Robust estimation S and MM methods

[barryoli]

I'm trying to run S and MM robust methods and I get an error:
"Maximum number of singular subsamples reached".

The model runs ok in M-estimation (multivariare ols):
corruption100 regulation ideology100 gender age polsci australia china c

the last three are dummy variables...I presume the method can't handle dummies if it is a method to reduce outliers?
hence the error.

If this is correct is there anyway around it?

any help appreciated as always.
Cheers
Barry

[EViews Gareth]

You can try changing the settings on the S estimation, but if you have sparse dummies (i.e. only a handful of observations that differ), it will never estimate.

[Guy]

EViews Gareth, can you flesh out a bit what you mean. I am running into the exact same problem when running MM-estimation. I have 217 observations, but I have numerous dummy variables that are binomial (1,0). Some have only eight observations representing 8 quarters over this long time series going back to 1960 when they are active (1). I suspect this is a problem for Robust Least Square MM-estimation. Thanks in advance for any clarifications.

[EViews Gareth]

http://www.eviews.com/help/helpintro.ht ... .2.html%23

If you look at the section S-Estimator Calculation, you can see that to perform S estimation (and thus MM estimation too), EViews estimates using sub-samples of the data. If you have sparse dummies, then chances are that some of those subsamples will have no variance in the dummy variable, which will mean it is perfectly colinear with a constant.

[fella shufa nm]

is it possible to estimate robust s estimation of the SUR model in eviews?

[EViews Gareth]

Not built in

[fella shufa nm]

But the result of beta in robust s same as beta in GLS. Is that still work in eviews 8?