Feb. 2027
| Intervenant : | Angelina Roche |
| Institution : | Université Paris-Cité |
| Heure : | 15h30 - 16h30 |
| Lieu : | 3L15 |
The Functional Autoregressive (FAR) model extends the classical multivariate AR(1) model to functional data. As one of the earliest models introduced in functional time series analysis, it remains a fundamental and widely used framework for modeling dependent functional data. The resulting process is a Markov chain taking values in a Hilbert space.
In this talk, we investigate minimax estimation rates for the FAR model under a prediction error criterion. We establish that these rates can be achieved by a ridge-type estimator with a suitably chosen regularization parameter. We will also briefly outline the main arguments underlying the minimax lower bound. Finally, we illustrate our theoretical findings through a simulation study.
[ Joint work with André Mas (Université de Montpellier). ]