Two-Stage Portfolio Optimization Framework Using CNN-Based Stock Chart Analysis

다국어 초록 (Multilingual Abstract)

Portfolio optimization in the stock market is a critical research area in finance, and advancements in deep learning have opened new opportunities for improvement. While recent studies have focused on analyzing visual patterns in stock charts and forecasting based on time series data, many have emphasized the predictive power of signals for stock selection without adequately addressing their integration with portfolio optimization. This separation often limits the synergy between the two stages. To address this limitation, this study proposes a Two-Stage Portfolio Optimization Model that integrates asset selection and portfolio optimization using signals derived from stock chart images. In the first stage, a Convolutional Neural Network (CNN) analyzes chart images to learn visual patterns and identify promising stocks by predicting the probability of price increases. In the second stage, portfolio optimization is applied to the selected assets to evaluate performance and refine investment strategies. This research highlights the potential of combining chart image signals with portfolio optimization, demonstrating the effectiveness of an integrated approach in enhancing both asset selection accuracy and the profitability of investment strategies.

목차 (Table of Contents)

• 1 Introduction 1
• 1.1 Contributions 3
• 1.2 Organization 3
• 2 Literature Review 4
• 2.1 Classical Portfolio Optimization 4
• 2.2 Momentum and Technical Analysis 5
• 2.3 Artificial Intelligence in Finance 7
• 2.4 Deep Learning for Portfolio Optimization 7
• 3 Method 8
• 3.1 Data Collection and Preprocessing 8
• 3.2 Generating Chart Data 8
• 3.3 Configuring a Data Universe 9
• 3.4 CNN-based Chart Image Model 9
• 3.5 Time-Series Neural Network Model 11
• 3.5.1 Score Block 12
• 3.5.2 Portfolio Block 12
• 3.5.3 Constraints 12
• 3.5.4 Model and Training Hyperparameters 13
• 4 Experiment 15
• 4.1 Experimental Settings 15
• 4.2 CNN Model Result 15
• 4.3 Backtest 17
• 4.4 Discussion 21
• 5 Conclusion 25
• References 26