restricted lag

[uvt9]

Hello,

I do have a problem to constrain my lag to exceed 1. When you see the output table below the sum of the lags (c(3)+...+c(10)) still exceeds one. I'm using the equation = C(1) + C(2)*OG +C(3)*I(-1) +C(4)*I(-2) +C(5)*I(-3) +C(6)*I(-4)
+C(7)*I(-5) +C(8)*I(-6) +C(9)*I(-7) +C(10)*I(-8)+(1-C(3)-C(4)-C(5)-C(6)
-C(7)-C(8)-C(9)-C(10))*I(-9).

Is there anybody how could help me with this problem.

Thank you very much,

Hans.


System: UNTITLED
Estimation Method: Least Squares
Date: 05/20/09 Time: 11:05
Sample: 10 123
Included observations: 114
Total system (balanced) observations 114

Coefficient Std. Error t-Statistic Prob.

C(1) -3.74E-05 0.000192 -0.194639 0.8461
C(2) 0.054478 0.034829 1.564142 0.1208
C(3) 0.169801 0.097054 1.749545 0.0831
C(4) 0.226616 0.095004 2.385336 0.0189
C(5) 0.039464 0.098689 0.399884 0.6901
C(6) 0.541811 0.099129 5.465703 0.0000
C(7) -0.104555 0.111760 -0.935530 0.3517
C(8) -0.096825 0.096894 -0.999286 0.3200
C(9) -0.000318 0.099079 -0.003208 0.9974
C(10) 0.275346 0.097910 2.812229 0.0059

Determinant residual covariance 3.82E-06


Equation: I= C(1) + C(2)*OG +C(3)*I(-1) +C(4)*I(-2) +C(5)*I(-3) +C(6)*I(-4)
+C(7)*I(-5) +C(8)*I(-6) +C(9)*I(-7) +C(10)*I(-8)+(1-C(3)-C(4)-C(5)-C(6)
-C(7)-C(8)-C(9)-C(10))*I(-9)
Observations: 114
R-squared 0.576425 Mean dependent var 0.003014
Adjusted R-squared 0.539770 S.D. dependent var 0.003018
S.E. of regression 0.002047 Sum squared resid 0.000436
Prob(F-statistic) 2.015860

[trubador]

As I tried to point you in the right direction before (http://forums.eviews.com/viewtopic.php?f=4&t=840), the following link really contains detailed information on the solution to your problem: http://forums.eviews.com/viewtopic.php? ... ction#p103

For instance, you can try replacing the coefficient of I(-9) with this one: (1-exp(c(3))-exp(c(4))-...-exp(c(10)))
Or, if you also want to restrict the coefficient of each lag to be between -1 and 1, then you can replace c(3), c(4), ...,and c(10) coefficients with (2*@logit(c(3))-1), (2*@logit(c(4))-1),...,and (2*@logit(c(10))-1). Of course, in this case you'll have to define the coefficent of I(-9) accordingly (with respect to @logit(.), not to exp(.)).

[uvt9]

Thank you for your quick reply.
I did try the equation with the exp in front of i(-9), but it still does not work. It might be a problem that i need the sum of the lags not to exceed one. I'm using the same specification as in the eviews guide, but i get no result.
I would appreciate it a lot, If you could take a second look.

Thank you in advance,

Hans.

[trubador]

Could you please post your workfile?

[uvt9]

The workfile is in the attachment. To test the sum of the lag I take the sample from 1 123 for the case of austria.
equation:
I= C(1) + C(2)*OG +C(3)*I(-1) +C(4)*I(-2) +C(5)*I(-3) +C(6)*I(-4)+C(7)*I(-5) +C(8)*I(-6) +C(9)*I(-7) +C(10)*I(-8)+(1-C(3)-C(4)-C(5)-C(6)
-C(7)-C(8)-C(9)-C(10))*I(-9)

[uvt9]

I'm not sure whether it works alright with the eviews file, so here the file in excel:

[trubador]

Workfile works fine, here is what I think: If there is no particular reason to restrict the sum of first eight lags to 1 other than ensuring the sum of first nine lags to be less than unity, then you really do not have to put constraints on any of your lagged variables. When you estimate the equation with assigning an additional coefficient (i.e. c(11)) to I(-9) variable, the sum of all coefficent values (from c(3) to c(11)) of the lagged variables equals to 0.754. Besides, each value already falls into the range between -1 and 1.

PS: I wonder why you are using a system object, since it seems you only have one equation.