Chiang Mai Journal of Science

Print ISSN: 0125-2526 | eISSN : 2465-3845

1,647
Articles
Q3 0.80
Impact Factor
Q3 1.3
CiteScore
7 days
Avg. First Decision

Effect of Preliminary Unit Root Tests on Prediction Intervals for Gaussian Autoregressive Process with Additive Outliers

Wararit Panichkitkosolkul and Sa-aat Niwitpong
* Author for corresponding; e-mail address: snw@kmutnb.ac.th
Volume: Vol.39 No.1 (JANUARY 2012)
Research Article
DOI:
Received: 25 May 2011, Revised: -, Accepted: 30 September 2011, Published: -

Citation: Panichkitkosolkul W. and Niwitpong S., Effect of Preliminary Unit Root Tests on Prediction Intervals for Gaussian Autoregressive Process with Additive Outliers, Chiang Mai Journal of Science, 2012; 39(1): 8-29.

Abstract

The  preliminary  unit  root  test  has  been found  to be  a  useful tool  for  improving  the  accuracy  of a  one-stop-ahead  predictor  and  prediction  interval  for  the first-order  autoregressive  process  when  an  autoregressive  coefficient  is  close  to  one. This paper  applies  the  aforementioned  concepts  of the  preliminary  unit root  test  in  order  to  improve the efficiency  of  prediction  intervals  for  the  Gaussian  autoregressive  process  with additive  outliers. The  preliminary unit  root  tests  considered  are  the  augmented  Dickey-Fuller (ADF) test  and  the  Shin  et  al.’s (SSL)  test. In  addition, the  analytic  expressions  of  the  coverage  probability  of  prediction  intervals  are  derived, and the  structure  of  the coverage  probability  was  proved  to  be  independent  from  the mean of  the  process  and the  parameter  of  the  innovation, it  is  a function  of  the  autoregressive  coefficients  only. For  the  parameter  estimation  of processes  we  use  the generalized  M-estimates. The  coverage  probabilities  and the widths  of  the  standard  prediction  interval, the prediction  interval  following  the ADF  test, and  the prediction  interval  following  the  SSL  test  are  also  compared  via  simulation  studies. Simulation  results  have shown  that the SSL  test  can help  to  improve  the accuracy  of  the  prediction  intervals  with  additive  outliers, especially  when  the  sample  size is  large. The  performance  of the  proposed  prediction  intervals  is  illustrated  with  an  empirical  application.  

Keywords: additive outlier, autoregressive process, prediction interval, preliminary unit root test

Related Articles

On Preliminary-prediction Intervals for the Difference Between Two Means with Missing Data
page: 21 - 28

Sa-aat Niwitpong, Pawat Paksaranuwat, and Suparat Niwitpong

Vol.37 No.1 (JANUARY 2010)
Research Article View: 908 Download: 280
New Estimator for an Unknown Mean Gaussian AR(1) Process with Additive Outliers
page: 14 - 20

Wararit Panichkitkosolkul

Vol.37 No.1 (JANUARY 2010)
Research Article View: 967 Download: 220
Outline
Figures