I have a few questions regarding in-sample size for volatility forecasting in EGARCH(1,1). I'm currently sitting with a dataset consisting of 1387 trading days of the S&P-500 index. I would like to forecast the volatility for a giving 3 month hedging period.
However, for now I've split the dataset into 20 non-overlapping portfolios of 3 months, then using an in-sample size of 126 days, to predict for each.
But I was wondering if there was a formal rule of this? And if you can refer me to a paper where this is described?
$\begingroup$
$\endgroup$
2
-
$\begingroup$ There does not exist a general rule for the amount of data you need, in order to procure stable estimates in the GARCH model, since it is dependent on the data itself. From this paper they recommend using a minimum of 1000 samples for a GARCH fit. In my opinion, an in-sample size of 126 days is definitely in the low end. $\endgroup$– PlebMay 19, 2021 at 10:07
-
$\begingroup$ @Pleb yeah that was generally my concern, I also read that paper. However, I went with the in-sample size as Hull's rule of thumb for calculating historical volatility is of 90-180 days, that is of course not the same as a good in-sample size for GARCH models. $\endgroup$– Sebastian Strauss HansenMay 19, 2021 at 11:27
Add a comment
|