Workshop Activities: Week 10

Week 10 Mini-Studio

Note

Back to Week 10 overview

This page follows week10.R. Focus on regression-based forecasting, variable selection, and harmonic terms.

A. Multiple regression for us_change

Core objective: forecast US consumption growth using economic regressors.

fit_cons <- us_change |>
      model(TSLM(Consumption ~ Income + Savings + Production + Unemployment))

report(fit_cons)
fit_cons |> gg_tsresiduals()

Then run the full model-comparison and scenario sections in week10.R.

NoteWhat to notice
  • Which regressors are consistently useful across candidate models?
  • How do AICc/CV trade-offs affect your final model choice?
  • Do residual plots suggest missing nonlinear terms or interactions?

B. Harmonic regression for us_gasoline

Core objective: model weekly seasonality with Fourier terms.

us_gasoline |>
      model(
            K06 = TSLM(log(Barrels) ~ trend(knots = yearweek(c("2006 W1", "2011 W1"))) + fourier(K = 6)),
            K10 = TSLM(log(Barrels) ~ trend(knots = yearweek(c("2006 W1", "2011 W1"))) + fourier(K = 10))
      ) |>
      glance()

Use week10.R to inspect diagnostics and produce forecasts.

NoteWhat to notice
  • How does increasing K change fit quality vs complexity?
  • Which model balances smooth seasonality and residual behavior best?
  • Explain why harmonic terms are useful for high-frequency seasonal data.

Exam Practice

Important2023 Exam Questions
  • Section A: Q4, Q5
  • Section B: Q3 (g-h)

Write responses in model-selection language: data features, candidate models, diagnostics, and forecast implications.