This paper compares the tractability of four discordancy statistics for modelling outliers based on extremeness. They are: the Generlaized Extreme Studentized Deviate (ESD), Grubb’s test, Hampel’s method and the quartile method. The last two methods are seen to detect outliers even for datasets that are not approximately normal, although Hampel outperforms the quartile method in some cases. However, a multiplier effect of 2.2 is proposed for the quartile method in addition to the robust statistics for accommodating the outliers.
Keywords: Autocorrelation function (ACF)., Generalized Extreme Studentized Deviate (ESD), Outliers, Time Series
