本計畫提出新檢定方法以比較不同模型預測績效的穩定性。不同於Giacomini and Rossi (2010) 根據歷史資料探討兩模型樣本外預測表現的穩定性，本計畫以Leisch et al. (2000) 一般化波動檢定為基礎，利用前瞻(forward-looking) 方法偵測不同模型樣本外預測表現是否因參數變動而發生改變，並推導檢定統計量的極限性質。為了避免因長期變異數或參數估計不確定造成過高的型一偏誤(size distortion)，本計畫亦參考Kiefer and Vogelsang (2005)與Ibragimov and Müller (2010) 的方法，建構頑強(robust) 樣本外預測能力的穩定性檢定。由於檢定統計量極限分配的boundary functions 可能不具有明確形式(closed form)，本計畫將利用模擬方法求導其臨界值，並分析檢定統計量的小樣本表現。為了驗證本計畫方法的有效性，我們將比較不同匯率基準(fundamental) 模型與random walk 模型預測績效，評估樣本外預測能力是否發生改變。 In this project I propose new generalized tests to compare the out-of-sample forecasting performance of two competing models in the presence of possible instabilities. In contrast with Giacomini and Rossi (2010), I use the forward-looking methods to monitor out-of-sample predictive ability and to see if a model's forecasting ability may disappear over time. Following Leisch et al. (2000), I develop the maximum and range of recursive- and moving-estimates tests based on general conditional loss functions. Because of sampling variability of HAC estimators and parameter estimation uncertainty, existing tests usually suffer from the size distortion problem. In this project I will also use different HAC estimators, such as Kiefer and Vogelsang (2005) and Ibragimov and Müller (2010), in the proposed tests to provide more robust inference on evaluating out-of-sample forecasting performance. The asymptotic distributions of the proposed tests will be established in the project. Because it may not have closed-form result for the crossing probability of the asymptotic distributions, I will simulate the appropriate critical values for the corresponding boundary functions. Some simulations will be conducted to show the finite-sample properties (size and power) of the proposed methods. To illustrate the usefulness of the new techniques in analyzing out-of-sample predictive ability, an empirical application for foreign exchange rate forecasting models will be presented. 研究期間：10008 ~ 10107
