Paper: arXiv 2512.24526

Authors: Alina Voronina, Oleksandr Romanko, Ruiwen Cao, Roy H. Kwon, Rafael Mendoza-Arriaga

Abstract

This paper investigates how Large Language Models (LLMs) from leading providers (OpenAI, Google, Anthropic, DeepSeek, and xAI) can be applied to quantitative sector-based portfolio construction. We use LLMs to identify investable universes of stocks within S&P 500 sector indices and evaluate how their selections perform when combined with classical portfolio optimization methods. Each model was prompted to select and weight 20 stocks per sector, and the resulting portfolios were compared with their respective sector indices across two distinct out-of-sample periods: a stable market phase (January-March 2025) and a volatile phase (April-June 2025). Our results reveal a strong temporal dependence in LLM portfolio performance. During stable market conditions, LLM-weighted portfolios frequently outperformed sector indices on both cumulative return and risk-adjusted (Sharpe ratio) measures. However, during the volatile period, many LLM portfolios underperformed, suggesting that current models may struggle to adapt to regime shifts or high-volatility environments underrepresented in their training data. Importantly, when LLM-based stock selection is combined with traditional optimization techniques, portfolio outcomes improve in both performance and consistency. This study contributes one of the first multi-model, cross-provider evaluations of generative AI algorithms in investment management. It highlights that while LLMs can effectively complement quantitative finance by enhancing stock selection and interpretability, their reliability remains market-dependent. The findings underscore the potential of hybrid AI-quantitative frameworks, integrating LLM reasoning with established optimization techniques, to produce more robust and adaptive investment strategies.

Complexity vs Empirical Score

  • Math Complexity: 2.5/10
  • Empirical Rigor: 6.0/10 (capped from 7.0: transaction costs not modelled; see penalty flags)
  • Quadrant: Street Traders — empirical and applied, lower math complexity (effective empirical rigor capped from 7.0)

Why this score: The paper focuses on applying existing LLMs and classical portfolio optimization (like mean-variance) with minimal novel mathematical derivations, but employs rigorous out-of-sample backtesting across multiple models and market regimes with performance metrics like Sharpe ratios.

Research Flowchart

  flowchart TD
  A["<b>Research Goal</b><br>How do LLMs enhance sector-based portfolio construction?"] --> B["<b>Methodology</b><br>Prompt 5 LLM providers for<br>stock selection & weighting (20/sector)"] --> C["<b>Hybrid Processing</b><br>Combine LLM selections with<br>Classical Portfolio Optimization"] --> D["<b>Out-of-Sample Testing</b><br>Compare vs Sector Indices<br>in 2 market regimes:"] --> E["<b>Key Findings</b><br>1. Stable Market: LLM portfolios outperform<br>2. Volatile Market: LLM portfolios struggle<br>3. Hybrid Frameworks improve robustness"]

  D --> D1["Stable Phase<br>Jan-Mar 2025"]
  D --> D2["Volatile Phase<br>Apr-Jun 2025"]

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper evaluates applying large language models to quantitative sector‑based portfolio construction and combines their stock selection with classical mean‑variance optimization.

Source sentence
This study set out to evaluate how effectively large language models (LLMs) from leading AI providers (OpenAI, Google, Anthropic, DeepSeek, and xAI) can be applied to the task of quantitative sector-based portfolio construction.

Assumes 6

  • LLMs have training cut‑off dates before 2025, so analysis is split into in‑sample and out‑of‑sample periods. model
    Source sentence
    Because most of the large language models (LLMs) used in this study have training cut‑off dates before 2025, we divided the analysis into one in‑sample period and two out‑of‑sample periods
  • Each LLM is prompted to select and weight exactly 20 stocks per sector. execution
    Source sentence
    Each model was prompted to select and weight 20 stocks per sector
  • Portfolio evaluation uses a mean‑variance optimization framework. statistical
    Source sentence
    The resulting portfolios were then evaluated using the mean–variance optimization framework to compare LLM‑generated results with those produced through traditional quantitative methods
  • Adjusted closing prices from Yahoo Finance are used for returns and risk calculations. data
    Source sentence
    Market data were obtained primarily from Yahoo Finance, which provided adjusted closing prices for both individual company stocks and the S&P 500 sector indices
Show 2 more
  • Risk‑free rate is set to 2% for Sharpe ratio calculations. statistical
    Source sentence
    The risk‑free rate is set to 2% (0.02) as per standards
  • Sector constituents are validated against Wikipedia and TradingView listings. data
    Source sentence
    list from Wikipedia [12] was used to identify which companies belong to each sector according to the GICS classification. • For an additional layer of validation, the resulting lists were compared with the sector compositions available on TradingView [13]

Data

  • Yahoo Finance adjusted closing prices · Yahoo Finance market-prices public
    five years preceding Jan 2025 and Jan‑Ju · S&P 500 sector constituents and sector indices
    Source sentence
    Market data were obtained primarily from Yahoo Finance, which provided adjusted closing prices for both individual company stocks and the S&P 500 sector indices
  • Wikipedia GICS sector list · Wikipedia benchmark public
    S&P 500 sector constituents
    Source sentence
    list from Wikipedia [12] was used to identify which companies belong to each sector according to the GICS classification
  • TradingView sector composition · TradingView benchmark public
    S&P 500 sector constituents
    Source sentence
    For an additional layer of validation, the resulting lists were compared with the sector compositions available on TradingView [13]

Recognised sources: Yahoo Finance

Costs

Not modeled A trading or execution result is reported without costs.

Finds 3

  • During the stable market period (Jan‑Mar 2025), LLM‑weighted portfolios frequently outperformed sector indices on cumulative return and Sharpe ratio. performance
    Source sentence
    During stable market conditions, LLM‑weighted portfolios frequently outperformed sector indices on both cumulative return and risk‑adjusted (Sharpe ratio) measures
  • In the volatile period (Apr‑Jun 2025), many LLM portfolios underperformed their sector benchmarks. negative
    Source sentence
    during the volatile period, many LLM portfolios underperformed
  • Combining LLM‑based stock selection with traditional mean‑variance optimization improves both performance and consistency of the portfolios. performance
    Source sentence
    Importantly, when LLM‑based stock selection is combined with traditional optimization techniques, portfolio outcomes improve in both performance and consistency

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 29 related papers). Coverage is partial, so this is neither replication nor refutation.

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.