Title: High-dimensional curve estimation in time-varying models
Authors: Stefan Richter - Heidelberg University (Germany) [presenting]
Jonas Krampe - University of Mannheim (Germany)
Jens-Peter Kreiss - Technische Universitaet Braunschweig (Germany)
Efstathios Paparoditis - University of Cyprus (Cyprus)
Abstract: Curve estimation for locally stationary processes is considered. We allow the parameter curves describing the non-stationarity to be high-dimensional. Under usual sparsity assumptions we derive concentration inequalities for a likelihood-based lasso estimator. Furthermore, we propose a desparsified lasso approach which allows for a Gaussian limit distribution and hypotheses testing via a bootstrap procedure. The finite-sample behavior of the estimation procedure is analyzed in simulations with tvVAR and tvARCH processes.