A0802
Title: Bootstrap inference for VAR models under rank uncertainty
Authors: Stephan Smeekes - Maastricht University (Netherlands) [presenting]
Lenard Lieb - Maastricht University (Netherlands)
Abstract: In impulse response analysis using VAR models it is common to construct confidence intervals using bootstrap techniques. However, in many practical applications, uncertainty regarding the true (unknown) cointegration rank is typically ignored, and impulse responses and their confidence intervals are constructed as if the cointegration rank were known, whether the rank has been estimated or simply assumed to be equal to a certain rank. Recently many methods have been proposed to robustify impulse responses to an unknown rank or order of integration. These studies however do not consider how the bootstrap is affected when constructing confidence intervals. We therefore investigate how bootstrap inference for VAR models, such as used for impulse response analysis, is affected by a misspecified cointegration rank. We derive theoretical results on the asymptotic validity of the bootstrap in this setting, and analyze finite sample effects through Monte Carlo simulation. This allows us to quantify how serious the problems are for empirical work if the uncertainty regarding the rank is ignored. We also consider modifications of the bootstrap that provide better guard against misspecification of the cointegration rank. Finally our results are illustrated with an empirical example.
