Consider the processes $X_t = \phi X_{t-1} + v_t$ and $Y_t = \phi Y_{t-1} + X_t + e_t$, in which $|\phi| < 1$ and $v_1$ and $e_t$ are non-correlated random errors with zero mean and variances equal to $\sigma^2$. Based on these informations, find the autocovariance function of the process $Y_t$.

I already know that if $|\phi| < 1$ than $Y_t$ should follow a stationary process and it’s autocovariance is:

$\begin{equation} E[Y_{t+h}Y_t] - \mu^2 \end{equation}$

but I do not know how to go further


Note that all the processes have zero mean, i.e. $E[Y_t]=E[X_t]=E[v_t]=E[\epsilon_t]=0$.

Let $h>0$: $$\begin{align}E[Y_{t+h}Y_t]&=E[(\phi Y_{t+h-1}+X_{t+h}+\epsilon_{t+h})Y_t]=\phi E[Y_{t+h-1}Y_t]+E[X_{t+h}Y_t]+E[\epsilon_{t+h}Y_t]\\&=\phi r_y(h-1)+\phi^h r_{xy}(0)+E[\epsilon_{t+h}]E[Y_t]=\phi r_y(h-1)+\phi^h r_{xy}(0)\end{align}$$

Note that: $r_{xy}(h)=E[X_{t+h}Y_t]=E[\phi^hX_tY_t+(\sum_{j=0}^{h-1} v_{t+h-j}\phi^j)Y_t]=\phi^h r_{xy}(0)$

For $h=0$: $$\begin{align}r_y(0)&=E[Y_t^2]=E[(\phi Y_{t-1}+X_t+\epsilon_t)^2]\\&=\phi ^2r_y(0)+E[X_t^2]+E[\epsilon_t^2]+2\phi \underbrace{E[Y_{t-1}X_t]}_{r_{xy}(1)}+2E[X_t\epsilon_t]+2E[Y_{t-1}\epsilon_t]\\&=\phi^2r_y(0)+r_x(0)+r_e(0)+2\phi^2 r_{xy}(0)\rightarrow r_y(0)=\frac{r_x(0)+\sigma^2+2\phi^2 r_{xy}(0)}{1-\phi^2}\end{align}$$

Similarly, we can find that $$r_x(0)=\frac{\sigma^2}{1-\phi^2}, \ \ r_x(h)=\phi r_x(h-1)$$

It just remains to find $r_{xy}(0)$: $$\begin{align}r_{xy}(0)&=E[X_tY_t]=[(\phi X_{t-1}+v_t)(\phi Y_{t-1}+X_t+\epsilon_t)]\\&=\phi^2E[X_{t-1}Y_{t-1}]+\phi r_x(1)+E[X_tv_t]\\&=\phi^2r_{xy}(0)+\phi r_x(1)+\sigma^2\end{align}$$

  • $\begingroup$ Why did you say that $E[X_{t+h}Y_t] = E[X_{t+h}][Y_t]$ and $E[\epsilon_{t+h} Y_t] = E[\epsilon_{t+h}] E[Y_t]$? $\endgroup$ – Tullio Oct 14 '19 at 20:39
  • $\begingroup$ $\epsilon_{t+h}$ doesn't depend on $\epsilon_t$, but $X_{t+h}$ has a separate recursion, therefore $E[X_{t+h}Y_t]\neq E[X_{t+h}]E[Y_t]$, which was a mistake; now corrected. it was a fast deploy, and I'll review the algebra again. The solution implies joint stationarity by the way. $\endgroup$ – gunes Oct 15 '19 at 11:01
  • $\begingroup$ why $\phi^h r_{xy}(0) = 0$? $\endgroup$ – Tullio Oct 15 '19 at 11:25
  • $\begingroup$ It's not. Where do you see that? $\endgroup$ – gunes Oct 15 '19 at 11:30
  • $\begingroup$ For $\phi r_y(h-1)+\phi^h r_{xy}(0)+E[\epsilon_{t+h}]E[Y_t]=\phi r_y(h-1)$ then $\phi^h r_{xy}(0)$ must be zero. $\endgroup$ – Tullio Oct 15 '19 at 11:55

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.