Structural Break KPSS Test
Also known as: KPSS test with breaks, structural break stationarity test, KPSS break test, SB-KPSS
The structural break KPSS test extends the standard Kwiatkowski-Phillips-Schmidt-Shin (KPSS) stationarity test to allow for one or more known or unknown structural breaks in the level or trend of a time series. Under the null hypothesis the series is stationary around a broken deterministic component, enabling researchers to distinguish genuine unit-root behaviour from apparent non-stationarity caused by regime shifts.
Key highlights
- Accounts for structural breaks so that genuine stationarity around a shifted level or trend is not mistakenly rejected.
- Complements ADF and PP break tests by approaching the unit-root question from the stationarity side, helping resolve conflicting test outcomes.
- Extends naturally to multiple breaks, providing flexibility for series with several regime changes.
- Bootstrap and simulation-based critical values are available, making inference valid even in moderately small samples.
- Directly identifies the break dates as part of the testing procedure when breaks are treated as unknown.
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
Use the structural break KPSS test when you have a univariate time series that exhibits a visible shift in its mean or trend at one or more points — such as after a financial crisis, a policy change, or a structural reform — and you want to test stationarity without those shifts biasing the conclusion. It is preferable to the standard KPSS test whenever exploratory analysis (Bai-Perron tests, visual inspection) suggests at least one structural break. Do not use it when the series length is very short (fewer than ~60 observations), when the number and timing of breaks are completely unknown and the sample is too small for reliable break detection, or when the data generating process is clearly nonlinear rather than piecewise-linear.
Strengths & limitations
- Accounts for structural breaks so that genuine stationarity around a shifted level or trend is not mistakenly rejected.
- Complements ADF and PP break tests by approaching the unit-root question from the stationarity side, helping resolve conflicting test outcomes.
- Extends naturally to multiple breaks, providing flexibility for series with several regime changes.
- Bootstrap and simulation-based critical values are available, making inference valid even in moderately small samples.
- Directly identifies the break dates as part of the testing procedure when breaks are treated as unknown.
- Critical values depend on the number of breaks, the break model (level vs. trend shift), and whether breaks are known or unknown, requiring careful tabulation or simulation.
- With multiple unknown breaks, the break-date estimation step adds uncertainty and computational burden.
- The test is designed for univariate series; multivariate or panel extensions require separate methods.
- Small-sample power can be low, particularly when breaks occur near the beginning or end of the sample.
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
How does this differ from the Zivot-Andrews test?
The Zivot-Andrews test has a unit root under the null and stationarity under the alternative, whereas the structural break KPSS test reverses this: stationarity is the null and a unit root is the alternative. Using both together — with one confirming stationarity and the other failing to reject it — strengthens the evidence.
How many breaks should I allow?
Start with economic reasoning: major crises, policy shifts, or institutional changes suggest natural break candidates. For a data-driven selection, information criteria (BIC) or sequential break tests (Bai-Perron) can determine the number. With shorter series, limit the number to one or two breaks.
What if the standard KPSS rejects stationarity but the break version does not?
This outcome suggests the apparent non-stationarity was driven by a structural shift rather than a genuine unit root. The series is stationary once the break is modelled. This is a common finding for macroeconomic series around recessions or major policy changes.
Do I need to know the break date in advance?
No. The unknown-break version estimates the break date(s) from the data. However, when break dates are estimated rather than specified, the asymptotic distribution of the test statistic changes, so you must use simulation-based or bootstrap critical values rather than standard tables.
What long-run variance estimator should I use?
A kernel-based (HAC) estimator — typically Bartlett or quadratic spectral — is standard. Bandwidth selection (e.g., Andrews automatic or Newey-West fixed rules) affects the test statistic, so it is good practice to check robustness across a few bandwidth choices.
Sources
- 1.Carrion-i-Silvestre, J. L., Del Barrio, T., & Lopez-Bazo, E. (2005). Breaking the panels: An application to the GDP per capita. Econometrics Journal, 8(2), 159-175.
- 2.Kurozumi, E. (2002). Testing for stationarity with a break. Journal of Econometrics, 108(1), 63-99.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Structural Break KPSS Test. ScholarGate. https://scholargate.app/econometrics/structural-break-kpss-test