Module 1 Discussion #2 — Addendum (Configuration Modulation)

1 Addendum — Configuration modulation

This addendum uses the same FRED series as the original post, but varies a small set of configuration choices to show how diagnostics and forecasts change under different assumptions. The intent is to treat the pipeline as a system and make configuration effects explicit.

1.1 Configuration (edit + re-render)

# Configuration toggles
cfg <- list(
  series_id = "RSAFS",
  start_date = as.Date("1992-01-01"),
  transform = "none",   # options: "none", "log"
  window    = "all",    # options: "all", "recent"
  model     = "both"    # options: "ets", "arima", "both"
)

1.2 Setup + data retrieval (minimal, required for a standalone addendum)

library(fredr)
library(dplyr)
library(tsibble)
library(feasts)
library(fable)
library(ggplot2)
library(ggtime)

# FRED API key should already be in your environment as FRED_API_KEY
fredr_set_key(Sys.getenv("FRED_API_KEY"))

raw <- fredr(
  series_id = cfg$series_id,
  observation_start = cfg$start_date
)

fred_ts <- raw |>
  transmute(
    Month = yearmonth(date),
    value = value
  ) |>
  as_tsibble(index = Month)

1.3 Apply configuration (transform + window)

fred_cfg <- fred_ts |>
  mutate(y = dplyr::case_when(
    cfg$transform == "log" ~ log(value),
    TRUE ~ value
  ))

fred_view <- fred_cfg
if (cfg$window == "recent") {
  fred_view <- fred_cfg |> filter_index("2010 Jan" ~ .)
}

2 Diagnostics (modulated)

2.1 Time plot

fred_view |>
  autoplot(y) +
  labs(
    title = paste0("Time plot (transform=", cfg$transform, ", window=", cfg$window, ")"),
    x = "Month", y = "y"
  )

2.2 ACF

fred_view |>
  ACF(y) |>
  autoplot() +
  labs(
    title = paste0("ACF (transform=", cfg$transform, ", window=", cfg$window, ")"),
    x = "Lag", y = "Autocorrelation"
  )

3 Forecasts (model-family modulation)

fit_cfg <- fred_cfg |>
  model(
    ets   = if (cfg$model %in% c("ets","both")) ETS(y) else NULL,
    arima = if (cfg$model %in% c("arima","both")) ARIMA(y) else NULL
  )

fc_cfg <- fit_cfg |> forecast(h = "24 months")

3.1 Forecast plot

fc_cfg |>
  autoplot(fred_cfg) +
  labs(
    title = paste0("Forecasts (model=", cfg$model, ", transform=", cfg$transform, ")"),
    x = "Month", y = "y"
  )

4 Citation

Hyndman and Khandakar discuss automatic forecasting workflows and model selection in R (hyndman2008?).

@article{hyndman2008,
  title   = {Automatic Time Series Forecasting: The forecast Package for R},
  author  = {Hyndman, Rob J. and Khandakar, Yeasmin},
  journal = {Journal of Statistical Software},
  year    = {2008},
  volume  = {27},
  number  = {3},
  pages   = {1--22},
  doi     = {10.18637/jss.v027.i03}
}