Rfssa (version 3.1.0)

fpredinterval: FSSA Forecasting Bootstrap Prediction Interval

Description

Calculate the bootstrap prediction interval for functional singular spectrum analysis (FSSA) forecasting predictions of univariate functional time series (funts) observed over a one-dimensional domain.

Usage

fpredinterval(
  Y,
  O = floor(Y$N * 0.7),
  L = floor((Y$N * 0.7)/12),
  ntriples = 10,
  Bt = 100,
  h = 1,
  alpha = 0.05,
  method = "recurrent",
  tol = 10^-3
)

Value

a list of numeric vectors: point forecast, lower, and upper bounds.

Arguments

Y

an object of class funts.

O

a positive integer specifying the training set size.

L

a positive integer specifying the window length.

ntriples

the number of eigentriples to use for forecasts.

Bt

a positive integer specifying the number of bootstrap samples.

h

an integer specifying the forecast horizon.

alpha

a double (0 < alpha < 1) specifying the significance level.

method

a character string: "recurrent" or "vector" forecasting.

tol

a double specifying tolerated error in the approximation.

Examples

if (FALSE) {
data("Callcenter")
pred_interval <- fpredinterval(
  Y = Callcenter, O = 310,
  L = 28, ntriples = 7, Bt = 10000, h = 3
)

# Plot the forecast and prediction interval using ggplot
df <- data.frame(
  x = 1:240,
  y = pred_interval$forecast,
  lower = pred_interval$lower,
  upper = pred_interval$upper
)
require(ggplot2)
# Create the ggplot
ggplot(df, aes(x = x, y = y)) +
  geom_line(linewidth = 1.2) +
  scale_x_continuous(
    name = "Time",
    breaks = c(1, 60, 120, 180, 240),
    labels = c("00:00", "06:00", "12:00", "18:00", "24:00"),
  ) +
  scale_y_continuous(name = "Sqrt of Call Numbers") +
  ggtitle("Prediction Intervals for Jan. 3, 2000") +
  geom_ribbon(aes(ymin = lower, ymax = upper), fill = "darkolivegreen3", alpha = 0.3) +
  theme_minimal()
}