Some basic tests on time series outliers

Abstract

Outliers are common place in applied time series analysis and various types of structural changes occur frequently and raises the question of efficiency and adequacy in fitting models. The methods under consideration for the tests of time series outliers are the Peirce’s criterion, Chauvenet’s criterion and Grubbs’ test. A set of data was considered and later on tested for outliers. From the findings, the Peirce’s criterion identified two outliers in the data set while the Chauvenet’s and Grubbs’ tests both identified only one outlier. In the Peirce’s criterion, the result of two outliers were opposed by the Chauvenet’s criterion and Grubb’s Test because Peirce’s criterion accounts for the case where there is more than one suspect data point at once.

Journal of the Nigerian Association of Mathematical Physics, Volume 15 (November, 2009), pp 101 - 106