The RATS Software Forum

Hi,

I hope this finds you well,

I try to apply the test of Andrews-Quandt on a linear regression containing 3 lags for each variable, and I find that the test refuses the hypothesis of coefficients constancy for the 2nd lag of a variable.

Does this mean that the coefficients of that variable "in general" aren't stable? And how could I fix this if you don't recommend the split of the sample?

Thanks in advance

Ahmed Sahloul

Does the model reject for "all coefficients"? If the overall test seems OK, I wouldn't worry about the test on the middle coefficient. If the overall test also rejects and the only coefficient that's individually a problem is the middle lag, you might need to look at a threshold autoregression. The A-Q test is looking for stability based upon the time sequence, but might be able to pick up some other form of instability.