par_init_ssm: Initialize the parametric terms using MLEs

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

Description

Compute initial values for the factors and nonlinear parameter (if necessary) of the parametric component using the Kalman Filter.

Usage

par_init_ssm(Y, tau, f_p, orthogonalize = TRUE)

Arguments

Y	the T x m data observation matrix, where T is the number of time points and m is the number of observation points (NAs allowed)
tau	vector of observation points (m-dimensional)
f_p	a function to compute the parametric component, which must return a m x K_p matrix for K_p the number of parametric curves; may include a (scalar) nonlinear parameter argument
orthogonalize	logical; when TRUE, orthogonalize the loading curve matrix

Details

Compute initial values via the following algorithm:

1. Impute missing values in Y

2. Initialize lambda_p = 1 and compute F_p; orthogonalize if specified

3. Estimate Beta_p via least squares

4. Estimate an initial standard deviation sigma_0 via conditional MLE

5. Estimate lambda_p via conditional MLE and recompute F_p; orthogonalize if specified

Value

a list containing

• Beta_p the T x K_p matrix of parametric factors

• F_p the m x K_p matrix of parametric loading curves

• lambda_p the scalar nonlinear parameter

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