In robjhyndman/hts: Hierarchical and Grouped Time Series

knitr::opts_chunk$set(collapse = TRUE, comment = "#>", fig.path = "man/figures/")

hts

hts is retired, with minimum maintenance to keep it on CRAN. We recommend using the fable package instead.

The R package hts presents functions to create, plot and forecast hierarchical and grouped time series.

Installation

You can install the stable version on R CRAN.

install.packages('hts', dependencies = TRUE)

You can also install the development version from Github

# install.packages("devtools")
devtools::install_github("earowang/hts")

Usage

Example 1: hierarchical time series

library(hts)

# hts example 1
print(htseg1)
summary(htseg1)
aggts1 <- aggts(htseg1)
aggts2 <- aggts(htseg1, levels = 1)
aggts3 <- aggts(htseg1, levels = c(0, 2))
plot(htseg1, levels = 1)
smatrix(htseg1)  # Return the dense mode

# Forecasts
fcasts1.bu <- forecast(
  htseg1, h = 4, method = "bu", fmethod = "ets", parallel = TRUE
)
aggts4 <- aggts(fcasts1.bu)
summary(fcasts1.bu)
fcasts1.td <- forecast(
  htseg1, h = 4, method = "tdfp", fmethod = "arima", keep.fitted = TRUE
)
summary(fcasts1.td)  # When keep.fitted = TRUE, return in-sample accuracy
fcasts1.comb <- forecast(
  htseg1, h = 4, method = "comb", fmethod = "ets", keep.fitted = TRUE
)
aggts4 <- aggts(fcasts1.comb)
plot(fcasts1.comb, levels = 2)
plot(fcasts1.comb, include = 5, levels = c(1, 2))

Example 2: hierarchical time series

# hts example 2
data <- window(htseg2, start = 1992, end = 2002)
test <- window(htseg2, start = 2003)
fcasts2.mo <- forecast(
  data, h = 5, method = "mo", fmethod = "ets", level = 1,
  keep.fitted = TRUE, keep.resid = TRUE
)
accuracy.gts(fcasts2.mo, test)
accuracy.gts(fcasts2.mo, test, levels = 1)
fcasts2.td <- forecast(
  data, h = 5, method = "tdgsa", fmethod = "ets", 
  keep.fitted = TRUE, keep.resid = TRUE
)
plot(fcasts2.td, include = 5)
plot(fcasts2.td, include = 5, levels = c(0, 2))

Example 3: grouped time series

# gts example
plot(infantgts, levels = 1)

fcasts3.comb <- forecast(infantgts, h = 4, method = "comb", fmethod = "ets")
agg_gts1 <- aggts(fcasts3.comb, levels = 1)
agg_gts2 <- aggts(fcasts3.comb, levels = 1, forecasts = FALSE)
plot(fcasts3.comb)
plot(fcasts3.comb, include = 5, levels = c(1, 2))

fcasts3.combsd <- forecast(
  infantgts, h = 4, method = "comb", fmethod = "ets",
  weights = "sd", keep.fitted = TRUE
)

fcasts3.combn <- forecast(
  infantgts, h = 4, method = "comb", fmethod = "ets",
  weights = "nseries", keep.resid = TRUE
)