The RATS Software Forum

Dear Tom,

is there a possibility to estimate a trivariate GARCH DCC with regressors affecting the mean in 1 step?
The tentative below shows that the coefficients and standard errors on jpn{1} xfra{1} xsui{1} cannot be computed.

garch(p=1,q=1,mv=dcc,method=bfgs,regressors) / xjpn xfra xsui
# constant xjpn{1} xfra{1} xsui{1}

MV_GARCH, DCC - Estimation by BFGS
Convergence in 43 Iterations. Final criterion was 0.0000020 <= 0.0000100
Usable Observations 6235
Log Likelihood -11815.32247128

Variable Coeff Std Error T-Stat Signif
*******************************************************************************
1. Constant -0.000987057 0.003996127 -0.24700 0.80490564
2. XJPN{1} 0.000000000 0.000000000 0.00000 0.00000000
3. XFRA{1} 0.000000000 0.000000000 0.00000 0.00000000
4. XSUI{1} 0.000000000 0.000000000 0.00000 0.00000000
5. C(1) 0.008483495 0.001133582 7.48380 0.00000000
6. C(2) 0.012373852 0.001296432 9.54454 0.00000000
7. C(3) 0.016467521 0.001806103 9.11771 0.00000000
8. A(1) 0.151577581 0.009289633 16.31685 0.00000000
9. A(2) 0.137614307 0.008072588 17.04711 0.00000000
10. A(3) 0.123296394 0.006950694 17.73872 0.00000000
11. B(1) 0.852136087 0.008320991 102.40800 0.00000000
12. B(2) 0.849418080 0.008055446 105.44643 0.00000000
13. B(3) 0.858515861 0.007713709 111.29741 0.00000000
14. DCC(1) 0.053339362 0.003239076 16.46746 0.00000000
15. DCC(2) 0.938923490 0.003809009 246.50076 0.00000000

The REGRESSORS option doesn't work with the multivariate model. Instead, use the MODEL option, first defining a 1-lag VAR:

Dear Tom,

the answer is very clear.

I have a second question related to a loop.
Assume that I want to estimate two trivariate DCC Garch as below {xjpn xfra xsui} and {xjpn xfra x}.
To save space and time I need to construct a loop given that xjpn xfra are common to both models.
Is there a simple way to do it? Thank you very much.

Is there something more to this than what you've done? If you're changing both one of the variable and the name of the model, this is as simple as anything you could do with a loop. If you could use the same model name, you could do:
However, to me, writing it out is clearer when the code block is that small.