FMOLS vs. DOLS vs. CCR vs. VEC

[z_high0]

I really need help in this:

So I want to check whether two variables (for which I have yearly data for about 60-70 years) are cointegrated or not. It seems that Eviews7 has plenty of options to check that.

1- VEC analysis (which I think is estimating the cointegrating equation with OLS, right?)

2- FMOLS

3- DOLS

4- CCR

now, 1 and 3 more or less give essentially the same results (same signs of coefficients in cointegrating regressions) but 2 and 4 are completely different. which one should I choose?

[z_high0]

any help?

[EViews Glenn]

There is no right answer to this question. I will note that the VEC and the DOLS are "lag based" estimators while the other two are "kernel based" estimators. Playing with the lags and/or kernel bandwidths might get the two sets to line up. The Johansen testing has the advantage of being system based instead of single equation based...Many would argue that that is a good thing.