2
$\begingroup$

Let us start from AR(1)-GARCH(1,1) model,

$r_t=\phi_0+\phi_1r_{t-1}+a_t$

$a_t=\epsilon_t\sigma_t$

$\sigma_t^2=\alpha_0+\alpha_1a_{t-1}^2+\beta_1\sigma_{t-1}^2$

where {$\epsilon_t$} is Gaussian white noise series with mean 0 and variance 1.

Assume that we can observe the returns $r_1,r_2,...,r_T$, then how to derive the conditional likelihood? Many books have no details about it.

The first question: I think the parameters we need to estimate include $\phi_0,\phi_1,\alpha_0,\alpha_1,\beta_1$. The conditional distribution of $r_t$ given $r_1,r_2,...,r_{t-1}$ is $Normal(\phi_0+\phi_1r_{t-1},\sigma_t^2)$. By multiplying them together we can estimate $\phi_0,\phi_1$ and calculate all $\sigma_t$ and $a_t$. But how should we go a step further to estimate $\alpha_0,\alpha_1,\beta_1$ by MLE. Given the observation up to time $t-1$, $\sigma_t$ is already measurable without any randomness.

The second question: Do we have a method that can derive the "joint" likelihood of all 5 parameters in one step, unlike the above one?

$\endgroup$
0
$\begingroup$

You can follow the steps in this question. The only thing you need to remember is that

$\varepsilon_t = r_t - \phi_0 - \phi_1 r_{t-1}$.

You are very close to the answer - using the fact that

$r_t \vert r_{t-1},...,r_{1} \sim N(\phi_0 + \phi_1 r_{t-1}, \sigma_t^2)$

you are basically done.

$\endgroup$

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.