Variance Estimation in a Random Coefficients Model

This papers describes an estimator for a standard state-space model
with coefficients generated by a random walk that is statistically
superior to the Kalman filter as applied to this particular class
of models. Two closely related estimators for the variances are
introduced: A maximum likelihood estimator and a moments estimator
that builds on the idea that some moments are equalized to their
expectations. These estimators perform quite similar in many cases.
In some cases, however, the moments estimator is preferable both to
the proposed likelihood estimator and the Kalman filter, as
implemented in the program package Eviews.