We focus on creating different algorithmic trading strategies that predict the movement of various stock prices. Thus, we propose a convolutional neural network (CNN) approach that will convert time series into 2-D financial time series images to analyze the movement of stock prices along with different trading strategies. The method will estimate the price at which we train the model to recognize the pattern and figure out which trade signal is generated. Thus, we used two proposed models, whichone whose specific assets are trained and tested in S&P500 and APPLE'sAPPLE, namely CNN-S, whereasand a group model trained on eight financial assets but used out-of-sample only on S&P500 index and APPLE stocks, namely CNN-G.
Moreover, our study test methods include a CNN model with the inputs of the model image Dimensionsimage's dimensions - 20 (height) x 20 (breadth) x 3 (number of channels, e.g.) - obtained by transforming the return series of historical stock prices. As a result, 20 x 20 sized20x20-sized 2-D images are built. Next, every image gets labeled as "Buy", "Sell" or "Hold," depending on the upward and downwarddownward trend of the original time series. Hence, three image transformations used Recurrence Plots, Gramian Angular Fields, and Markov Transition Fields in this research. We then createcreated traditional algorithm strategies such as Relative Strength Index (RSI), Bollinger Bands (BB), and B&H, to compare them with the CNN model throughout 2006-2007, 2008-2009, and 2017-2019 time periodperiods used for testing purposes.
The rest of the paper is divided as follows: After this brief introduction, a theoretical background, and the literature review are presented. In the second section, the choice of dataset and applied methodologies are explained. The third section is devoted to the empirical results of the research. The fourth section describes sensitivity analysis results by changing the hyperparameter that is set in the methodology part. We end this paper with a conclusion and recommendation for future research.