2
$\begingroup$

I'm working with the really nice rugarch-package and currently have an issue with respect to the news impact curves (NIC).

In an attempt to plot several NIC into the same plot I realized that while the NIC for the sGARCH, the gjrGARCH and the eGARCH are given with respect to the $\epsilon_{t-1}$, the NICs for the submodels of the fGARCH model are given in terms of $z_{t-1}$, with $ z_t = \frac{r_t-\mu_t}{\sigma_t}= \frac{\epsilon_t}{\sigma_t} $ being the standardized innovations. So they can't be plotted together in a single plot using the same scales.

I am a bit confused since just like the fGARCH model, the eGARCH model (as to the eGARCH-model-setup in the vignette) is also given in terms of the $ z_{t-k} , k={1,...,q}$. But nevertheless the NIC of the eGARCH is given in terms of $\epsilon_{t-1}$ . At least this is what the "newsimpact(fit)"-output tells me. I think that this might be a typo and it should mean "$z_{t-1}"$. Can anyone please help?

The eGARCH vola equation is given by

$ln(\sigma_t^2)= \omega + \sum_{k=1}^p \Big( \alpha_k \ z_{t-k} + \gamma_k\ (|z_{t-k}| - E|z_{t-k}|) \Big) + \sum_{k=1}^q \beta_k \ ln(\sigma^2_{t-k})$

And we obviously have

$\alpha_k \ z_{t-k} + \gamma_k\ (|z_{t-k}| - E|z_{t-k}|) = \begin{cases} (\alpha_k + \gamma_k)z_{t-k} + C & z_{t-k} \ge 0 \\ (\alpha_k - \gamma_k)z_{t-k} + C & z_{t-k}<0 \end{cases} $

with the constant $C=-\gamma_k E[|z_{t-k}|]$. So the news impact depends on $\sigma_{t-1}$ which is nonconstant and therefore cannote be ignored or set to a fixed value for the NIC to make a general statement. Rather it has to be considered as being random along with the $\epsilon_{t-1}$ and so we have a NIC that depends on $ z_t = \frac{\epsilon_t}{\sigma_t} $. This is why I think that there might be a typo in the rugarch output.

Is there an error in my thinking?

Many thanks and kind regards, Jo

$\endgroup$

1 Answer 1

1
$\begingroup$

There was a typo indeed, but in the fGARCH-output. The author of the package is already informed. It must be $\epsilon_{t-1}$ in all cases.

This is achieved by plugging the UNCONDITIONAL volatility into the formula as a proxy for the time varying conditional volatility $\sigma_{t-1}$ where needed - e.g. for the eGARCH and fGARCH models.

$\endgroup$

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.