Time-Varying Parameter SVAR Model (TVP-SVAR)
Time-Varying Parameter Structural Vector Autoregression Model · Also known as: TVP-SVAR, time-varying SVAR, drifting-parameter SVAR, TVP structural VAR
The Time-Varying Parameter Structural VAR (TVP-SVAR) model extends classical structural VARs by allowing both the reduced-form coefficients and the structural impact matrix to evolve continuously over time. Estimated via Bayesian MCMC, it captures shifting transmission mechanisms and heteroscedastic volatility — making it the workhorse for empirical macroeconomics when policy regimes and economic relationships change.
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-SVAR when you need to trace how the dynamic effects of structural shocks (monetary policy, oil price, fiscal) have changed over a long sample spanning multiple policy regimes or crises. It is appropriate when preliminary tests suggest parameter instability, or when economic theory predicts that transmission channels are not stable. Prefer it over a standard SVAR when stochastic volatility is also suspected. Avoid it with short samples (fewer than ~80 quarterly observations), since the state-space requires many periods to identify smooth drifts; also avoid it when only a single breakpoint is hypothesised — in that case a simple Chow test and a split-sample SVAR are cheaper and equally valid.
Strengths & limitations
- Captures continuously evolving structural relationships without requiring pre-specified breakpoints.
- Simultaneously estimates time-varying impulse responses and stochastic volatility, providing a rich picture of macroeconomic dynamics.
- Full Bayesian posterior gives honest uncertainty bands around every time period's coefficients and impulse responses.
- Widely accepted in central bank and academic research as the benchmark for empirical monetary policy analysis.
- Flexible identification: compatible with Cholesky, sign restrictions, and external instrument approaches.
- Computationally intensive — MCMC estimation with thousands of draws can take hours even on modern hardware for moderate-size systems.
- Requires long time series (ideally 100+ quarterly observations) for the random-walk state to be reliably identified.
- The number of free parameters grows quickly with the number of variables, making large systems (>5 variables) impractical without dimension-reduction priors.
- Inference is sensitive to the prior on Q (parameter drift variance); a poorly chosen prior can force near-constant or excessively volatile paths.
Frequently asked
How does TVP-SVAR differ from a Markov-switching SVAR?
A Markov-switching SVAR allows parameters to jump between a small number of discrete regimes governed by a hidden Markov chain. TVP-SVAR instead lets parameters drift continuously as a random walk. The choice depends on the economic question: discrete regime changes (e.g. fixed vs. floating exchange rate) suit Markov-switching; gradual evolution of policy rules suits TVP-SVAR.
How many variables can the TVP-SVAR handle?
Standard implementations typically handle three to five variables. With six or more, the parameter count explodes and MCMC becomes very slow. Factor-augmented VARs or large Bayesian VARs with Minnesota priors are better choices for large information sets.
What software can estimate TVP-SVAR?
Primiceri's original MATLAB code is publicly available and widely used. The R package 'bsvars' and the Dynare toolbox also support TVP-SVAR estimation. Python implementations exist but are less standardised.
Can I use quarterly and monthly data together?
Not directly in a single TVP-SVAR; all variables must share the same frequency. Mixed-frequency extensions (MF-VAR combined with TVP) exist in the research literature but are not yet routine in practice.
What minimum sample size is needed?
The rule of thumb is at least 80 quarterly (or 240 monthly) observations. Shorter samples leave the random-walk drift unidentified, causing the posterior to collapse onto the prior and the time-varying paths to appear spuriously smooth or noisy depending on prior choice.
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 ↗
- Nakajima, J. (2011). Time-Varying Parameter VAR Model with Stochastic Volatility: An Overview of Methodology and Empirical Applications. IMES Discussion Paper Series 2011-E-9, Bank of Japan. link ↗
How to cite this page
ScholarGate. (2026, June 3). Time-Varying Parameter Structural Vector Autoregression Model. ScholarGate. https://scholargate.app/en/econometrics/time-varying-parameter-svar-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
- Time-varying parameter VAR modelEconometrics↔ compare