Hi, I'm trying to use dynamic panel model (system GMM) using Lewbel (2012) estimator.
With the web dataset abdata, I use xtabond2 for dynamic panel model without Lewbel's method.
1. Now, I assume w is endogenously determined and want to instrument w with a new IV derived from k (Lewbel's method). So, I estimate the residual from the first stage and generate a new instrument lewbel_iv by multiplying k-mean_k. Then, I just put lewbel_iv into iv option of xtabond2. Is this correct way to do it?
2. What if I want to instrument w with instruments derived from both k and ys? I have no idea how to write a code for this issue.
3. I also tried to use ivreg2h to generate an instrument and apply it into xtabond2. But the result is totally different from the above one.
Any suggestions or ideas would be much appreciated.
With the web dataset abdata, I use xtabond2 for dynamic panel model without Lewbel's method.
Code:
webuse abdata xtabond2 n L.n w k ys yr*, gmm(L.n w k, laglimits(2 2) collapse) iv(ys, equation(diff)) iv(ys yr*, equation(level)) small robust twostep artest(2)
Code:
* first stage xtreg k L.n w ys yr*, fe predict ehat, residuals egen mean_k = mean(k) gen lewbel_iv = (k-mean_k)*ehat * System GMM with Lewbel estimator xtabond2 n L.n w k ys yr*, gmm(L.n w, laglimits(2 2) collapse) iv(lewbel_iv ys, equation(diff)) iv(lewbel_iv ys yr*, equation(level)) small robust twostep artest(2)
2. What if I want to instrument w with instruments derived from both k and ys? I have no idea how to write a code for this issue.
3. I also tried to use ivreg2h to generate an instrument and apply it into xtabond2. But the result is totally different from the above one.
Code:
ivreg2h n L.n w k ys yr* (w=), fe small gmm2s z(k) cluster(id) robust gen(, replace) xtabond2 n L.n w k ys yr*, gmm(L.n w, laglimits(2 2) collapse) iv(w_k_g ys, equation(diff)) iv(w_k_g ys yr*, equation(level)) small robust twostep artest(2)
Any suggestions or ideas would be much appreciated.
