are used to model high-frequency responses by means of low-frequency variables. However, due to the periodic structure of RU-MIDAS regressions, the dimensionality grows quickly if the frequency mismatch between the high- and low-frequency variables is large. Additionally the number of high-frequency observations available for estimation decreases. We propose to counteract this reduction in sample size by pooling the high-frequency coefficients and further reduce the dimensionality through a sparsity-inducing convex regularizer that accounts for the temporal ordering among the different lags. To this end, the regularizer prioritizes the inclusion of lagged coefficients according to the recency of the information they contain. We demonstrate the proposed method on an empirical application for daily realized volatility forecasting where we explore whether modeling high-frequency volatility data in terms of low-frequency macroeconomic data pays off.

Original language	English
Publisher	Cornell University - arXiv
Number of pages	23
Publication status	Published - 2023

Publication series

Series	arXiv.org
Number	arXiv:2301.10592
ISSN	2331-8422

Keywords

mixed-frequency models
MIDAS MODELS
forecasting
group lasso

Access to Document

https://arxiv.org/abs/2301.10592

Cite this

@techreport{1032e5128ce44f4cb675f2efd1d134c7,

title = "Hierarchical Regularizers for Reverse Unrestricted Mixed Data Sampling Regressions",

abstract = "Reverse Unrestricted MIxed DAta Sampling (RU-MIDAS) regressions are used to model high-frequency responses by means of low-frequency variables. However, due to the periodic structure of RU-MIDAS regressions, the dimensionality grows quickly if the frequency mismatch between the high- and low-frequency variables is large. Additionally the number of high-frequency observations available for estimation decreases. We propose to counteract this reduction in sample size by pooling the high-frequency coefficients and further reduce the dimensionality through a sparsity-inducing convex regularizer that accounts for the temporal ordering among the different lags. To this end, the regularizer prioritizes the inclusion of lagged coefficients according to the recency of the information they contain. We demonstrate the proposed method on an empirical application for daily realized volatility forecasting where we explore whether modeling high-frequency volatility data in terms of low-frequency macroeconomic data pays off.",

keywords = "mixed-frequency models, MIDAS MODELS, forecasting, group lasso",

author = "Alain Hecq and Marie Ternes and Ines Wilms",

note = "FRED-MD dataset: https://research.stlouisfed.org/econ/mccracken/fred-databases/. The Economic Policy Uncertainty Index: https://www.policyuncertainty.com. The Oxford–Man Institute of Quantitative Finance",

year = "2023",

language = "English",

series = "arXiv.org",

number = " arXiv:2301.10592",

publisher = "Cornell University - arXiv",

address = "United States",

type = "WorkingPaper",

institution = "Cornell University - arXiv",

}

TY - UNPB

T1 - Hierarchical Regularizers for Reverse Unrestricted Mixed Data Sampling Regressions

AU - Hecq, Alain

AU - Ternes, Marie

AU - Wilms, Ines

N1 - FRED-MD dataset: https://research.stlouisfed.org/econ/mccracken/fred-databases/. The Economic Policy Uncertainty Index: https://www.policyuncertainty.com. The Oxford–Man Institute of Quantitative Finance

PY - 2023

Y1 - 2023

N2 - Reverse Unrestricted MIxed DAta Sampling (RU-MIDAS) regressions are used to model high-frequency responses by means of low-frequency variables. However, due to the periodic structure of RU-MIDAS regressions, the dimensionality grows quickly if the frequency mismatch between the high- and low-frequency variables is large. Additionally the number of high-frequency observations available for estimation decreases. We propose to counteract this reduction in sample size by pooling the high-frequency coefficients and further reduce the dimensionality through a sparsity-inducing convex regularizer that accounts for the temporal ordering among the different lags. To this end, the regularizer prioritizes the inclusion of lagged coefficients according to the recency of the information they contain. We demonstrate the proposed method on an empirical application for daily realized volatility forecasting where we explore whether modeling high-frequency volatility data in terms of low-frequency macroeconomic data pays off.

AB - Reverse Unrestricted MIxed DAta Sampling (RU-MIDAS) regressions are used to model high-frequency responses by means of low-frequency variables. However, due to the periodic structure of RU-MIDAS regressions, the dimensionality grows quickly if the frequency mismatch between the high- and low-frequency variables is large. Additionally the number of high-frequency observations available for estimation decreases. We propose to counteract this reduction in sample size by pooling the high-frequency coefficients and further reduce the dimensionality through a sparsity-inducing convex regularizer that accounts for the temporal ordering among the different lags. To this end, the regularizer prioritizes the inclusion of lagged coefficients according to the recency of the information they contain. We demonstrate the proposed method on an empirical application for daily realized volatility forecasting where we explore whether modeling high-frequency volatility data in terms of low-frequency macroeconomic data pays off.

KW - mixed-frequency models

KW - MIDAS MODELS

KW - forecasting

KW - group lasso

M3 - Preprint

T3 - arXiv.org

BT - Hierarchical Regularizers for Reverse Unrestricted Mixed Data Sampling Regressions

PB - Cornell University - arXiv

ER -