predict.arfima {arfima}

Predicts from a fitted object.

Description

Performs prediction of a fitted arfima object. Includes prediction for each mode and exact and limiting prediction error standard deviations. NOTE: the standard errors in beta are currently not taken into account in the prediction intervals shown. This will be updated as soon as possible.

Usage

## S3 method for class 'arfima'
predict(
  object,
  n.ahead = 1,
  prop.use = "default",
  newxreg = NULL,
  predint = 0.95,
  exact = c("default", T, F),
  setmuhat0 = FALSE,
  cpus = 1,
  trend = NULL,
  n.use = NULL,
  xreg = NULL,
  ...
)

Arguments

object

A fitted arfima object

n.ahead

The number of steps ahead to predict

prop.use

The proportion (between 0 and 1) or percentage (between >1 and 100) of data points to use for prediction. Defaults to the string "default", which sets the number of data points n.use to the minimum of the series length and 1000. Overriden by n.use.

newxreg

If a regression fit, the new regressors

predint

The percentile to use for prediction intervals assuming normal deviations.

exact

Controls whether exact (based on the theoretical autocovariance matrix) prediction variances are calculated (which is recommended), as well as whether the exact prediction formula is used when the process is differenced (which can take a fair amount of time if the length of the series used to predict is large). Defaults to the string "default", which is TRUE for the first and FALSE for the second. A Boolean value (TRUE or FALSE) will set both to this value.

setmuhat0

Experimental. Sets muhat equal to zero

cpus

The number of CPUs to use for prediction. Currently not implemented

trend

An optional vector the length of n.ahead or longer to add to the predictions

n.use

Directly set the number mentioned in prop.use.

xreg

Alias for newxreg

...

Optional arguments. Currently not used

Value

A list of lists, ceiling(prop.use * n)one for each mode with relavent details about the prediction

Author(s)

JQ (Justin) Veenstra

References

Veenstra, J.Q. Persistence and Antipersistence: Theory and Software (PhD Thesis)

See Also

arfima, plot.predarfima, print.predarfima

Examples

set.seed(82365)
sim <- arfima.sim(1000, model = list(dfrac = 0.4, theta=0.9, dint = 1))
fit <- arfima(sim, order = c(0, 1, 1), back=TRUE)
fit
pred <- predict(fit, n.ahead = 5)
pred
plot(pred, numback=50)
#Predictions aren't really different due to the
#series.  Let's see what happens when we regress!

set.seed(23524)
#Forecast 5 ahead as before
#Note that we need to integrate the regressors, since time series regression
#usually assumes that regressors are of the same order as the series.
n.fore <- 5
X <- matrix(rnorm(3000+3*n.fore), ncol = 3)
X <- apply(X, 2, cumsum)
Xnew <- X[1001:1005,]
X <- X[1:1000,]
beta <- matrix(c(2, -.4, 6), ncol = 1)
simX <- sim + as.vector(X%*%beta)
fitX <- arfima(simX, order = c(0, 1, 1), xreg = X, back=TRUE)
fitX
#Let's compare predictions.
predX <- predict(fitX, n.ahead = n.fore, xreg = Xnew)
predX
plot(predX, numback = 50)
#With the mode we know is really there, it looks better.
fitX <- removeMode(fitX, 2)
predXnew <- predict(fitX, n.ahead = n.fore, xreg = Xnew)
predXnew
plot(predXnew, numback=50)