LLM-Enhanced Black-Litterman Portfolio Optimization

TL;DR

This work tackles the fragility of traditional mean-variance optimization by embedding Large Language Model (LLM) forecasts and their predictive uncertainty into the Black-Litterman framework. It introduces a systematic pipeline that converts LLM outputs into the BL inputs $\mathbf{q}$, $\mathbf{P}$, and $\boldsymbol{\Omega}$, producing posterior returns $\boldsymbol{\mu}$ for single-period portfolio optimization. Through backtesting on 50 S&P 500 constituents, the authors demonstrate that LLM-driven BL portfolios, especially those based on Qwen and Llama, outperform standard baselines in both absolute and risk-adjusted terms, with performance driven by each model’s distinctive investment style and its alignment with market regimes. The study highlights the practical potential of combining structured prompting, view uncertainty, and BL optimization to create scalable, data-driven asset allocation strategies, and provides open-source code and data for replication and extension.

Abstract

The Black-Litterman model addresses the sensitivity issues of tra- ditional mean-variance optimization by incorporating investor views, but systematically generating these views remains a key challenge. This study proposes and validates a systematic frame- work that translates return forecasts and predictive uncertainty from Large Language Models (LLMs) into the core inputs for the Black-Litterman model: investor views and their confidence lev- els. Through a backtest on S&P 500 constituents, we demonstrate that portfolios driven by top-performing LLMs significantly out- perform traditional baselines in both absolute and risk-adjusted terms. Crucially, our analysis reveals that each LLM exhibits a dis- tinct and consistent investment style which is the primary driver of performance. We found that the selection of an LLM is therefore not a search for a single best forecaster, but a strategic choice of an investment style whose success is contingent on its alignment with the prevailing market regime. The source code and data are available at https://github.com/youngandbin/LLM-BLM.

Figures (8)

• Figure 1: The Proposed Framework for Integrating LLM Views into Black-Litterman Model
• Figure 2: Comparative Analysis of Cumulative Returns for BLM and Benchmark Portfolios. The results highlight the significant outperformance of the BLM-Qwen and BLM-Llama strategies, which consistently generated higher returns compared to both the market index and conventional quantitative models.
• Figure 3: LLM-generated Views Over Time at Rebalancing Intervals. The figure visualizes the distribution of all generated return forecasts for each LLM at every two-week rebalancing interval.
• Figure 4: Correlation Between LLM Predictive Sentiment and Portfolio Performance. This figure compares the cumulative return of each LLM-driven portfolio (line graphs) with the underlying sentiment of the LLM's views at each rebalancing point (background color).
• Figure 5: The structure of the system prompt used to elicit return forecasts from the LLMs.