Dear Statalists,
Recently, I have started looking into Quantile regression, and I found a couple of packages, including but not limited to: xtqreg, ivqreg2, qregpd, genqreg, mmqreg, and so on. They are designed to do different things of course, but intrinsically are quantile regressions (QR).
The type that piqued my interest was QR with instrumental variables, including the ivqreg2 and qregpd (Since I'm working with panel data). These QR types allow for a lot of regression specifications of the same Y X Control set through the switching of instrumented variables, yet for my case, I find that all my coefficients are insignificant across all Quantiles? With P-values from 0.9... to 1.0. I wonder if I'm doing something wrong, here is a sample code structure with qregpd (Called QRPD):
local Step = 0
forval X = 1/10 {
local Step = `Step' + 0.1
qregpd Y L1_Y X1 X1_Sq L1_X2 L1_X3 L1_X4, q(`Step') id(ID) fix(Year) instr (Z1 X1 X1_Sq L1_X2 L1_X3 L1_X4)
}
In which I instrumented all the RHS (Right-Hand-Side) variables along with another Z1 variable. If I don't instrument anything, the QRPD becomes a restricted QR, and all RHS Variables are considered exogenous, else the specified variables in instr are exogenous, the rest not in it are endogenous.
Also, if anyone is in the position to elaborate, are we allowed to use lags of RHS Variables as additional instrument should be suspect the variable itself to be endogenous? Like in the case of xtabond2 (Two-step system GMM) where the specified variables in the gmmstyle are endogenous and instrumented with its own lags.
Thank you very much for your time!
Recently, I have started looking into Quantile regression, and I found a couple of packages, including but not limited to: xtqreg, ivqreg2, qregpd, genqreg, mmqreg, and so on. They are designed to do different things of course, but intrinsically are quantile regressions (QR).
The type that piqued my interest was QR with instrumental variables, including the ivqreg2 and qregpd (Since I'm working with panel data). These QR types allow for a lot of regression specifications of the same Y X Control set through the switching of instrumented variables, yet for my case, I find that all my coefficients are insignificant across all Quantiles? With P-values from 0.9... to 1.0. I wonder if I'm doing something wrong, here is a sample code structure with qregpd (Called QRPD):
local Step = 0
forval X = 1/10 {
local Step = `Step' + 0.1
qregpd Y L1_Y X1 X1_Sq L1_X2 L1_X3 L1_X4, q(`Step') id(ID) fix(Year) instr (Z1 X1 X1_Sq L1_X2 L1_X3 L1_X4)
}
In which I instrumented all the RHS (Right-Hand-Side) variables along with another Z1 variable. If I don't instrument anything, the QRPD becomes a restricted QR, and all RHS Variables are considered exogenous, else the specified variables in instr are exogenous, the rest not in it are endogenous.
Also, if anyone is in the position to elaborate, are we allowed to use lags of RHS Variables as additional instrument should be suspect the variable itself to be endogenous? Like in the case of xtabond2 (Two-step system GMM) where the specified variables in the gmmstyle are endogenous and instrumented with its own lags.
Thank you very much for your time!
Comment