過濾自體抽樣預測績效檢定量
A Sieve Bootstrap Test for Predictive Ability
國立暨南大學經濟學系(Department of Economics, National Chi Nan University);國立高雄應用科技大企業管理學系(Department of Business Administration, National Kaohsiung University of Applied Science);國立政治大學國際貿易學系(Department of International Trade, National Cheng-Chi University)
- 卷期
- 34卷3期
- 出版
- 2006年09月
- 頁碼
- 317–334
- 編號
- NTUEJ00006880
摘要
預測準確性與預測涵蓋性檢定同是評估預測績效的有效方法。欉清全等(2005)提出一種新的檢定程序,建構出AR-tc統計量並導出其漸近分配。該程序雖然計算容易,但在樣本外預測個數有限時,隨預測期間的增長,AR-tc存在型一誤差扭曲的問題。本文提出一種自體抽樣程序用以估計AR-tc統計量的小樣本分配,冀以改善型一誤差扭曲的問題。我們提供該程序正確性的證明。模擬結果的確顯示,該程序可以有效改善統計量的錯誤拒絕比率,同時具有不錯的檢力表現。
Abstract
The easy-to-implement AR-tc tests for forecast accuracy and encompassing proposed by Tsong, Lee and Kuo (2005) are subject to considerable size distortions in a finite sample. The problem tends to increase as the forecast horizon increases. A sieve bootstrap scheme for the tests is thus suggested and found by simulation to have excellent control over the size under the null. The validity of the bootstrap tests is justified in an asymptotic sense.