A New Algorithmic Trading Approach Based on Ensemble Learning and Candlestick Pattern Recognition in Financial Assets

Abstract

Financial assets considered as time series are chaotic in nature. The main goal of investors is to take a position at the right time and in the right direction by making predictions about the future of this chaotic series. These time series consist of the opening, low, high, and closing prices of a certain period. The approaches used to make predictions about trend direction and strength using moving averages and indicators based on them have noise and lag problems as they are obtained statistically. Candlestick charts, on the other hand, reflect the price-based psychology of bear and bull investors, and facilitate the interpretation of price movements by consolidating the said opening, closing, lowest and highest prices in a single image. It is known that it was applied to Japanese rice markets for the first time in history and there are more than 100 candle patterns. In this study, an extensible architecture software framework using factory patterns and an object-oriented approach is proposed for defining candlestick patterns and developing intelligent learning algorithms based on them. In the studies carried out for financial assets, the profit factor, which shows the portfolio gain of the strategy, is used. It is desirable that this number of wins be greater than 1. When the proposed approach is tested for 5 major financial assets, this value was obtained as greater than 1 for all assets. The proposed software framework can also be used in the development of new robotic approaches in terms of being applicable to all kinds of financial assets in every period.