A formal approach to candlestick pattern classification in financial time series

Patterns with varying numbers of candlestick-shaped features are commonly used by analysts to predict future price trends in financial markets. Although general descriptions of candlestick patterns have been reported in literature, they are usually described in natural languages. Such descriptions are prone to ambiguity and misinterpretation by users. Hence, these descriptions written in natural language cannot be easily adopted for use in computational technical analysis. Since there is also no agreed-upon standard on describing the definitions of these patterns, inconsistencies can easily occur during the applications of these patterns. To alleviate these problems, we propose a comprehensive formal specifications of 103 known candlestick patterns. Our goal is to establish an unambiguous reference model which can be used in future pattern classification research without significant modifications. The formal specifications of these patterns are formulated in the form of the first-order logic, which is comprehensive, extensible, and reusable. To evaluate the proposed specifications, extensive experiments are performed for classifying candlestick patterns from synthetic and real datasets. The experimental results show that the proposed specifications can be used to effectively generate synthetic datasets for selecting best classifiers for candlestick pattern identification.