Mathematics Ph.D. Dissertations

Empirical Likelihood For Change Point Detection And Estimation In Time Series Models

Date of Award

2017

Document Type

Dissertation

Degree Name

Doctor of Philosophy (Ph.D.)

Department

Statistics

First Advisor

Wei Ning (Committee Co-Chair)

Second Advisor

Arjun Gupta (Committee Co-Chair)

Third Advisor

Junfeng Shang (Committee Member)

Fourth Advisor

Robert Midden (Committee Member)

Abstract

Empirical Likelihood (EL) method introduced by Owen (1988) is a widely used nonparametric tool for constructing confidence regions due to its appealing asymptotic distribution of the likelihood-ratio-type statistic which is same as the one under the parametric settings. However, the EL method was introduced to be used for independent data, hence it becomes difficult to apply it to dependent data such as time series data. Owen (2001) suggested using the conditional likelihood to remove the dependence structure and generate the estimating equations. Monti (1997) developed the idea of extending the EL method to short-memory time series models using the Whittle’s (1953) estimation method to obtain an M-estimator of the periodogram ordinates of a time series which are asymptotically independent. This reduces a dependent data problem into an independent data problem. Nordman and Lahiri (2006) also formulated a frequency domain empirical likelihood (FDEL) using spectral estimating equations which can be used for short- and long- range dependent data. FDEL applies a data transformation which weakens the dependence structure of the data hence, allowing to use the EL method for the transformed data which is considered to be asymptotically independent.

Unfortunately, there is a good chance that the solution to the profile empirical likelihood function computation which involves constrained maximization does not exist which raises some computational issues as mentioned by Chen et al. (2008). To overcome this difficulty, Chen et al. (2008) proposed an adjusted empirical likelihood (AEL) ratio function by adding a pseudo term to guarantee the zero to be an interior point of the convex hull, therefore, the required numerical maximization is guaranteed to have a solution always. This dissertation focuses on developing novel nonparametric tests based on the empirical likelihood to estimate and detect changes in parameters of various times series models. First part is focused on the AEL for short-memory time series models such as autoregression (AR), moving average (MA), autoregressive moving average (ARMA), etc. I incorporated Monti’s (1997) approach along with Nordman and Lahiri’s (2006) formulation, to propose an AEL for short-memory dependence data. In the second part, an AEL-type statistic has been established for long-memory time series models suggested by Yau (2012). The third part of the dissertation focuses on the detection of changes in structures of time series models based on the EL method. Real data sets are used in each section to illustrate the performance of the proposed methods.

Link to Full Text

Share

COinS