FMOLS Summary

[Pedro_PS]

Hello everybody,

my question is regarding the summary of the function cointreg(). The Eviews unser's guide states "The summary statistic portion of the output is relatively familiar but does require a bit of comment. First, all of the descriptive and fit statistics are computed using the original data, not the FMOLS transformed data. Thus, while the measures of fit and the Durbin-Watson stat may be of casual interest, you should exercise extreme caution in using these measures."

I took two dara sets and ran an regular OLS and the FMOLS. The fit statistic R-squared was still diffrerent for both estimations. According to the statement I believe they should be the same values. Can I trust the fit values of the FMOLS or not? What is the reason for not computing the fit statistics with the FMOLS trasformed data.

Thank you in advance

Pedro

[EViews Glenn]

The R2 are different because the OLS and FMOLS coefficient estimates are presumably different, as are the residuals.

You may use them, but in your case since the FMOLS objective is very different from the OLS estimates, I'm not certain what one would conclude from the differences.

[Pedro_PS]

Dear Glenn,

thank you for your answer. I understand now why the R squared is different.

I am just confused: Why write that "the fit statistics are computed using the original data, not the FMOLS transformed data." When R2 is a fit statistic?
Solely the "Mean dependent var" and "S.D. dependent var" are indeed the same.

Best wishes

Pedro

[EViews Gareth]

R^2 is calculated using residuals.

Residuals are Y-Xb.

Thus to calculate R2 you need two different pieces of information - the data (Y,X) and the coefficients.

The statement in the documentation is stating that the original, untransformed data is used, but the coefficients come from fmols