Re: [Gretl-users] convergence tolerance
In your case 1.0E-02 may be sufficient. In many other cases it will
not. In your case the algorithm may have found an accurate value for
the optimum by accident and both convefrgence criteria may have been
satisfied. In general you can not expect to be so lucky. A good
textbook on numeical methods in optimisation will explain a lot more.
John,
On Sunday, 20 March 2011, Sam Sam <dear.sam(a)livemail.tw> wrote:>
Dear all:
I estimate ARIMA(2,1,1,)(1,1,1) by conditional maximum likelihood using convergence
tolerance 1.0E-6 and 1.0E-2
The two outcome using different convergence tolerance are similar.
So why convergence tolerance 1.0E-2 may be too high? (the outcome is similar to 1.0E-6)
Is it any resonable range of convergence tolerance?
Is the model must misspecificated when convergence tolerance is 1.0E-2 ?
--
John C Frain
Economics Department
Trinity College Dublin
Dublin 2
Ireland
www.tcd.ie/Economics/staff/frainj/home.html
mailto:frainj@tcd.ie
mailto:frainj@gmail.com