Finance-Insurance
Time: 11.00 am
Date: 13th of March 2024
Room 3001

Valentina CORRADI (University of Surrey)”Sparsity Tests for High-Dimensional Linear Regression Models in Time Series”

Abstract :Penalised Regression methods and in particular the Least Absolute Shrinkage and Selection Operator (LASSO) have become an integral part of modern-day time series analysis. As the performance of LASSO crucially hinges on the assumption of sparsity, which is unknown in practice, we propose a Hausman type test to assess it. The null hypothesis of our test is that there are at most k0 relevant regressors with non-zero coefficients. A key distinction between our test and existing methods, such as Information Criteria, is that we allow the number of regressors to be (much) larger than the sample size. We further propose a modified version of the test that employs critical values from a moving block bootstrap procedure based on an asymptotic linear representation of the (desparsified) LASSO estimator. We provide two key applications for our test: in the first one, the test helps to assess the validity of standard Gaussian inference in out-of-sample forecast comparisons with LASSO, which is inherently related to the presence of sparsity through parameter estimation error. In the second application, we consider pretesting the validity of Gaussian inference on Impulse Response Functions in high-dimensional (Structural) Vector Autoregressions.

Organizers:  Zakoian Jean-Michel