|
|
|
|
|
|
|
|
|
|
| By Year | By Month | By Week | Today | Search | Jump to month | ||||
|
|
|||||||||
SEMINARIO DE ESTADÍSTICA
SEMINARIO DE ESTADÍSTICA
Florian Heinrichs
RuHr Universität Bochum
"A new approach to detect changes in the mean of a time series"
Viernes, 15 de marzo, 10:30 h.
Sala 520, Departamento de Matemáticas
Resumen: When investigating the classic change point problem of one
single change point, the CUSUM principle is a good choice. The basic
idea is to compare the empirical mean of the first observations with the
empirical mean of the remaining ones. If the difference is big at some
time instance, we would derive that the mean changes over time.
Tests based on the CUSUM principle have been proven to be consistent
against a broad range of alternatives. For example, this kind of tests
are consistent against smooth and abrupt changes under certain
conditions. Contrarily to the theoretic result of consistency, in
applications with finite samples, the CUSUM principle may lead to little
power under some alternatives, like an oscillating mean function.
In this talk, a new approach to detect changes in the mean function of a
time series is introduced that is suspected to avoid the aforementioned
problem.
single change point, the CUSUM principle is a good choice. The basic
idea is to compare the empirical mean of the first observations with the
empirical mean of the remaining ones. If the difference is big at some
time instance, we would derive that the mean changes over time.
Tests based on the CUSUM principle have been proven to be consistent
against a broad range of alternatives. For example, this kind of tests
are consistent against smooth and abrupt changes under certain
conditions. Contrarily to the theoretic result of consistency, in
applications with finite samples, the CUSUM principle may lead to little
power under some alternatives, like an oscillating mean function.
In this talk, a new approach to detect changes in the mean function of a
time series is introduced that is suspected to avoid the aforementioned
problem.
Location Viernes, 15 de marzo, 10:30 h. Sala 520, Departamento de Matemáticas