The RATS Software Forum

Hi,

I am estimating a multivariate IGARCH model and have when I use the BEKK specification, my model doesn't converge. I think this is because my matrix is near singular.

The constant conditional correlation method works, however the assumption of constant correlation for my model may not be realistic. I have tried a DCC GARCH model and it also fails to converge.

Is there any way of getting a BEKK specification for this?
I have 740 observations and 8 series.

Thanks

You can sometimes fix convergence problems in multivariate GARCH models by multiplying the data by 100. The magnitudes of the constant parameters in the variance equation are often rather small without the rescaling, which makes it difficult to test for convergence. However, BEKK models are often quite difficult to fit, and can require quite a bit of fussing with optimization settings.