GMM estimation

[atbui]

I am following Chinn et al(2004) to reestimate UIP condition. However, Chinn uses GMM estimator of Hansen(1982) to correct standard errors of the parameter estimates for MA serial correlation. Could anyone help me to write the code of UIP model using GMM estimator of Hansen(1982)

Thank you and best regards

[TomDoan]

Please give complete citations. I assume you mean "Testing Uncovered Interest Parity at Short and Long Horizons
during the Post-Bretton Woods Era" which is now a 2005 working paper, not 2004.

The single equation estimates are just done with LINREG(LWINDOW=NEWEY,LAGS=maturity-1). (According to footnote 6: A Bartlett window is used
instead, to guarantee positive semi-definiteness of the variance-covariance matrix. Bartlett and Newey-West are synonymous). The joint (fixed effects) is done with SUR(ROBUSTERRORS,LWINDOW=NEWEY,LAGS=maturity-1) with parameters EQUATE 2 to force the coefficients to be equal while allowing the intercepts to vary.

[atbui]

Thank you very much Tom

The paper I am following is Chinn et al (2004) "Monetary Policy and Long-Horizon Uncovered Interest Parity" IMF staff papers, Vol 51, No3. In page 414 he said that " we used the GMM estimator of Hansen (1982) to correct the standard errors of the paprameter estimates for MA serial correlation of order k-1"

Could you please explain me (1) the difference between GMM and OLS for this model; is it simply the use of Newey-West correction with lags=k-1?

Many thank

[TomDoan]

What he's describing wouldn't generally be called "GMM" nowadays, but using OLS with HAC or autocorrelation robust or Newey-West or ... standard errors. Both the GMM estimators and the correction for "sub-optimal" estimators date to the same Hansen GMM paper. If someone says he is using Hansen or Hansen-Hodrick standard errors, that will mean the flat window; LAGS=m-1,LWINDOW=FLAT. However, those have an annoying tendency to produce non-positive-definite covariance matrices. In the subsequent paper, he shifted to Newey-West/Bartlett standard errors.