Robust Zivot-Andrews Test
Robust Zivot-Andrews Structural Break Unit Root Test · Also known as: robust ZA test, ZA test with robust inference, Zivot-Andrews test with heteroscedasticity-robust critical values, structural break unit root test
The Robust Zivot-Andrews test extends the classic Zivot-Andrews (1992) unit root test to provide reliable inference when the error term may be heteroscedastic or non-normal. It tests whether a time series has a unit root while endogenously identifying a single structural break in the level, trend, or both, without requiring the researcher to pre-specify the break date.
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 the Robust Zivot-Andrews test when you suspect a time series has a single structural break of unknown timing and you cannot assume homoscedastic, normally distributed errors — common in macroeconomic, financial, or energy price series subject to crises or regime changes. It is preferable to the standard ZA test whenever residual heteroscedasticity or ARCH effects are detected (e.g., by a Breusch-Pagan or ARCH-LM test). Do not use it when there are two or more structural breaks (prefer Lumsdaine-Papell or Lee-Strazicich), when the series is very short (fewer than roughly 50 observations), or when robust critical values have not been tabulated for the chosen break model and sample size.
Strengths & limitations
- Endogenously identifies the break date, eliminating pre-testing bias from researcher-chosen break points.
- Provides correct asymptotic size in the presence of heteroscedastic or ARCH errors, unlike the original ZA test.
- Handles the three standard break models (level shift, trend shift, and combined) within a single framework.
- Wild-bootstrap implementation is feasible in small-to-moderate samples and does not require symmetric error distributions.
- Nests the standard Zivot-Andrews test as a special case when errors are homoscedastic.
- Assumes at most one structural break; multiple breaks reduce power and can cause spurious findings.
- Endogenous break selection can still yield a misleading break date when the true break is near the sample boundary.
- Robust critical values depend on the break model and sample size; simulation or wild-bootstrap is needed when published tables do not apply.
- The test has low power against near-unit-root alternatives or when the break magnitude is small relative to error variance.
- Does not accommodate smooth transitions or gradual regime changes, for which smooth-transition unit root tests are more appropriate.
Frequently asked
What makes the Robust Zivot-Andrews test different from the standard Zivot-Andrews test?
The standard ZA test derives its critical values under homoscedastic errors. The robust variant uses heteroscedasticity-consistent standard errors or a wild bootstrap to construct the test statistic, so the nominal size is maintained even when errors have ARCH effects or heavy tails. The search procedure and break-model logic are otherwise identical.
How do I choose between Model A, B, and C?
Model A allows only a level shift (intercept break), Model B allows only a trend-slope break, and Model C allows both simultaneously. In practice, Model C is the most common default because it nests the other two. If economic reasoning rules out a slope change, Model A may have more power by imposing the correct restriction.
What should I do if the test fails to reject but I still suspect stationarity?
Failure to reject may reflect low power rather than a true unit root. Consider whether there could be two or more structural breaks (use Lumsdaine-Papell or Lee-Strazicich), or whether the series undergoes a smooth transition (use ESTAR-based unit root tests). Also check lag length selection and sample size, as small samples reduce power substantially.
Is the wild bootstrap valid for small samples?
The wild bootstrap performs well in moderate samples (n ≥ 50) and outperforms asymptotic critical values when errors are heteroscedastic. For very small samples it can be somewhat undersized, so interpret borderline p-values cautiously and report the bootstrap p-value alongside the t-statistic.
Can I apply this test to panel data?
No — the Robust Zivot-Andrews test is a univariate test. For panel data with structural breaks, use panel unit root tests that allow cross-section-specific breaks, such as Im-Pesaran-Shin with break corrections or panel Lee-Strazicich procedures.
Sources
- Zivot, E., & Andrews, D. W. K. (1992). Further evidence on the great crash, the oil-price shock, and the unit-root hypothesis. Journal of Business & Economic Statistics, 10(3), 251–270. DOI: 10.1080/07350015.1992.10509904 ↗
- Zivot-Andrews test. Wikipedia. link ↗
How to cite this page
ScholarGate. (2026, June 3). Robust Zivot-Andrews Structural Break Unit Root Test. ScholarGate. https://scholargate.app/en/econometrics/robust-zivot-andrews-test
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.
- Bai-Perron TestEconometrics↔ compare
- Lee-Strazicich TestEconometrics↔ compare
- Lumsdaine-Papell TestEconometrics↔ compare
- Phillips-Perron TestEconometrics↔ compare
- Zivot-Andrews TestEconometrics↔ compare