A1420
Title: Model averaging predictor in an infinite-order autoregressive process
Authors: Hsin-Chieh Wong - National Taipei University (Taiwan) [presenting]
Abstract: Least-squares forecast combination is investigated for a stationary infinite-order autoregressive process under same-realization prediction. One-step-ahead predictors fitted from a growing collection of finite-order AR models are averaged, allowing both the candidate orders and the size of the model family to increase with the sample size. An explicit asymptotic expression for the mean-squared prediction error (MSPE) of the least-squares averaging predictor in the increasing-order setting is obtained. The main technical advance is a uniform convergence theorem establishing that the MSPE approximation holds uniformly across the candidate models, providing a theoretical basis for validating model averaging criteria.
