EViews.com

I am trying to test for simultaneity by estimating a 2SLS instrumental variable procedure. I estimated the two steps manually and then compared them to the Eviews preprogrammed 2SLS estimation and got extremely different results. I am obviously making a mistake in one, if not both, of the procedures. Can someone help correct my estimation mistakes?

Eviews Estimation
P (price) is the potential endogenous explanatory variable and othrsrvcs is the instrumental variable.
Equation Specification:


Instrument List:




Manual Estimation
I first regressed the potential simultaneous variable (p) on all of the other explanatory variables and the instrument (othrsrvcs).


I saved the residuals as the series pfit. I then estimated the original specification with the pfit series instead of p:

The problem is probably that you want the fitted values--not the residuals.

Thank you. I want to include both the residuals and fitted value of the possible endogenous variable correct?

I generated the fitted values by forecasting the historical observations (phat). I estimated the following:





Eviews Estimation

No.

startz wrote:No.

You are referring to including the residuals and the fitted values of the endoeneous variable? I got that from pages 6 and 7 from following source:
http://www.nyu.edu/classes/nagler/quant ... tes_oh.pdf. Is this methodology incorrect or just different?

But below are the results only using the fitted values of the endogenous variable. Is the discrepancy between methods a result of a mistake in my use of the Eviews 2SLS method?

1. What is the discrepancy you are seeing?
2. The link you posted says to use the fitted values.

startz wrote:1. What is the discrepancy you are seeing?

For example, when I use the preprogrammed 2SLS on Eviews I get the following results, which are much different than the ones I just posted.

startz wrote: 2. The link you posted says to use the fitted values.

It says that X=X^+v^.

And then says to substitute X^ into the original equation for X: Y=a+b1(X^+v^)+b2X2+u.

It says substitute in X^.
Which results are different?

startz wrote:Which results are different?

The standard errors, and therefore t-Statistics, and the summary statistics differ.

That's right. Doing the two stages manually requires a correction to the standard errors, as the document you posted states.

startz wrote:Doing the two stages manually requires a correction to the standard errors.

How can I correct the standard errors?

Since the Eviews 2SLS estimation results are poor, this would suggest that X^ is a noisy predictor of X?

If by "poor" you mean low t-statistics, yes that could be due to a noisy predictor.

Okay. Thank you very much for all your help and for walking me through this step by step.