Estimación de la función de autocorrelación en modelos AR(1) con Métodos de Replicaciones.

The autocorrelation function (ACF) is a fundamental tool in the analysis of linear time series, among other things, for model identification. The sample estimate of the ACF is highly sensitive to the presense of outliers. The aim of this study is to compare differents estimator of the ACF proposed i...

Descripción completa

Guardado en:
Detalles Bibliográficos
Autores principales: Bussi, Javier, Marí, Gonzalo Pablo Domingo, Méndez, Fernanda
Otros Autores: Secretaría de Ciencia y Tecnología. Facultad de Ciencias Económicas y Estadística. Universidad Nacional de Rosario
Formato: conferenceObject documento de conferencia
Lenguaje:Español
Publicado: 2017
Materias:
Función de autocorrelación
Estimadores robustos
JACKKNIFE, BOOTSTRAP
JACKKNIFE
BOOTSTRAP
Acceso en línea:http://hdl.handle.net/2133/7525
http://hdl.handle.net/2133/7525
Aporte de:
Repositorio Hipermedial de la Universidad Nacional de Rosario (UNR) de Universidad Nacional de Rosario
Descripción
Sumario:The autocorrelation function (ACF) is a fundamental tool in the analysis of linear time series, among other things, for model identification. The sample estimate of the ACF is highly sensitive to the presense of outliers. The aim of this study is to compare differents estimator of the ACF proposed in the literature with five estimators based on resampling methods through the bias and the Mean Square Error (MSE). Four of these estimatos are variations based on Jackknife for time series with moving blocks. The other is an adaptation of the estimator through the method Fast and Robust Bootstratp (FRB) in order to estimate the ACF. The estimates of the lag of order 1 of the ACF of an AR(1) model were compared. The estimator based on the FRB method seems to be a serious competitor of the highly robust estimator MG as it is better in terms of bias and MSE in the cases without outliers, and found to have a similar behaviour in simulated cases with one outlier if Φ=±0.9; ±0.6. The same is true when compared with the Trun2 estimator, which has a good behavior when there are outliers. For cases with small absolute values of Φ, such as 0.3, the FRB method performance declines significantly. Estimates based on the Jackknife method behave reasonably well in the presence of outliers but canot, in any case, overcome the performance achieved by the MG estimator.

Ejemplares similares