Nonparametric identification of autoregressions

This paper considers the problem of constructing probability models for processes of an autoregressive type, which means estimating unknown parameters of processes and constructing one-step predictions of observations based on these estimates and reconstructing the limit probability density of prediction errors.