ABSTRACT
A shrinkage estimator for the regression parameter in a dynamic input-output linear model is presented as an alternative to the maximum likelihood estimator (MLE). It is shown that the shrinkage estimator has a lower risk than the maximum likelihood estimator when the coefficient of the lagged dependent variable is known. When the autoregressive parameter is zero or known, it is the independence that is exploited in previous work, via Stein’s lemma, which allows for finite sample dominance of the shrinkage estimator over the MLE. It happens that the dependence is the stumbling block which makes it difficult for anything other than large sample results for the case in which the autoregressive parameter is unknown.
