Development of algorithmic trading strategies for financial assets using ensemble learning methods

Financial assets, which are considered time series, are chaotic in nature. It is the main goal of investors to take a position at the right time and in the right direction by making future predictions on this chaotic series. These time series consist of the opening, closing, lowest and highest price in a certain period. Statistical approaches used to make predictions about trend direction and strength using moving averages or technical indicators can have noise and lag problems. Candlestick charts consolidate the mentioned opening, closing, lowest and highest prices in a single image, making it easier to interpret price movements, as well as reflecting the price-based psychology of bear and bull investors. It is known that it was applied to the Japanese rice markets for the first time in history and there are more than 100 candle formations. In the thesis study, an extensible architecture software framework using factory pattern and object-oriented approach is proposed for defining candlestick formations and integrating different machine learning approaches for financial assets with ensemble learning and developing smart learning algorithms based on them. One of the integrated approaches in this context is to use selected technical indicators as attributes in financial forecasting. Finally, three main approaches have been developed by coding the strategies used by investors on a rule-based basis. These three main approaches were combined with the voting principle used in community learning approaches, and the results were compared. In the thesis study, when tested for financial assets on stocks of the Dow30 index, better results were obtained with the proposed ensemble learning approach than the Buy and Hold strategy and traditional machine learning approaches. These results are given in detail for metrics such as cumulative portfolio gain, maximum withdrawal, and percentage of profitable trades, and all results are compared.