Constraining estimation results

[jynterpek]

Hi there,

I hope ya'll are doing well. I am formulating a non-linear regression model and I'm getting an error when I try to estimate parameters. I've started using EViews only recently and so maybe I'm missing something basic.

Following is a sample of the model I'm trying to estimate -

y = c(1) + c(2)*log(1+x1) + c(3)*log(1+x2 + c(4)*x3) + [ar = c(5)]

If try to estimate the c(i), I keep getting an error message that says "Log of non positive number" I suspect it is because c(4) turns out to be negative in the estimation process. Is there a way I can constrain a variable to ensure that it remains positive?

If you have any other thoughts on the source of the error message, please let me know.

Thanks,
~D

[EViews Gareth]

Please see the following post:
viewtopic.php?f=4&t=48

[jynterpek]

Thank you for your reply. I had used couple of transformations but it yielded in different errors

Squared transformation of the variable -
y = c(1) + c(2)*log(1+x1) + c(3)*log(1+x2 + c(4)^2*x3) + [ar = c(5)] ---> yields an error "overflow"

Exponential transformation of the variable -
y = c(1) + c(2)*log(1+x1) + c(3)*log(1+x2 + exp(c(4))*x3) + [ar = c(5)] ---> yields an error "singular matrix"

Do you think it is perhaps because the function has too many parameters to estimate?

~D

[EViews Gareth]

Just sounds like your function is so non-linear that EViews is having trouble finding a solution. You could try changing the starting values (i.e. the values in your C vector) to see if it helps.