Time-Varying Parameter Granger Causality
Also known as: TVP Granger causality, rolling-window Granger causality, time-varying Granger test, dynamic Granger causality
Time-varying parameter Granger causality extends the classical Granger causality framework by allowing the predictive relationships between time series to evolve across time. Instead of assuming fixed causal effects, the model estimates causal coefficients that can shift, capturing structural breaks, regime changes, or gradual evolution in economic or financial relationships.
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 TVP Granger causality when you suspect the causal relationship between two time series is not constant — for example, across different monetary policy regimes, financial crises, or structural economic shifts. It is appropriate when theory or prior evidence suggests instability in the lag coefficients, and when the sample is long enough to estimate the time-varying process reliably (typically 100 or more observations). Do not apply this method when the sample is short (fewer than ~80 observations), when the relationship is plausibly stable throughout the period (use standard Granger causality instead), or when computational simplicity and reproducibility in small datasets take priority.
Strengths & limitations
- Captures structural breaks and regime shifts in causal relationships without requiring a priori knowledge of break dates.
- Provides a time-stamped picture of causality, showing when X predicts Y and when the relationship weakens or reverses.
- The Bayesian TVP-VAR framework naturally handles parameter uncertainty and produces credible intervals around the time-varying estimates.
- More informative than a single fixed-coefficient Granger test when economic or financial conditions change substantially over the sample.
- Can be combined with stochastic volatility to model both mean and variance dynamics simultaneously.
- Requires substantially longer samples than fixed-parameter Granger tests; short series lead to poorly identified time-varying coefficients.
- Bayesian TVP-VAR estimation is computationally intensive and requires specifying priors for the state covariances, making results sensitive to hyperparameter choices.
- The random-walk assumption for coefficient evolution may be too restrictive or too flexible depending on the true data-generating process.
- Multiplicity concerns arise when causality is tested at every time point; corrections or careful interpretation of the resulting time profiles are needed.
- Interpretation is more complex than for a single Granger F-statistic, demanding higher econometric literacy from the analyst and the audience.
Frequently asked
How is TVP Granger causality different from rolling-window Granger causality?
Rolling-window tests re-estimate a fixed-coefficient VAR over a moving sub-sample, treating each window as independent. TVP Granger causality uses a state-space model that lets coefficients evolve continuously, using all available data efficiently and producing smoother, more statistically coherent coefficient paths.
How long does my time series need to be?
As a practical rule, at least 100 observations are recommended for stable TVP-VAR estimation. With fewer than 80 observations, the Kalman filter has insufficient data to track meaningful variation, and the results become dominated by prior assumptions.
Do I need to account for cointegration?
Yes. If the series are integrated (I(1)) and cointegrated, the TVP model should include an error-correction term or be specified in levels with care, following the Toda-Yamamoto or Dolado-Lutkepohl approach to avoid invalid inference.
What software can estimate TVP Granger causality?
R packages such as bvarsv and BVAR, MATLAB code distributed with Primiceri (2005), and Python libraries with Kalman filter routines all support TVP-VAR estimation. Rolling-window alternatives are straightforward with any standard VAR package.
How do I choose the number of lags for the TVP-VAR?
Standard information criteria (AIC, BIC, HQIC) applied to a fixed-coefficient VAR on the same data provide a reasonable starting point. Sensitivity checks with adjacent lag lengths are recommended because TVP models are more parameter-rich than their fixed-coefficient counterparts.
Sources
- Granger, C. W. J. (1969). Investigating causal relations by econometric models and cross-spectral methods. Econometrica, 37(3), 424-438. DOI: 10.2307/1912791 ↗
- Primiceri, G. E. (2005). Time varying structural vector autoregressions and monetary policy. Review of Economic Studies, 72(3), 821-852. DOI: 10.1111/j.1467-937X.2005.00353.x ↗
How to cite this page
ScholarGate. (2026, June 3). Time-Varying Parameter Granger Causality. ScholarGate. https://scholargate.app/en/econometrics/time-varying-parameter-granger-causality
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.
- Granger CausalityEconometrics↔ compare
- Kalman FilterBayesian↔ compare
- Structural VAREconometrics↔ compare
- Vector AutoregressionEconometrics↔ compare