Seminar in applied mathematics and statistics
SPEAKER: Hiroki Masuda (Kyushu University, Japan)
TITLE: Model selection for LAQ models
ABSTRACT: Model selection is a crucial step in making relative comparison between several candidate statistical models. In this talk, we will overview the recent theoretical findings about model selection for a family of locally asymptotically quadratic (LAQ) statistical models. The proposed criteria include an extension of the classical Schwarz's BIC, which can deal with both ergodic and non-ergodic stochastic process models, where the corresponding quasi-maximum likelihood estimator may of multi-scaling type (i.e. may have a different rates of convergence) and the asymptotic quasi-information matrix may be random. Some illustrative examples will be given.
Most contents are based on joint works with Dr. Shoichi Eguchi (Bernoulli, 2019).
Wednesday, March 27 at 11.15: Kathryn Colborn