Time-Varying Parameter Zivot-Andrews Unit Root Test
Also known as: TVP Zivot-Andrews test, time-varying Zivot-Andrews unit root test, TVP-ZA test
The time-varying parameter Zivot-Andrews test extends the classic Zivot-Andrews (1992) structural break unit root test by allowing the regression coefficients to evolve over time. Rather than assuming fixed parameters across the full sample, this approach lets the autoregressive dynamics and break timing adapt through a state-space or rolling framework, improving robustness when economic relationships shift gradually.
Key highlights
- Combines endogenous break detection with parameter flexibility, reducing misspecification from ignored coefficient drift.
- Retains the Zivot-Andrews property of not requiring the break date to be pre-specified by the researcher.
- Provides a continuous portrait of how persistence evolves over the sample, useful for narrative economic analysis.
- Outperforms fixed-parameter unit root tests when structural change is gradual rather than abrupt.
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 this procedure when you suspect that a series contains both a unit root and gradual parameter instability, and you want to test for stationarity allowing for an unknown structural break without assuming fixed dynamics. It is particularly suited to long macroeconomic series spanning policy regime changes, financial crises, or technological transitions. Do not use it on short samples (fewer than 60–80 observations) where the Kalman filter has insufficient observations to estimate time-varying paths reliably, nor when theory strongly supports a single abrupt break with stable dynamics — in that case, the standard Zivot-Andrews test is more powerful and interpretable.
Strengths & limitations
- Combines endogenous break detection with parameter flexibility, reducing misspecification from ignored coefficient drift.
- Retains the Zivot-Andrews property of not requiring the break date to be pre-specified by the researcher.
- Provides a continuous portrait of how persistence evolves over the sample, useful for narrative economic analysis.
- Outperforms fixed-parameter unit root tests when structural change is gradual rather than abrupt.
- Requires moderately long time series; sparse samples leave the Kalman filter poorly initialised and critical values unreliable.
- Tabulated Zivot-Andrews critical values are not strictly valid under TVP dynamics; bootstrap inference adds computational cost.
- Sensitive to the assumed state-equation variance (signal-to-noise ratio), which is typically calibrated rather than estimated.
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 standard Zivot-Andrews test?
The standard Zivot-Andrews test assumes fixed coefficients in each sub-period before and after the detected break. The TVP version relaxes this by letting coefficients evolve continuously via a state-space model, reducing misspecification when dynamics shift gradually rather than abruptly.
Which critical values should I use?
The original Zivot-Andrews (1992) critical values are an approximation. For the TVP variant, bootstrapped or simulated critical values tailored to your sample size and state-equation variance are more reliable and are reported in recent applied studies.
Can I apply this test to panel data?
The test as described is univariate. For panel settings, consider panel unit root tests with structural breaks (e.g., Im-Pesaran-Shin with break corrections) rather than running individual TVP-Zivot-Andrews tests for each cross-section.
What software implements this procedure?
There is no single canonical package. The procedure is typically coded in R (using the strucchange and KFAS packages together) or in Stata/GAUSS with custom scripts. The standard Zivot-Andrews test is available in R via urca::ur.za and in Stata via zandrews.
What is the trimming window and why does it matter?
The trimming window excludes the first and last fraction (typically 15%) of the sample from the search for break dates, ensuring enough observations on each side to estimate reliable regressions. Too little trimming inflates size; too much reduces the power to detect early or late breaks.
Sources
- 1.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.
- 2.Cooley, T. F., & Prescott, E. C. (1976). Estimation in the Presence of Stochastic Parameter Variation. Econometrica, 44(1), 167–184.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Time-varying parameter Zivot-Andrews test. ScholarGate. https://scholargate.app/econometrics/time-varying-parameter-zivot-andrews-test