Clustering techniques applied to outlier detection of financial market series using a moving window filtering algorithm

In this study we combine clustering techniques with a moving window algorithm in order to filter financial market data outliers. We apply the algorithm to a set of financial market data which consists of 25 series selected from a larger dataset using a cluster analysis technique taking into account the daily behaviour of the market; each of these series is an element of a cluster that represents a different segment of the market. We set up a framework of possible algorithm parameter combinations that detect most of the outliers by market segment. In addition, the algorithm parameters that have been found can also be used to detect outliers in other series with similar economic behaviour in the same cluster. Moreover, the crosschecking of the behaviour of different series within each cluster reduces the possibility of observations being misclassified as outliers.