ARCH(1)的参数估计——公式推导

ARCH(1)

$y_t={ϵ}_t$

其中，${ϵ}_t={\eta }_t\sqrt{h_t}$，$h_t={\alpha }_0+{\alpha }_1{ϵ}_{t-1}^2$
${\eta }_t\stackrel{i.i.d.}{\sim }N(0,1)$

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$\theta =({\alpha }_0,{\alpha }_1{)}^{\prime }$
$$\begin{array}{rl}L(\theta )& =f_{\theta }(y_1,\dots ,y_T)\\ & =\prod _{t=2}^T{f}_{\theta }(y_t|y_{t-1},\dots ,y_1)f_{\theta }(y_1)\\ & =\prod _{t=2}^T{f}_{\theta }(y_t|y_{t-1})f_{\theta }(y_1)\\ & =\prod _{t=2}^T{f}_{\theta }(y_t|{ϵ}_{t-1})f_{\theta }(y_1)\end{array}$$

$y_1\sim N(0,\frac{{\alpha }_0}{1-{\alpha }_1})$

由$y_t={ϵ}_t$可知，$y_t$服从正态分布，且$E{y}_t=0$

$Var(y_t)=E{y}_t^2=E{\eta }_t^2({\alpha }_0+{\alpha }_1{ϵ}_{t-1}^2)=E{\eta }_t^2({\alpha }_0+{\alpha }_1{y}_{t-1}^2)={\alpha }_0+{\alpha }_{1}E{y}_{t-1}^2$
由$y_t$平稳可知二阶矩不随时间变化，即$E{y}_t^2=E{y}_{t-1}^2$，故有
$Var(y_t)=\frac{{\alpha }_0}{1-{\alpha }_1}$

$y_t|y_{t-1}\sim N(0,{\alpha }_0+{\alpha }_1{y}_{t-1}^2)$

$E{y}_t|y_{t-1}=E{y}_t|{ϵ}_{t-1}=E{\eta }_t\sqrt{{\alpha }_0+{\alpha }_1{ϵ}_{t-1}^2}|{ϵ}_{t-1}=E{\eta }_t|{ϵ}_{t-1}E\sqrt{{\alpha }_0+{\alpha }_1{ϵ}_{t-1}^2}|{ϵ}_{t-1}=0$

$Var(y_t|y_{t-1})=({\alpha }_0+{\alpha }_1{y}_{t-1}^2)Var({\eta }_t)={\alpha }_0+{\alpha }_1{y}_{t-1}^2$

综上可得：
$$L(\theta )=\prod _{t=2}^T\frac{1}{\sqrt{2\pi ({\alpha }_0+{\alpha }_1{y}_{t-1}^2)}}exp\left(-\frac{y_t^2}{2({\alpha }_0+{\alpha }_1{y}_{t-1^2})}\right)\frac{1}{\sqrt{2\pi {\alpha }_0/(1-{\alpha }_1)}}exp\left(-\frac{y_1^2}{2{\alpha }_0/(1-{\alpha }_1)}\right)$$

对$-lnL(\theta )$采用梯度下降求解得参数估计

条件似然估计

在给定$y_0$的条件下，似然函数为：
$$\begin{array}{rl}L(\theta )& =f_{\theta }(y_1,\dots ,y_T|y_0)\\ & =\prod _{t=1}^T{f}_{\theta }(y_t|y_{t-1})\\ & =\prod _{t=1}^T\frac{1}{\sqrt{2\pi ({\alpha }_0+{\alpha }_1{y}_{t-1}^2)}}exp\left(-\frac{y_t^2}{2({\alpha }_0+{\alpha }_1{y}_{t-1}^2)}\right)\end{array}$$

$y_t|y_{t-1}\sim N(0,{\alpha }_0+{\alpha }_1{y}_{t-1}^2)$

同样对$-lnL(\theta )$利用梯度下降求解得参数估计。