148  Introduction to Box-Jenkins Analysis

This part of the handbook presents the Box-Jenkins workflow (Box and Jenkins 1970) for equi-distant quantitative time series: identification, estimation, diagnostics, and forecasting. The aim is practical model building: transform a raw series to approximate stationarity, identify plausible ARIMA orders, estimate competing models, and keep only models whose residuals behave like white noise.

The chapters are organised as an applied sequence:

  1. theoretical concepts (stationarity, ACF/PACF, ARMA behavior),
  2. stationarity induction (differencing and variance-stabilizing transforms),
  3. ARIMA identification and estimation,
  4. forecasting,
  5. extensions for external information (intervention, CCF, transfer function, and GtS/ECM).

The methodology is illustrated with five datasets that each highlight a different issue (seasonality, structural breaks, external drivers, or web-traffic dynamics). The examples are intentionally reused across chapters so readers can compare methods on the same data rather than relearn a new dataset at each step.

The following time series are used in the following chapters:

The run sequence plots of the first three time series are displayed here:

Monthly Unemployment

Interactive Shiny app (click to load).
Open in new tab

Monthly Births

Interactive Shiny app (click to load).
Open in new tab

Monthly US Soldiers Killed In Action (Iraq)

Interactive Shiny app (click to load).
Open in new tab