-
Notifications
You must be signed in to change notification settings - Fork 1
flat_favar_onestep
Jaromír Beneš edited this page Mar 20, 2026
·
1 revision
The One-Step FAVAR with Flat Prior implements simultaneous estimation of factors and VAR parameters using uninformative flat priors in a unified Bayesian framework. This estimator jointly models factor extraction and VAR dynamics without strong prior assumptions, providing theoretically optimal inference that accounts for factor uncertainty while letting the data drive parameter estimates.
The Flat FAVAR One-Step specification features:
- Simultaneous factor extraction and VAR parameter estimation
- Full Bayesian treatment of factor uncertainty
- Uninformative flat priors on VAR parameters
- Joint posterior inference over factors and parameters
- Data-driven parameter estimation without strong prior beliefs
This approach provides theoretical rigor by properly accounting for all sources of uncertainty while maintaining prior neutrality.
-
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 (used for initialization)
-
Burnin: Number of burn-in draws
- Range: [0, ∞)
- Default: 5000
- Required burn-in for joint factor-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
- Prior neutrality desired: When avoiding strong prior assumptions about parameters
- Large datasets: When data are sufficiently informative without priors
- Exploratory analysis: When investigating data-driven relationships without prior beliefs
- Methodological research: When studying effects of different prior specifications
- Small samples: When data are insufficient for reliable estimation without priors
- Ill-conditioned problems: When regularization through priors is needed
- Real-time applications: When speed is more important than theoretical optimality
- Forecasting applications: When shrinkage through priors typically improves performance
- Most computationally intensive FAVAR approach
- Requires extensive MCMC diagnostics and convergence assessment
- May suffer from overfitting in small samples
- Benefits from large, informative datasets
Measurement Equation
X_t = Λ F_t + R_t + u_t
Transition Equation
[Y_t] [Y_{t-1}]
[F_t] = A [F_{t-1}] + ε_t
All parameters {Λ, F_t, A, Σ} estimated jointly accounting for all uncertainties.
A ~ Flat (improper uniform prior)
Σ ~ Flat (improper uniform prior on covariance)
F_t ~ Flat (improper uniform prior on factors)
Λ ~ Flat (improper uniform prior on loadings)
- Draw factors F_t given parameters and data
- Draw loadings Λ given factors and data
- Draw VAR coefficients A given factors and data
- Draw covariance Σ given all other parameters and data
- Speed: Slowest among all FAVAR approaches due to joint estimation and non-conjugate structure
- Memory: Highest requirements for storing all parameter draws
- Convergence: Most challenging convergence due to flat priors and complexity
- Scalability: Very limited scalability, practical only for small to medium systems
- Flat priors lead to slow mixing and convergence issues
- May require very long chains for proper convergence
- High sensitivity to initialization
- Needs sophisticated sampling schemes for efficiency
- Factors and parameters determined entirely by data
- No prior assumptions about factor dynamics or loadings
- Most flexible approach to factor-augmented modeling
- Results reflect pure data-driven relationships
- Most comprehensive uncertainty treatment with prior neutrality
- Posterior reflects only data information without prior influence
- May lead to very wide credible intervals in small samples
- Better suited for large, informative datasets
- Complete data dependence: Results depend only on data, not priors
- Optimal uncertainty treatment: Properly accounts for all sources of uncertainty
- Prior sensitivity: Reveals what data alone tell us about relationships
- Methodological purity: Most theoretically neutral approach
- Extreme computational cost: Most expensive FAVAR implementation
- Convergence difficulties: Flat priors create sampling challenges
- Overfitting risk: May overfit in small samples without regularization
- Implementation complexity: Most difficult to implement and diagnose
- Best theoretical properties for prior neutrality but highest computational cost
- Most complete uncertainty treatment but poorest convergence properties
- Pure data-driven results but may perform poorly in forecasting
- Maximum flexibility but minimum practical applicability
- Critical importance of good starting values due to flat priors
- Use two-step flat estimates or other methods for initialization
- Initialize factors using multiple approaches and compare
- Consider multiple starting points to assess sensitivity
- Requires state-of-the-art sampling algorithms
- Use adaptive schemes to improve mixing
- Consider gradient-based methods for difficult blocks
- Implement sophisticated convergence monitoring
- Essential to use highly optimized implementations
- Consider parallel computing strategies
- May require specialized software or custom implementations
- Budget substantial computational resources
- Use battery of advanced convergence diagnostics
- Monitor all parameter blocks with multiple chains
- Assess effective sample sizes carefully
- Validate posterior stability across very long runs
- Compare results with informative prior specifications
- Assess sensitivity to starting values
- Evaluate robustness across different initialization strategies
- Test consistency with large-sample theoretical properties
- Compare forecasting performance with shrinkage approaches
- Evaluate in-sample fit and out-of-sample predictive ability
- Assess density forecast calibration
- Test stability of structural interpretations
- Studies investigating prior sensitivity in FAVAR models
- Comparison of different FAVAR estimation approaches
- Academic research requiring maximum prior neutrality
- Investigation of data-driven factor relationships
- Applications with very large, informative datasets
- Settings where data clearly dominate any reasonable priors
- Exploratory analysis of new or unfamiliar economic relationships
- Situations where prior specification is highly controversial
- Checking sensitivity of results to prior assumptions
- Establishing baseline results free from prior influence
- Testing whether priors or data drive key conclusions
- Validating findings from informative prior approaches
- Requires high-end computational resources
- Essential to have many-core processors for parallel sampling
- Needs very substantial RAM for parameter storage
- May require HPC or cloud computing resources
- Requires highly sophisticated MCMC implementation
- Benefits from specialized numerical libraries and optimization
- May need custom sampling schemes for efficiency
- Consider using cutting-edge research implementations
- Expect extremely long computation times
- Plan for weeks or months of computation for larger systems
- Budget substantial resources for multiple long runs
- Allow extensive time for convergence diagnostics and validation
- Maximum theoretical purity and prior neutrality
- Most complete treatment of all sources of uncertainty
- Reveals what data alone say about factor-augmented relationships
- Best suited for methodological research and robustness analysis
- Extremely high computational requirements and convergence challenges
- Poor performance in small samples due to overfitting
- Very limited practical applicability
- May produce unreliable results without sufficient data
- Should be used primarily for research and robustness checking
- Requires expertise in advanced MCMC methods
- Not suitable for routine forecasting or policy applications
- Most appropriate when very large datasets are available
- "Flat" priors are typically implemented as very diffuse proper priors
- May use very large variances rather than truly improper priors
- Specification affects computational properties and convergence
- Requires careful attention to prior implementation details
- Flat priors may exacerbate identification problems
- Factor rotation and scale identification become more challenging
- May require additional identification restrictions
- Benefits from careful attention to factor identification strategies
- Bernanke, B.S., J. Boivin, and P. Eliasz (2005). "Measuring the Effects of Monetary Policy: A Factor-Augmented Vector Autoregressive (FAVAR) Approach"
- Sims, C.A. (1980). "Macroeconomics and Reality"
- Stock, J.H. and M.W. Watson (2002). "Forecasting Using Principal Components From a Large Number of Predictors"
- Lopes, H.F. and M. West (2004). "Bayesian Model Assessment in Factor Analysis"