The impact of duration response measurement error on
parameter estimates is investigated by using small variance approximations. The
inconsistency of GMM estimators that ignore it is studied for special cases of
single spell models, allowing for the presence of independent right censoring,
and for two spell lagged duration dependence model. The results suggest an
approximate bias corrected GMM estimator for the parameters of the error free
and measurement error distributions. When the error free density is known
identification of the extended parameter vector is achieved by using the score for
the variance of the measurement error under the null employed in a measurement
error sensitive specification test as an additional moment condition. The results
are applied to unemployment duration data from the BHPS.
Presented by:
Montezuma Dumangane (ISEG, Technical University of Lisbon)
Date & time:
January 30, 2006 4:00 pm - January 30, 2006 12:00 am
External seminars home