Output on Eviews changes

[ElineP]

Hello,

when we put our data into eviews, one day we get a totally different result than another day.
Also, given a time series of +- 50 observations, a result appears. If we add the next observation (which does not deviate that much from the previous ones), again our results change drastically.
How is this possible? The time series are all the same and the estimation is also exactly the same.
How can we fix this problem?

We also noticed that when we estimate our equation, this gives an error, but when we afterwards give another equation by specifying the equation, get results and then use the first equation again (by specifying) the first equation works. Is there some kind of memory when you specify the equation?

We are using Eviews 7.

Thanks

[ElineP]

Maybe it's best if I elaborate.The equation we are using is this:

RMB = C(1)*GDP - C(2)*SPREAD + C(3)*TIME + C(1)^(-3)*d(GDP(-3)) + C(1)^(-2)*d(GDP(-2)) + C(1)^(-1)*d(GDP(-1)) + d(GDP) + C(1)*d(GDP(1)) + C(1)^(2)*d(GDP(2)) + C(1)^(3)*d(GDP(3)) + C(2)^(-3)*d(SPREAD(-3)) + C(2)^(-2)*d(SPREAD(-2)) + C(2)^(-1)*d(SPREAD(-1)) + d(SPREAD) + C(2)*d(SPREAD(1)) + C(2)^(2)*d(SPREAD(2)) + C(2)^(3)*d(SPREAD(3))

and we are trying to estimate the values of C(1), C(2) and C(3).

We don't know anymore which outcome is right and which is wrong. Possible mistakes from our side:
- Is it allowed to write C(1)^(-3) or do we have to transform to 1/(C(1)^3)?
- We get the error "^ argument is out of range"
- Maybe the equation itself is not correct.
- We are using LS and we have noticed that the starting values in the c-vector change whenever we specify the estimation. Probably that's the mistake but what do we have to do to get the right estimates? I'm guessing that the values in the c-vector are used when the estimation is made but we need to estimate the values of C(1), C(2) and C(3) themself.

I hope someone understands what I mean with this problem and can help me really soon. Also, it's kind of urgent...

Thank you

We are using Eviews 7 (Student version)

[startz]

EViews always starts a nonlinear estimation using whatever values happen to be in the C vector. And when it's done, it leaves the final results in the C vector. If you always start the estimate with the same C values (which you can set by hand) you should always end up in the same place.

[EViews Glenn]

Note that the equation object has a proc that will put the current estimated values into the coefficients in the workfile. You can use this prior to reestimation to provide additional stability to your estimates.

Click on Proc/Update Coefs from Equation. Then reestimate using the existing or an extended sample.