BVAR: bv_metropolis – R documentation

Become an expert in R — Interactive courses, Cheat Sheets, certificates and more!

bv_metropolis

Metropolis-Hastings settings

Description

Function to provide settings for the Metropolis-Hastings step in bvar. Options include scaling the inverse Hessian that is used to draw parameter proposals and automatic scaling to achieve certain acceptance rates.

Usage

bv_metropolis(
  scale_hess = 0.01,
  adjust_acc = FALSE,
  adjust_burn = 0.75,
  acc_lower = 0.25,
  acc_upper = 0.45,
  acc_change = 0.01
)

bv_mh(
  scale_hess = 0.01,
  adjust_acc = FALSE,
  adjust_burn = 0.75,
  acc_lower = 0.25,
  acc_upper = 0.45,
  acc_change = 0.01
)

Arguments

`scale_hess`	Numeric scalar or vector. Scaling parameter, determining the range of hyperparameter draws. Should be calibrated so a reasonable acceptance rate is reached. If provided as vector the length must equal the number of hyperparameters (one per variable for `psi`).
`adjust_acc`	Logical scalar. Whether or not to further scale the variability of parameter draws during the burn-in phase.
`adjust_burn`	Numeric scalar. How much of the burn-in phase should be used to scale parameter variability. See Details.
`acc_lower, acc_upper`	Numeric scalar. Lower (upper) bound of the target acceptance rate. Required if adjust_acc is set to `TRUE`.
`acc_change`	Numeric scalar. Percent change applied to the Hessian matrix for tuning acceptance rate. Required if adjust_acc is set to `TRUE`.

Details

Note that adjustment of the acceptance rate by scaling the parameter draw variability can only be done during the burn-in phase, as otherwise the resulting draws do not feature the desirable properties of a Markov chain. After the parameter draws have been scaled, some additional draws should be burnt.

Value

Returns a named list of class bv_metropolis with options for bvar.

Examples

# Increase the scaling parameter
bv_mh(scale_hess = 1)

# Turn on automatic scaling of the acceptance rate to [20%, 40%]
bv_mh(adjust_acc = TRUE, acc_lower = 0.2, acc_upper = 0.4)

# Increase the rate of automatic scaling
bv_mh(adjust_acc = TRUE, acc_lower = 0.2, acc_upper = 0.4, acc_change = 0.1)

# Use only 50% of the burn-in phase to adjust scaling
bv_mh(adjust_acc = TRUE, adjust_burn = 0.5)

BVAR

Hierarchical Bayesian Vector Autoregression

v1.0.1

GPL-3 | file LICENSE

Authors

Nikolas Kuschnig [aut, cre] (<https://orcid.org/0000-0002-6642-2543>), Lukas Vashold [aut] (<https://orcid.org/0000-0002-3562-3414>), Michael McCracken [dtc], Serena Ng [dtc]

Initial release

2020-09-26