Dear Gretl community,
I am working with a VAR in which one of my endogenous variables is a
generated regressor. Specifically, I have estimated a time-varying
volatility measure from an EGarch model, and I'm using this mesure to
investigate the effects of volatility shocks.
I'm fully aware that alternative (and more fancy) econometric models would
do better than this 2-step procedure. Also because the estimated volatility
belongs to another endogenous VAR variable. However, ignoring this fact, my
question is whether gretl provides an option to account for this
measurement error problem when estimating a simple linear VAR system.
I had a look at Murphy and Topel's (1985) paper, and also tried to
find other references on how to correct the potential bias, but the
solution is not clear to me yet. I would appreciate any comments and
suggestions on this issue.
Thanks a lot,
Show replies by thread