Reproduce “Global Stock Market Prediction Based on Stock Chart Images Using Deep Q-Network” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jinho Lee, Raehyun Kim, Yookyung Koh, Jaewoo Kang
Researchers at Korea University developed a Deep Q-Network (DQN) model that uses Convolutional Neural Networks (CNNs) to analyze stock chart images and predict market movements. The model, trained solely on US market data, demonstrated profitability within the US and generalized its predictive capabilities to a majority of 30 other global stock markets, generating annual returns of 10-100% per annum before transaction costs.
Key insights
- Distinct, predictive patterns exist within stock chart images (displaying past daily closing prices and volume) and these patterns exhibit cross-market universality.
- Reinforcement Learning, through DQN, offers advantages for financial prediction by optimizing for long-term cumulative rewards, providing graded rewards, and quantifying action values to differentiate signal strength.
- AI-based stock forecasting models, once trained on data from a large and liquid market, can be effectively deployed in smaller, data-scarce emerging markets, suggesting a path to resource efficiency.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.