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

In dlmodeler: Generalized Dynamic Linear Modeler

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.

Details

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

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

Note

State representations are not unique, so other forms could be used to achieve the same goals.

Author(s)

Cyrille Szymanski <[e-mail]>

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)

Example output

dlmodeler documentation built on May 29, 2017, 11:33 a.m.