Structural breaks are abrupt changes in the data-generating process that invalidate constant-parameter models; their detection is essential in finance because undetected breaks can bias forecasts, misestimate volatility, and mis-price risk. Causes include policy shifts, market regulation changes, political events, or financial crises; consequences affect portfolio allocation, stress testing, and territorial market comparisons where regulatory regimes differ.
Classical tests for known breakpoints
The Chow test is the standard approach when a candidate breakpoint is known a priori; it compares regression fits before and after the date and is effective for single, exogenously dated changes but loses power if the date is misspecified. For endogenous dating and cases with a single unknown break, the Zivot-Andrews unit-root test developed by Eric Zivot at University of Washington addresses structural breaks in integrated series by allowing the break date to be chosen from the data, improving inference about persistence and trend properties in financial time series.Methods for unknown or multiple breaks
For multiple or unknown breakpoints, the work of Jushan Bai at Johns Hopkins University and Pierre Perron at Boston University established a rigorous framework. The Bai-Perron methodology estimates multiple break dates by minimizing the sum of squared residuals with sequential and global tests (sup-Wald, sup-F), accommodating heteroskedasticity and serial correlation. Cumulative sum procedures such as CUSUM and CUSUMSQ monitor parameter stability over time and are useful for real-time surveillance; they are simple to implement but can be sensitive to serial dependence and may require bootstrap calibration in small samples. The Quandt likelihood-ratio approach and sup-type statistics locate unknown single breaks by scanning candidate dates, trading off computational cost against detection power.Markov-switching models, notably developed and applied to macro and finance by James D. Hamilton at University of California San Diego, model regime changes as probabilistic transitions and capture persistent shifts in mean and volatility without explicit breakpoint estimation. Bayesian change-point and state-space methods offer flexible inference with posterior uncertainty on both number and timing of breaks, often preferred when prior information or hierarchical structure is relevant. Robust practice combines exploratory diagnostics, multiple tests (unit-root-aware and structural), and model validation across subperiods. Detecting breaks reliably improves risk management, regulatory assessment, and cross-border comparisons where cultural and institutional shifts drive structural change.