Nonlinear Structural Vector Autoregression (NL-SVAR) Model
Nonlinear Structural Vector Autoregression Model · Also known as: nonlinear structural VAR, NL-SVAR, threshold SVAR, regime-switching SVAR
The Nonlinear Structural VAR model extends the standard SVAR framework to allow structural relationships and dynamic responses to vary across economic regimes or states of the world. By imposing nonlinear transition mechanisms — such as threshold switching or smooth regime change — it captures asymmetric responses to shocks that a linear SVAR cannot detect.
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 Nonlinear SVAR when theory or preliminary evidence suggests that the economy responds asymmetrically to shocks — for example, if fiscal multipliers, monetary transmission, or financial contagion differ between expansions and recessions, or between low- and high-volatility periods. It is well suited for medium-sized macroeconomic systems (3–8 variables) with at least 100 quarterly observations. Do not use it when the sample is too short to identify multiple regimes reliably, when a linear SVAR passes standard linearity tests, or when the research question does not concern regime-dependent dynamics.
Strengths & limitations
- Captures asymmetric impulse responses across economic regimes, which a linear SVAR masks by averaging over states.
- Maintains the structural identification framework of the standard SVAR, so economic shocks retain a clear causal interpretation.
- Bayesian estimation handles short samples and propagates parameter uncertainty into regime-conditional impulse response bands.
- Regime probabilities are themselves informative outputs that can date recession, crisis, or high-volatility episodes.
- Compatible with proxy-SVAR and sign-restriction identification strategies used in modern empirical macroeconomics.
- Parameter proliferation: each additional regime multiplies the number of coefficients, requiring larger samples to achieve reliable estimates.
- Regime identification is sensitive to the choice of threshold variable or the number of Markov states; results can be fragile if these choices are ad hoc.
- Generalised impulse response functions depend on the history and the initial regime, making inference computationally intensive and presentation more complex than in a linear SVAR.
- Structural identification assumptions must be re-argued within each regime; researchers sometimes import linear-SVAR identifying assumptions without verifying their validity in a nonlinear context.
Frequently asked
What is the difference between a Nonlinear SVAR and a standard SVAR?
A standard SVAR assumes constant structural coefficients across all time periods; a Nonlinear SVAR allows those coefficients — and hence the economy's response to shocks — to vary depending on economic regime or state. The identification logic is the same, but the impulse responses differ across regimes.
How do I choose between a threshold SVAR, a smooth-transition SVAR, and a Markov-switching SVAR?
Threshold SVAR is natural when regime shifts are tied to an observable variable crossing a known threshold. Smooth-transition SVAR is preferred when regime change is gradual. Markov-switching SVAR is appropriate when the regime is latent and switches abruptly with unknown timing. Linearity tests and information criteria can guide the choice.
How many observations do I need for a Nonlinear SVAR?
At least 100 observations (quarterly) are recommended to identify two regimes reliably. With monthly data the threshold is lower, but each additional regime or variable substantially increases the parameter count and thus the minimum required sample.
Can I apply sign restrictions in a Nonlinear SVAR?
Yes. Sign restrictions — requiring, for example, that a contractionary monetary shock reduces output and inflation — can be imposed within each regime separately. Bayesian algorithms (e.g., rejection sampling over the rotation matrix) extend naturally to nonlinear settings.
Are generalised impulse responses comparable across different Nonlinear SVAR studies?
Comparisons require care because GIRFs depend on the chosen initial history and regime. Studies that report simulation-averaged GIRFs over the empirical distribution of histories are more comparable than those conditioned on a single arbitrary initial state.
Sources
- Koop, G., & Korobilis, D. (2010). Bayesian multivariate time series methods for empirical macroeconomics. Foundations and Trends in Econometrics, 3(4), 267–358. DOI: 10.1561/0800000013 ↗
- Auerbach, A. J., & Gorodnichenko, Y. (2012). Measuring the output effects of fiscal policy. American Economic Journal: Economic Policy, 4(2), 1–27. DOI: 10.1257/pol.4.2.1 ↗
How to cite this page
ScholarGate. (2026, June 3). Nonlinear Structural Vector Autoregression Model. ScholarGate. https://scholargate.app/en/econometrics/nonlinear-svar-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.
- Nonlinear ARDLEconometrics↔ compare
- Nonlinear VAR ModelEconometrics↔ compare
- Nonlinear VECMEconometrics↔ compare
- Structural VAREconometrics↔ compare
- Vector AutoregressionEconometrics↔ compare
- Vector Error Correction ModelEconometrics↔ compare