Decentralized optimal control of Markov chains with a common past information set