Reproduce “Investigating Market Strength Prediction with CNNs on Candlestick Chart Images” 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
Thanh Nam Duong, Trung Kien Hoang, Quoc Khanh Duong, Quoc Dat Dinh, Duc Hoan Le, Huy Tuan Nguyen, Xuan Bach Nguyen, Quy Ban Tran
The authors test whether candlestick-chart images can classify upcoming market trends as strong or weak, and whether detected patterns improve predictions. Their convolutional neural network (CNN) processes chart images, while an added pattern-focused stream supplies explicit pattern information. Pattern augmentation performs worse on EUR/USD and AAPL but better on BTC-USD, so results show no consistent gain across these tested assets.
Key insights
- VGG16 proved to be an effective feature extractor for learning representations directly from candlestick chart images.
- Explicitly including detected candlestick patterns did not consistently improve market strength prediction over using raw chart images, and sometimes led to lower performance.
- Purely image-based CNN models can achieve a stable predictive accuracy of about 0.7 across various financial markets, but face limitations in reaching higher accuracies compared to some complex time-series models.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.