Abstract: | 這項計劃旨在研究一種時間序列回歸,特點為序列存在一個單位根,誤差服從可能重尾的 GARCH(1,1)模型。 GARCH(1,1) 隨機過程中的正規變化指數(index of regular variation) α 決定兩邊尾部的分布。 文獻中證得,誤差的方差有限,當且僅當 α > 2。 和 Phillips(1987, 1990)一樣,我們的時間序列回歸只有一個滯後量;不同的是,我們的誤差項是 GARCH(1,1) 誤差的線性過程(linear process)。 三種情況會被考慮:(一)0 < α < 2; (二)α = 2; 及(三) α > 2。 (二)是一個邊緣情況,其誤差也稱為無限方差 IGARCH(1,1)。 概言之,這個計 劃從 Chan and Zhang (2010) 的 I(1)-GARCH(1,1)過程擴展到的 I(1)與 GARCH(1,1) 誤差的線性過程。 當 0 < α < 2,估計參數的逼近分布為兩個互相依賴的穩定過程(stable process)的函數(functional),穩定過程的指數分別為 α 及 α/ 2。 當 α ≥ 2,估計參數的逼 近分布為一個標準布朗運動(standard Brownian motions)的函數。 在三種情況下,逼近分 布皆涉及多餘參數(nuisance parameters),而估計(一)的多餘參數是困難甚至不可能的。 鑑 於此,我們修訂 Paparoditis and Politis (2003)提出的殘差為本區塊拔靴法(residual-based block bootstrap,RBB),以此逼近估計參數的逼近分布。 使用 RBB 的另一個優點,是我們只需少 量關於 α 的先有知識。 This project considers the time series regression with a unit root and possibly heavy-tailed GARCH(1,1) errors. The heavy tails are characterized by α > 0, where a is the index of regular variation of the GARCH(1,1) errors. Proven in the literature, the errors have finite variance iff α > 2. Following the lines in Phillips (1987, 1990), we consider a time series regression with only one lag, but here the error term follows a linear process of GARCH(1,1) errors. Three cases are considered: (i) 0 < α < 2; (ii) α = 2; and (iii) α > 2. Case (ii) is a borderline case in which the error process is also known as infinite-variance IGARCH(1,1). All in all, we generalize the results in Chan and Zhang (2010) from an I(1)-GARCH(1,1) process to an I(1) process with a linear process of GARCH(1,1) errors. For 0 < α < 2, the estimated parameter converges in distribution to a functional of two dependent stable processes, one with index α and the other with index α/2. For α ≥ 2, it converges in distribution to a functional of a standard Brownian motion. In all three cases, nuisance parameters are involved and for case (i), the nuisance parameters are difficult, if not impossible, to estimate. In view of this, we approximate the distribution by a method modified upon the residual-based block bootstrap (RBB) suggested by Paparoditis and Politis (2003). An additional merit of using bootstrap is that little prior knowledge on α is required. |