EViews.com

I'm struggling to make a Hausman-Taylor Estimation (TSLS estimation) from my panel data. I have more then 7000 units observed over 120months (unbalanced). As suggested I use instruments for time-varying and time-invariant variables. I want to control for period and unit effects (as suggested by the tests used). That is the reason I have chosen time series dummies on a yearly basis (total 10 years from 2000 to 2009, starting d00 to d09). When I include one more regressor (let’s call it hwm), which is time-invariant, Eviews points out near singular matrix (I guess it is because of the dummy trap). I would appreciate if someone could give me the code I need that I can estimate my equation with the additional time-invariant regressor hwm of the 7000 units:

Equations specification without the time-invariant regressor hwm:
return aum age d01 d02 d03 d04 d05 d06 d07 d08 d09

Instruments without the time-invariant regressor hwm:
iaum iage d01 d02 d03 d04 d05 d06 d07 d08 d09

Panal options:
Effect specification:
Cross section: fixed

Hi,

FEM could not be used in the case of time-invariant explanatory variable. By design of construction, The FEM will automatically eliminate the coefficient estiamte of time-invariant variable. Alternatively, you can employ Husman-Taylor model (1981) or Fixed Effect Vector Decomposition Model (FEVD). You can see FEVD application in Plumper and Troeger paper (2007) which is similar with a method proposed by Hsiao (2003). The FEVD is quite simple and can be manually perfomed.