EViews.com

Hello!

I try to estimate cointreg in E7 and don't understand why in "Actual, Fitted, Residual Table" figures are started at point 2. Why? When I make forecast at the same period, it calculate me the forecast value for first observation.
And summary stats are calculated not from 1st point, but from 2nd, even mean.

Please, tell me why it is so.

Thank you

Moreover, SSR that E7 shows is calculated on period [1,T], but 1st resid is never shown. I think something is wrong. Statistics are inconcistent.

Please post your workfile with the estimated equation. In general, you lose the first observation since the specification for the cointegrating regressors is in differences. But I''d have to see what you are doing to give you a more precise answer.

File is attached. There eq01 is COINTREG. In it Sum squared resid = 1.673508. But if we'll get resids and calculate SSR on it, it is 1.421286838.
yf is forecast for period 2000-2011. We could see, there is no NA at 2000. Res is calculated as res = y - yf, this is residuals from first point. And its SSR is 1.673508.
So, I see that E7 tells me, that it used 11 observations, but actually it use 12 to calculate some stats.

Thanks for providing the workfile. You've found an inconsistency that I've fixed and will be updated in the next patch of EViews.

Some background. There are two samples floating around in FMOLS estimation: the sample for the cointegrating equation and the sample for the I1 regressor equations. The latter generally has fewer observations than the former due to lags dropping off. The regressor equation sample is for FMOLS since the estimator uses components that derived from the residuals of the regressor equations.

As you suspected, we were using the former for the SSR calculation and the latter when providing the residuals. You could argue (I wouldn't) that one might want to use different samples for the two calculations, but even if you were, you'd probably do it in the other direction since the FMOLS estimation uses the restricted sample. I've modified the equation code so that the SSR uses the restricted sample.

You'll have to re-estimate your equation to get the new results. Sorry for the inconvenience.

Thanks a lot!

I've found another inconsistency. =(
In the workfile R2 is calculated using resids from 1st point, but (y-mean) from 2nd point. Why? According your logic, you should calculate all stats from 2nd point.

The R2 uses the SSR calculated above. Since the first used the wrong sample so did the second. I've put the fix into the code base and it should show up in the next patch.

Patch is up, should fix this.

Ok. Thank you again)