trendModelArb: Univariate trend components

In dajmcdon/myFKF: Fast Kalman Filtering and Smoothing

Description

This function can generate the appropriate state and transition matrices (Tt, Zt, GGt, HHt) for different possible trend components. Possible univariate trends are given with:

y(t) = mu(t) + extra stuff

(1 - phi1 L)^d mu(t) = nu(t-1) + xi(t-1)

(1 - phi2 L)^r nu(t) = eta(t)

Noise terms eta and xi are iid normal mean 0 with variances sig2eta etc. This function generates all possible combinations of these. See Chapter 3 in (Durbin and Koopman) for more information. "Extra stuff" can be simply an error variance or may include seasonal, cyclical, or other components.

Usage

trendModelArb(parmlist)

Arguments

parmlist

The input contains named elements d, phi1, r, phi2, sig2eta, sig2xi. All are scalars parameterizing the above model. It also contains a logical for nu, which determines whether to include the nu component (default is FALSE). All of these can be set with parmList. The default values (generated by parmList) are d = 2 phi1 = 1 r = 2 phi2 = 1 sig2eta = 1 sig2xi = 1 nu = FALSE

Value

A list of matrices with components Z, Tt, Q, and R.

dajmcdon/myFKF documentation built on May 3, 2019, 5:16 p.m.