general_tv

VAR with Time-Varying Parameters and Stochastic Volatility (GeneralTV)

Overview

The General Time-Varying VAR implements the most flexible time-varying specification, combining time-varying coefficients with time-varying error covariances (stochastic volatility). This estimator allows for simultaneous evolution of both structural relationships and error volatilities, making it suitable for modeling complex time-varying economic systems with changing volatility patterns.

Methodology

The GeneralTV specification features:

• Time-varying VAR coefficients following random walk processes
• Time-varying error covariances with stochastic volatility
• State-space representation for both parameters and volatilities
• MCMC estimation with specialized algorithms for high-dimensional systems
• Most comprehensive time-varying VAR specification available

This approach captures both structural parameter changes and evolving uncertainty/volatility patterns in economic data.

Parameters

MCMC Settings

• Burnin: Number of burn-in draws

  • Range: [0, ∞)
  • Default: 0
  • Recommended: Extensive burn-in (10,000+) due to model complexity

• StabilityThreshold: Threshold for maximum eigenvalue magnitude

  • Range: (0, ∞)
  • Default: null (no stability check)
  • Ensures VAR stability by rejecting draws with eigenvalues above threshold

• MaxNumUnstableAttempts: Maximum number of unstable sampling attempts

  • Range: [1, ∞)
  • Default: 1000
  • Maximum attempts to find stable draw before stopping

Heteroskedasticity Parameters

• HeteroskedasticityAutoRegression: AR coefficient on residual variance

  • Range: [0, 1)
  • Default: 1
  • Controls persistence in stochastic volatility process

• HeteroskedasticityShape: IG shape parameter on residual variance

  • Range: (0, ∞)
  • Default: 0.001
  • Controls prior beliefs about volatility distribution shape

• HeteroskedasticityScale: IG scale parameter on residual variance

  • Range: (0, ∞)
  • Default: 0.001
  • Controls prior beliefs about volatility level

Usage Guidelines

When to Use

• Comprehensive time-variation: When both parameters and volatility change over time
• Financial crises: When modeling periods with changing volatility and relationships
• Long time series: When sufficient data exists for complex time-varying patterns
• Research applications: When understanding sources of time-variation is crucial

When Not to Use

• Computational constraints: Extremely computationally demanding
• Short time series: Insufficient data for complex time-varying estimation
• Simple applications: When constant-parameter models are adequate
• Real-time applications: Too slow for high-frequency updating

Practical Considerations

• Requires very long time series (preferably 200+ observations)
• Extremely computationally intensive - plan for long estimation times
• Monitor convergence extensively across all parameter dimensions
• Consider simpler alternatives unless full time-variation is essential

Model Specification

State Equations

Coefficients evolve as:

β_t = β_{t-1} + ν_t

Volatilities evolve as:

log(σ²_{i,t}) = log(σ²_{i,t-1}) + η_{i,t}

Observation Equation

y_t = X_t β_t + A_t^{-1} Σ_t^{1/2} ε_t

where both coefficients and covariances are time-varying.

Computational Characteristics

Performance

• Speed: Very slow - can take hours or days for estimation
• Memory: Extremely high memory requirements
• Convergence: Requires extensive monitoring and very long chains
• Parallelization: Limited due to sequential filtering requirements

Technical Requirements

• High-performance computing recommended
• Substantial RAM requirements (8GB+ recommended)
• Consider distributed computing for large systems
• Plan for multi-day estimation runs

Implementation Strategy

Initialization

• Use simpler models (BetaTV, stochastic volatility only) for starting values
• Careful tuning of proposal distributions in MCMC
• Consider sequential estimation approaches
• Monitor acceptance rates closely

Convergence Assessment

• Use multiple chains with overdispersed starting values
• Monitor all parameter dimensions for convergence
• Check effective sample sizes across parameter space
• Consider thinning to reduce autocorrelation

Comparison with Simpler Alternatives

vs. BetaTV

• Adds: Time-varying volatility
• Cost: Much higher computational burden
• When to upgrade: When volatility changes are important

vs. Stochastic Volatility Models

• Adds: Time-varying coefficients
• Cost: Higher computational complexity
• When to upgrade: When parameter instability is also present

Diagnostic Considerations

Model Validation

• Compare time-varying parameters to structural break tests
• Assess volatility evolution against known economic events
• Validate forecasting performance against simpler alternatives
• Check parameter identification across different subsamples

Robustness Checks

• Sensitivity to prior specifications
• Robustness to different initialization strategies
• Comparison with alternative time-varying approaches
• Assessment of model parsimony vs. flexibility

References

• Primiceri, G.E. (2005). "Time Varying Structural Vector Autoregressions and Monetary Policy"
• Cogley, T. and T.J. Sargent (2005). "Drifts and Volatilities: Monetary Policies and Outcomes in the Post WWII US"
• Koop, G. and D. Korobilis (2013). "Large Time-Varying Parameter VARs"
• Carriero, A., T.E. Clark, and M. Marcellino (2019). "Large Vector Autoregressions with Stochastic Volatility and Non-Conjugate Priors"