Risk-sensitive filters for identification of hidden Markov models

We derive risk-sensitive filters which can be used for both online and off-line identification of hidden Markov models. The identification is achieved by taking risk-sensitive conditional mean estimates of the number of state transitions (jumps) and occupation times and then using these values to estimate the parameters of the system. Furthermore, we demonstrate that the risk-sensitive filters approach the existing asymptotically optimal (risk-neutral) filters in the limit of the risk-sensitive parameter