Reproduce “Aligning Multilingual News for Stock Return Prediction” 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
Yuntao Wu, Lynn Tao, Ing-Haw Cheng, Charles Martineau, Yoshio Nozawa, John Hull, Andreas Veneris
Researchers at the University of Toronto developed a method to improve stock return prediction by semantically aligning sentences in multilingual financial news using optimal transport. This technique significantly boosted the annualized Sharpe ratios of long-short trading strategies, achieving 4.36 for Japanese news and 3.42 for English news, by leveraging shared cross-lingual information.
Key insights
- Optimal transport produces sparse and interpretable cross-lingual sentence alignments, effectively identifying a smaller, more relevant set of semantically similar pairs compared to dense methods.
- Information consistently conveyed across different languages (aligned content) provides stronger and more reliable signals for predicting stock returns.
- Granular, sentence-level alignment can uncover latent predictive signals that are often obscured in document-level analyses, even when using powerful LLM embeddings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.