Abstract-Similarity metric is of fundamental importance for similarity matching and subsequence query in time series applications. Most existing approaches measure the similarity by calculating and aggregating the point-to-point distance, few of them take the segment trend duration into account. In this paper, upon analyzing the properties of financial time series, we define a time series notation which is more intuitive and expressive. Base on that, a new similarity model is proposed. Experiments on both real foreign currency exchange rate data and stock market data are performed. The result shows the effectiveness and good accuracy of our method. The similarity model is also proved to be segmentation algorithm independent thus can be combined with other segmentations for similarity query, pattern matching, classification, and clustering.

No abstract

No abstract

No abstract