Applying Deep Learning for Stock Chart-Based Stock Market Trend Forecasting

This research explores an innovative three-step approach to stock market forecasting through deep learning, using Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks. The first step lies in utilizing stock chart images as inputs, a method that mirrors the visual analysis commonly performed by traders. Secondly, we shift from traditional regression to binary classification to predict market trends, leveraging the strengths of image-based analysis. The third step involves optimizing prediction confidence by selecting the model's threshold based on validation set performance, enhancing forecast precision. This calibration fine-tunes the decision-making boundary for issuing the buy, sell or hold signals of the models. To the best of our knowledge this is the first time that the above three steps have been combined in this manner. We evaluate this approach on historical data from 2010, assessing its efficacy on a 2021–2022 test set. Our findings reveal the significant potential of these models in stock market analysis, offering a novel fusion of visual representation and classification techniques. The comprehensive analysis includes both single stock and multi-stock evaluations, providing insights into model versatility and the financial implications of incorporating these strategies into trading systems. This approach underscores the power of deep learning in enriching investment strategies within the dynamic stock market landscape.