Structural Vector Autoregression (SVAR)
Also known as: SVAR, structural vector autoregression, identified VAR, structural VAR model
Structural VAR extends the reduced-form VAR by imposing economic theory-based restrictions that identify orthogonal structural shocks. This allows researchers to disentangle the causal effects of distinct economic disturbances — such as supply versus demand shocks — and trace their dynamic propagation through a system of variables via impulse response functions and forecast error variance decompositions.
Key highlights
- Enables causal inference about distinct structural shocks without requiring a full structural DSGE model.
- Impulse response functions provide intuitive, policy-relevant dynamic narratives.
- Flexible identification: supports Cholesky, long-run (Blanchard-Quah), and sign restrictions.
- Forecast error variance decomposition reveals the relative importance of each shock to each variable's variability.
- Handles simultaneous equation bias that plagues single-equation approaches.
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 SVAR when you have a multivariate time series and want to identify and quantify the causal dynamic effects of distinct structural shocks grounded in economic theory. It is well suited for macroeconomic policy analysis — for example, isolating monetary policy shocks, oil price shocks, or technology shocks and studying their propagation. The method requires stationarity (or appropriate cointegration treatment), a defensible identification scheme, and sufficient time-series length (typically at least 80-100 observations per variable). Do not use SVAR when you cannot justify the identification restrictions theoretically, when the sample is very short (fewer than 50 observations), or when the variables exhibit strong nonlinearity that standard VAR dynamics cannot approximate.
Strengths & limitations
- Enables causal inference about distinct structural shocks without requiring a full structural DSGE model.
- Impulse response functions provide intuitive, policy-relevant dynamic narratives.
- Flexible identification: supports Cholesky, long-run (Blanchard-Quah), and sign restrictions.
- Forecast error variance decomposition reveals the relative importance of each shock to each variable's variability.
- Handles simultaneous equation bias that plagues single-equation approaches.
- Results are sensitive to the chosen identification scheme; different restrictions can yield substantively different impulse responses.
- Requires a sufficient number of observations relative to the number of variables and lags to estimate reliably.
- Assumes linearity and time-invariant coefficients; may miss regime changes or nonlinear dynamics.
- Identification restrictions must be justified by economic theory; ad hoc choices undermine credibility.
- Confidence bands for IRFs can be wide, especially at long horizons.
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
What is the difference between a VAR and an SVAR?
A reduced-form VAR estimates the predictive relationships among variables but cannot identify the causal direction of shocks because its residuals are correlated. An SVAR imposes restrictions — derived from economic theory — on the contemporaneous relationships to recover orthogonal structural shocks that can be given a causal interpretation.
How do I choose an identification scheme?
The choice should be guided by economic theory. Cholesky (short-run) restrictions assume a recursive causal ordering; they are easy to implement but require strong ordering assumptions. Blanchard-Quah long-run restrictions tie the identification to theoretical long-run neutrality properties. Sign restrictions are more agnostic and reflect qualitative theoretical predictions. Document and defend your choice explicitly.
What sample size do I need for SVAR?
As a rule of thumb, aim for at least 80-100 time observations for a small SVAR (2-4 variables). Each additional variable and lag multiplies the number of parameters to estimate. With short samples the IRF confidence bands become very wide and identification may be fragile.
Can I use SVAR with non-stationary data?
Not directly. Non-stationary variables must first be transformed (differenced) or, if cointegrated, modelled within a structural VECM framework. Applying SVAR to levels of I(1) variables without cointegration adjustment yields spurious results.
How do I construct confidence bands for impulse responses?
The most common approaches are recursive bootstrap (resample reduced-form residuals), wild bootstrap (heteroscedasticity-robust), and Bayesian methods with conjugate or Minnesota priors. Asymptotic delta-method bands are available but tend to be less reliable in small samples.
Sources
- 1.Blanchard, O. J., & Quah, D. (1989). The dynamic effects of aggregate demand and supply disturbances. American Economic Review, 79(4), 655-673.
- 2.Sims, C. A. (1980). Macroeconomics and reality. Econometrica, 48(1), 1-48.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Structural VAR. ScholarGate. https://scholargate.app/econometrics/structural-var