The RATS Software Forum

Hello Tom,

I was wondering if one could estimate a multivariate GARCH-in-mean model such as Elder and Serletis (2010) or Rahman and Serletis (2012) with Bayesian methods (and with more than two variables).

i) If so, what would be a good starting point to learn it, if you have only very basic knowledge of Bayesian estimation? (Which algorithm would you think is most suitable for such models?)
ii) What would be potential drawbacks and problems of this estimation approach?

Isn't that what the error band calculation in the Elder Serletis replication is doing? It's not concentrating on the parameters directly (rather the IRF's that are derived from them) but it's the same thing.

The bigger the model, the more tuning will probably be necessary to get the jump distribution right.

Ok, thanks. I will have a closer look at it.