dlm: dlmForecast – R documentation

Become an expert in R — Interactive courses, Cheat Sheets, certificates and more!

dlmForecast

Prediction and simulation of future observations

Description

The function evaluates the expected value and variance of future observations and system states. It can also generate a sample from the distribution of future observations and system states.

Usage

dlmForecast(mod, nAhead = 1, method = c("plain", "svd"), sampleNew = FALSE)

Arguments

`mod`	an object of class `"dlm"`, or a list with components `m0`, `C0`, `FF`, `V`, `GG`, and `W`, defining the model and the parameters of the prior distribution. `mod` can also be an object of class `"dlmFiltered"`, such as the output from `dlmFilter`.
`nAhead`	number of steps ahead for which a forecast is requested.
`method`	`method="svd"` uses singular value decomposition for the calculations. Currently, only `method="plain"` is implemented.
`sampleNew`	if `sampleNew=n` for an integer `n`, them a sample of size `n` from the forecast distribution of states and observables will be returned.

Value

A list with components

`a`	matrix of expected values of future states
`R`	list of variances of future states
`f`	matrix of expected values of future observations
`Q`	list of variances of future observations
`newStates`	list of matrices containing the simulated future values
	of the states. Each component of the list corresponds
	to one simulation.
`newObs`	same as `newStates`, but for the observations.

The last two components are not present if sampleNew=FALSE.

Note

The function is currently entirely written in R and is not particularly fast. Currently, only constant models are allowed.

Author(s)

Giovanni Petris GPetris@uark.edu

Examples

## Comparing theoretical prediction intervals with sample quantiles
set.seed(353)
n <- 20; m <- 1; p <- 5
mod <- dlmModPoly() + dlmModSeas(4, dV=0)
W(mod) <- rwishart(2*p,p) * 1e-1
m0(mod) <- rnorm(p, sd=5)
C0(mod) <- diag(p) * 1e-1
new <- 100
fore <- dlmForecast(mod, nAhead=n, sampleNew=new)
ciTheory <- (outer(sapply(fore$Q, FUN=function(x) sqrt(diag(x))), qnorm(c(0.1,0.9))) +
             as.vector(t(fore$f)))
ciSample <- t(apply(array(unlist(fore$newObs), dim=c(n,m,new))[,1,], 1,
                    FUN=function(x) quantile(x, c(0.1,0.9))))
plot.ts(cbind(ciTheory,fore$f[,1]),plot.type="s", col=c("red","red","green"),ylab="y")
for (j in 1:2) lines(ciSample[,j], col="blue")
legend(2,-40,legend=c("forecast mean", "theoretical bounds", "Monte Carlo bounds"),
       col=c("green","red","blue"), lty=1, bty="n")

dlm

Bayesian and Likelihood Analysis of Dynamic Linear Models

v1.1-5

GPL (>= 2)

Authors

Giovanni Petris [aut, cre], Wally Gilks [ctb] (Author of original C code for ARMS)

Initial release

2018-05-30