|
Seminarium 25.04.2023
dodano: 21/04/2023
przez: Dariusz Zawisza |
- Referent : Jarosław Duda.
- Tytuł referatu : Moving estimator alternative for ARMA-ARCH, also for evolution of Student’s t shape parameter \nu.
- Abstrakt : While ARMA-ARCH assume arbitrary dependence type, I would talk about agnostic alternative: just shifting local estimator – with exponentially weakening weights (EMA). Besides, usually leading to better log-likelihood evaluation, it allows to estimate evolution e.g. of all 3 parameters of Student’s t-distribution, including nu degrees of freedom defining tail
shape: \rho(x) ~ |x|^{-\nu-1}. This \nu parameter evaluates probability of extreme events, which might be potentially catastrophic – destabilizing the market. Surprisingly, for Dow Jones 1900-2007, while usually \nu~4, turns out only in 1970s there was nearly Gaussian behavior. Article link. - Termin: Wtorek 25.04.2023, 12:15, sala 1093.
