Structural Vector Autoregression (SVAR)
Also known as: Structural VAR, Identified VAR, SVAR Model, Yapısal Vektör Otoregresyon
Structural Vector Autoregression (SVAR) is a multivariate time-series model, developed by Christopher Sims (1980), that extends the reduced-form VAR by imposing economically motivated identifying restrictions on contemporaneous relationships among variables. SVAR enables researchers to isolate orthogonal structural shocks and trace their causal dynamic effects through impulse response functions and forecast error variance decompositions, making it a cornerstone of modern empirical macroeconomics.
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
SVAR is appropriate when the research goal is causal inference about dynamic shock transmission rather than mere forecasting. It requires stationary or cointegrated multivariate time series (usually three to eight variables), a theoretical rationale for the identifying restrictions, and sufficient sample length (commonly at least 100 quarterly observations). SVAR is the preferred tool for studying monetary policy transmission, fiscal multipliers, and business-cycle decomposition. When cointegration is present, the VECM-based SVAR (Blanchard-Quah or King et al. framework) should be considered. Pure forecasting without causal claims favors unrestricted VAR or factor models.
Strengths & limitations
- Allows economically interpretable causal identification of structural shocks from observational time-series data
- Produces impulse response functions and forecast error variance decompositions that directly answer policy-relevant questions
- Flexible identification: short-run, long-run, or sign restrictions can be combined to match diverse theoretical frameworks
- Grounded in the Nobel-recognized methodological tradition of Sims (1980), ensuring wide acceptance in peer-reviewed publication
- Results are sensitive to the choice of identifying restrictions; different orderings or assumptions can yield substantially different impulse responses
- Curse of dimensionality: the number of parameters grows quadratically with system size, requiring large samples for reliable estimation
- Standard SVAR assumes linear, time-invariant dynamics and may miss structural breaks, regime changes, or nonlinearities
- Exact identification is not testable; over-identification tests (LR) exist but cannot validate the economic interpretation itself
Frequently asked
How many restrictions are needed to identify an SVAR?
For a system of n variables, the reduced-form residual covariance matrix provides n(n+1)/2 unique moments. The structural matrix B_0 has n^2 free elements, but n normalizations fix shock variances to unity, leaving n(n-1) parameters. Exact identification requires n(n-1)/2 additional restrictions. A three-variable system therefore needs three zero or other restrictions on B_0.
What is the difference between a reduced-form VAR and an SVAR?
A reduced-form VAR models each variable as a linear function of lags of all variables, producing correlated residuals that are statistical composites of multiple simultaneous shocks. An SVAR imposes economic theory in the form of restrictions on the contemporaneous matrix B_0 to recover orthogonal structural shocks. Only the structural form supports causal interpretation of impulse responses.
Can SVAR be used with non-stationary data?
SVAR estimated in levels can be valid if variables are cointegrated, exploiting Sims, Stock, and Watson's (1990) result that OLS estimation remains consistent under cointegration. However, when long-run neutrality restrictions are theoretically motivated, the Blanchard-Quah SVAR in first differences with cumulated responses is more appropriate. Unit-root pre-testing and cointegration analysis should precede model specification.
Sources
- Sims, C. A. (1980). Macroeconomics and reality. Econometrica, 48(1), 1–48. DOI: 10.2307/1912017 ↗
How to cite this page
ScholarGate. (2026, June 2). Structural Vector Autoregression (SVAR). ScholarGate. https://scholargate.app/en/econometrics/svar
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.
- Impulse Response FunctionEconometrics↔ compare
- VAR ModelEconometrics↔ compare