Supplementary material to "Technical note: A procedure to clean, decompose and aggregate time series"
François Ritter · 2021
Supplementary material of "A procedure to clean, decompose and aggregate univariate time series" Part I: Outliers Fig. S1.Location of the 600 random distributions picked among the Pearson family in the (π ππ₯ , π 2 ) space (panel a.1).Example of the three upper thresholds based on the boxplot rule (π’ 3π , π’ 4π and π’ 5π ) to detect outliers for a Betaprime distribution with a skewness of 5 and a kurtosis of β3 (panel a.2).The Generalized Lambda Distribution (GLD) system is shown for comparison in the (π ππ₯ , π 2 ) space (panel b.1).Similar figure than a.2 except for a distribution coming from the GLD system with a skewness of 5 and a kurtosis of β3 (panel b.2).β’ Figure 1 shows that the Generalized Lambda Distribution (GLD) system produces non-realistic outlier thresholds because each distribution from this system is bounded.The Pearson family is therefore preferred over the GLD system to model outlier behaviors.