Yaping Wang - Universitat Pompeu Fabra

Abstract

This paper introduces a high-dimensional quantile regression that bridges the dense and sparse modeling perspectives by allowing conditional quantiles to depend densely on latent factors capturing pervasive comovements and sparsely on idiosyn- cratic components reflecting heterogeneous, localized shocks. The resulting frame- work combines the interpretability and variable selection advantages of sparse mod- els with the stability and dimension reduction of factor models. Theoretically, we establish convergence rates for the proposed estimator under weak temporal depen- dence and allow for both strong and weak factors. Simulation studies demonstrate favorable finite-sample performance and highlight a trade-off under weak factors, where the need to retain idiosyncratic components increases as the precision of their estimation deteriorates. In an empirical application to a large macro financial panel, the estimator achieves lower check loss than sparse quantile regression and factor only specifications, with the largest gains in the lower tail.

Additional information:

• Speaker: Yaping Wang
• Time: Thursday, 15.01.2026; 16:00
• Location: Faculty Room  (U 1.040)
• Further links:
  • Speaker's webpage
• Organizer: Statistics Group
• Contact:
  • Almut Lunkenheimer
  • +49 228 73-9228
  • ifs@uni-bonn.de