-

Applications of markov chain models in science and technology are varied and numerous. Successful applications depend on the accurate knowledge of chain parameters, i. e. initial distribution of states, transition probability matrix, and stationary distribution.
In this work, we intend to provide good estimates of such parameters. In order to avoid the vagaries of maximum likelihood estimators which heavily depend on the frequency counts of consecutive states visited by the chain, as well as the subjective nature of Bayesian estimators, we propose the empirical Bayes procedure which utilizes the information contained in the past data to identify the prior distribution of parameters.