Keywords

Online Monitoring, Changepoint Detection, High-Dimensional, Factor Model, Statistical Analysis, Economic

Abstract

This dissertation presents the Factor-Augmented Detection (FAD) algorithm, a robust methodological framework for online change point detection in high-dimensional data streams, grounded in classical factor model analysis. To tackle the challenges of monitoring large-scale datasets, the FAD algorithm decomposes data into a parsimonious set of common factors and sparsely changing idiosyncratic residuals, facilitating independent, targeted analysis of low-dimensional factor dynamics and sparse high- dimensional residual evolution.

Following decomposition, structural changes in the factor components are identified using established multivariate sequential monitoring techniques, specifically the Multivariate Cumulative Sum (mCUSUM) and Multivariate Exponentially Weighted Moving Average (mEWMA) procedures. Simultaneously, sparse alterations in the residual components are tracked using Maximum-Norm Moving Sum (MOSUM) statistics, designed to detect localized deviations. The efficacy and adaptability of the FAD method are rigorously validated through extensive simulation studies, which evaluate its performance across a range of hyperparameters, including the reference value for mCUSUM, the smoothing parameter for mEWMA, and the rolling window size for MOSUM statistics.

The algorithm is applied to a dataset of 40 U.S. macroeconomic time series. The results demonstrate its ability to accurately detect significant structural breaks in factor components, corresponding to major recessionary periods. Additionally, the MOSUM statistic effectively identifies changes in residuals, often linked to specific economic indicators, which are less consistently associated with macroeconomic downturns.

Completion Date

2025

Semester

Summer

Committee Chair

Mengyu Xu

Degree

Doctor of Philosophy (Ph.D.)

College

College of Sciences

Department

Department of Statistics & Data Science

Format

PDF

Identifier

DP0029591

Language

English

Document Type

Thesis

Campus Location

Orlando (Main) Campus

Download

Share

COinS