Structural Break Johansen Cointegration Test
Johansen Cointegration Test with Structural Breaks · Also known as: Johansen cointegration with breaks, break-robust Johansen test, cointegration test with regime shifts, structural change Johansen VECM
The structural break Johansen cointegration test extends the standard maximum-likelihood Johansen procedure to settings where the multivariate time series exhibits level shifts or trend breaks. By incorporating dummy variables or shift regressors into the VECM, the test determines the cointegrating rank without confounding genuine long-run relationships with regime changes.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use this test when you have two or more I(1) time series and suspect a structural break — for example, financial crises, monetary regime changes, or major policy reforms that shift levels or trends. The method is appropriate when a prior test (Zivot-Andrews, KPSS, or Chow-type) suggests a break, and when the sample is long enough (at least 100 observations is advisable) to estimate both the break location and the cointegrating rank reliably. Avoid it when no prior evidence of a break exists, since over-parameterisation from unnecessary break terms reduces power. Also avoid when the series are stationary or I(2), or when you have fewer than three variables and prefer the simpler Engle-Granger approach.
Strengths & limitations
- Prevents spurious conclusions: accounts for breaks that would otherwise bias the rank test toward over- or under-rejecting cointegration.
- Multivariate framework tests for multiple cointegrating vectors simultaneously, unlike the single-equation Engle-Granger test.
- Allows explicit estimation of the break timing and type within a unified likelihood framework.
- Produces a fully specified VECM for out-of-sample forecasting and impulse-response analysis after the break is accommodated.
- Break-adjusted critical values are tabulated and available in standard software (JMulTi, EViews, Stata sbvar package).
- Requires pre-specification or prior estimation of break date and type, introducing a two-step uncertainty that inflates effective Type I error.
- Break-adjusted asymptotic distributions assume a single break; multiple breaks require further extensions and are more demanding.
- Demands longer samples than the standard Johansen test; rank identification is unreliable with fewer than roughly 100 observations.
- Sensitive to lag length choice: too few lags leave residual autocorrelation; too many consume degrees of freedom that small samples cannot afford.
- Results depend on whether the break is placed in the cointegrating space, the deterministic trend, or the short-run part, which requires careful economic reasoning.
Frequently asked
How do I choose the break date?
Use economic reasoning first (a known crisis or policy event). If the break date is unknown, apply a structural-break unit-root test such as Zivot-Andrews or the Saikkonen-Lütkepohl pre-test to each series, or search for the break that maximises a likelihood criterion — but account for the resulting pre-test uncertainty when interpreting significance levels.
Can I test for more than one break?
Single-break asymptotic critical values are the standard tabulated option. For two or more breaks, some software provides simulation-based critical values, but the procedure becomes considerably more demanding and small samples rarely support reliable rank identification with multiple breaks.
What is the difference between a level-shift and a trend-shift break?
A level shift is a one-time permanent jump in the mean of a variable, modelled by a step dummy. A trend shift is a change in the slope of a deterministic trend, modelled by a kinked trend term. The type of break affects which critical-value table to use and how to specify the deterministic component in the VECM.
How does this differ from the Engle-Granger test with a break?
The Johansen framework is multivariate and tests for multiple cointegrating vectors jointly using maximum likelihood, while the Engle-Granger test is a two-step single-equation procedure limited to one cointegrating vector. The Johansen approach is generally preferred when there are three or more variables.
Which software implements this test?
JMulTi (free) offers a dedicated structural-break Johansen module following Saikkonen-Lütkepohl. EViews supports break specifications within the VECM framework. Stata users rely on community-contributed routines such as the sbvar package. R users can combine the urca and strucchange packages with custom critical-value tables.
Sources
- Johansen, S. (1988). Statistical analysis of cointegration vectors. Journal of Economic Dynamics and Control, 12(2–3), 231–254. DOI: 10.1016/0165-1889(88)90041-3 ↗
- Saikkonen, P., & Lütkepohl, H. (2000). Testing for the cointegrating rank of a VAR process with structural shifts. Journal of Business and Economic Statistics, 18(4), 451–464. DOI: 10.1080/07350015.2000.10524884 ↗
How to cite this page
ScholarGate. (2026, June 3). Johansen Cointegration Test with Structural Breaks. ScholarGate. https://scholargate.app/en/econometrics/structural-break-johansen-cointegration
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Structural Break ARDL Bounds TestEconometrics↔ compare
- Structural break Engle-Granger cointegrationEconometrics↔ compare
- Structural break VECMEconometrics↔ compare
- Vector Error Correction ModelEconometrics↔ compare
- Zivot-Andrews Structural Break TestEconometrics↔ compare