Starting values for truncated Tobit estimation

[Raupel]

Dear all,

dealing with a truncated dependent variable, I run several tobit-regressions using Eviews 6.0. Fortunately, my results are robust against different starting values. To check that, I used several options to determine these values, starting with "Eviews Supplied" and then testing ".8 x Eviews", ".5 x Eviews" and Zero. However, I cannot really find additional information about the built-in algorithms behind "Eviews Supplied" which provide the starting values.

Could somebody please give me some information about these procedures? I would like to mention this in my paper and I'm therefore keen to provide some rough details.

THX!

Raupel

[EViews Glenn]

They're basically the OLS results for the equation with no correction for censoring, with the starting variance coefficient taken from the residual variance. An oft-mentioned alternative would be to take the regression on the truncated sample.

That said, I wouldn't view taking fractions of the default coefficients as particularly great way of showing robustness since all your comparisons lie on the same hyperplane.

[Raupel]

Thx Glenn!