Workshop Activities: Week 9

Week 9 Mini-Studio

Note

Back to Week 9 overview

This page follows week9.R. The goal is to practise complete ARIMA workflows: identify, estimate, diagnose, then forecast.

A. ARIMA practice sets: usmelec, leisure, h02

usmelec <- as_tsibble(fpp2::usmelec) |>
  rename(Month = index, Generation = value)

leisure <- us_employment |>
  filter(Title == "Leisure and Hospitality", year(Month) > 2000) |>
  mutate(Employed = Employed/1000) |>
  select(Month, Employed)

h02 <- PBS |>
  filter(ATC2 == "H02") |>
  summarise(Cost = sum(Cost))

Run the corresponding sections in week9.R, including diagnostics and forecasts.

NoteWhat to notice
  • For each series, state the differencing decision and seasonal structure.
  • Compare manual model choices with automated ARIMA output.
  • Check whether residual autocorrelation remains and what that means for intervals.

B. ARIMA vs ETS: aus_economy and cement

aus_economy <- global_economy |>
  filter(Code == "AUS") |>
  mutate(Population = Population / 1e6)

cement <- aus_production |>
  select(Cement) |>
  filter_index("1988 Q1" ~ .)

Use the train/test and forecasting sections in week9.R to compare ARIMA and ETS.

NoteWhat to notice
  • Which model wins on forecast accuracy for each dataset?
  • Is the numerical win practically meaningful or marginal?
  • How do model assumptions differ between ARIMA and ETS here?

Exam Practice

Important2024 Exam Questions
  • Section A: Q5
  • Section B: Q3 (g-h)
  • Section D: all questions

Frame each answer using the workflow: identify -> estimate -> diagnose -> forecast -> justify.