Structural Break VAR Model
Vector Autoregression Model with Structural Breaks · Also known as: VAR with structural breaks, break-point VAR, regime-switching VAR, SB-VAR
The Structural Break VAR model extends the standard Vector Autoregression (VAR) framework by allowing coefficient matrices and error covariance to shift at one or more unknown break dates. It is designed for multivariate time series where economic relationships change abruptly due to policy shifts, financial crises, or major structural events.
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 a structural break VAR when theory or visual inspection suggests that the dynamic relationships among your variables have shifted at least once during the sample period — for example, around a financial crisis, a major policy change, or a structural reform. It is appropriate when you have a multivariate system with at least 50–60 observations per regime so that each segment can be estimated reliably. Do not use it when the sample is short and suspected breaks would leave fewer than 30 observations in a segment, when breaks occur frequently (consider Markov-switching VAR instead), or when a single reduced-form VAR with dummy intercepts adequately captures the shift.
Strengths & limitations
- Captures genuine regime changes in multivariate dynamics without forcing a single set of coefficients on the full sample.
- Regime-specific impulse responses and variance decompositions provide more accurate pictures of shock transmission in each period.
- Formal break-point tests (Bai-Perron) control size and can identify multiple breaks simultaneously.
- Prevents omitted-break bias that can distort Granger causality and cointegration findings.
- Remains interpretable: each segment is a standard VAR whose coefficients have direct economic meaning.
- Each regime requires sufficient observations; too many breaks relative to sample size leads to imprecise estimates.
- Break dates are estimated with uncertainty, and inference conditional on estimated breaks understates true variability.
- The number of parameters grows quickly with the number of variables, lags, and regimes, increasing overfitting risk.
- Assumes breaks are abrupt (level-shift); gradual or smooth parameter changes require time-varying parameter VAR instead.
Frequently asked
How do I decide how many breaks to allow?
Use the Bai-Perron sequential sup-F testing procedure, which starts from zero breaks and adds breaks one at a time while controlling the overall significance level. Information criteria such as BIC can serve as a cross-check. In practice, two or three breaks are typical for post-war macroeconomic samples.
What is the minimum number of observations per regime?
A common rule of thumb is at least 15% of the total sample or a minimum of 30 observations per segment — whichever is larger — so that each regime's VAR can be estimated with reasonable precision. In practice, 50 or more observations per segment is preferable for reliable impulse-response inference.
Should I use the structural break VAR or a Markov-switching VAR?
Use the structural break VAR when breaks are believed to be permanent and datable (one-time policy changes, crises). Use Markov-switching VAR when the economy cycles between recurring regimes (e.g., expansion and recession) with no fixed transition dates.
Can I test Granger causality across the full sample if I find breaks?
No. Full-sample Granger causality tests are invalid if coefficients are not stable. You should test Granger causality within each regime separately and compare results across regimes.
Does finding a structural break in the VAR invalidate a prior cointegration analysis?
It may. Standard Johansen and Engle-Granger tests have low power or distorted size in the presence of structural breaks. You should use cointegration tests that allow for breaks (e.g., Gregory-Hansen) or re-run the analysis within each regime.
Sources
- Bai, J., & Perron, P. (1998). Estimating and testing linear models with multiple structural changes. Econometrica, 66(1), 47–78. DOI: 10.2307/2998540 ↗
- Sims, C. A. (1980). Macroeconomics and reality. Econometrica, 48(1), 1–48. DOI: 10.2307/1912017 ↗
How to cite this page
ScholarGate. (2026, June 3). Vector Autoregression Model with Structural Breaks. ScholarGate. https://scholargate.app/en/econometrics/structural-break-var-model
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 ARIMA ModelEconometrics↔ compare
- Structural break VECMEconometrics↔ compare
- Structural VAREconometrics↔ compare
- Vector AutoregressionEconometrics↔ compare
- Vector Error Correction ModelEconometrics↔ compare
- Zivot-Andrews Structural Break TestEconometrics↔ compare