5.6 Summary: stationarity testing

The basic stationarity diagnostics are the following

  • Plot your data. Look for
    • An increasing trend
    • A non-zero level (if no trend)
    • Strange shocks or steps in your data (indicating something dramatic changed like the data collection methodology)
  • Apply stationarity tests
    • adf.test() p-value should be less than 0.05 (reject null)
    • kpss.test() p-value should be greater than 0.05 (do not reject null)
  • If stationarity tests are failed, then try differencing to correct
    • Try ndiffs() in the forecast package or manually try different differences.