Reproduce “From Text to Returns: Using Large Language Models for Mutual Fund Portfolio Optimization and Risk-Adjusted Allocation” on OpenResearch

About this paper

Abrar Hossain Mufakir Qamar Ansari Haziq Jeelani Monia Digra Fayeq Jeelani Syed, Mufakir Qamar Ansari, Haziq Jeelani, Monia Digra, Fayeq Jeelani Syed

Researchers integrated Large Language Models (LLMs) into a retrieval-augmented generation (RAG) framework, combined with blended optimization, to generate risk-adjusted sectoral allocation strategies for mutual funds. Advanced LLMs like Zypher 7B and Mistral 7B consistently delivered superior returns and risk-adjusted returns compared to original allocations and smaller LLMs like Microsoft Phi 2.

Key insights

• More advanced Large Language Models like Zypher 7B and Mistral 7B demonstrate superior contextual understanding and reasoning for financial optimization tasks.
• Integrating LLMs with a retrieval-augmented generation (RAG) pipeline effectively bridges unstructured information to actionable, data-driven investment strategies.
• The blended optimization approach provides a flexible mechanism to fine-tune LLM recommendations, allowing for tailored risk-return profiles based on investor preferences.

Summary by alphaXiv.