Error Correction Model based on Dynamic OLS

[ftc1000]

Hi. I'm using Eviews 7. My question is an abridged version of a string in the Estimation Forum (http://forums.eviews.com/viewtopic.php?f=4&t=2032). Given the evolution of the discussion Glenn suggested I repost here.

I'm wrestling with the best way to estimate an error correction model – specifically, one that utilizes Dynamic Ordinary Least Squares for the cointegrating relationship. My question has two parts, one more general to the ECM and the other more specific to the DOLS/ECM approach.

1) When I estimate an ECM, is it better to proceed in one step or two? In other words, should I first estimate the cointegrating relationship, then collect residuals from this estimation and use their (lagged) values in the ECM step? Or, is it better to proceed straight to the ECM estimation and fully specify the cointegrating relationship inside the brackets? Or, is there really no practical difference? I ask because when I follow the latter approach there sometimes appear to be convergence problems related to the nonlinear specification, which tempts me to think that going in two steps is better.

2) This question is related to 1). Should the lead/lag terms from a DOLS be included in the cointegrating relationship part of an ECM? The literature I've read on DOLS notes that these terms are included to generate asymptotically efficient estimates, but that they are typically discarded when forecasting. Consequently, I am uncertain whether the cointegrating relationship in the ECM should include these terms or whether they should be left out. Any insight would be appreciated. Thanks.