Durbin-Watson for pooled model

[kiber_master]

Hello!

I'm trying to estimate polled model using Least Squares Method without effects for pooled sample versus ols for the same values. As the result, all statistics are the same except one - Durbin-watson. Why? Is there some special formula for it?

OLS
Variable Coefficient
X 0.712793

R-squared -0.769297
Adjusted R-squared -0.769297
S.E. of regression 0.413207
Sum squared resid 10.07368
Log likelihood -31.60375
Durbin-Watson stat 1.739714 <---------------------

Pooled model
Variable Coefficient
X1 0.712793

R-squared -0.769297
Adjusted R-squared -0.769297
S.E. of regression 0.413207
Sum squared resid 10.07368
Log likelihood -31.60375
Durbin-Watson stat 1.740643 <---------------------

[startz]

Hello!

I'm trying to estimate polled model using Least Squares Method without effects for pooled sample versus ols for the same values. As the result, all statistics are the same except one - Durbin-watson. Why? Is there some special formula for it?

Durbin-Watson stat 1.739714 <---------------------

Durbin-Watson stat 1.740643 <---------------------

Those are the same except round-off error.

[kiber_master]

Why there could be such error? Does tolerance is different for these two methods?
Beside, I've checked the residuals and they are absolutely the same! I've calculated DW by hand and got the same result for pooled model as for ols, even for formatted residuals (with 5 decimal points). I don't believe that there is such error only in calculation!

[startz]

I suppose it's possible that there is some minor difference in the formulas. In particular, remember that the Durbin-Watson has to drop one observation for the lag. Perhaps in a pooled model EViews is dropping one for each pool member?

In any event, computer arithmetic isn't exact. And there's no substantive difference in the Durbin-Watsons you show.

[kiber_master]

Perhaps in a pooled model EViews is dropping one for each pool member?

I checked this hypothesis. ) But it doesn't give me the result. Ok, I've get what I want - formulas aren't different. Thank you.