A0237
Title: Model averaging factor-augmented quantile regressions with smooth structural change
Authors: Siwei Wang - Hunan University (China) [presenting]
Yundong Tu - Peking University (China)
Abstract: Quantile regression is an effective tool in modelling data with heterogeneous conditional distribution. The time-varying coefficient quantile predictive regressions with factor-augmented predictors are considered to capture smooth structural changes and incorporate high-dimensional data information in prediction simultaneously. Theoretical results are established, including the uniform consistency and the asymptotic normality of the quantile estimators under misspecification. A novel time-varying jackknife model averaging method, which utilizes the local leave-one-out cross-validated weight, is developed to improve the forecast accuracy. The averaging estimator is asymptotically optimal in the sense of out-of-sample final prediction error. Numerical results demonstrate the superior performance of the averaging estimators.
