"overflow of analytical derivatives" error

[purdued]

Hi all,

I'm currently trying to estimate an non-linear STR with the following:

smpl @all

series dr_uk = r_uk - r_uk(-1)
series ldr_uk = dr_uk(-1)
series t = @trend + 1

coef(1) gamma
coef(1) lambda
coef(7) beta
coef(7) theta


LS R_UK = (1 - 1/(1 + @exp(-gamma(1)*(t - lambda(1)))))*(BETA(1) + BETA(2)*R_UK(-1) + BETA(3)*LDR_UK + BETA(4)*PI_GUK + BETA(5)*Y_GUK + BETA(6)*PI_UK + BETA(7)*CLI_UK) + (1/(1 + @exp(-gamma(1)*(t - lambda(1)))))*(THETA(1) + THETA(2)*R_UK(-1) + THETA(3)*LDR_UK + THETA(4)*PI_GUK + THETA(5)*Y_GUK + THETA(6)*PI_UK + THETA(7)*CLI_UK)

I get an error saying "overflow of analytical derivatives" although I get estimates for my coefficients. What I don't get is a equation object.

Anyone knows what's happening here?

I'm new to eviews so I'm hoping its something stupid.

Thanks
Dave

[EViews Glenn]

Can you post your workfile?

[purdued]

Hi Glenn,

Here's the workfile, sorry for the delay.

I have a feeling all this is due to the spec of the pc I'm working on.

str_data.WF1

(123.58 KiB) Downloaded 740 times

[trubador]

Try assigning initial values to coefficients other than zero:

Code: Select all

coef(1) gamma = .1 coef(1) lambda = .1 coef(7) beta = .1 coef(7) theta = .1

[purdued]

Thanks Trubador.

Changing initial values helps alright. the model is very sensitive to what you set them to though.

I can set the initial values to non-zero numbers and have gotten equation objects while other non-zero numbers will result in the error will pop up.

Is it possible a figure is tending to zero or infinity in the iterative process and that cuts it off (i'm not sure - just throwing ideas out there).

Dave

[dsokhi]

Hi,

I am getting the same "overflow in analytical derivatives" error as soon as I add a dummy variable. Number of observations is greater than the number of variables on RHS. Also I tried to set the initial values of coefficients in 'c' to non-zero numbers (1 and -1).

Could you please let me know the reason why I get this error and how can I overcome it?

Thanks.

Note: Sorry I can't post the eviews file due to confidentiality of data!

[EViews Glenn]

Can you look at the Derivatives display to see the analytic derivatives? You can try to evaluate these expressions one by one to see which one is giving you problems.