Robust Vector Autoregression (Robust VAR) Model
Robust Vector Autoregression Model · Also known as: robust VAR, outlier-robust VAR, heavy-tailed VAR, RVAR
The Robust VAR model extends the classical Vector Autoregression framework by replacing ordinary least squares estimation with robust estimators — such as M-estimators or median-based methods — to reduce the influence of outliers, structural breaks, and heavy-tailed shocks common in financial and macroeconomic time series.
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 Robust VAR when your multivariate time series contains known or suspected outliers, structural breaks, or heavy-tailed distributions — situations common in financial returns, commodity prices, or macroeconomic data during crisis periods. It is also appropriate when a standard VAR passes specification tests poorly due to non-normality or when exploratory analysis reveals extreme observations that may distort OLS-based VAR estimates. Do not use Robust VAR when data are well-behaved and approximately normally distributed without outliers — in that case, standard VAR is equally valid and more widely understood; also avoid it when the primary goal is testing Granger causality with established critical values, since the asymptotic distribution of test statistics under robust estimators may differ.
Strengths & limitations
- Yields consistent and more efficient coefficient estimates when the error distribution is heavy-tailed or contains outliers.
- Reduces distortion of impulse response functions caused by extreme observations or financial crises.
- Allows valid inference in the presence of heteroscedastic or non-Gaussian innovations that violate classical VAR assumptions.
- Preserves the intuitive multivariate structure and interpretation of a standard VAR.
- Compatible with bootstrap-based inference that is also robust to unknown heteroscedasticity.
- Robust estimators are computationally more demanding than OLS and may not be available in all statistical packages.
- The choice of robustness tuning constant (e.g., Huber's c) affects results and requires justification.
- Asymptotic theory for hypothesis tests (Granger causality, lag selection) is less developed than for standard VAR.
- May perform similarly to standard VAR when the data are genuinely well-behaved, providing no net gain.
Frequently asked
How is a Robust VAR different from a standard VAR?
A standard VAR is estimated by OLS, which minimises squared residuals and is sensitive to outliers. A Robust VAR replaces OLS with an estimator — such as an M-estimator or least trimmed squares — that down-weights or discards observations with unusually large residuals, making the coefficient estimates less sensitive to extreme data points.
Can I still perform Granger causality tests in a Robust VAR?
Yes, but the test statistics may not follow their standard chi-squared or F distributions asymptotically when the robust estimator is used. Simulation-based critical values or bootstrap p-values are recommended to ensure valid inference.
Which software packages support Robust VAR estimation?
R packages such as robVAR and MTS provide robust multivariate time-series estimation. Stata users can combine robust regression commands with system estimators, though dedicated plug-ins are limited. Python users can combine statsmodels VAR infrastructure with scipy robust estimators.
Should I check for cointegration before estimating a Robust VAR?
Yes. If the variables are non-stationary and cointegrated, a Robust Vector Error Correction Model (VECM) is more appropriate. Standard cointegration tests — Johansen trace and maximum eigenvalue — should be run first, though robust versions of these tests exist for outlier-contaminated samples.
How do I choose the tuning constant for the M-estimator?
Huber's M-estimator tuning constant c is commonly set to 1.345 times the estimated scale, which achieves about 95% efficiency relative to OLS under normality while providing substantial robustness protection. For heavily contaminated samples, lower values of c give stronger downweighting at some cost to efficiency.
Sources
- Goncalves, S., & Kilian, L. (2004). Bootstrapping autoregressions with conditional heteroskedasticity of unknown form. Journal of Econometrics, 123(1), 89-120. DOI: 10.1016/j.jeconom.2003.10.030 ↗
- Lutkepohl, H. (2005). New Introduction to Multiple Time Series Analysis. Springer, Berlin. ISBN: 978-3540401728
How to cite this page
ScholarGate. (2026, June 3). Robust Vector Autoregression Model. ScholarGate. https://scholargate.app/en/econometrics/robust-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.
- Panel VAREconometrics↔ compare
- Quantile VAREconometrics↔ compare
- Structural VAREconometrics↔ compare
- VAR ModelEconometrics↔ compare
- VECMEconometrics↔ compare