How do I calculate the VaR in case of using an ARMA-GARCH approach? I am not good at time series, so I am more or less confused with the different possible notations of an ARMA-GARCH process. I hope the notation I use is correct.

Suppose I use the following AR(1)-GARCH(1,1):

\begin{align*} r_t=\delta + \phi_1 r_{t-1} + z_t\\ z_t = \sigma_t \epsilon_t\\ \epsilon_t \text{ is } N(0,1)\\ \sigma_t^2=\alpha_0+\alpha_1 z_{t-1}^2+\beta \sigma_{t-1}^2 \end{align*}

I mean, I would have to calculate a quantile at timepoint t. Normally in case of a more simpler approach: $r_t=\mu + \sigma_t*\epsilon_t$ and the $\epsilon_t$ is N(0,1) this would give $VaR=\mu + \hat{\sigma_t}*z_\alpha$ where $z_\alpha$ is the quantile of the standard normal. How to do this in case of the ARMA-GARCH approach? I now have no $\mu$, but an AR process?

I read some papers, they all give more or less the ARMA-GARCH formula and say that they generate the forecast, but they give no formula, how to do this.


1 Answer 1


To estimate $VaR$ of the one-step-ahead return (which is what you seem to need to do) you should do the following ($E_t$ denotes the conditional expectation given the values of the time series up to and including the moment $t$ have been observed):

  1. Compute the forecast of the mean. In your case of $AR(p)$ process you should do the following: $$r_{t+1|t} = E_t(r_{t+1}) = E_t( \delta + \sum^p_{k=1} \phi_k r_{t+1-k} + \sigma_{t+1} \epsilon_{t+1})$$ which reduces to $r_{t+1|t} = \delta + \sum^p_{k=1} \phi_k r_{t+1-k}$, since $\epsilon_{t+1}$ has zero mean, and $\sigma^2_{t+1}$ is known given the observations up to $t$. In your case $$r_{t+1|t} = \delta + \phi_1 r_t$$

  2. The volatility forecast is $\sigma^2_{t+1|t} = \alpha_0 + \alpha_1 \hat{z}^2_t + \beta \sigma^2_t$ where $\hat{z}_t = r_t - \delta - \phi_1 r_{t-1}$, and can be obtained similarly.

  3. Now to get $VaR$ at the moment $t+1$, given the history up to $t$, you should just take the quantile of the standard normal, multiply it by the square root of $\sigma^2_{t+1|t}$ and then add $r_{t+1|t}$: $$VaR = r_{t+1|t} + \sqrt{\sigma^2_{t+1|t}} z_\alpha$$


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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