EViews.com

My version is Eviews 7. I am using DOLS to estimate a cointegrating relationship, and intend to include some short run dynamics and generate forecasts by utilizing an error-correction model. My primary concern is that I proceed through this process in an appropriate sequence, but I continue to stumble over the best way to include the cointegrating relationship in the ECM.

I would like to use the Eviews DOLS functionality because of its user-friendliness. However, in discussing DOLS, the Eviews help files note that "The short-run dynamics are also used in computing the residuals used by various equation views and procs such as the residual plot or the gradient view."

My understanding is that the DOLS short-run dynamics (i.e. the lead/lag terms) are "nuisance" parameters used to generate efficient estimates, but should not be used in forecasting. If that is the case, would collecting the error terms from the Eviews DOLS estimation and using their lags as the RHS variable for the cointegrating relationship in the ECM be inappropriate?

If the answer is "yes", would it then be better to collect the coefficients on the variables that cointegrate in the DOLS step and then use them to fully specify the cointegrating relationship in the ECM equation? Or, would a better solution be to estimate the Eviews DOLS, then use the model object to generate error terms (since the help files note that "forecasting and model solution using an equation estimated by DOLS are also based on the long-run relationship"), and then finally use these error terms in the ECM? Thanks.

I won't comment on what is the best approach, but will note that the reason that the residuals use the short-run dynamics is that we wanted the residuals to be the residuals from the specification that you provided. In this view, the residuals are those from estimation, not from forecasting.

If you want the residuals purged of the short-run dynamics you may simply forecast from the equation (you don't need to use the model object, just hit the forecast button), which will give you the long-run forecast, and take the actual/fitted difference.

Glenn, thanks. I was hoping for a bit of insight on the pros and cons of the various approaches I listed, so if you have any thoughts I'd love to hear them. I'm quite new to this and am still trying to resolve some uncertainties related to the methodology. Apologies if this belonged in the Econometric Discussions forum...my question related partly to the program and partly to the methodology so I wasn't sure which place to go.

You are better off with this one in the methodologies forum.

I will note that typically in an Engle-Granger two-step framework the first step isn't doing DOLS, it's just doing the long-run estimation and then estimating the short-run dynamics in the second step. This is why I'm not quite clear as to what you are trying to do.

Incidentally, I highly recommend the Phlllips and Loretan article which does a nice job of fleshing out a number of different specifications in this context.