VAR models multiple series jointly, capturing cross‑effects and enabling impulse response analysis and coherent multivariate forecasts.
Preparation
- Align timestamps and ensure consistent sampling
- Check stationarity; difference/detrend or consider VECM if cointegrated
- Consider transformations (logs) to stabilize variance
Modeling Steps
- Select lag order via AIC/BIC with holdout checks
- Fit VAR; verify stability (roots) and residual diagnostics
- Run Granger causality tests for directional predictability
Interpretation Tools
- Impulse Response Functions (IRFs): impact of shocks over time
- Forecast Error Variance Decomposition (FEVD): which variables explain forecast errors
When to Use VECM
- When series are non‑stationary but cointegrated (long‑run relations)
- VECM captures both long‑run equilibrium and short‑run dynamics