Seasonal Adjustment via the CAMPLET Method

Seasonal adjustment with CAMPLET (Abeln and Jacobs (2023) ), a transparent, adaptive method that corrects for seasonality and outliers via arithmetic-progression updates, without model identification or optimisation, and without revising earlier adjusted values when new observations arrive. A pure R port of the original 'Excel' 'VBA' implementation, validated against its output.

camplet

A pure R implementation of CAMPLET, a transparent, adaptive seasonal adjustment method that corrects for seasonality and outliers via arithmetic-progression updates, without model identification or optimisation, and without revising earlier adjusted values when new observations arrive. The method is described in Abeln & Jacobs (2023), Seasonal Adjustment Without Revisions.

Ported from the original VBA macro (D200603-CampletExcel-v4.2, basCBFherzien.bas) and validated against its own output (see tests/testthat/test-camplet.R). A Java port for JDemetra+ v3 lives at jdplus-camplet; this package has no dependency on it or on Java.

Installation

install.packages("camplet")

# development version
# install.packages("remotes")
remotes::install_git("https://gitrepo.service.rug.nl/team-data-science/camplet-r.git")

Usage

library(camplet)

res <- camplet(AirPassengers)
res$accuracy
head(as.data.frame(res))

camplet() is univariate: one series in, one result out. To adjust many series, loop over them yourself (e.g. lapply()/purrr::map() over a list of series, or dplyr::group_by() |> group_modify() on a long data frame) - the same way you would with seasonal::seas() or stats::stl().

The returned object's $data has raw/sa/seasonal columns; $diagnostics (or as.data.frame(res, diagnostics = TRUE), the default) adds the per-period adaptive-parameter trace that the original VBA tool wrote to its "Adj_" worksheets (ca, mu, pa, le, ti, expo_e, abs_pct_e, years, weight, growth).

Input requirements

CAMPLET reads observations by position: the k-th value is taken to be the period after the (k-1)-th. A skipped period therefore shifts every later observation into the wrong season while still producing plausible-looking output. To catch this, camplet():

• stops if x contains NA, NaN or Inf (fill or trim the gaps first);
• stops if dates has duplicated, out-of-order or irregularly spaced timestamps, e.g. a missing month;
• warns if an explicit periodicity contradicts monthly or quarterly data, e.g. camplet(AirPassengers, periodicity = 4).

A plain numeric vector without dates can't be checked, so making sure it has no gaps is up to you. See ?camplet, section "Input checks".