Time series outlier detection using non parametric test. This methodology is based on three points tendence to fall or not to fall into line. In grouped time series we use the grouped tendence to align or not with a very small time range. If we use the linear analysis we use the general variance inside the time series to test if a single variation is too heavy or not. For other informations see: A. Venturini (2011) (https://rivista-statistica.unibo.it/article/download/3617/2968) but also (https://sites.google.com/site/andreaventurini65/home/outlier-detection) or (https://www.r-bloggers.com/time-series-outlier-detection-a-simple-r-function/) for an informal explanation.
Andrea Venturini "Time Series Outlier Detection: A New Non Parametric Methodology (washer)" Statistica — Universita' di Bologna, Vol. 71, 2011, pp. 329-344.