2
$\begingroup$

In Kim et al. (1998) stochastic volatility model is specified as:

$y_t=\beta\exp({\frac{h_t}{2}})\varepsilon_t,\quad t\geqslant1$

$h_{t+1}=\mu+\phi(h_t-\mu)+\sigma_\eta\eta_t$

$h_1\sim N(\mu,\frac{\sigma^2}{1-\phi^2})$

and $\varepsilon_t, \eta_t$ are independent standard normal.

Later on they write (page 365) that it can be shown that

$f(h_t|h_{\backslash t},\theta)=f(h_t|h_{t+1}, h_{t-1}, \theta)=f_N(h_t|h^*_t, \nu^2)$

where $h^*_t=\mu+\frac{\phi( (h_{t-1}-\mu)+(h_{t+1}-\mu) )}{1+\phi^2},\quad\nu^2=\frac{\sigma^2_\eta}{1+\phi^2}$

Also, $\theta=(\mu,\sigma^2,\phi)$, $f_N$ - normal pdf, $h_{\backslash t}$ - values of the volatility vector without the value at $t$

I do not understand how they derive this $f_N(h_t|h^*_t, \nu^2)$. Somehow they play around with AR(1) equations for $h_{t+1}$ and $h_t$, but how? They plug what into what?

$\endgroup$
  • $\begingroup$ Let me know if you are not able to follow the answer below and I can make it a little more detailed. $\endgroup$ – tchakravarty Oct 19 '14 at 18:21
3
$\begingroup$

Some fairly simple algebra will get you there.

From the forward equation, you have: $$ \begin{align} h_{t+1} &= \mu + \phi(h_t - \mu) + \sigma_\eta \eta_t \\ \phi^2(h_t - \mu) &= \phi(h_{t+1} - \mu) - \phi\sigma_\eta \eta_t \\ \end{align} $$

Similarly, from the backward equation, you have: $$ \begin{align} h_t - \mu &= \phi(h_{t-1} - \mu) + \sigma_\eta \eta_{t-1} \\ \end{align} $$

Adding the two expressions together gives $$ \begin{align} (1+\phi^2)(h_t - \mu) &= \phi((h_{t-1} - \mu) + (h_{t+1} - \mu)) + \sigma_\eta (\eta_{t-1}- \phi \eta_t) \\ (h_t - \mu) &= \frac{\phi}{1+\phi^2}((h_{t-1} - \mu) + (h_{t+1} - \mu)) + \frac{\sigma_\eta}{1+\phi^2} (\eta_{t-1}- \phi \eta_t) \\ h_t &= \mu + \frac{\phi}{1+\phi^2}((h_{t-1} - \mu) + (h_{t+1} - \mu)) + \frac{\sigma_\eta}{1+\phi^2} (\eta_{t-1}- \phi \eta_t) \end{align} $$

It is easy to see that if the $\{\eta_t\}$ are standard normal, then $h_t$ is (conditionally on $h_{t-1}$ and $h_{t+1}$) normally distributed, and $$ \begin{align} \mathbb{E}(h_t\mid h_{t-1}, h_{t+1}) &= \mu + \frac{\phi}{1+\phi^2}((h_{t-1} - \mu) + (h_{t+1} - \mu)) \\ &\equiv h_t^* \\ \mathbb{V}(h_t\mid h_{t-1}, h_{t+1}) &= \frac{\sigma_\eta^2}{1+\phi^2}\\ &\equiv \nu^2 \end{align} $$

$\endgroup$

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.