Are there performance benchmark of ARCHModels.jl vs. rugarch & rmgarch?
rugarch is pretty slow, and I kind of remember a while ago someone mentioned ARCHModels has a performance issue.
Yes: Archmodels.Jl: Estimating Arch Models in Julia by Simon A. Broda, Marc S. Paolella :: SSRN
ArchModels can be used for more than just volatility models
1 Like
In my experience, it’s very fast! I was messing around earlier this year with adding additional methods and the internals are impressive. I’m really not surprised that they find it being the most performant out of the standard toolboxes (in the SSRN paper linked above).
2 Likes
Thanks for the link to the paper. They used a dataset that only has about 2000 observations for the benchmark. I understand that it makes sense to use this original data of Bollerslev and Ghysels (1996) to show the correctness of estimates. However, to show the performance, I believe it is better to use larger data.
2 Likes
Not sure. The performance should scale more or less linearly with the size of the data set, so I’m not sure how much difference it would make. If anything, it would probably favor ARCHModels.jl even more.
Another idea might be to show a backtesting exercise. That’s what I did in a seminar I gave a while ago, see here: Jupyter Notebook Viewer It shows ARCHModels.jl estimating 7334 GARCH models in 20 seconds 
Granted, the window size is just 1000, but that’s pretty standard in backtesting.
@Albert_Zevelev Tangential: I currently have two students working on new models, so there should be some nice updates soon 
3 Likes