EViews.com

Hi masters of EViews and Statistics and Econometrics,

I am replicating some forecasting models. The one i am dealing with now is: Y=17139*x^-1.093
How can i estimate the same equation on Eviews. As it is a power function, apart from y and x variables to upload on eviews, how can i make sure that i define the constant=17139 and the exponent -1,093, so that the estimation recognizes my formula then?

Thanks a lot!
p.s. with regression analysis on excel, i found out that the residuals are not randomly distributed, i.e. strong autocorrelation, which might indicated that the model is not an appropriate one. In order to run more analysis, i would like to do all this with eviews, so that i can get more feedback and much quickly.
I would appreciate any suggestions on validating, evaluating existing forecasting models (linear, non-linear, etc.). This could be a great topic for me to read through!

What are you estimating? There are no unknown coefficients in that expression.

You're right!
What i mean, is how can i reproduce the same equation with Eviews? I want to estimate Y with the explanatory variable X, with the condition that i get the same coefficient and exponent, as in the original model.
I am trying to reproduce on Eviews the power function (as given above) that the authors have estimated in 2003. Like this i can run more analysis on this model an try to find any incovenience or misspecification.
I hope this makes more sense now.

You can only replicate it if you have exactly the same data. Assuming you do, then create a new equation and enter the specification as:

Thank you so much. I have almost the exact function, so i will consider this as ok.


Dependent Variable: IU_AT
Method: Least Squares
Date: 05/07/14 Time: 00:58
Sample: 1977 2000
Included observations: 24
Convergence achieved after 33 iterations
IU_AT=C(1)*GDP_CAP_AT^C(2)

Coefficient Std. Error t-Statistic Prob.

C(1) 17124.90 3611.923 4.741213 0.0001
C(2) -1.092480 0.072092 -15.15406 0.0000

R-squared 0.914266 Mean dependent var 682.9475
Adjusted R-squared 0.910369 S.D. dependent var 101.3751
S.E. of regression 30.35009 Akaike info criterion 9.743132
Sum squared resid 20264.82 Schwarz criterion 9.841303
Log likelihood -114.9176 Hannan-Quinn criter. 9.769176
Durbin-Watson stat 0.821953


And this is how the residuals look like in Excel, and Eviews. Do you see the difference and what i meant? Where am i going wrong? How could i verify that this model is a good one or has misspecifications?

on excel:
- no randomness of residuals
- durbin-watson = 0,19

hence, i was thinking that there was a misleading model.

Did Eviews, maybe adjust something automatically from this model? Or do you still see any inconvenience in the residuals graph? Histogram attached.

First of all, the two results are not "close". Nonlinear models are sensitive to initial conditions as well as to the changes in parameter values. And they are also susceptible to local optima. In this case, however, you can linearize the model by applying logarithmic transformation to the variables. You can then run a simple OLS and this should yield the same output for EViews and Excel.