-
Notifications
You must be signed in to change notification settings - Fork 1
beta_tv_favar
Jaromír Beneš edited this page Mar 20, 2026
·
1 revision
The Beta Time-Varying FAVAR combines factor-augmented VAR analysis with Beta Time-Varying parameter modeling in a two-step framework. This estimator first extracts factors via principal components, then applies time-varying coefficient priors following beta distributions to model parameter evolution in the factor-augmented VAR, allowing coefficients to evolve smoothly over time within bounded ranges.
The Beta Time-Varying FAVAR specification features:
- Two-step estimation: factor extraction followed by time-varying VAR estimation
- Principal component factor extraction in first step
- Beta-distributed time-varying coefficients with bounded evolution
- Smooth coefficient evolution within [0,1] intervals after transformation
- Minnesota-style priors combined with time-varying parameter structure
This approach captures both common factor dynamics and smooth parameter evolution with natural bounded behavior.
-
Sigma: Method of calculating priors on covariance matrix
- Values: "ar" (autoregressive) or "eye" (identity-based)
- Default: "ar"
- Controls how the prior covariance matrix is constructed
-
Autoregression: Prior on first-order autoregression
- Range: [0, 1]
- Default: 0.8
- Controls the belief in random walk behavior for own lags
-
Lambda1: Overall tightness of priors
- Range: (0, ∞)
- Default: 0.1
- Smaller values impose tighter priors (more shrinkage)
-
Lambda2: Variable weighting
- Range: (0, ∞)
- Default: 0.5
- Controls relative weights between own-variable and cross-variable lags
-
Lambda3: Lag decay
- Range: (0, ∞)
- Default: 1
- Controls how quickly prior importance decays with lag length
-
Lambda4: Exogenous variable tightness
- Range: (0, ∞)
- Default: 100
- Controls prior tightness on exogenous variable coefficients
-
Lambda5: Block exogeneity shrinkage
- Range: (0, ∞)
- Default: 0.001
- Controls shrinkage when block exogeneity is imposed
-
TVPVolPrior: Prior variance for time-varying parameter innovations
- Range: (0, ∞)
- Default: 10000
- Controls prior belief about parameter variation over time
-
TVPVolPriorDims: Prior for parameter evolution variance
- Range: (0, ∞)
- Default: 2
- Controls tightness of priors on parameter change rates
-
Exogenous: Priors on exogenous variables flag
- Default: false
- Enables specialized priors for exogenous variables
-
BlockExogenous: Block exogeneity flag
- Default: false
- Imposes block exogeneity restrictions
-
Burnin: Number of burn-in draws
- Range: [0, ∞)
- Default: 5000
- Required burn-in for time-varying parameter sampling
-
It: Total number of iterations
- Range: [1, ∞)
- Default: 20000
- Total MCMC iterations including burn-in
-
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
- Smooth parameter evolution: When coefficients are expected to change gradually
- Bounded parameter behavior: When parameters should remain within reasonable ranges
- Structural change analysis: When investigating gradual structural changes
- Policy regime analysis: When modeling smooth transitions between policy regimes
- Constant parameters assumed: When coefficients are believed to be stable
- Discrete breaks preferred: When abrupt structural breaks are expected
- Small sample sizes: When insufficient data for reliable time-variation estimation
- High-frequency applications: When parameter constancy is more plausible
- Requires sufficient time series length for parameter evolution identification
- Beta distribution naturally bounds parameter evolution
- More interpretable than unbounded time-varying approaches
- Benefits from periods with gradual economic changes
Step 1: Factor Extraction
F_t = Λ X_t + e_t
Factors extracted via principal components from observable dataset X_t.
Step 2: Beta Time-Varying FAVAR
[Y_t]
[F_t] = A_t [Y_{t-1}] + ε_t
[F_{t-1}]
A_t = f(B_t)
B_{i,j,t} ~ Beta(α_{i,j,t}, β_{i,j,t})
logit(B_{i,j,t}) = logit(B_{i,j,t-1}) + ω_{i,j,t}
where f(·) transforms beta-distributed parameters to appropriate coefficient ranges.
- Parameters evolve on logit scale to maintain beta distribution
- Smooth evolution with natural bounds at 0 and 1 (after transformation)
- Allows for different evolution patterns across coefficients
- Can accommodate both persistent and mean-reverting parameter behavior
- Speed: Moderate speed due to time-varying parameter sampling complexity
- Memory: Higher requirements for storing parameter paths over time
- Convergence: Good convergence due to bounded parameter space
- Scalability: Scales moderately with system size and sample length
- Requires sampling of time-varying parameter paths
- Beta distribution constraints aid convergence
- Parameter transformation handled through Jacobian adjustments
- Benefits from efficient path sampling algorithms
- Factor loadings in VAR equations can evolve over time
- Different factors may have time-varying importance
- Evolution patterns may differ across variables
- Useful for identifying changing factor transmission mechanisms
- Smooth parameter evolution captures gradual structural change
- Beta distribution provides natural economic bounds
- Time-varying patterns may correlate with policy regimes
- Factor importance changes reveal evolving economic relationships
- Parameters: Time-varying vs. constant coefficients
- Flexibility: More flexible but computationally more demanding
- Applications: Better for periods with gradual structural change
- Interpretation: Reveals parameter evolution patterns
- Distribution: Beta vs. normal time-varying parameters
- Bounds: Natural bounds vs. potentially unrealistic parameter values
- Stability: More stable due to bounded parameter space
- Interpretation: More economically interpretable parameter ranges
- Change pattern: Smooth vs. discrete regime changes
- Identification: Gradual vs. sharp transition points
- Applications: Different types of structural change
- Complexity: Similar computational complexity
- Standardize data and handle missing observations
- Select number of factors using information criteria
- Validate factor stability across time periods
- Assess factor interpretability and economic meaning
- Initialize parameter paths using constant parameter estimates
- Use efficient sampling schemes for bounded parameter evolution
- Monitor convergence of parameter paths
- Validate parameter bounds and economic interpretation
- Handle logit transformation for beta-distributed parameters
- Ensure proper Jacobian adjustments in MCMC sampling
- Map beta parameters to economically meaningful coefficient ranges
- Monitor transformation stability and numerical accuracy
- Examine parameter paths for economic interpretation
- Test parameter evolution patterns against known events
- Compare parameter change points with structural break tests
- Assess parameter evolution predictability
- Compare in-sample fit with constant parameter alternatives
- Assess out-of-sample forecasting performance over time
- Evaluate parameter evolution forecasting accuracy
- Test structural interpretation consistency across periods
- Examine time-varying factor importance in different equations
- Assess factor loading evolution patterns
- Compare factor contribution changes over time
- Validate factor interpretation stability
- Central bank policy rule evolution over different regimes
- Changing monetary policy transmission mechanisms
- Interest rate setting behavior evolution
- Factor-based policy rule estimation with time variation
- Changing relationships between financial and real variables
- Evolution of market efficiency and information processing
- Risk factor importance changes over time
- Financial integration and globalization effects
- Gradual productivity growth changes
- Evolving labor market dynamics
- Energy price transmission mechanism evolution
- International trade relationship changes
- Time-varying business cycle propagation mechanisms
- Changing factor loadings over different cycle phases
- Evolution of economic relationships during expansions vs. recessions
- Long-term structural transformation analysis
- Natural parameter bounds through beta distribution
- Smooth parameter evolution captures gradual changes
- More stable than unbounded time-varying approaches
- Economically interpretable parameter ranges
- Requires sufficient data for reliable parameter evolution estimation
- May not capture abrupt structural breaks well
- Parameter transformation adds computational complexity
- Beta distribution may be restrictive for some applications
- Beta distribution naturally bounds parameters within economically reasonable ranges
- Parameter values near bounds indicate strong relationships or constraints
- Evolution within bounds shows changing relationship strength
- Bound-approaching behavior may signal structural stress
- Gradual parameter changes suggest evolving economic relationships
- Cyclical parameter patterns may reflect business cycle effects
- Trending parameter evolution indicates long-term structural change
- Parameter volatility shows relationship stability over time
- Bernanke, B.S., J. Boivin, and P. Eliasz (2005). "Measuring the Effects of Monetary Policy: A Factor-Augmented Vector Autoregressive (FAVAR) Approach"
- Primiceri, G.E. (2005). "Time Varying Structural Vector Autoregressions and Monetary Policy"
- Eisenstat, E., J.C.C. Chan, and R.W. Strachan (2016). "Stochastic Model Specification Search for Time-Varying Parameter VARs"
- Koop, G. and D. Korobilis (2010). "Bayesian Multivariate Time Series Methods for Empirical Macroeconomics"