Package {rpanelauto}

Title:	Perform Automatic Estimation on Time Series in Multidimensional Panels
Version:	1.0.0
Description:	Provides a framework for automatic estimation of time-series models for univariate and multidimensional panel data. A user-supplied estimation function is applied independently to each time series, with optional processing before and after estimation. The function returns the transformed data together with the fitted models.
Language:	en-US
License:	MIT + file LICENSE
Encoding:	UTF-8
Depends:	R (≥ 4.3)
Suggests:	forecast, testthat (≥ 3.0.0)
Config/testthat/edition:	3
URL:	https://github.com/econcz/rpanelauto
BugReports:	https://github.com/econcz/rpanelauto/issues
Config/roxygen2/version:	8.0.0
NeedsCompilation:	no
Packaged:	2026-07-25 21:59:32 UTC; ilyabolotov
Author:	Ilya Bolotov [aut, cre]
Maintainer:	Ilya Bolotov <ilya.bolotov@vse.cz>
Repository:	CRAN
Date/Publication:	2026-08-05 07:10:02 UTC

Perform Automatic Estimation on Time Series in Multidimensional Panels

Description

Applies a user-supplied automatic estimation function independently to the selected time-series variables in a data frame. Estimation is performed separately for every combination of the preceding dimensions, while the final dimension determines the ordering of observations within each time series.

Usage

panelauto(
  data,
  vars,
  dimensions,
  estimate,
  frequency = 1,
  preestimation = NULL,
  postestimation = NULL,
  ...
)

Arguments

Details

When only one dimension is supplied, each selected variable is treated as a single ungrouped time series. Optional pre-estimation and post-estimation functions can be evaluated before and after each automatic estimation.

The final variable in dimensions determines the ordering of observations within each time series. All preceding dimension variables define independent groups.

For example, with

dimensions = c("country", "sector", "year")

observations are ordered by year, and separate models are estimated for every variable-country-sector combination.

With

dimensions = "year"

each variable listed in vars is treated as one ungrouped time series.

The dimension variables themselves are not modified, and the original row order of data is preserved.

Missing values in variables listed in vars are passed unchanged to the estimation function. Their admissibility and treatment therefore depend on the selected estimator.

Missing values in dimension variables are not permitted. Each combination of the dimension variables must uniquely identify one observation.

The order of operations within each panel group is:

1. Order observations by the final dimension.

2. Apply preestimation to the ordered group data.

3. Convert each selected variable to a ts object.

4. Apply estimate independently to each time series.

5. Apply postestimation to the group data and fitted model.

6. Store the fitted model and copy the transformed data back to the original row positions.

The preestimation and postestimation functions must preserve the number of observations and the values of all dimension variables within each group.

When several variables are modelled, postestimation should generally use variable-specific output names to avoid overwriting quantities created for another model. When the function declares a variable argument, panelauto supplies the name of the current variable automatically.

Any estimation function can be used, provided that it accepts the time series as its first argument. Functions with different interfaces can be supplied through a user-defined wrapper.

Value

A named list containing at least the following components:

data

The complete data frame after application of the optional preestimation and postestimation functions. The original row order is preserved.

models

A named list containing one model record for every selected variable and every combination of the grouping dimensions. Each model record contains the following components:

variable

Name of the modelled variable.

group

Named list identifying the values of the grouping dimensions. This component is an empty list when no grouping dimensions are supplied.

model

The fitted model returned by estimate, or the model component returned by postestimation.

See Also

arima, ts, auto.arima, ets

Examples

  ## Example: user-defined automatic autoregressive estimation

  set.seed(123456789)

  data <- data.frame(
      country = rep(c("A", "B"), each = 20L),
      year    = rep(2001:2020, 2L),
      y       = c(
          cumsum(stats::rnorm(20L)),
          cumsum(stats::rnorm(20L))
      )
  )

  automatic_ar <- function(series, orders = 1:3) {
      models <- lapply(
          orders,
          function(p) {
              stats::arima(
                  series,
                  order = c(p, 0L, 0L)
              )
          }
      )

      aic <- vapply(
          models,
          stats::AIC,
          numeric(1L)
      )

      models[[which.min(aic)]]
  }

  result <- panelauto(
      data       = data,
      vars       = "y",
      dimensions = c("country", "year"),
      estimate   = automatic_ar,
      orders     = 1:2
  )

  print(result$data)
  print(result$models[[1L]]$model)

  ## Example: automatic ARIMA estimation

  if (requireNamespace("forecast", quietly = TRUE)) {
      result <- panelauto(
          data       = data,
          vars       = "y",
          dimensions = c("country", "year"),
          estimate   = forecast::auto.arima,
          seasonal   = FALSE
      )

      print(result$models[[1L]]$model)
  }

  ## Pre-estimation, estimation, and fitted values

  result <- panelauto(
      data       = data,
      vars       = "y",
      dimensions = c("country", "year"),
      preestimation = function(data) {
          data$y <- data$y - mean(data$y)
          data
      },
      estimate = automatic_ar,
      orders   = 1:2,
      postestimation = function(data, model, variable) {
          fitted_name   <- paste0(variable, "_fitted")
          residual_name <- paste0(variable, "_residual")

          data[[residual_name]] <- as.numeric(stats::residuals(model))
          data[[fitted_name]]   <- data[[variable]] - 
                                   as.numeric(stats::residuals(model))

          list(
              data  = data,
              model = model
          )
      }
  )

  print(result$data)