This is a page that curates AI-related papers published worldwide. All content here is summarized using Google Gemini and operated on a non-profit basis. Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.
THEME: Enhancing Thematic Investing with Semantic Stock Representations and Temporal Dynamics
Created by
Haebom
Author
Hoyoung Lee, Wonbin Ahn, Suhwan Park, Jaehoon Lee, Minjae Kim, Sungdong Yoo, Taeyoon Lim, Woohyung Lim, Yongjae Lee
Outline
This paper proposes THEME, a novel framework for constructing semantic representations of investment themes from text data, to address the challenges of thematic investment, which involves structuring portfolios based on structural trends. We highlight the inadequacy of existing large-scale language model (LLM) embedding models in capturing the nuanced characteristics of financial assets. We present THEME, a framework that fine-tunes embeddings through hierarchical contrastive learning. THEME aligns hierarchical relationships between themes and constituent stocks and improves embeddings by incorporating stock returns, generating representations that are effective for retrieving thematically aligned assets with high return potential. Empirical results demonstrate that THEME outperforms leading LLMs in thematic asset retrieval and also outperforms the constructed portfolios. Combining textual thematic relationships with market dynamics of returns, we generate stock embeddings specifically designed for a variety of real-world investment applications.
Takeaways, Limitations
•
Takeaways:
◦
We present the possibility of generating effective stock embeddings specialized for thematic investing using hierarchical contrastive learning.
◦
Achieve improved thematic asset search performance and portfolio performance compared to existing LLMs.
◦
A new thematic investment approach that integrates text and market data.
◦
High applicability to various real-world investment applications
•
Limitations:
◦
Further research is needed on the generalization performance and long-term performance of the THEME framework presented in this paper.
◦
Consideration should be given to the limitations of the dataset used and its dependence on specific market conditions.
◦
The need for greater transparency and interpretability in the process of establishing hierarchical relationships and integrating stock returns.
◦
Comparative analysis with other thematic investment strategies is necessary.
