dfmMS: Dynamic factor model with Markov-switching states

In SebKrantz/dynfacto_R: Dynamic Factor Model Estimation for Nowcasting

Description

This function extends in a sense dfm in that it allows observation or transition matrices to follow a Markov-switching process, so that they are state-dependent.

Usage

dfmMS(data, nf = 1, ns = 2, x0 = rep(0, nf), P0 = diag(1, nf), init,
  ...)

Arguments

data	Data matrix
nf	Number of factors
ns	Number of states
x0	Initial value for state vector
P0	Initial value for state covariance, i.e. uncertainty of the initial state value vector
init	List with initial values for A, F, R, p. If not supplied, dfm function will be run and the estimates will be used as initial values.

Details

Note that this method does not implement Chang-Jin Kim algorithm. It is based on the same idea as the dfm function with the adaptation that Kalman filter is replaced by Kim filter which is able to deal with multiple states. Optimization is done over a list of parameters by a simple call to optim instead of an EM-algorithm.

Currently, state equation covariance matrix is restricted to identity.

SebKrantz/dynfacto_R documentation built on Dec. 31, 2020, 4:30 p.m.