-----Original Message----- From: gretl-devel-bounces(a)lists.wfu.edu [mailto:gretl-devel- bounces(a)lists.wfu.edu] On Behalf Of Riccardo (Jack) Lucchetti Sent: Tuesday, May 02, 2017 5:17 AM To: Gretl development Subject: Re: [Gretl-devel] menus AR(1) / autoregr. / vs. "ordinary" AR On Tue, 2 May 2017, Sven Schreiber wrote: > Hi, > > although I do a lot of time series, I don't usually use the methods > that come under the heading of "time-series AR(1)" or "autoregressive" > (that is Cochrane-Orcutt etc.). Indeed I would claim that those are in > general not much used anymore -- either people use ARIMA(X) estimators > (with AR as special cases), or they estimate with autocorrelation and > then apply Newey-West standard errors, or they follow the dictum of > Wooldridge and others and try to introduce more regressors to capture the dynamics. > > Having this as a background, I have two questions: > First, is it really necessary to have two separate menu entries in the > time-series submenu (of Model) for AR(1) and Autoregressive? AFAICS > it's the same stuff, and that submenu is pretty long already. > Secondly, I'd suggest to put the keyword "errors" or "innovations" in > those menu labels, because for me an AR(1) model is just "ols y const > y(-1)", not the other stuff. > > So, instead of "AR(1)" and "Autoregressive estimation" I'd have "With > AR errors". I agree. In fact, I think some restructuring of the Model>Time Series menu may be a good idea. Perhaps, we could move Cochrane-Orcutt etc, which nobody uses anymore, but are nice to have if only for completeness, to "Other linear models" (because that's what they are). I would also like to have a "Multivariate" submenu for all VAR/VECM things.

I like the idea of a Multivariate submenu, and that's now in git. Note that this means the menu attachment for SVAR should be changed for the next release.

But a couple of other thoughts here: might we be better off with two menus: "Univariate time series" and "Multivariate time series"? As things stand we have a pretty deep cascade of Time series/Multivariate/Cointegration test/Johansen. Or alternatively, move the two cointegration options up a level? Cointegration test (Johansen) Cointegration test (Engle-Granger)

As for "AR(1)" and "Autoregressive estimation", I agree they want cleaning up in some way but I'm not sure yet what's best. I'm not so keen on parking them under "Other linear models" because they are time-series specific. Maybe a single entry "AR errors (GLS)" with a fork to first-order vs general?

While we're at it, I'm not sure we need the trailing "..." on lots of the model menu items (though trashing the ellipses would make some work for translators).