From Text to Returns: Using Large Language Models for Mutual Fund Portfolio Optimization and Risk-Adjusted Allocation

TL;DR

The paper investigates using Retrieval-Augmented Generation with open-source LLMs to optimize mutual fund sector allocations, aiming to improve returns and risk management. It compares Microsoft Phi-2, Mistral 7B, and Zypher 7B within a synthetic, RAG-enabled framework that converts data to embeddings, retrieves contextual signals, and applies a blended optimization step. Results show Zypher 7B delivering the strongest returns and risk-adjusted performance across Funds A and C, with Mistral 7B also performing well, while Phi-2 underperforms notably. The study demonstrates the viability of GenAI-driven, data-driven portfolio optimization while acknowledging limitations (synthetic data, interpretability, and computational demands) and outlining future work on real-time data integration and enhanced transparency for AI-assisted finance.

Abstract

Generative AI (GenAI) has enormous potential for improving two critical areas in investing, namely portfolio optimization (choosing the best combination of assets) and risk management (protecting those investments). Our study works at this intersection, using Large Language Models (LLMs) to upgrade how financial decisions are traditionally made. This research specifically tested how well advanced LLMs like Microsoft Phi 2, Mistral 7B, and Zypher 7B can create practical, risk-aware strategies for investing mutual funds in different sectors of the economy. Our method is sophisticated: it combines a Retrieval-Augmented Generation (RAG) pipeline, which enables the LLM to check external, real-time data with standard financial optimization methods. The model's advice is context-aware because we feed it large economic signals, like changes in the global economy. The Zypher 7B model was the clear winner. It consistently produced strategies that maximized investment returns while delivering better risk-adjusted results than the other models. Its ability to process complex relationships and contextual information makes it a highly powerful tool for financial allocation. In conclusion, our findings show that GenAI substantially improves performance over basic allocation methods. By connecting GenAI to real-world financial applications, this work lays the groundwork for creating smarter, more efficient, and more adaptable solutions for asset management professionals.

Figures (4)

• Figure 1: Overview of the LLM-based blended optimization framework. The process begins with financial data being transformed into vector embeddings, which are stored in a vector database. When a query is made, relevant context is retrieved from the database and fed into Large Language Models (LLMs). The LLMs generate recommended sectoral allocations, which are further refined using a blended optimization approach to produce final allocations. This workflow integrates data-driven context retrieval, generative AI capabilities, and optimization techniques to improve portfolio performance.
• Figure 2: Performance comparison of mutual fund portfolios optimized by Large Language Models (LLMs) — Microsoft Phi-2, Mistral-7B, and Zypher-7B — against the original allocations. (a) Return (%): Zypher-7B demonstrates the highest improvements in Fund A and Fund C, outperforming other models, while Mistral-7B performs well in Fund A. Microsoft Phi-2 underperforms across all funds. (b) Risk-Adjusted Return (%): Zypher-7B achieves superior risk-adjusted returns for Funds A and C, reflecting its strength in balancing returns and risks. Mistral-7B also shows competitive performance, whereas Microsoft Phi-2 struggles across all funds.
• Figure 3: (a) Return Improvement (%): Comparison of return improvements across three models — Microsoft Phi-2, Mistral-7B, and Zypher-7B. Mistral-7B and Zypher-7B demonstrate significant positive improvements for Fund A, while Microsoft Phi-2 underperforms across all funds, particularly in Fund A and Fund C, where returns drop drastically. (b) Risk-Adjusted Return (RAR) Improvement (%): Mistral-7B and Zypher-7B achieve substantial RAR improvements in Fund A, with Mistral-7B reaching close to 100%. However, Microsoft Phi-2 shows negative RAR improvements across all funds, indicating poor performance in balancing returns and risk.
• Figure 4: Comparison of Large Language Models (LLMs) — Microsoft Phi-2, Mistral-7B, and Zephyr-7B — on Volatility (left) and Risk Reduction (right) across varying exposure ratios. Volatility shows a decreasing trend as ratios increase, with Microsoft Phi-2 and Mistral-7B exhibiting similar stability at mid-range ratios. Zephyr-7B achieves the highest risk reduction overall, particularly at higher ratios, showcasing its superior ability to mitigate risks while maintaining portfolio performance.