GMM

[Hshadmani]

Hi I am trying to replicate two papers 1. Foreign exchange market intervention and reserve accumulation in emerging Asia by Pontines and Rjan (2011)
2. Preference asymmetry and international reserve in India. by Srinivasan, Mohambare and Rmachandran (2009). I am using the command below to estimate a GMM with the same set of instruments they used in their paper , optimal weighting , and fourth lag of Newey West matrix. However, my results are not even close to those of the above papers. I want to make sure that my codes are correct. Can anyone help me figure out the issue. Thanks!

INSTRUMENTS R{1 2 3 4 5 6 7 8 9 10 12 15} e{1 2 3 4 5 6 7 8 9 10 12 15 } $
FEDFUND{0 1 2 3 4 8 15 }
LINREG(INST,OPTIMALWEIGHTS,LWINDOW=NEWEY,LAGS=4) R 2000:01 2009:07
# Constant e e2
Summarize(title="Asymmetry Parameter") 2*%beta(3)/%beta(2)

[TomDoan]

Include CONSTANT in your instruments and see what that does. Note also that "GMM" isn't necessarily well-defined for an over-identified model---particularly with a linear model, you can see people do one iteration (do 2SLS, compute the weights using the NW windows, do one GMM calculation), or can do multiple iterations. With that many instruments, the results can (unfortunately) be somewhat sensitive to the choice.

[Hshadmani]

Thank you Tom! I did the following still got similar results.
LINREG R 2000:01 2008:01
# Constant e e2
Compute %zz= %WMATRIX
INSTRUMENTS constant R{1 2 3 4 5 6 7 8 9 10 11 12 } e{0 1 2 3 4 5 6 7 8 9 10 11 12 } $
FEDFUND{1 2 3 4 5 6 7 8 9 10 11 12 }
LINREG(INST,WMATRIX=%zz, LWINDOW=NEWEY,LAGS=4) R 2000:01 2008:01
# Constant e e2
Summarize(title="Asymmetry Parameter") 2*%beta(3)/%beta(2)

[TomDoan]

INSTRUMENTS constant R{1 2 3 4 5 6 7 8 9 10 11 12 } e{0 1 2 3 4 5 6 7 8 9 10 11 12 } $
FEDFUND{1 2 3 4 5 6 7 8 9 10 11 12 }
LINREG(INST,OPTIMALWEIGHTS, LWINDOW=NEWEY,LAGS=4) R 2000:01 2008:01
# Constant e e2

would be the typical way to do that. (I originally wrote that wrong---it's 2SLS as the first step, not OLS). The two step optimal GMM is done automatically if you do OPTIMALWEIGHTS.

Do you really want E{0} as an instrument? Note that that's a LOT of instruments for 97(?) data points.

[sanjeev]

Dear Tom,
Could you please send me the code for GMM estimation for time series?
Please reply soon!


Thanks and Regards.

[TomDoan]

Dear Tom,
Could you please send me the code for GMM estimation for time series?
Please reply soon!

GMM estimation of what model?