其他分享
首页 > 其他分享> > ACVF of ARMA(1, 1)

ACVF of ARMA(1, 1)

作者:互联网

\(ARMA(1, ~ 1)\) process is a time series \(\left\{ X_{t} \right\}\) defined as:

\[X_{t} - \phi X_{t-1} = Z_{t} + \theta Z_{t-1} \]

where \(|\phi| < 1\) and \(\left\{ Z_{t} \right\} \sim WN(0, ~ \sigma^{2})\)。


它的 ACVF (autocovariance function) 可以通过改写为 linear process 的形式的方法求出,其中 linear process 定义为一个 time series \(\left\{ X_{t} \right\}\) which can be written as:

\[X_{t} = \sum\limits^{\infty}_{j = -\infty} \varphi_{j} Z_{t - j} \]

其中对于\(\forall j \in \mathbb{Z}\),系数 \(\varphi_{j}\) 为常数,并且\(\left\{ Z_{t} \right\} \sim WN(0, ~ \sigma^{2})\)。


对于上述的一个 linear process,它的 ACVF 为:

\[\gamma (h) = \sigma^2 \sum\limits^{\infty}_{j = -\infty} \varphi_{j}\varphi_{j+h} \]

这是因为:

\[\begin{align*} \gamma(h) & = Cov(X_{t+h}, ~ X_{t})\\ & = Cov(\sum\limits^{\infty}_{j = -\infty} \varphi_{j} Z_{t + h - j}, ~ \sum\limits^{\infty}_{j = -\infty} \varphi_{j} Z_{t - j}) \\ & = Cov(\sum\limits^{\infty}_{i = -\infty} \varphi_{i} Z_{t + h - i}, ~ \sum\limits^{\infty}_{j = -\infty} \varphi_{j} Z_{t - j}) \end{align*} \]

由于 \(Z_{t} \sim WN(0, ~ \sigma^2)\),那么:

\[Cov(Z_{t}, ~ Z_{t}) = Var(Z_{t}) = \sigma^2 \]

并且对于 \(\forall s \neq t\):

\[Cov(Z_{s}, ~ Z_{t}) = 0 \]

因此观察 \(Cov(\sum\limits^{\infty}_{i = -\infty} \varphi_{i} Z_{t + h - i}, ~ \sum\limits^{\infty}_{j = -\infty} \varphi_{j} Z_{t - j})\),将它逐项展开后,当且仅当 \(Z_{t+h-i}\) 和 \(Z_{t-j}\) 为同一个随机变量时,有:

\[Cov(Z_{t+h-i}, ~ Z_{t-j}) = \sigma^{2} \]

否则:

\[Cov(Z_{t+h-i}, ~ Z_{t-j}) = 0 \]

也就是说,当且仅当下标满足 \(t + h - i = t - j \implies i = j + h\) 时,展开后的该项不为 \(0\)。那么不为零的各项前面的系数应是 \(\varphi_{j}\) 和 \(\varphi_{j+h}\),且值为 \(\sigma^{2}\),即:

\[\gamma (h) = \sigma^2 \sum\limits^{\infty}_{j = -\infty} \varphi_{j}\varphi_{j+h} \]


推导如下:

\(ARMA(1, 1)\) 的 LHS:

\[\begin{align*} X_{t} - \phi X_{t-1} & = X_{t} - \phi B X_{t}\\ & = (1 - \phi B) X_{t} \end{align*} \]

其中 \(B\) 为 backward shift operater,那么 \(ARMA(1, ~ 1)\) process 可以继续做如下变换:

\[\begin{align*} X_{t} & = \frac{1}{1 - \phi B} (Z_{t} + \theta Z_{t-1})\\ & = (1 + \phi B + \phi^{2} B^{2} + \phi^{3} B^{3} + \cdots) (Z_{t} + \theta Z_{t-1})\\ & = (Z_{t} + \phi B Z_{t} + \phi^{2} B^{2} Z_{t} + \phi^{3} B^{3} Z_{t} + \cdots) + (\theta Z_{t-1} + \theta \phi B Z_{t-1} + \theta \phi^{2} B^{2} Z_{t-1} + \theta \phi^{3} B^{3} Z_{t-1} +\cdots)\\ & = (Z_{t} + \phi Z_{t-1} + \phi^{2} Z_{t-2} + \phi^{3}Z_{t-3} + \cdots) + (\theta Z_{t-1} + \theta \phi Z_{t-2} + \theta \phi^{2} Z_{t-3} + \cdots)\\ & = Z_{t} + (\phi + \theta) Z_{t-1} + \phi (\phi + \theta) Z_{t-2} + \phi^{2} (\phi + \theta) Z_{t-3} + \cdots \end{align*} \]

因此它可以写作 linear process 的形式:

\[X_{t} = \sum\limits^{\infty}_{j = 0} \varphi_{j} Z_{t - j} \]

其中,\(\varphi_{0} = 1\),\(\varphi_{j} = \phi^{j-1}(\phi + \theta)\) for \(j \geq 1\)。

因此它的 ACVF 为:

标签:ACVF,infty,phi,limits,sum,varphi,theta,ARMA
来源: https://www.cnblogs.com/chetianjian/p/16436386.html