I am currently using stata 7.0 SE. My concern is that the ARCH command in stata maybe not stable. I have run some garch estimation many times. the estimation will give error message occationly, depend the sample size. When i add changed more samples, even a few, the ML works. I wonder if i can preset the initial values for the coefficient but failed. I am writing this
letter to see if any of you have also encounter such problems and whether stata 8.0 is more powerful in GARCH estimation.

I think S-plus with S+FinMetrics is best choise.