Least squares identification of autoregressive models with time-varying parameters

The estimate of the parameters of time-varying autoregressive models is often performed with the recursive least squares algorithm equipped with exponential forgetting. In this paper, we study the properties of these estimates when the parameter variation is governed by a stable equation subject to an L²-bounded drift. Under suitable assumptions, we show that if the forgetting factor is large enough then the tracking error keeps bounded, and has an interesting expression