Workshop Activities: Week 8

Week 8 Mini-Studio

Note

Back to Week 8 overview

This page follows the same sequence as week8.R. Use it as a stable seminar companion: run each code block in RStudio, then answer the prompts in your own words.

TipHow to use this page
  1. Open week8.R beside this page.
  2. Work through sections A to C in order.
  3. Explain what each model is doing, not just what command ran.

A. Egyptian exports: identify an ARIMA model

egypt <- global_economy |>
  filter(Code == "EGY")

egypt |> autoplot(Exports) +
  labs(y = "% of GDP", title = "Egyptian Exports")

egypt |>
  features(Exports, unitroot_ndiffs)

egypt |>
  gg_tsdisplay(Exports, plot_type = "partial")

fit <- egypt |>
  model(ARIMA(Exports))

report(fit)
gg_tsresiduals(fit)
NoteWhat to notice
  • Decide if the raw series looks stationary and whether differencing is required.
  • Compare visual evidence (plot, ACF/PACF) with unit root outputs.
  • Check whether residual diagnostics support the fitted ARIMA model.

B. Central African Republic exports: compare candidate models

caf <- global_economy |>
  filter(Code == "CAF")

caf |> autoplot(Exports)

caf_fit <- caf |>
  model(
    arima210 = ARIMA(Exports ~ pdq(2,1,0)),
    arima013 = ARIMA(Exports ~ pdq(0,1,3)),
    stepwise = ARIMA(Exports),
    fullsearch = ARIMA(Exports, stepwise = FALSE, approximation = FALSE)
  )

glance(caf_fit) |>
  arrange(AICc) |>
  select(.model:BIC)

caf_fit |>
  select(fullsearch) |>
  gg_tsresiduals()
NoteWhat to notice
  • Which models are competitive by AICc and why?
  • Does the best information-criterion model also pass residual checks?
  • Explain your final model choice for forecasting.

C. Continue the guided examples in week8.R

Use the remaining script sections (us_change, mink, web_usage, elecequip) to practise manual vs automatic ARIMA selection.

ImportantPractice focus
  • Write one paragraph per example explaining: stationarity decision, candidate models, chosen model, and residual conclusion.
  • Keep a shortlist of plausible alternatives, not only one “best” model.