dlmodeler (version 1.4-2)

dlmodeler.build.structural: Build a structural time series model

Description

Builds a DLM for a structural time series, consisting of a polynomial term (level, trend, ...), a "dummy seasonal" pattern, a trigonometric cycle term, and an observation disturbance.

Usage

dlmodeler.structural(pol.order = NULL, dseas.order = NULL, tseas.period = NULL, tseas.order = NULL, sigmaH = NA, pol.sigmaQ = 0, dseas.sigmaQ = 0, tseas.sigmaQ = 0, name = "structural")
dlmodeler.build.structural(pol.order = NULL, dseas.order = NULL, tseas.period = NULL, tseas.order = NULL, sigmaH = NA, pol.sigmaQ = 0, dseas.sigmaQ = 0, tseas.sigmaQ = 0, name = "structural")

Arguments

pol.order

order of the polynomial (0=constant, 1=linear, 2=cubic...), or NULL to ignore polynomial component.

dseas.order

period of the dummy seasonal pattern, or NULL to ignore dummy seasonal component.

tseas.period

period of the trigonometric seasonal pattern, or NULL to ignore trigonometric component.

tseas.order

number of harmonics in the trigonometric seasonal pattern, or NULL. Optional when tseas.period is an integer (a default value is used), mandatory otherwise.

sigmaH

std dev of the observation disturbance (if unknown, set to NA and use dlmodeler.fit to estimate it). Default = NA.

pol.sigmaQ

std dev of the polynomial state disturbances (if unknown, set to NA and use dlmodeler.fit to estimate it). Default = 0.

dseas.sigmaQ

std dev of the dummy seasonal state disturbances (if unknown, set to NA and use dlmodeler.fit to estimate it). Default = 0.

tseas.sigmaQ

std dev of the trigonometric seasonal state disturbances (if unknown, set to NA and use dlmodeler.fit to estimate it). Default = 0.

name

an optional name to be given to the resulting DLM.

Value

An object of class dlmodeler representing the structural model. This object can have the following components:

level

component representing the level (when pol.order = 0)

level+trend

component representing the level+trend (when pol.order = 1)

polynomial

component representing the level, trend, ... (when pol.order > 1)

seasonal

component representing the dummy seasonal pattern

trigonometric

component representing the trigonometric seasonal pattern

Details

The initial value P0inf is parametered to use exact diffuse initialisation (if supported by the back-end).

References

Durbin, and Koopman, Time Series Analysis by State Space Methods, Oxford University Press (2001), pages 38-45.

See Also

dlmodeler, dlmodeler.build, dlmodeler.build.polynomial, dlmodeler.build.dseasonal, dlmodeler.build.tseasonal, dlmodeler.build.arima, dlmodeler.build.regression

Examples

## Not run: 
# require(dlmodeler)
# 
# # generate some quarterly data
# n <- 80
# level <- 12
# sigma <- .75
# season <- c(5,6,8,2)
# y <- level + rep(season,n/4) + rnorm(n, mean=0, sd=sigma)
# 
# # deterministic level + quarterly seasonal
# mod <- dlmodeler.build.structural(pol.order=0, dseas.order=4,
#                                   sigmaH=sigma)
# f <- dlmodeler.filter(y, mod)
# 
# # show the one step ahead forecasts
# par(mfrow=c(2,1))
# plot(y,type='l')
# lines(f$f[1,],col='light blue')
# 
# # show the filtered level and seasonal components
# c <- dlmodeler.extract(f,mod,type="state")
# lines(c$level[1,],col='blue')
# plot(c$seasonal[1,],type='l',col='dark green')
# ## End(Not run)