Closed sondalex closed 1 year ago
I close the issue. After experimenting, it seems to apply only to just identified case. There might be computational benefit with high dimensional fixed effect (HDF) in the just identified case, (not tested), but HDF is not covered by the package anyway.
Currently the addition of time effects to an Hausman-Taylor model estimation requires adding the time dummies to the model and treating those dummies as instruments.
In my current understanding, this case could be optimized using the Two-Way fixed effects and modifying slightly the code.
The calculation of the weight $\theta$ used for WLS transformation would remain the same because the one-way within with time dummies and the two way are equivalent.
The W2SLS would be of the form:
$$ \widehat{\Omega}^{-1 / 2}WT y{i t}=\widehat{\Omega}^{-1 / 2}WT X{i t} \beta+\widehat{\Omega}^{-1 / 2}W_T Z_i \gamma+\widehat{\Omega}^{-1 / 2}WT u{i t} $$
and estimated using the same set of instruments as in the one-way case:
$$Z = [W_NX, B_NX_1 , Z_1]$$ for
random.method == "ht"
.