Reproduce “THEME: Enhancing Thematic Investing with Semantic Stock Representations and Temporal Dynamics” 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
Hoyoung Lee, Wonbin Ahn, Suhwan Park, Jaehoon Lee, Minjae Kim, Sungdong Yoo, Taeyoon Lim, Woohyung Lim and 1 more
Researchers from Ulsan National Institute of Science and Technology (UNIST) and LG AI Research developed "THEME," a hierarchical contrastive learning framework that integrates semantic stock representations from textual data with temporal market dynamics. This system achieves significant improvements in thematic stock retrieval precision and enhances investment portfolio performance, outperforming real-world thematic ETFs and generic large language models for this specialized financial application.
Key insights
- Using abstract thematic descriptions as anchors in contrastive learning provides more effective supervision and better generalization for stock retrieval than stock-to-stock alignment.
- The comprehensive Thematic Representation Set (TRS), which augments ETF data with sector and news information, is crucial for achieving broad thematic coverage and improved retrieval gains.
- An integrated approach combining textual semantics and temporal market performance signals, as implemented in THEME, proves more effective for nuanced financial tasks than relying on general-purpose large language models.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.