Glossary

This page defines the main acronyms and terms used throughout the toolkit.

ELB

Effective lower bound. In this toolkit, ELB refers to a censoring constraint applied to selected observed series (typically short-term policy rates). When the observed series is at or below the bound (up to a tolerance), the model treats the “true” value as a latent shadow value subject to the constraint.

Shadow rate

A shadow rate is a latent (unobserved) policy rate intended to represent the stance of monetary policy when observed policy rates are constrained by an ELB.

VAR

Vector autoregression. A multivariate time-series model where each variable is regressed on its own lags and the lags of the other variables.

BVAR

Bayesian VAR. A VAR estimated with Bayesian priors over coefficients and innovation covariance.

NIW

Normal-Inverse-Wishart. A conjugate prior for VAR coefficients and the innovation covariance matrix.

Minnesota prior

A structured shrinkage prior for VAR coefficients, typically shrinking toward a random-walk / white-noise baseline with lag decay and cross-variable shrinkage.

In srvar-toolkit, PriorSpec.niw_minnesota_legacy(...) is the historical compatibility path, while PriorSpec.niw_minnesota_canonical(...) provides explicit equation-wise Minnesota shrinkage for homoskedastic and diagonal SV models.

SSVS

Stochastic search variable selection. A spike-and-slab prior with inclusion indicators that stochastically include/exclude predictor rows.

SV / SVRW

Stochastic volatility. A model for time-varying residual variances.

SVRW means the log-variance follows a random walk. In this toolkit, volatility is currently diagonal (series-specific variances, no time-varying covariances).

KSC

Kim-Shephard-Chib auxiliary mixture approximation for the log-(\chi^2) distribution used in many stochastic volatility samplers.

Gibbs sampler

An MCMC algorithm that iteratively samples from conditional distributions of each parameter block.