An easy way to restrict regression coefficients

For making suggestions and/or requests for new features you'd like added to EViews.

Moderators: EViews Gareth, EViews Moderator

wc_guy
Posts: 1
Joined: Thu Apr 03, 2014 10:52 am

An easy way to restrict regression coefficients

Postby wc_guy » Thu Apr 03, 2014 11:13 am

Back in grad school when I used SAS, there was an easy way to estimate a regression that could restrict a coefficient to a range. There is at least one forum post that asks about this and the answer is a kludgy log transformation of the underlying series. This would be an especially nice feature to have in pool and panel tools. As an example, I work with a lot of geographical time series (metro area, countries, etc); the wonder of the pool tools in Eviews is the ability to quickly make and modify regression/model specs, but invariably specs with say, cross-section specific coefficients, will give "wrong signs" in some annoyingly small percentage of cases. Rather than going to single-equation solutions, it would be great to have a simple way to prevent wacky results that come because real world data is annoyingly messy. As someone who thinks OLS is plenty good enough for most industry uses (I've done economic consulting and I'm currently in asset management), I'd love to see more tools for generating well-behaved models and forecasts (the robust regression technique that was added in 8.0 is a great example of one) versus the latest technique from Econometrica. But to be fair, I may be a minority case.

EViews Glenn
EViews Developer
Posts: 2671
Joined: Wed Oct 15, 2008 9:17 am

Re: An easy way to restrict regression coefficients

Postby EViews Glenn » Thu Apr 03, 2014 4:24 pm

It is a more than sensible request and one that we have on our to-do list. Thanks for your input.

Incidentally, our recommended method is not a transformation of the underlying series, but a conventional transformation of the parameters to an unrestricted space. This is a common method. What would be nice is us making it so that the user didn't have to do the transformation themselves. In some cases (e.g., pools), the latter is not even possible.

trubador
Did you use forum search?
Posts: 1518
Joined: Thu Nov 20, 2008 12:04 pm

Re: An easy way to restrict regression coefficients

Postby trubador » Fri Apr 04, 2014 4:05 am

I agree with Glenn as this is a genuine need. And also share the same feelings with wc_guy on robust regression tools.

This may be off topic, but I'd like to further elaborate on OLS-type methods. I have been working in a highly dynamic emerging market and have had a similar career path as wc_guy. Although the usefulness and effectiveness of any method cannot be separated from its user, I have witnessed very poor performances of such simple methods when applied to these markets. Regime changes, structural breaks, outliers, time-variation, etc., which are very common issues in emerging markets, do require more complex analytical tools. It does not have to be the latest, but the techniques that proved useful and gained widespread popularity/influence should be considered top priority for implemention in EViews. Bai-Perron and Switching Regression are very nice examples of this.

Since it is not possible to access and modify built-in objects, providing the user with additional flexibility is equally important. In that respect, I find Optimize feature very useful and necessary. I understand the trade-off between assortment and variety, but more advanced programming features are needed. For instance, EViews not handling multidimensional arrays or complex numbers put a certain limit on its progress. I think I should stop here and discuss my suggestions under a separate topic...


Return to “Suggestions and Requests”

Who is online

Users browsing this forum: No registered users and 13 guests