Restricted OLS with robust (HAC) standard errors

[yelena]

Dear Tom,
Could you please help me? Do you know how to run a RESTRICTED OLS regression (say, we restricted three coefficients) with robust (HAC) standard errors? Thank you very much in advance.

[TomDoan]

LINREG(ROBUSTERRORS) followed by RESTRICT(CREATE) estimates the coefficients by finding the closest coefficient vector which meets the restriction, where "close" is defined in terms of the distance using the HAC covariance matrix. That will not be the same as the restricted least squares estimate which is the closest using the OLS covariance matrix, but will be a reasonable estimator if the restrictions aren't strongly rejected. If you want the actual least square estimator with HAC standard errors, you have to either use NLLS or ENCODE/UNRAVEL to directly estimate the restricted model and apply ROBUSTERRORS to those.

[yelena]

Thank you, Tom, very much for your help.