Changing constant term – Full Information Maximum Likelihood

[Barry Golding]

Changing constant term – Full Information Maximum Likelihood
Using a system of three equations with time series data with years numbered one to fifty, I estimate the equations using Full Information Maximum likelihood (FIML) method which provides a constant term. If I then evaluate using OLS I get a very different constant term. I then re-estimate using FIML and the constant term changes to one similar to the OLS estimate. Is this a glitch in the EViews FIML estimation Program or a glitch in my understanding of FIML?

Cheers
Barry

[EViews Gareth]

Seems likely that your specification is sensitive to starting values. When you re-estimate the system using FIML the second time, you have set the starting values to be those estimated by OLS, and thus the results are close to those estimated by OLS.

FIML can be quite difficult to estimate. You could try tightening the convergence criteria of your system to see if it is more stable.