sampleMGP: Multiplicative Gamma Process (MGP) Sampler

In drkowal/FDLM: A Bayesian Multivariate Functional Dynamic Linear Model

Description

Sample the global parameters, delta.h, with tau.h = cumprod(delta.h), from the MGP prior for factor models

Usage

sampleMGP(theta.jh, delta.h, a1 = 2, a2 = 3)

Arguments

theta.jh	the p x K matrix with entries theta.jh ~ N(0, tau.h^-1), j=1:p, h=1:K
delta.h	the K-dimensional vector of previous delta.h values, tau.h[h] = prod(delta.h[1:h])
a1	the prior parameter for factor 1: delta.h[1] ~ Gamma(a1, 1)
a2	the prior parameter for factors 2:K: delta.h[h] ~ Gamma(a2, 1) for h = 2:K

Value

delta.h, the K-dimensional vector of multplicative components, where tau.h[h] = prod(delta.h[1:h])

Note

The default a1 = 2 and a2 = 3 appears to offer the best performance in Durante (2017).

drkowal/FDLM documentation built on May 20, 2019, 5:20 p.m.