Clustering financial time series

Abstract

Time series clustering is heavily based on choosing a proper dissimilarity measure between a pair of time series. We present several dissimilarity measures and use two synthetic datasets to evaluate their performance. Hierarchical clustering and network analysis methods are used to perform cluster analysis on stock price time series of 594 US-based companies in order to verify whether stock prices of companies operating within an industry have common uctuations. The results of the thesis show that some companies within the same industry do form clusters, while others are relatively scattered.