Using a dynamic linear equation that has a conditionally homoscedastic
moving average disturbance, we compare two parameterizations of a com
monly used instrumental variables estimator to one that is asymptotica
lly optimal in a class of estimators that includes the conventional on
e. We find that, for some plausible data-generating processes, the opt
imal one is distinctly more efficient asymptotically. Simulations indi
cate that in samples of size typically available, asymptotic theory de
scribes the distribution of the parameter estimates reasonably well bu
t that test statistics sometimes are poorly sized.