The RATS Software Forum

Hi all

Can anyone explain me the J-statistic when using GMM model. For example. I am running GMM

instruments x1 x2 x3 x4
Linreg(inst, optimalweights) y
# constant x1

The significant level of J is, for example, 0.72 is good or not?

Thank you very much

That would indicate that you have a compatible set of instruments. The model is fit using the "optimal" weighting on the instruments, which sets those linear combinations of conditions to zero, but to get a .72 significance level, the remaining linear combinations are close to zero as well.

TomDoan wrote:That would indicate that you have a compatible set of instruments. The model is fit using the "optimal" weighting on the instruments, which sets those linear combinations of conditions to zero, but to get a .72 significance level, the remaining linear combinations are close to zero as well.

Thank you very much Tom. Anyways, could you please advise me the critical p-value of J-statistic to pass the Hansen J-test?

It's not a particularly powerful test (particularly when the degree of overidentification is high), so the conventional .05 might be a bit too stringent. I'd probably be a bit concerned if it were less than .10.