Chronopoulos, Ι., " Uniform Inference for Penalised High-Dimensional Linear Time-Series Models"

Title: "Uniform Inference for Penalised High-Dimensional Linear Time-Series Models"

Speaker: Assistant Professor Ilias Chronopoulos, University of Essex

Host:  Assistant Professor Alexopoulos Angelos, Department of Economics, Athens University of Economics and Business

Room:  76, Patission Str., Antoniadou Wing, 3rd floor, Room A36

Abstract: This paper develops a statistical inference procedure for random designed high-dimensional linear models, allowing their regressors and residuals to exhibit heteroskedasticity, dependence, and nonstationarity. To this end,  we introduce a debiased elastic net estimator that does not rely on node-wise regressions to estimate the precision matrix of regressors, thereby avoiding structural assumptions on regressors. A dependent wild bootstrap procedure is developed to construct simultaneous confidence intervals and perform hypothesis tests for  diverging numbers of linear combinations of parameters. From a theoretical perspective, we derive a Gaussian approximation theorem for the proposed estimator and demonstrate the consistency of the bootstrap algorithm. Simulations and empirical applications confirm the theoretical properties and demonstrate improved coverage relative to existing high-dimensional inference methods, including those based on nodewise-Lasso debiasing, when the structural assumptions of regressors fail.