An empirical Bayes approach to modeling and control of stochastic systems with time-varying parameters

An empirical Bayes approach is proposed for modeling the dynamics of unknown parameters, which may undergo both regular fluctuations and erratic changes over time, in stochastic regression models and linear stochastic difference equations. A rich and flexible class of empirical Bayes models of parameter dynamics is shown to lead to tractable recursive algorithms for estimating the time-varying parameters with good statistical properties. Applications of these recursive estimators to developing adaptive controllers of certainty-equivalence type are also discussed.<
>