arXiv · math/0603047
On recursive estimation for time varying autoregressive processes
Abstract
This paper focuses on recursive estimation of time varying autoregressive processes in a nonparametric setting. The stability of the model is revisited and uniform results are provided when the time-varying autoregressive parameters belong to appropriate smoothness classes. An adequate normalization for the correction term used in the recursive estimation procedure allows for very mild assumptions on the innovations distributions. The rate of convergence of the pointwise estimates is shown to be minimax in $β$-Lipschitz classes for $0<β\leq1$. For $1<β\leq 2$, this property no longer holds. This can be seen by using an asymptotic expansion of the estimation error. A bias reduction method is then proposed for recovering the minimax rate.
Explore related subjects
Keep this discovery
Explore connections, maps & timelines
Eric Moulines, Pierre Priouret, François Roueff. 2006-03-02. On recursive estimation for time varying autoregressive processes. https://doi.org/10.1214/009053605000000624. https://arxiv.science/abs/math/0603047
Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite arXiv Science by Indiciae: BibTeX · RIS.