Website CV Paper
Fields
Econometrics, Macroeconomics, Finance, Health
Job market paper
Time-Varying Jackknife Model Averaging for Quantile Regression
We propose a time-varying jackknife model averaging (TV-JMA) estimator for quantile regression that allows model averaging weights to adapt smoothly over time via local kernel estimation. A centering correction to the cross-validation criterion eliminates the bias from the non-zero conditional mean of the quantile regression error, enabling MSE-based weight selection. We establish convergence rates, asymptotic normality with diverging model dimensions, and asymptotic optimality of the selected weights under a scale-bias orthogonality condition that is automatically satisfied for median regression, pure location models, and heteroskedastic models with appropriately specified candidates. Monte Carlo simulations show that TV-JMA dominates constant-weight and constant-coefficient alternatives, with the gains increasing at tail quantiles and larger sample sizes. An empirical application to conditional quantile forecasting of excess stock returns and GDP growth-at-risk demonstrates the practical gains.