Trend Analysis of Structural Shifts in U.S. Industry Indicators: A Change-Point Approach
Rahman, Zareen (2025)
2025
Master's Programme in Computing Sciences and Electrical Engineering
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2025-11-10
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-2025111010486
https://urn.fi/URN:NBN:fi:tuni-2025111010486
Tiivistelmä
This thesis investigates how key U.S. industries have changed over the past decade by looking at both their overall trends and the moments when those trends transformed. Instead of describing long-term patterns, the study focuses on detecting the timing and stability of these turning points. Official datasets from the Bureau of Economic Analysis (BEA) and the Bureau of Labor Statistics (BLS) were analyzed using three change-point detection methods: the non-parametric Pettitt test for identifying a single change, and the Bai-Perron and PELT algorithms for detecting multiple shifts. To control the risk of false discoveries when testing multiple industries, both the Benjamini-Hochberg false discovery rate (FDR) and the Bonferroni correction were applied.
Following change-point identification, the Mann-Kendall test and Theil-Sen slope estimation were employed to examine pre- and post-change trends, highlighting whether industry growth accelerated, slowed, or reversed direction. A sliding-window framework (5-9-year windows with one-year shifts) was implemented to assess the stability of detected breaks across varying time spans, and simulation experiments were used to evaluate method sensitivity and validity under different noise conditions.
Results indicate concentrated turning points for many industries around 2014-2016 and 2020, with method agreement highest when segment lengths exceed 6-7 years. The study establishes a compact and reproducible workflow for change detection, significance control, and trend interpretation that prioritizes linking statistical evidence to real-world economic context. Overall, it presents a clear, systematic framework for identifying and validating structural changes in short economic time series, combining trend analysis with change-point detection, to present a transparent view of when and how U.S. industries have evolved, supporting better and informed decision-making for economists, policymakers, and business leaders.
Following change-point identification, the Mann-Kendall test and Theil-Sen slope estimation were employed to examine pre- and post-change trends, highlighting whether industry growth accelerated, slowed, or reversed direction. A sliding-window framework (5-9-year windows with one-year shifts) was implemented to assess the stability of detected breaks across varying time spans, and simulation experiments were used to evaluate method sensitivity and validity under different noise conditions.
Results indicate concentrated turning points for many industries around 2014-2016 and 2020, with method agreement highest when segment lengths exceed 6-7 years. The study establishes a compact and reproducible workflow for change detection, significance control, and trend interpretation that prioritizes linking statistical evidence to real-world economic context. Overall, it presents a clear, systematic framework for identifying and validating structural changes in short economic time series, combining trend analysis with change-point detection, to present a transparent view of when and how U.S. industries have evolved, supporting better and informed decision-making for economists, policymakers, and business leaders.