Time-Varying Quantile VAR Models: Estimation and Inference

We study time-varying quantile VAR models in which the conditional quantile does not propagate linearly through the recursion, and develop a toolkit for impulse response analysis, which to the best of our knowledge has not been established elsewhere. For example, we establish Bahadur representation for quantile VAR regression, infer model parameters via smoothed check function, and define and investigate the best linear projection of the time-varying tau-quantile impulse response via local projections. An application to U.S. industrial production growth and credit spreads from 1919-2025 uncovers structural shifts in tail relationships, demonstrating downside asymmetry diminishes at longer forecast horizons.