Time-Varying Parameter VAR Model (TVP-VAR)
Time-Varying Parameter Vector Autoregression Model · Also known as: TVP-VAR, time-varying VAR, TV-VAR, drifting-coefficient VAR
The Time-Varying Parameter VAR (TVP-VAR) model extends the standard vector autoregression by allowing the coefficients and error covariances to evolve gradually over time. Estimated via Bayesian methods and MCMC simulation, it captures how dynamic relationships between macroeconomic or financial variables shift across different economic regimes without requiring pre-specified break points.
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-VAR when you want to study how the dynamic interrelationships among several macroeconomic or financial variables change over a long time series — for example, shifting monetary policy transmission, evolving international spillovers, or time-changing exchange-rate pass-through. The method is most appropriate when structural change is suspected to be gradual and continuous rather than a single sharp break, and when the sample spans multiple decades. It is less suitable for short samples (fewer than ~80 quarterly observations), where the number of parameters easily exceeds the information available and priors dominate. Avoid it as a pure forecasting exercise without substantive interest in parameter evolution: a constant-coefficient BVAR is computationally cheaper and often forecasts as well or better.
Strengths & limitations
- Captures gradual, continuous structural change without requiring pre-specified break dates.
- Simultaneously models time-varying volatility (stochastic volatility), providing a richer picture of macroeconomic risk.
- Bayesian posterior yields credible intervals for every parameter path, making uncertainty explicit.
- Impulse response functions can be computed at any historical date, enabling before/after regime comparisons.
- Does not require stationarity pre-testing: the random-walk specification accommodates persistent dynamics naturally.
- Computationally demanding: MCMC chains for a moderate VAR with stochastic volatility can take hours even on fast hardware.
- Requires a sufficiently long time series; with fewer than ~80 observations the prior heavily governs the results.
- Results are sensitive to prior hyperparameter choices, particularly the variance of the random-walk innovations (Q). Robustness checks across prior settings are essential.
- The model can overfit in-sample, so it is best evaluated on out-of-sample predictive performance or via log predictive scores.
Frequently asked
How is TVP-VAR different from a standard VAR with dummy variables for structural breaks?
A standard VAR with dummy variables assumes abrupt, discrete breaks at known dates. TVP-VAR lets every parameter drift continuously throughout the sample without requiring the researcher to specify when or how many breaks occurred. This makes it more flexible but also more demanding in terms of data length and computation.
Can I compute impulse response functions from a TVP-VAR?
Yes. Because the coefficient matrices differ at each date t, you can compute a separate set of impulse response functions (IRFs) for any historical episode. Comparing IRFs across dates — for example, the 1980s versus the 2000s — is the primary way researchers use TVP-VAR to document structural change in dynamic responses.
How long a time series do I need?
A common rule of thumb is at least 80 quarterly observations (20 years) for a small to medium system. Shorter samples mean the priors dominate posteriors, so the apparent parameter evolution largely reflects the prior rather than the data. Sensitivity analysis to prior hyperparameters is especially important in small samples.
What software can estimate TVP-VAR?
MATLAB code by Primiceri (2005) is widely circulated. In R the 'bvarsv' and 'shrinkTVP' packages implement Bayesian TVP-VAR with stochastic volatility. Stata and EViews lack built-in support, so custom code is typically required.
Is TVP-VAR the same as a Markov-switching VAR?
No. Markov-switching VAR assumes a finite number of discrete regimes governed by a hidden Markov chain, so parameters switch abruptly between a small set of values. TVP-VAR assumes parameters drift continuously over time via a random walk. Both capture structural change but with very different assumptions about its nature.
Sources
- 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 ↗
- Cogley, T., & Nason, J. M. (1995). Effects of the Hodrick-Prescott filter on trend and difference stationary time series: Implications for business cycle research. Journal of Economic Dynamics and Control, 19(1-2), 253-278. DOI: 10.1016/0165-1889(93)00781-X ↗
How to cite this page
ScholarGate. (2026, June 3). Time-Varying Parameter Vector Autoregression Model. ScholarGate. https://scholargate.app/en/econometrics/time-varying-parameter-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.
- Bayesian VAR modelEconometrics↔ compare
- Kalman FilterBayesian↔ compare
- State Space ModelEconometrics↔ compare
- Structural VAREconometrics↔ compare
- Time-varying parameter ARDL bounds testEconometrics↔ compare
- Vector AutoregressionEconometrics↔ compare