Stochastic Model Solution Inversion

[cel]

I want to perform the following operation on a model object. Say I have an estimated non-linear model:

y1 = f(y2,x1,x2; coefs)+u1
y2 = g(y1,x1,x2; coefs)+u2
….
Assume x1 is a simple exog vble, in my case a parameter times a trend: x1=gamma*t.

But now assume I want to solve the previously estimated model so that the composite variable x1 is endogenous. It’s like inverting the (albeit non-linear) model so as to solve what x1 would look like against the actual data. In so doing I can see the plausibility for my original assumption about x1 (namely that it was a scaled trend).

Is it possible to do this? The manual doesn’t quite touch on this (I think?!). If I can do this, can I simulate the (inverted) model in such a way that I can do all the normal things I like, e.g., stochastic simulation with coefficient uncertainty so that when I derive the solved x1 I can also get confidence bounds around it?

Thanks!

[EViews Chris]

It sounds like you're trying to change which variables are endogenous in the model.

You talk about making z1 endogenous, but you would also need to change one of the variables that is currently endogenous (y1 or y2) into an exogenous variable (otherwise you would have two equations in three unknowns).

EViews uses the first variable name that appears within each equation to determine the endogenous variables. So if you wanted to make y1 exogenous and x1 endogenous, you could 'fool' eviews by changing your equation specification to this:

x1*0 + y1 = f(y2,x1,x2; coefs)+u1
y2 = g(y1,x1,x2; coefs)+u2

The additional x1*0 doesn't change the numbers -- it just means that x1 appears before y0 so EViews will treat it as an endogenous variable.

[cel]

Thanks. This worked a treat.

It sounds like you're trying to change which variables are endogenous in the model.

You talk about making z1 endogenous, but you would also need to change one of the variables that is currently endogenous (y1 or y2) into an exogenous variable (otherwise you would have two equations in three unknowns).

EViews uses the first variable name that appears within each equation to determine the endogenous variables. So if you wanted to make y1 exogenous and x1 endogenous, you could 'fool' eviews by changing your equation specification to this:

x1*0 + y1 = f(y2,x1,x2; coefs)+u1
y2 = g(y1,x1,x2; coefs)+u2

The additional x1*0 doesn't change the numbers -- it just means that x1 appears before y0 so EViews will treat it as an endogenous variable.