MA(1)的参数估计——公式推导

Problem

---

设$MA(1)$模型为：$y_t={\theta }_0+{\theta }_1{ϵ}_{t-1}+{ϵ}_t$，其中${ϵ}_t\stackrel{i.i.d.}{\sim }N(0,{\sigma }_{ϵ}^2)$

极大似然估计

---

假设观测数据集为 { y 1 , … , y T } \{y_1,\ldots,y_T\} {y1​,…,yT​}，令$\theta =({\theta }_0,{\theta }_1,{\sigma }_{ϵ}^2{)}^{\prime }$，$y=(y_1,\dots ,y_T{)}^{\prime }$，
我们有$(y_1,\dots ,y_T)\sim N_T(\mu ,\Sigma )$

在精确似然估计中我们假设$(y_1,\dots ,y_T)$服从多元正态分布

$\mu =({\theta }_0,\dots ,{\theta }_0{)}^{\prime }$

$E{y}_t=E{\theta }_0+E{\theta }_1{ϵ}_{t-1}+E{ϵ}_t={\theta }_0$

$$\Sigma =\left(\begin{array}{ccccc}Cov(y_1,y_1)& Cov(y_1,y_2)& Cov(y_1,y_3)& \dots & Cov(y_1,y_T)\\ Cov(y_2,y_1)& Cov(y_2,y_2)& Cov(y_2,y_3)& \dots & Cov(y_2,y_T)\\ Cov(y_3,y_1)& Cov(y_3,y_2)& Cov(y_3,y_3)& \dots & Cov(y_3,y_T)\\ ⋮& ⋮& ⋮& \ddots & ⋮\\ Cov(y_T,y_1)& Cov(y_T,y_2)& Cov(y_T,y_3)& \dots & Cov(y_T,y_T)\end{array}\right)=\left(\begin{array}{ccccc}(1+{\theta }_1^2){\sigma }_{ϵ}^2& {\theta }_1{\sigma }_{ϵ}^2& 0& \dots & 0\\ {\theta }_1{\sigma }_{ϵ}^2& (1+{\theta }_1^2){\sigma }_{ϵ}^2& {\theta }_1{\sigma }_{ϵ}^2& \dots & 0\\ 0& {\theta }_1{\sigma }_{ϵ}^2& (1+{\theta }_1){\sigma }_{ϵ}^2& \dots & 0\\ ⋮& ⋮& ⋮& \ddots & ⋮\\ 0& 0& 0& \dots & {\theta }_1{\sigma }_{ϵ}^2\\ 0& 0& 0& \dots & (1+{\theta }_1){\sigma }_{ϵ}^2\end{array}\right)$$

$Cov(y_t,y_t)=Var(y_t)=Var({\theta }_0+{\theta }_1{ϵ}_{t-1}+{ϵ}_t)=(1+{\theta }_1){\sigma }_{ϵ}^2$
$Cov(y_t,y_{t-1})=Cov({\theta }_0+{\theta }_1{ϵ}_{t-1}+{ϵ}_t,{\theta }_0+{\theta }_1{ϵ}_{t-2}+{ϵ}_{t-1})=Cov({\theta }_1{ϵ}_{t-1},{ϵ}_{t-1})={\theta }_1{\sigma }_{ϵ}^2$
$Cov(y_t,y_{t-k})=0(k>1)$

那么，似然函数便可得到：
$$\begin{array}{rl}L(\theta )& =f_{\theta }(y_1,\dots ,y_T)\\ & =\frac{1}{\sqrt{(2\pi {)}^T|\Sigma |}}exp\left(-\frac{1}{2}(y-\mu {)}^{\prime }{\Sigma }^{-1}(y-\mu )\right)\end{array}$$

条件极大似然估计

---

给定${ϵ}_0$的情况下，可以迭代地计算出${ϵ}_1,\dots ,{ϵ}_{T-1}$，因此：
$$\begin{array}{rl}L(\theta )& =f_{\theta }(y_1,\dots ,y_T|{ϵ}_0)\\ & =f_{\theta }(y_1,\dots ,y_T|{ϵ}_{T-1},\dots ,{ϵ}_0)\\ & =\prod _{t=2}^T{f}_{\theta }(y_t|y_{t-1},\dots ,y_0,{ϵ}_{T-1},\dots ,{ϵ}_0)f_{\theta }(y_1|{ϵ}_{T-1},\dots ,{ϵ}_0)\\ & =\prod _{t=2}^T{f}_{\theta }(y_t|{ϵ}_{t-1})f_{\theta }(y_1|{ϵ}_0)=\prod _{t=1}^T{f}_{\theta }(y_t|{ϵ}_{t-1})\end{array}$$其中，$y_t|{ϵ}_{t-1}\sim N({\theta }_0+{\theta }_1{ϵ}_{t-1},{\sigma }_{ϵ}^2)$

$E(y_t|{ϵ}_{t-1})={\theta }_0+{\theta }_1{ϵ}_{t-1}+E{ϵ}_t={\theta }_0+{\theta }_1{ϵ}_{t-1}$
$Var(y_t|{ϵ}_{t-1})=Var({ϵ}_t)={\sigma }_{ϵ}^2$

因此，似然函数可以写为：
$$L(\theta )=\prod _{t=1}^T\frac{1}{\sqrt{2\pi {\sigma }_{ϵ}^2}}exp(-\frac{(y_t-{\theta }_0-{\theta }_1{ϵ}_{t-1}{)}^2}{2{\sigma }_{ϵ}^2})$$