Package {icomb}

Produce coherent forecasts using the information combination method

Description

The forecast function allows you to generate coherent future predictions for hierarchical or grouped time series based on fitted models.

Usage

## S3 method for class 'mdl_icomb_lst'
forecast(
  object,
  new_data = NULL,
  h = NULL,
  point_forecast = list(.mean = mean),
  ...
)

Arguments

Details

The forecasts returned contain both point forecasts and their distribution. A specific forecast interval can be extracted from the distribution using the distributional::hilo() function. These intervals are stored in a single column using the hilo class, to extract the numerical upper and lower bounds you can use fabletools::unpack_hilo().

Value

A fable containing the following columns:

• .model: The name of the model used to obtain the forecast. Taken from the column names of models in the provided mable.

• The forecast distribution. The name of this column will be the same as the dependent variable in the model(s). If multiple dependent variables exist, it will be named .distribution.

• Point forecasts computed from the distribution using the functions in the point_forecast argument. If bootstrap = TRUE, point forecasts are generated using bootstrapped sample paths.

Examples

library(fable)
library(fabletools)
library(tsibble)
library(distributional)

tourism_hts <- tourism |>
  aggregate_key(State,
                Trips = sum(Trips))

fit <- tourism_hts |>
  model(base = ETS(Trips)) |>
  reconcile(ols = min_trace(base, method = "ols"),
            icomb = icomb(base, train_size = 75))


fit |>
  forecast(bootstrap = TRUE, times = 1000) |>
  hilo(level = c(80, 95))

Glance a mable which contains an information combination reconciliation method

Description

It uses the models within a mable to produce a one row summary statistics of their fits.

Usage

## S3 method for class 'mdl_icomb_lst'
glance(x, ...)

Arguments

Value

The tibble contains the output of glance() for that model, with an added logical column .included indicating whether the corresponding node is present in the reconciliation process when the information combination method is used.

Examples

library(fable)
library(fabletools)
library(tsibble)
library(dplyr)

tourism_hts <- tourism |>
  aggregate_key(State,
                Trips = sum(Trips))

fit <- tourism_hts |>
  model(base = ETS(Trips)) |>
  reconcile(ols = min_trace(base, method = "ols"),
            icomb = icomb(base, train_size = 75))

fit |>
  glance()

Information combination reconciliation

Description

Implements the Information Combination (IComb) method for forecast reconciliation, combining information from multiple base forecasts through a regression-based framework that can be estimated using penalized regression techniques. The penalty parameter is estimated using the rolling forecast origin cross-validation.

Usage

icomb(
  models,
  train_size,
  alpha = 1,
  standardize = FALSE,
  standardize_response = FALSE,
  intercept = TRUE,
  lambda = NULL,
  lambda_min_ratio = "expand",
  nlambda = 100,
  maxit = 1e+07,
  thresh = 1e-07,
  exact = TRUE
)

Arguments

Value

A 'global' model which is icomb coherent

Parallel computing

The model fitting and icomb cross-validation procedure can be parallelized using the future package. By specifying a future::plan() prior to model estimation or the reconciliation step, the computations will be carried out according to the chosen future strategy.

Progress

Progress on the cross-validation procedure can be obtained by wrapping the code with progressr::with_progress(). Further customization on how progress is reported can be controlled using the progressr package.

Note

Missing values are removed prior to applying the information combination method.

The default value of lambda_min_ratio may result in very small values of \lambda, which can increase computation time. It is therefore recommended to choose this parameter according to the specific requirements of your application.

Author(s)

Shanika L Wickramasuriya

References

Nguyen, M., Vahid, F., & Wickramasuriya, S. L. (2025). Hierarchical Forecasting: The Role of Information Combination (Working Paper No. 11/25). Department of Econometrics and Business Statistics, Monash University.

See Also

fabletools::reconcile(), fabletools::aggregate_key()

Examples

library(fable)
library(fabletools)
library(tsibble)
library(dplyr)
library(lubridate)
library(ggtime)

tourism_hts <- tourism |>
  aggregate_key(State,
                Trips = sum(Trips))

fit <- tourism_hts |>
  model(base = ETS(Trips)) |>
  reconcile(ols = min_trace(base, method = "ols"),
            icomb = icomb(base, train_size = 75))

fit |>
  forecast(h = "3 years")

# extracting results from icomb cross-validation
fit |>
  pull(icomb) |>
  attr("icombfit")


# Extracting probabilistic forecasts
fit |>
  forecast(h = "3 years", bootstrap = TRUE, times = 1000) |>
  filter(State == "Victoria") |>
  autoplot(filter(tourism_hts,
                  State == "Victoria", year(Quarter) > 2010))

# grouped structure
tourism_gts <- tourism |>
  aggregate_key(State * Purpose,
                Trips = sum(Trips))

fit <- tourism_gts |>
  model(base = ETS(Trips)) |>
  reconcile(ols = min_trace(base, method = "ols"),
            icomb = icomb(base, train_size = 75))

# Parallelizing cross-validation
library(future)
plan(multisession, workers = 2)

tourism_gts |>
  model(base = ETS(Trips)) |>
  reconcile(ols = min_trace(base, method = "ols"),
            icomb = icomb(base, train_size = 75)) |>
  forecast(h = "3 years")
plan(sequential)

fit <- tourism_gts |>
  model(base = ETS(Trips))

# progress
progressr::with_progress(
fit |>
  reconcile(ols = min_trace(base, method = "ols"),
            icomb = icomb(base, train_size = 35))
  )

Calculate a sequence of penalty parameters

Description

Generates a decreasing sequence of regularization parameters for penalized multivariate regression by estimating the maximum lambda from the data and constructing a log-spaced path to a minimum value, with options for standardization and handling of zero-variance predictors and responses.

Usage

lambda_path(
  x,
  y,
  alpha = 1,
  standardize = FALSE,
  standardize_response = FALSE,
  intercept = TRUE,
  lambda_min_ratio = "expand",
  nlambda = 100
)

Arguments

Value

A sequence of penalty parameters

Author(s)

Shanika L Wickramasuriya

Examples

set.seed(2024)
n <- 100
p <- 50
k <- 2
x <- matrix(rnorm(n * p), ncol = p)
y <- matrix(rnorm(n * k), ncol = k)
glmnet::glmnet(x, y, family = "mgaussian", standardize = FALSE,
               standardize.response = FALSE, intercept = TRUE)$lambda
lambda_path(x, y, lambda_min_ratio = "glmnet")

Objects exported from other packages

Description

These objects are imported from other packages. Follow the links below to see their documentation.

generics

forecast, glance