Novel pattern matching methods for stock data analysis

Abstract

Efficient retrieval and matching of time series data are required in the stock pattern matching process. A new pattern-matching scheme called Visually and Practically Important Point (VPIP) is proposed. By adopting the method, encouraging experiment is reported from the tests that there is an association between the bid sequences and transaction price time series in the selected Chicago Stock Exchange, while currently rarely had any relevant study covered the relational effect of the bidder factor on stock price. The contribution is that it provides more reference information for the decision-making when trading stock data. For stock data, it has its special technical features such as the Head and Shoulder pattern or the peak values. Therefore, a suitable pattern matching method is needed for considering the characteristics of stock data. A flexible real-time hybrid pattern-matching algorithm is proposed. This method with combination of algorithms outperforms others in differentiating the prototype stock patterns or even distorted patterns. The efficiency and effectiveness of the method were demonstrated via extensive experiments on subsequence matching queries against the real stock price dataset as well as a synthetic dataset. The contributions of this work are as follows. Firstly, for the technical contribution, a novel pattern-matching scheme called Visually and Practically Important Point (VPIP) is proposed. By using this scheme, most of the meaningful features can be kept in the extracted points. A flexible real-time hybrid pattern-matching algorithm is also proposed. The combination of Spearman’s Rank Correlation Coefficient and rule sets algorithms outperforms other methods in differentiating the prototype stock patterns or even distorted patterns, both effectively and efficiently. Secondly, for the applied contribution, by adopting the VPIP algorithm, an association between bid and transaction price time series is found. We can make an earlier decision on whether to transact stock or not. It can provide pre-noticing signal for those investors when large fluctuation in transaction price will occur, and then support users to make their short-term trading decisions.

Date of Award	15 Jul 2008
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Huai Qing WANG (Supervisor)