Date of Award

August 2020

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

School of Mathematical and Statistical Sciences

Committee Member

Robert Lund

Committee Member

Xin Liu

Committee Member

Yu-Bo Wang

Abstract

Changepoint analysis has played an important role in modern time series study. Detection of changepoints helps modelling and prediction of time series and is found in applications of many fields. This dissertation focuses on the detection of mean structure changes in correlated time series. It consists of the results of three research projects on changepoint problems: (1) the comparison of changepoint techniques; (2) autocovariance estimation of an AR(p) time series with changepoints; and (3) L1-regularization in changepoint analysis.

In chapter 2 the single changepoint techniques, or At-Most-One-Changepoint (AMOC) tests are reviewed. A new AMOC test, Sum of Squared CUSUMz is developed and is shown to be the most powerful AMOC test through simulation studies on the time series with various ARMA(p,q) structures. Multiple changepoint techniques that are applicable to correlated time series are discussed in chapter $3$, which includes an in-depth discussion on the wild binary segmentation. A new distance metric is also proposed in this chapter for comparing the multiple changepoint techniques. Next in the chapter 4 a Yule-Walk moment estimator based on the first order difference is proposed for autocovariance estimation of an AR(p) time series with a small number of changepoints. The last chapter simply reviews the L1- regularization and its application to changepoint analysis.

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.