Dear Tom Doan,
I am estimating a GARCH-BEKK model and one of the series appears to have an outlier, causing a rather bizarre results for that series. In particular, the series named as variable 2 has a significantly higher coefficient for it's own past shocks on it's volatility, i.e. A(2, 2) is -1.6 which is considerably higher. I want to ask if there is a way I can remove this outlier before proceeding with the estimation of the model. When I cut the data short to end before the date of the outlier, I obtained accepatable results but this shortens the entire sample period.
Thank you
Dealing with Outliers in GARCH-BEKK
Re: Dealing with Outliers in GARCH-BEKK
You might want to look at the advice to another user:
https://estima.com/forum/viewtopic.php?p=12698#p12698
Much depends upon whether the outlier is really an "outlier" in the sense that it doesn't fit the behavior of a GARCH model. The main question for that is whether the data are in fact extremely volatile subsequently or not.
There are more complex "jump" models which allow for isolated data spikes that don't have the same implications for future volatility as ordinary movements.
https://estima.com/forum/viewtopic.php?p=12698#p12698
Much depends upon whether the outlier is really an "outlier" in the sense that it doesn't fit the behavior of a GARCH model. The main question for that is whether the data are in fact extremely volatile subsequently or not.
There are more complex "jump" models which allow for isolated data spikes that don't have the same implications for future volatility as ordinary movements.
Re: Dealing with Outliers in GARCH-BEKK
Dear Tom,
I notice that some of my series have outliers just by looking at their graphs. I did this check after I found that my model was not converging whenever I include these series.Could you tell me how should I treat them? Also should I first determine if they are really outliers or not(As you mentioned in one of the earlier posts on the topic)?) If so, how should I do it?
Please help me.
Thanks.
Regards.
I notice that some of my series have outliers just by looking at their graphs. I did this check after I found that my model was not converging whenever I include these series.Could you tell me how should I treat them? Also should I first determine if they are really outliers or not(As you mentioned in one of the earlier posts on the topic)?) If so, how should I do it?
Please help me.
Thanks.
Regards.
Re: Dealing with Outliers in GARCH-BEKK
This is your data. Look into the situation at those data points. Is there a reason for odd behavior that might be outside of the ability of a model to explain? Is there any reason to doubt the value in the data set? The new edition of the GARCH e-course includes a section looking at odd behavior of some of the series in the Lee JIMF paper where it appears there are typos in the data set---there are a handful of inexplicable blips in both future and in spot prices that seem to have no effect on subsequent values for either.
If the problems are near the beginning or end of the data range, just cut the data set off. Most papers will never describe precisely why their analysis starts and stops at particular dates, and sometimes, it may simply be because the model fails outside of that (information which probably should be included at least as a footnote, but often isn't).
If the problems are near the beginning or end of the data range, just cut the data set off. Most papers will never describe precisely why their analysis starts and stops at particular dates, and sometimes, it may simply be because the model fails outside of that (information which probably should be included at least as a footnote, but often isn't).