Paper: arXiv 2503.09647

Authors: Ryan Quek Wei Heng, Edoardo Vittori, Keane Ong, Rui Mao, Erik Cambria, Gianmarco Mengaldo

Abstract

This paper introduces a methodology leveraging Large Language Models (LLMs) for sector-level portfolio allocation through systematic analysis of macroeconomic conditions and market sentiment. Our framework emphasizes top-down sector allocation by processing multiple data streams simultaneously, including policy documents, economic indicators, and sentiment patterns. Empirical results demonstrate superior risk-adjusted returns compared to traditional cross momentum strategies, achieving a Sharpe ratio of 2.51 and portfolio return of 8.79% versus -0.61 and -1.39% respectively. These results suggest that LLM-based systematic macro analysis presents a viable approach for enhancing automated portfolio allocation decisions at the sector level.

Complexity vs Empirical Score

  • Math Complexity: 2.0/10
  • Empirical Rigor: 6.0/10 (capped from 6.5: no out-of-sample test; see penalty flags)
  • Quadrant: Street Traders — empirical and applied, lower math complexity (effective empirical rigor capped from 6.5)

Why this score: The paper presents a methodology using existing LLM frameworks with a backtest showing specific performance metrics like Sharpe ratios, indicating strong empirical implementation, but the math is primarily descriptive of concepts like cross-sectional momentum and top-down investment without dense mathematical derivations.

Research Flowchart

  flowchart TD
  A["Research Goal: Top-Down Sector Allocation via LLMs"] --> B["Data Collection: Policy, Economic, & Sentiment Data"]
  B --> C["Methodology: LLM Processing of Macro Conditions"]
  C --> D["Computational Process: Sector-Level Allocation Model"]
  D --> E["Outcome: Sector Portfolio Construction"]
  E --> F["Key Findings: Sharpe 2.51 & Return 8.79%"]
  E --> G["Comparison: Outperforms Cross Momentum -0.61%"]

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 proposes a novel LLM‑based methodology for top‑down sector allocation that integrates macroeconomic and market sentiment data.

Source sentence
We introduce a novel LLM‑based methodology that emphasizes top‑down sector allocation through integrated analysis of macroeconomic and market sentiment data.

Assumes 5

  • Backtesting is limited to the period January 2019 through June 2019 due to computational constraints. data
    Source sentence
    The backtesting period was limited to January 2019 through June 2019 due to constraints on computational resources and cost.
  • The universe consists exclusively of S&P 500 constituents, updated dynamically to avoid survivorship bias. market
    Source sentence
    Our backtesting framework focused exclusively on S&P 500 constituents to maintain a well‑defined set of liquid, large‑cap stocks.
  • Each trade size is fixed at 1% of the portfolio. execution
    Source sentence
    the size of the position of each trade is fixed at 1% of the portfolio.
  • A commission of 10 basis points per trade and a market‑impact cost of 10 basis points are applied. execution
    Source sentence
    A commission rate of 10 basis points per trade was applied to reflect standard institutional brokerage fees, while market impact costs of 10 basis points were included to account for price slippage during execution.
Show 1 more
  • LLM ranking agents can accurately synthesize policy documents, economic indicators, and sentiment patterns for sector allocation. model
    Source sentence
    Our framework emphasizes top‑down sector allocation by processing multiple data streams simultaneously, including policy documents, economic indicators, and sentiment patterns.

Data

Backtest sample: 2019-01 – 2019-06 · equities · daily

  • Alpha Vantage market data (OHLC and volume) · Alpha Vantage market-prices commercial
    Jan‑2019 to Jun‑2019 · S&P 500 constituents
    Source sentence
    Historical market data, including OHLC (open, high, low, close) prices and trading volumes, was sourced from the Alpha Vantage (2025) API.
  • Macro economic indicators (CPI, PPI, PCE, employment, etc.) · U.S. Bureau of Labor Statistics / U.S. Bureau of Economic An macro public
    Jan‑2019 to Jun‑2019
    Source sentence
    Consumer Price Index (CPI) (U.S. Bureau of Labor Statistics, 2025a) – Producer Price Index (PPI) (U.S. Bureau of Labor Statistics, 2025c) – Personal Consumption Expenditures (PCE) (U.S. Bureau of Economic Analysis, 2025)
  • Policy documents and market sentiment text streams text-news unknown
    Jan‑2019 to Jun‑2019
    Source sentence
    Our framework emphasizes top‑down sector allocation by processing multiple data streams simultaneously, including policy documents, economic indicators, and sentiment patterns.

Recognised sources: Alpha Vantage

Costs

Modeled Costs are part of the reported headline results.

Commission 10 bps per trade plus market‑impact cost 10 bps.

Source sentence
A commission rate of 10 basis points per trade was applied to reflect standard institutional brokerage fees, while market impact costs of 10 basis points were included to account for price slippage during execution.

Finds

Reported performance (Sector-Allocation): Sharpe 2.51 (benchmark -0.61) · total return 8.8%

Checked by others

No other paper in the archive has been found to support or contradict these findings. 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.

Cited References (41)

  1. Augmenting Equity Factor Investing with Global Macro Regimes (2024) — link
  2. Large Language Model Agent in Financial Trading: A Survey — → 2408.06361
  3. Explainable Natural Language Processing for Corporate Sustainability Analysis (2024) — link
  4. LLMFactor: Extracting Profitable Factors through Prompts for Explainable Stock Movement Prediction (2024) — link
  5. Quantitative stock portfolio optimization by multi-task learning risk and return (2024) — link
  6. A Multimodal Foundation Agent for Financial Trading: Tool-Augmented, Diversified, and Generalist — → 2402.18485
  7. Financial Sentiment Analysis: Techniques and Applications (2024) — link
  8. Designing Heterogeneous LLM Agents for Financial Sentiment Analysis (2024) — link
and 33 more
  1. Can Large Language Models beat wall street? Evaluating GPT-4’s impact on financial decision-making with MarketSenseAI — → 2401.03737
  2. FinMem: A Performance-Enhanced LLM Trading Agent With Layered Memory and Character Design — → 2311.13743
  3. Towards Mitigating Hallucination in Large Language Models via Self-Reflection (2023) — link
  4. A comprehensive review on financial explainable AI — → 2309.11960
  5. Can ChatGPT Forecast Stock Price Movements? Return Predictability and Large Language Models — → 2304.07619
  6. Generative Agents: Interactive Simulacra of Human Behavior (2023) — link
  7. FinXABSA: Explainable Finance through Aspect-Based Sentiment Analysis (2023) — link
  8. Multi-source aggregated classification for stock price movement prediction (2022) — link
  9. An Empirical Study of Macroeconomic Factors and Stock Returns in the Context of Economic Uncertainty News Sentiment Using Machine Learning (2022) — link
  10. Building Cross-Sectional Systematic Strategies by Learning to Rank (2020) — link
  11. Financial Sentiment Analysis: An Investigation into Common Mistakes and Silver Bullets (2020) — link
  12. Public Mood–Driven Asset Allocation: the Importance of Financial Sentiment in Portfolio Management (2018) — link
  13. WWW'18 Open Challenge: Financial Opinion Mining and Question Answering (2018) — link
  14. US Stock Market And Macroeconomic Factors (2015) — link
  15. Adapting boosting for information retrieval measures (2010) — link
  16. Learning to rank: from pairwise approach to listwise approach (2007) — link
  17. Learning to rank using gradient descent (2005) — link
  18. Macroeconomic Influence on the Stock Market: Evidence from an Emerging Market in South Asia (2004) — link
  19. Returns to Buying Winners and Selling Losers: Implications for Stock Market Efficiency (1993) — link
  20. Portfolio Performance Based on LLM News Scores and Related Economical Analysis (2024) — link
  21. Tough Talk: The Fed and the Risk Premium (2023) — link
  22. DOES INFLATION MATTER FOR SECTORAL GROWTH IN PAKISTAN? An Empirical Analysis (2013)
  23. Science Research Network electronic library at www.ssrn.com. Monetary Policy and the Cyclicality of Risk ∗ (2010)
  24. U.S. Bureau of Labor Statistics — link
  25. Meta AI
  26. Consumer price index (cpi) for all urban consumers (cpi-u), u.s. city average, all items, seasonally adjusted, 2025a
  27. Investing.com
  28. Producer price index by commodity: Final demand, 4-digit level, 2025c
  29. DeepSeek-AI
  30. Only provide factual analysis based on the data given
  31. Alpha Vantage
  32. Table 2.8.7. percent change from preceding period in prices for personal consumption expenditures by major type of product, monthly
  33. NewsAPI

6 of 41 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.