GARCH with asymmetric volatility

[timkrause]

Hi,

I am interested in examining volatility spillovers among 3 stock price series, while accounting for asymmetric volatility. When I run each of the univariate series separately, the asymmetric coefficient is negative and significant (as expected). But when I try to run the multivariate estimation, the asymmetric coefficients are positive (and significant). I'm thinking I must be setting up the model incorrectly. Here is the code that I am using:

GARCH(P=1,Q=1,PMETHOD=SIMPLEX, PITERS=10,MV=CC,ROBUST,LAGS=5,VARIANCES=SPILLOVER,ASYMMETRIC, DISTRIB=T) / R1E R2E R3E

Does this look correct and does anyone have an idea why this might be happening?

Thanks,

Tim

[moderator]

I don't see any issues with the setup. You might try switching to the diagonal model (MV=DIAGONAL) as a test. That should produce essentially the same results as doing the univariate models with separate GARCH instructions.

If the results do match, that would be a sign your basic setup is comparable, in which case the differences are presumably a result of your choice of multivariate model form, and/or any issues with getting the model to converge to a global max. If you haven't already done so, you'll want to experiment with different initial values, changing the number of preliminary simplex iterations, and examining the TRACE output carefully to try to assess model convergence.

Regards,
Tom Maycock

[timkrause]

Thanks Tom,

I will try that.

Tim