normal_wishart_favar_onestep

One-Step FAVAR with Normal-Wishart Prior (NormalWishartFAVAROnestep)

Overview

The One-Step FAVAR with Normal-Wishart Prior implements simultaneous estimation of factors and VAR parameters within a unified Bayesian framework. This estimator jointly models factor extraction and VAR dynamics using conjugate Normal-Wishart priors, providing theoretically optimal inference by accounting for factor uncertainty in the estimation process.

Methodology

The Normal-Wishart FAVAR One-Step specification features:

• Simultaneous factor extraction and VAR parameter estimation
• Full Bayesian treatment of factor uncertainty
• Conjugate Normal-Wishart priors on VAR parameters
• Joint posterior inference over factors and parameters
• Optimal incorporation of factor estimation uncertainty

This approach provides theoretical advantages by properly accounting for all sources of uncertainty in the factor-augmented framework.

Parameters

Prior Configuration

• 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

Minnesota-Style Prior Parameters

• 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

Advanced Options

• Exogenous: Priors on exogenous variables flag

  • Default: false
  • Enables specialized priors for exogenous variables

• BlockExogenous: Block exogeneity flag

  • Default: false
  • Imposes block exogeneity restrictions

MCMC Settings

• Burnin: Number of burn-in draws

  • Range: [0, ∞)
  • Default: 5000
  • Requires 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

Usage Guidelines

When to Use

• Theoretical optimality: When proper treatment of factor uncertainty is crucial
• Small to medium datasets: When computational cost is manageable
• Inference precision: When accurate uncertainty quantification is needed
• Research applications: When methodological rigor is prioritized

When Not to Use

• Large datasets: When computational burden becomes prohibitive
• Real-time applications: When speed is more important than precision
• Exploratory analysis: When quick results are preferred
• Limited computational resources: When MCMC is not feasible

Practical Considerations

• Computationally more intensive than two-step approaches
• Requires careful MCMC diagnostics and convergence assessment
• Benefits from proper initialization of factors and parameters
• May require longer chains for convergence

Model Specification

Joint Factor-VAR Model

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

Simultaneous Estimation

All parameters {Λ, F_t, A, Σ} estimated jointly accounting for all uncertainties.

Prior Specifications

A | Σ ~ N(M, Σ ⊗ Ω)
Σ ~ IW(S, ν)
F_t ~ N(μ_F, Σ_F)
Λ ~ N(M_Λ, Σ_Λ)

Gibbs Sampling Steps

1. Draw factors F_t given parameters and data
2. Draw loadings Λ given factors and data
3. Draw VAR coefficients A given factors and data
4. Draw covariance Σ given all parameters and data

Computational Characteristics

Performance

• Speed: Slower than two-step approaches due to joint estimation
• Memory: Higher requirements for storing factor draws
• Convergence: May require longer chains due to complexity
• Scalability: Limited scalability with dataset size

MCMC Considerations

• Requires assessment of chain convergence
• May exhibit slow mixing due to factor identification
• Benefits from adaptive or optimized sampling schemes
• Needs careful initialization strategies

Factor-Augmented Analysis

Uncertainty Quantification

• Proper posterior distributions for all factor-related quantities
• Credible intervals account for factor estimation uncertainty
• More realistic uncertainty bands for impulse responses
• Better out-of-sample prediction intervals

Factor Identification

• Joint estimation helps with factor identification
• Rotation invariance addressed through identification restrictions
• More stable factor interpretation across samples
• Better handling of factor number uncertainty

One-Step vs. Two-Step Comparison

Theoretical Advantages

• Uncertainty: Properly accounts for factor estimation uncertainty
• Efficiency: Asymptotically more efficient parameter estimates
• Inference: Correct posterior distributions and credible intervals
• Consistency: Maintains theoretical coherence of Bayesian framework

Practical Trade-offs

• Computation: Significantly more expensive computationally
• Implementation: More complex to implement and diagnose
• Speed: Much slower than two-step alternatives
• Scalability: Limited applicability to very large datasets

Performance Differences

• Better uncertainty quantification but at computational cost
• More accurate inference in finite samples
• Superior out-of-sample density forecasts
• Better handling of model uncertainty

Implementation Strategy

Initialization

• Use two-step estimates for initial values
• Initialize factors using principal components
• Set reasonable starting values for all parameters
• Consider multiple starting points for robustness

MCMC Diagnostics

• Monitor convergence using standard diagnostics
• Check factor rotation and identification
• Assess effective sample sizes for all parameters
• Validate posterior concentration

Computational Optimization

• Use efficient sampling algorithms for factors
• Implement adaptive schemes where possible
• Consider parallel computing for factor sampling
• Monitor computational efficiency

Model Validation

Convergence Assessment

• Use multiple chain diagnostics (Gelman-Rubin, etc.)
• Check trace plots for all parameter blocks
• Assess effective sample sizes and autocorrelations
• Validate posterior stability across chains

Factor Analysis

• Examine factor loading posterior distributions
• Assess factor identification and interpretation
• Compare factor estimates across estimation methods
• Validate factor contribution to model fit

Predictive Performance

• Compare forecasting accuracy with two-step methods
• Evaluate prediction interval coverage
• Assess density forecast calibration
• Test structural interpretation stability

Applications

High-Stakes Forecasting

• Central bank policy applications requiring uncertainty quantification
• Financial risk management with proper interval forecasts
• Academic research requiring methodological rigor
• Applications where factor uncertainty is economically significant

Methodological Research

• Studies comparing one-step vs. two-step approaches
• Investigation of factor uncertainty effects
• Development of improved FAVAR methodologies
• Theoretical analysis of FAVAR properties

Computational Requirements

Hardware Considerations

• Requires substantial computational resources
• Benefits from multi-core processors for parallel sampling
• May need significant RAM for large factor systems
• Consider cloud computing for very large applications

Software Implementation

• Requires sophisticated MCMC implementation
• Benefits from optimized linear algebra routines
• May need specialized software for large-scale problems
• Consider using established FAVAR packages when available

References

• Bernanke, B.S., J. Boivin, and P. Eliasz (2005). "Measuring the Effects of Monetary Policy: A Factor-Augmented Vector Autoregressive (FAVAR) Approach"
• Pitt, M. and N. Shephard (1999). "Time-Varying Covariances: A Factor Stochastic Volatility Approach"
• Aguilar, O. and M. West (2000). "Bayesian Dynamic Factor Models and Portfolio Allocation"
• Lopes, H.F. and M. West (2004). "Bayesian Model Assessment in Factor Analysis"