Text only print
Full screen print
Colloquium
Seminar information archive ～04/28｜Next seminar｜Future seminars 04/29～
| Organizer(s) | ASUKE Taro, TERADA Itaru, HASEGAWA Ryu, MIYAMOTO Yasuhito (chair) |
|---|---|
| URL | https://www.ms.u-tokyo.ac.jp/seminar/colloquium_e/index_e.html |
2023/05/19
15:30-16:30 Room #大講義室 (Graduate School of Math. Sci. Bldg.)
If you do not belong to Graduate School of Mathematical Sciences, the University of Tokyo, please apply from the form at https://forms.gle/J4Wo8N6CbLmYiprUA.
Hiroki Masuda (Graduate School of Mathematical Sciences, the University of Tokyo)
Locally stable regression (日本語)
If you do not belong to Graduate School of Mathematical Sciences, the University of Tokyo, please apply from the form at https://forms.gle/J4Wo8N6CbLmYiprUA.
Hiroki Masuda (Graduate School of Mathematical Sciences, the University of Tokyo)
Locally stable regression (日本語)
[ Abstract ]
A non-ergodic model structure naturally emerges in estimating a stochastic process model observed at high frequency over a fixed period. The probability structure of the driving noise determines whether or not the characteristics of the model can be statistically estimated. However, it is difficult to describe the possible phenomena in general when the noise is non-Gaussian. Building on such backgrounds, we will present some recent results on non-ergodic regression modeling driven by a locally stable Lévy process: the construction of an explicit non-Gaussian quasi-maximum likelihood and the asymptotic distribution of the corresponding estimator. We will also present a method for relative model comparison and its theoretical property.
A non-ergodic model structure naturally emerges in estimating a stochastic process model observed at high frequency over a fixed period. The probability structure of the driving noise determines whether or not the characteristics of the model can be statistically estimated. However, it is difficult to describe the possible phenomena in general when the noise is non-Gaussian. Building on such backgrounds, we will present some recent results on non-ergodic regression modeling driven by a locally stable Lévy process: the construction of an explicit non-Gaussian quasi-maximum likelihood and the asymptotic distribution of the corresponding estimator. We will also present a method for relative model comparison and its theoretical property.
