EViews.com

Hi,

I'm trying to estimate the following regression :

y = c(1)*y(-1)^2 + c(2)*x

but the resulting R-squared (unadjusted) is much lower than when I regress the following equation :

y = c(2)*x.

How could it possibly happen if c(1) could be 0?

Any ideas to get around this problem would be greatly appreciated!

Sorry to insist here but I badly need an answer on this one... everytime I add a regressor to my equations the fit is reduced and this doesn't seem normal!

Quite possibly the sample is different.

Hi startz,

thank you very much for posting help on this one! It was badly needed. Have a look at my estimation outputs below.. sample looks the same

Okay, that is indeed weird.

The AR terms make the estimation very slightly nonlinear, but I don't think that accounts for what's going on.

You might post your workfile and see if anyone sees something.

here it is :)

thanks again!!

better use long version instead... sorry

Quelle malchance. C'est dommage.

The answer is that you're having very bad luck.

Including the AR terms makes the estimates very slightly nonlinear.

Move REGRESSEURJOUR to the end of the regression list. Then estimate the equation without REGRESSEURJOUR. Now estimate the longer equation, setting the option for starting values at user supplied. You'll get a higher R Squared.
Dependent Variable: VARLOGEPARGNEANNUEL
Method: Least Squares
Date: 08/04/11 Time: 12:17
Sample (adjusted): 2005M01 2009M12
Included observations: 60 after adjustments
Convergence achieved after 16 iterations

Variable Coefficient Std. Error t-Statistic Prob.

JEUDI 0.002660 0.002063 1.288910 0.2030
VENDREDI -0.003813 0.001034 -3.686594 0.0005
JEUDI_JOUR_8 -0.004328 0.001970 -2.197008 0.0324
VENDREDI_JOUR_8 0.001125 0.001840 0.611434 0.5435
REGRESSEURJOUR 0.154670 0.048400 3.195646 0.0024
AR(1) 0.949231 0.149572 6.346326 0.0000
AR(2) 0.048067 0.149842 0.320787 0.7496

R-squared 0.919392 Mean dependent var 0.070122
Adjusted R-squared 0.910267 S.D. dependent var 0.019208
S.E. of regression 0.005754 Akaike info criterion -7.368595
Sum squared resid 0.001755 Schwarz criterion -7.124255
Log likelihood 228.0578 Hannan-Quinn criter. -7.273020
Durbin-Watson stat 1.866333

Inverted AR Roots 1.00 -.05

wow! that worked just awesome!

c'est fantastique!

thank you forever...