I'm trying to model several financial time series and to get some forecast on them.
Considering the log-returns, I fitted an EGARCH(1,1) on the data and got the parameters, as known constraint-free because of the model:
$$
\text{log}(\sigma^2_t)=\omega + \alpha_1 g(\eta_{t-i}) + \beta \text{log}(\sigma^2_{t-j})
$$
which considers exponentation for the parameters.
So I fit the model with ugarchfit
function from rugarch
package in R and then I use the ugarchforecast
function to get the unconditional volatility of series. In the output I can notice many negative prediction!
Is that a problem of model-validity or does the ugarchforecast
function return the conditional log-volatility so I have to exponentiate my output?