Identifying Graphical Configurations in Technical Analysis Using Machine Learning

In this paper, we enhance Leigh’s procedure (Leigh 2002a) for identifying the bull-flag configuration within a specified timeframe. We improve the template construction method by introducing more flexibility and eliminating predefined weight choices. The optimization of parameters to maximize annualized return is achieved using a modified fireworks algorithm. Additionally, in our approach, we introduce a signal generator to enhance model robustness and account for trader risk attitudes. The approach is flexible, as it is suitable for other graphical configurations. Our proposal demonstrates superior performance compared to standard settings.