CMStatistics 2021: Start Registration
View Submission - CFE
Title: High-dimensional generalized least squares Authors:  Kaveh Salehzadeh Nobari - Lancaster University (United Kingdom) [presenting]
Anders Kock - Aarhus University and CREATES (Denmark)
Alex Gibberd - Lancaster University (United Kingdom)
Abstract: The finite sample properties of the GLS-LASSO estimator for high-dimensional regressions with potentially autocorrelated errors are studied. We further assess the performance of a feasible GLS-LASSO estimator and establish non-asymptotic oracle inequalities for estimation accuracy within this framework. We consider settings where the number of parameters is significantly greater than the sample size. We then illustrate the usefulness of the proposed estimators for application in high-dimensional temporal disaggregation, where disaggregation methods are used to disaggregate a time-series using an $n\times p$ indicator series matrix with $p\gg n$. Finally, we conduct a Monte Carlo Study to assess the finite sample performance of the feasible GLS-LASSO estimators in terms of estimation and variable selection accuracy using an array of error structures.