High-order Markov chain is approximated by mixture of low order Markov chains.

Corresponding Bayesian approximate recursive estimator is proposed.

Forgetting counteracting error accumulation is tailored to this case.

A simple example related to Ultimatum game indicates the power of the approach.