Effect of segmentation on financial time series pattern matching

• We evaluate financial time series pattern matching and segmentation methods.
• PIP, PAA, PLA, and TP segmentation methods are analyzed.
• TB, RB, HY, DT, and SAX pattern matching approaches are evaluated.
• PIP achieves better performance and is especially superior when used with RB and HY.

In financial time series pattern matching, segmentation is often performed as a pre-processing step to reduce the data points from the input sequence. The segmentation process extracts important data points and produces a time series with reduced data points. In this paper, we evaluate the effectiveness and accuracy of four approaches to financial time series pattern matching when used with four segmentation methods, the perceptually important points, piecewise aggregate approximation, piecewise linear approximation and turning points methods. The pattern matching approaches analysed in this paper include the template-based, rule-based, hybrid, decision tree, and Symbolic Aggregate approXimation (SAX) approaches. The analysis is performed twice, on a real data set (of Hang Seng Index prices from the Hong Kong stock market) and on a synthetic data set containing positive and negative cases of a technical pattern known as head-and-shoulders.

Figure optionsDownload as PowerPoint slide