Proceedings of Abstracts of the 6th. International Conference on Data - Algorithms - Decision Making. - Praha : ÚTIA AV ČR, v.v.i, 2010
S. 53-53
Poč.str.
1 s.
Akce
6th International Workshop on Data–Algorithms–Decision Making
Datum konání
2.12.2010-4.12.2010
Místo konání
Jindřichův Hradec
Země
CZ - Česká republika
Typ akce
EUR
Jazyk dok.
eng - angličtina
Země vyd.
CZ - Česká republika
Klíč. slova
robust ; bayesian ; auto-regression
Vědní obor RIV
BB - Aplikovaná statistika, operační výzkum
CEP
1M0572 GA MŠMT - Ministerstvo školství, mládeže a tělovýchovy
GA102/08/0567 GA ČR - Grantová agentura ČR
CEZ
AV0Z10750506 - UTIA-B (2005-2011)
Anotace
The problem of estimating parameters of an auto-regression model in a Bayesian paradigm has been solved before, when the model has innovations coming from exponential family. The main reason for choosing exponential family was the simplicity of computation and the fact that Gaussian distribution, often found in nature due to existence of limit theorems, is also a member of this family. Applications of modeling to data, where the distribution of innovations is known to be heavy-tailed calls for a method, more robust with respect to possible outliers. We choose the 1-D innovations of the model to be Laplace distributed, choose a Bayesian conjugate prior to such a model distribution and try to compute the resulting filtration, when new data of a realization of an adjacent random process arrive. The computation of the resultant posterior distribution of the parameters of the model is still computationally tractable as will be shown.