Panel Vector Autoregression (Panel VAR)
Also known as: PVAR, panel vector autoregression, Panel VAR (PVAR)
Panel VAR extends the vector autoregression model to panel data, modelling the dynamic interactions among several variables while controlling for cross-unit heterogeneity through fixed effects. It was introduced by Holtz-Eakin, Newey and Rosen in 1988 and produces impulse-response functions and variance decompositions at the panel level.
Key highlights
- Captures dynamic, bidirectional interactions among several variables while controlling for unobserved cross-unit heterogeneity through fixed effects.
- Pools information across many units, so reliable system dynamics can be estimated even when each unit has a short time series (N >> T micropanels).
- Delivers panel-level impulse-response functions (IRF) and forecast-error variance decompositions (FEVD), plus panel Granger causality tests, for interpreting how shocks propagate.
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 Panel VAR when you have continuous variables observed for many units over time and want to study dynamic, two-way interactions among them rather than a single dependent variable. It fits micropanels where the number of units is much larger than the number of time periods (N >> T) and needs at least about 50 observations. Key requirements are that fixed effects be removed with forward orthogonal (Helmert) deviations, the lag length be selected by MBIC/MAIC, and panel stationarity be checked with IPS and LLC tests. When T is large relative to N a single-unit VAR is the better tool.
Strengths & limitations
- Captures dynamic, bidirectional interactions among several variables while controlling for unobserved cross-unit heterogeneity through fixed effects.
- Pools information across many units, so reliable system dynamics can be estimated even when each unit has a short time series (N >> T micropanels).
- Delivers panel-level impulse-response functions (IRF) and forecast-error variance decompositions (FEVD), plus panel Granger causality tests, for interpreting how shocks propagate.
- Designed for micropanels with N >> T; when the time dimension is large a single-unit VAR is more appropriate.
- Fixed effects must be removed with forward orthogonal (Helmert) deviations rather than ordinary differencing, otherwise the lagged-dependent-variable bias contaminates the GMM estimates.
- Requires panel stationarity (IPS, LLC) and careful lag selection (MBIC/MAIC); mis-specifying these distorts the impulse responses.
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
How is Panel VAR different from an ordinary VAR?
A standard VAR works on a single time series of a system of variables. Panel VAR adds a cross-sectional dimension: the same system is observed for many units, and unit-specific fixed effects absorb persistent heterogeneity, so the shared dynamics are estimated by pooling information across all units.
Why are fixed effects removed with forward orthogonal (Helmert) deviations?
The lagged dependent variables are correlated with the fixed effects, so ordinary first-differencing leaves the transformed error correlated with the regressors and biases the estimates. Forward orthogonal (Helmert) deviations remove the fixed effects while keeping the transformed errors uncorrelated with valid lagged instruments used in GMM.
When should I prefer a single-unit VAR instead?
Panel VAR is built for micropanels where the number of units is much larger than the number of time periods (N >> T). If your time dimension T is large, a single-unit VAR for each series is usually more appropriate.
What do the IRF and FEVD outputs tell me?
The impulse-response function (IRF) traces how a shock to one variable moves the others over time, and the forecast-error variance decomposition (FEVD) reports what share of each variable's forecast error is attributable to each shock. Together they summarise the estimated dynamic interactions at the panel level.
Sources
- 1.Holtz-Eakin, D., Newey, W. & Rosen, H. S. (1988). Estimating Vector Autoregressions with Panel Data. Econometrica, 56(6), 1371-1395.
- 2.Abrigo, M. R. M. & Love, I. (2016). Estimation of Panel Vector Autoregression in Stata. Stata Journal, 16(3), 778-804.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 1). Panel VAR. ScholarGate. https://scholargate.app/econometrics/panel-var