Graduate Thesis Or Dissertation
 

Estimation and Sparse Selection of Conditional Probability Models for Vector Time Series

Öffentlich Deposited

Herunterladbarer Inhalt

PDF Herunterladen
https://ir.library.oregonstate.edu/concern/graduate_thesis_or_dissertations/k0698g13s

Descriptions

Attribute NameValues
Creator
Abstract
  • Diverse scientific fields collect multiple time series data to investigate the dynamical behavior of complex systems: atmospheric and climate science, geophysics, neuroscience, epidemiology, ecology, and environmental science. Identifying patterns of mutual dependence among such data generates valuable knowledge that can be applied either for inferential or forecasting purposes. Vector autoregressive (VAR) processes provide a flexible class of statistical models for multiple time series that are easy to estimate using regression techniques. However, scaling to large data sets and extension to more general processes stretch the framework's capacity: due to the dense parametrization of VAR models, which have one parameter for every possible pairwise relationship between components (i.e., between each univariate time series in a collection), high-dimensional data generate difficulties associated with model selection and parametric regularization; and modeling more general processes requires data transformations that complicate inference, forecasting, and model interpretation. Fields such as epidemiology, neuroscience, and ecology generate high-dimensional time series of count vectors, which incur both sets of challenges at once. Autoregressive conditional probability models --- models in which the conditional means of a time series follow an autoregressive structure in the process history --- are natural generalizations that preserve ease of estimation and, in conjunction with selection methods in regression, promise to address challenges associated with modeling large multiple time series of count (and other discrete) data. This thesis focuses on developing empirical methodology for sparse selection of nonlinear VAR-type conditional Poisson models. Chapter 1 provides an overview of related existing work. Chapter 2 develops an empirical method for sparse selection in VAR models based on resampling methods. Chapter 3 presents a conditional probability generalization of the VAR model and analyzes the stability properties of Poisson generalized vector autoregressive (GVAR) processes. Chapter 4 combines the work of the preceding two chapters and develops a resampling-based method for sparse estimation of Poisson GVAR models. Finally, Chapter 5 summarizes key findings, challenges, and future work.
License
Resource Type
Date Issued
Degree Level
Degree Name
Degree Field
Degree Grantor
Commencement Year
Advisor
Committee Member
Academic Affiliation
Urheberrechts-Erklärung
Publisher
Peer Reviewed
Language

Beziehungen

Parents:

This work has no parents.

In Collection:

Artikel

Miniaturansicht Titel Hochladedatum Sichtbarkeit Aktionen
file details RuizTrevorD2020.pdf 2020-09-01 Öffentlich Herunterladen