[R-SIG-Finance] Rugarch: How to do Iterated n-ahead Multistep Out-of-Sample GARCH Forecasts?

[Alexios Ghalanos]

Have a look at the ugarchfilter method (also have a look at the underlying code for ugarchroll for some possible pointers).

Alexios

> On 21 Sep 2014, at 07:56, Don Brady <dbrady at pobox.com> wrote:
> 
> What would be the most efficient way to make iterated n-ahead Multistep Out-of-Sample GARCH Forecasts using Rugarch?
> 
> Let me explain what I mean.
> 
> A number of papers on using GARCH describe such a method and call it the "iterated" method. (They also say that it works very).
> I am trying to use it.
> 
> For example, from
> "Multi-Period Forecasts of Volatility:Direct, Iterated, and Mixed-Data Approaches"
> by Eric Ghysels† Antonio Rubia‡ et al:
> 
> "Long horizon volatility forecasts can be constructed in three fundamentally different ways.
> ....... The second approach is to estimate a daily autoregressive volatility forecasting model and then
> iterate over the daily forecasts for the necessary number of periods to obtain weekly, monthly, or quarterly predictions
> of the volatility. The forecasting literature refers to the first approach as “direct” and
> the second as “iterated” (Marcellino, Stock, and Watson (2006))."
> http://www.unc.edu/~eghysels/papers/Var_9.pdf
> 
> I am looking to use this "iterated" approach to make a long term forecast.
> 
> The authors do not appear to be referring to a simulation, but rather are making an out of sample iterative forecast that ends up cumulatively giving them a forecast for up to 30 days ahead.
> 
> I can see that this could be done in rugarch by using an R loop, stepping forward out of sample one day at a time. At each step of the loop, one would call ugarchfit, then call ugarchforecast with a one day horizon. Then for the next iteration of the loop in R, one would augment the data by the result of the just-performed forecast, and re-fit and re-forecast etc.
> 
> However, this might be slow so I was just wondering if there is a rugarch-built-in way of doing this without needing the outer loop in R.
> 
> ugarchforecast does offer n.ahead forecasts, but states that n-step ahead (n>1) forecasts are based on the unconditional expectation of the models, which does not seem to be the same thing as these authors are suggesting.
> 
> I just have the feeling that I am missing something.
> 
> THANK YOU for any comments and also for providing this incredibly comprehensive package!
> 
> _______________________________________________
> R-SIG-Finance at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-finance
> -- Subscriber-posting only. If you want to post, subscribe first.
> -- Also note that this is not the r-help list where general R questions should go.
>