A1719
Title: Cross-validated estimation window averaging for multi-step forecasting
Authors: Jen-Che Liao - National Chengchi University (Taiwan) [presenting]
Abstract: A frequentist model averaging approach is developed to address estimation window uncertainty in multi-step forecast regressions. The proposed data-driven methods for window selection and averaging are based on a leave-h-out cross-validation criterion, where h is the forecast horizon. The procedures are robust to heteroskedasticity and serial correlation of unknown forms, both of which are inherent in multi-step forecast errors. Theoretical properties of the methods are established and their performance is demonstrated through simulations.
