EViews.com

Sometimes after editing the model sspace crashes in EViews 9. The estimation window doesn't even pop up. It happens with some models, not with all. No problem in EViews 8.

Thanks for the report. Is it after you edit and then agree to save the edits? Or following new estimation?

I edit the model, hit "estimate" and then it crashes and EViews stops. The edit is not saved after I restart EViews.

Thanks. We'll investigate.

Try today's patch.

It seems to work. Will do some more testing tomorrow. Thanks!

Hi everyone, the estimations of the SS models are troublesome: i.e. high instability of the results, program crashes. the same program does not show these problems with Eviews8. The new version (Eviews9X64 - 8 May) has had a very difficult start. I'm worried and a little bit disappointed.

There are two parts to your comment that I'd like to address in turn.

First, EViews shouldn't crash, and with the exception of the dialog problem for which a fix was posted earlier, you should not be seeing crashes. If you do continue to see problems, please let us know and we'll do our best to get a fix as soon as possible.

Second, as to instability of results. Estimation of Sspace models in EViews 9 was enhanced with the introduction of BFGS optimization, a feature for which we have had many, many requests over the past few years. BFGS has a number of nice properties, but of course, the use of a different optimization method means that you may get different results. The improvements to state space estimation were primarily motivated by the desire to offer users different approaches to estimation so that they could better their examine results for sensitivity. At the same time, we are acutely aware that individuals desire some measure of backward compatibility. Accordingly, all of the nonlinear optimization routines which have been updated allow for "legacy" estimation which should, for the same starting values and other settings, produce the same results as before.

Note that Sspace models are generally non-concave and quite sensitive to both starting values and estimation algorithm, and you may find that while the new optimizers produce coefficient singularities and other undesirable properties, that these results may be associated with better local optima. As always in nonlinear estimation (and especially with state space estimation) you should carefully examine results for sensitivity (to starting values, convergence tolerance, and algorithm). It sounds as though your results are particularly sensitive to method. You may, of course, use the legacy optimizer, which should behave exactly as in earlier versions. I do hope that you bear in mind that singularity of results does not mean that a local optima is inferior to the ones you observed previously.