Fast Recursive Algorithm for Tracking Time-Varying Autoregressive Models

In this paper we shall propose a fast recursive algorithm for tracking time-varying autoregressive models by the direct use of observed date sequences. Though this algorithm consisting of a triple of recurrences. (order-ascending, descending and time-updating one), in form resembles Morf's algorithm, it is also varid for tracking time-varying autoregressive models in contrast to Morf's one. It further wins an advantage over the usual adaptive filtering algorithm in that it can track not only fast time-varying model parameters but also their orders based on Akaike's information criterion. Finally a result of computer simulation will be given to illustrate the efficiency of this algorithm.