Elder-Serletis(2010) VAR-GARCH-M

[TomDoan]

If you're talking about the test described at https://estima.com/forum/viewtopic.php?p=7175#p7175, then those are on the VAR residuals. If there weren't ARCH effects in either set of residuals, it would call into question the need for the whole GARCH analysis.

[cczzwhy]

Thank you very much! One more question about the graph,I want to change the X labels and Y labels in the impulse response graph,so I tried this code

Code: Select all

@MCProcessIRF(model=basevar,center=median,lower=lower,upper=upper,irf=irf)

,but I got the errors

Code: Select all

## SX22. Expected Type RECTANGULAR[SERIES[REAL]], Got VECTOR[VECTOR[REAL]] Instead
>>>>pper=upper,irf=irf)<<<<

Can you help me?

[TomDoan]

The variable IRF is already in use for doing the calculation of the IRF's. Choose a different name for the target of the IRF option on @MCPROCESSIRF and anything you do after it. (When you get a message like that, it's almost always that you're trying to re-use a variable name).

[gorgorm]

Hello Tom,

Just some clarification.

From Table 2 in Elder and Serletis (2010), does G1(1) and G2 (1) correspond to the constant; G1(2) and G2(2) correspond to E_i(t-1)^2; and G1(3) and G2(3) correspond to H_t i (t-1)?

The Gs are estimates from RATS.

Thanks

[TomDoan]

Correct.

[n_khraief]

Dear Tom,

Have you an idea how to transform this code for a panel of many countries?

Thanks a lot,
N.K

[TomDoan]

I'm not sure what you have in mind. Multivariate GARCH models are almost always fit on what might be (in other contexts) described as a small N-big T panel. This model doesn't fit that mold, since it uses two very different series (oil prices and GDP) and looks more at the "VAR" behavior that the "GARCH" behavior. If you're talking about oil prices and a "panel" of GDP's, you could extend it to that, but since GDP has almost no "GARCH" properties, it's not clear what the point is. You would probably be better off estimating oil price volatility using a separate univariate model, and then feeding that in as an exogenous variable to a VAR on the GDP's.

[sarthak24]

Hi Tom

I am working on a similar bivariate GARCH-M model in which I need to estimate the impact of volatility of both series on the mean equation of both series. My question in context with this paper is
1. How can the code be modified to estimate the impact of oil price volatility on both Oil Price growth as well as GDP growth.
2. How can the code be modified to estimate the impact of GDP growth volatility on both Oil Price growth and GDP growth.

I am attaching the replication code for Elder-Serletis (2010) which I found on the RATS website.

Many thanks in advance.
Sarthak

[TomDoan]

I'm not sure this is the model that you want if that's what you're trying to do. This has a lot of pieces to it, one of which is that the multivariate GARCH model is a non-standard Cholesky factor model which uses univariate GARCH models on unobservable orthogonalized components. That rather intentionally is designed to treat the two series very differently.

[sarthak24]

Hi Tom

Thanks for your reply. Could you please point me towards the right model I should be looking at to accomplish this. I have been searching to implement this methodology but this is the model and software which I found to be most closely related to what I want to do. If not using this code specifically, do you think it can be done at all using RATS?

Many thanks for your help.
Sarthak

[TomDoan]

Look at Grier, Henry, Olekalns & Shields(2004). That does a VARMA rather than VAR mean model---you can get rid of the "MA" part by cutting the EPS's out of the VAR system. (VARMA mean models are very hard to estimate, so I would recommend sticking with the VAR). You can use any type of multivariate GARCH model for that---they did an asymmetric BEKK with a custom sign convention, but you can use just a standard BEKK if you want. BEKK tends to be the most common choice for multivariate GARCH with variables as different as yours.

[sarthak24]

Thanks a lot for the help Tom. I will go through the paper and the code you have provided. This will be very useful.

[Yeadly]

Dear Tom

I do have a question, if instead of a GARCH (1,1) as Elder's version. I try run a GARCH (2,2) what would be the code? Let's say, this would be the only modification I make to Elder's version.

could you help me? please.


Thanks a lot.

[TomDoan]

It's not that difficult but the big question is why? This model already needs one of the GARCH coefficients pegged because it goes negative. A (2,2) model will almost certainly have some major problems.

[Yeadly]

I'm new in RATS and I'm exploring the examples. I would simply like to know the code to A(2,2).
You could write the codes?


Thanks a lot.