In some cases, the VARMAX procedure may return either of the following two errors when fitting a multivariate GARCH model:

ERROR: Floating Point Overflow

or

ERROR: Invalid Operation.

These exceptions may occur when jumps or large changes in the objective function are observed in the iteration history returned by the estimation of the model.

Although there is no circumvention that applies to all cases in which this problem is observed, specifying an alternative model may allow the procedure to converge to a solution. Another possible alternative is to use a different optimization method by specifying the TECH= option on the NLOPTIONS statement in the PROC VARMAX step. TECH=QUANEW is the default quasi-Newton minimization method.