A multiplicative season did not fix the airline passengers series

changepoint-detector on the classic monthly airline passengers data: 21 false steps with an additive season, 22 with a multiplicative one, 8 after taking logs.

Tools and projects · 1 min read · Source
changepoint-detectorstatisticstime-series

The --multiplicative season in changepoint-detector was tested only on simulated steps. Today it met a real series with a season that grows with the level: monthly airline passengers, 1949 to 1960. The series has no documented step, so every changepoint the tool reports is a false alarm.

ModelSteps reported
mean, additive season21
mean, --multiplicative22
--trend bend, additive / multiplicative20 / 15
log series, --trend bend8
log series, --trend bend, meanvar5

The multiplicative season did not help, because the underlying rise is exponential and a level model cannot follow it. Taking logs first is the better route, which agrees with the earlier UK gas result. One series, so no claim about rates; the counts are pinned in airline-passengers.test.mjs.

More in Tools and projects

All writing · All in this topic · All tools and research