Reproduce “Discovering Language of the Stocks” 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
Marko Poženel, Dejan Lavbič
This research bridges technical analysis with natural language processing to predict stock price movements by treating Japanese candlestick patterns as a learnable language. The developed Word2Vec-based model achieved a higher financial yield compared to traditional technical indicators like Moving Average and MACD strategies when tested on individual stocks and a broader index.
Key insights
- Stock price patterns, specifically Japanese candlesticks, can be conceptualized as a 'language' whose 'words' and 'sentences' convey information about market conditions.
- Unsupervised learning (K-means clustering) can automatically identify distinct candlestick patterns, and Word2Vec can then learn meaningful contextual relationships between these patterns.
- Evaluating stock prediction models by financial yield, which accounts for real-world trading costs, provides a more practical measure of performance than prediction accuracy alone.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.