CUSUM Test: Detecting Parameter Instability in Regression Models
Also known as: Cumulative Sum Test, CUSUMSQ Test, Brown-Durbin-Evans Test, Kümülatif Toplam Testi
The CUSUM (Cumulative Sum) and CUSUMSQ (Cumulative Sum of Squares) tests, introduced by Brown, Durbin, and Evans (1975), assess whether the coefficients of a linear regression model remain constant over time. They are standard tools in econometrics for detecting structural breaks, policy shifts, or regime changes in time-series data without requiring prior knowledge of when a break occurs.
Key highlights
- Does not require a pre-specified breakpoint date, making it suitable for exploratory structural break analysis.
- Provides a visual diagnostic: plotting W_t or S_t against the sample shows where instability begins.
- Simultaneously tests for multiple or gradual breaks across the entire sample period.
- CUSUMSQ variant captures changes in error variance, complementing the mean-shift sensitivity of CUSUM.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Apply the CUSUM test when you have a time-series regression and suspect that coefficients may have shifted at an unknown date due to policy changes, economic crises, or structural reforms. It requires OLS-estimable linear models with at least k+1 observations. The test does not require you to specify a breakpoint in advance, unlike the Chow test. CUSUM is sensitive to gradual or mean shifts; use CUSUMSQ when volatility breaks are the primary concern. The test assumes normally distributed errors; in the presence of heteroskedasticity, robust alternatives such as the Bai-Perron test may be preferable.
Strengths & limitations
- Does not require a pre-specified breakpoint date, making it suitable for exploratory structural break analysis.
- Provides a visual diagnostic: plotting W_t or S_t against the sample shows where instability begins.
- Simultaneously tests for multiple or gradual breaks across the entire sample period.
- CUSUMSQ variant captures changes in error variance, complementing the mean-shift sensitivity of CUSUM.
- Power can be low when breaks occur near the beginning or end of the sample, where few recursive residuals are available.
- Assumes normally distributed recursive residuals; non-normality inflates Type I error.
- Detects but does not estimate the location or number of breakpoints, requiring follow-up with Bai-Perron or Quandt-Andrews tests.
- Both variants may signal instability even when a model is misspecified, confounding specification error with genuine structural breaks.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
What is the difference between the CUSUM and CUSUMSQ tests?
CUSUM accumulates standardized recursive residuals and is designed to detect shifts in the mean of the coefficients over time. CUSUMSQ accumulates the squared recursive residuals as a fraction of their total and is more sensitive to changes in residual variance or sudden jumps in coefficient magnitudes. Running both together gives a more complete picture of parameter instability.
Does the CUSUM test identify where the break occurs?
No. The CUSUM test indicates that a break exists somewhere in the sample by flagging when the cumulative statistic crosses the critical boundary, but it does not pinpoint the exact date. To estimate break locations, use the Quandt-Andrews test for a single unknown breakpoint or the Bai-Perron procedure for multiple breaks.
Can CUSUM tests be applied to non-linear or VAR models?
The classical Brown-Durbin-Evans CUSUM is designed for linear OLS regression. Extensions exist for generalized linear models and system equations, and some implementations adapt the recursive residual concept to VAR settings. However, these extensions require careful verification of the underlying distributional assumptions before conclusions are drawn.
Sources
- 1.Brown, R. L., Durbin, J., & Evans, J. M. (1975). Techniques for testing the constancy of regression relationships over time. Journal of the Royal Statistical Society: Series B, 37(2), 149–192.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 2). CUSUM Test. ScholarGate. https://scholargate.app/econometrics/cusum-test