QuantAgent: Price-Driven Multi-Agent LLMs for High-Frequency Trading

Fei Xiong, Xiang Zhang, Aosong Feng, Siqi Sun, Chenyu You · 2025

Read QuantAgent: Price-Driven Multi-Agent LLMs for High-Frequency Trading by Fei Xiong, Xiang Zhang, Aosong Feng, Siqi Sun, Chenyu You (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-09995.txt.

Introduction

Fei Xiong and 4 co-authors aim a team of language model agents at high-frequency trading in a 2025 finance paper. Earlier multi-agent trading systems reason over news and fundamentals for slow, strategic bets. High-frequency desks live elsewhere: decisions must land in tiny windows, driven by compact structured signals such as indicator readings, chart shapes, and short trends, with risk checked constantly.

QuantHarness splits the job among four specialists covering indicators, patterns, trends, and risk, each armed with its own tools and structured reasoning routine. Trials spanned nine instruments, from Bitcoin to Nasdaq futures, and the authors present the system as the first of its kind shaped for this speed class. The design bets that narrow, fast agents beat one broad thinker when the clock dominates.

Catalog record

CREATORFei Xiong, Xiang Zhang, Aosong Feng, Siqi Sun, Chenyu You
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE64 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-09995

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

QuantAgent: Price-Driven Multi-Agent LLMs for High-Frequency Trading Fei Xiong, Xiang Zhang, Aosong Feng, Siqi Sun, Chenyu You Abstract Recent advances in Large Language Models (LLMs) have shown remarkable capabilities in financial reasoning and market understanding. Multi-agent LLM frameworks such as TradingAgent and FINMEM augment these models to long-horizon investment tasks by leveraging fundamental and sentiment-based inputs for strategic decision-making. However, these approaches are ill-suited for the high-speed, precision-critical demands of High-Frequency Trading (HFT). HFT typically requires rapid, risk-aware decisions driven by structured, short-horizon signals, such as technical indicators, chart patterns, and trend features. These signals stand in sharp contrast to the long-horizon, text-driven reasoning that characterizes most existing LLM-based systems in finance. To bridge this gap, we introduce QuantAgent , the first multi-agent LLM framework explicitly designed for high-frequency algorithmic trading. The system decomposes trading into four specialized agents, Indicator , Pattern , Trend , and Risk , each equipped with domain-specific tools and structured reasoning capabilities to capture distinct aspects of market dynamics over short temporal windows. Extensive experiments across nine financial instruments, including Bitcoin and Nasdaq futures, demonstrate that QuantAgent consistently outperforms baseline methods, achieving higher predictive accuracy at both 1-hour and 4-hour trading intervals across multiple evaluation metrics. Our findings suggest that coupling structured trading signals with LLM-based reasoning provides a viable path for traceable, real-time decision systems in high-frequency financial markets. 1]Stony Brook University 2]Carnegie Mellon University 3]University of British Columbia 4]Yale University 5]Fudan University \contribution [†]Equal contribution \metadata [Github] https://github.com/Y-Research-SBU/QuantAgent \metadata [Website] https://Y-Research-SBU.github.io/QuantAgent/ \metadata [Corresponding Authors] 1 Introduction In quantitative finance, technical analysis treats historical price action as the most immediate and information-dense reflection of market conditions ( Pring 1991 ) . The central premise is that market dynamics, including fundamentals, macro events, institutional flows, and collective sentiment, are ultimately embedded in price movements ( Murphy 1999 ) . Each bar, defined by its open, high, low, and close (OHLC), provides a compact yet universal representation of short-horizon market behavior. This structure enables systematic detection of recurring setups such as trends, reversals, breakouts, and momentum shifts across asset classes ranging from equities and commodities to digital assets ( Moskowitz et al. 2012 ) . Under the efficient market hypothesis ( Fama 1970 ) , prices adjust rapidly to public information, making patterns in OHLC bars a natural substrate for short-term prediction without reliance on lagging textual inputs. Figure 1 : Workflows of IndicatorAgent, PatternAgent, and TrendAgent. IndicatorAgent interprets signals from MACD, RSI, ROC, and Williams %R; PatternAgent detects formations such as double bottoms; TrendAgent extracts directional flow via support and resistance channels. Large Language Models (LLMs) have recently demonstrated impressive capabilities in multi-step reasoning, tool use, and interpretable decision-making ( OpenAI et al. 2024 ) . These capabilities are directly relevant to quantitative trading ( Yang et al. 2023 ) , which heavily depends on integrating heterogeneous signals, applying systematic trading rules, and controlling execution risks. However, most existing LLM-driven financial frameworks operate primarily on textual inputs, such as news articles, social media streams, or earnings reports ( Nguyen et al. 2015 ; Xiao et al. 2025 ; Zakir et al. 2025 ; Zhang et al. 2023a ) . This reliance introduces two major limitations: (i) textual signals typically lag price discovery and are incorporated into markets only after the fact ( Chordia et al. 2013 ) , and (ii) such data is noisy, unstructured, and difficult to validate ( Liu et al. 2022a ) . Since short-horizon market dynamics are already encoded in OHLC bars, a more direct approach is to align LLM reasoning with structured price-based signals. To the best of our knowledge, no prior work has developed an LLM-based framework for high-frequency trading (HFT) that operates directly on OHLC data. In this paper, we propose QuantAgent (Figure 1 ), the first multi-agent LLM framework tailored to high-frequency algorithmic trading. Specifically, QuantAgent decomposes the trading process into four specialized agents – IndicatorAgent, PatternAgent, TrendAgent, and RiskAgent – each designed to capture a complementary dimension of technical analysis. IndicatorAgent condenses raw OHLC bars into robust technical indicators, providing a noise-resistant summary of recent market behavior. PatternAgent chart formations such as peaks, troughs, and consolidations, leveraging the multimodal reasoning abilities of LLMs ( Nison 2001 ) . TrendAgent identifies directional bias from short-horizon price dynamics, while RiskAgent integrates all signals into a coherent risk–reward profile. Final trade decisions emerge from the interaction of these agents, yielding traceable, language-native rationales that can be inspected alongside execution ( Schick et al. 2023 ) . We evaluate QuantAgent on a multi-asset benchmark spanning commodities, equities, cryptocurrencies, and volatility indices. At 1-hour and 4-hour bar resolutions, QuantAgent consistently outperforms baselines across both directional accuracy and return-based metrics, with particularly pronounced gains in equity markets. Rolling-window validation further demonstrates robust generalization, achieving up to 80% directional accuracy in forecasting short-term price movements. Besides its strong empirical performance, QuantAgent provides natural-language rationales for trading decisions, enabling a degree of traceability and interpretability often missing in traditional algorithmic strategies. 2 Related Works Agent-Based LLMs for Financial Decision-Making. The design of QuantAgent builds on recent work that organizes LLMs into multi-agent systems for financial decision-making. FINCON ( Yu et al. 2024 ) introduces a manager–analyst hierarchy trained via verbal reinforcement learning, while TradingAgents ( Xiao et al. 2025 ) models institutional workflows through agent communication, prioritizing interpretability over the low-latency, price-driven logic required in high-frequency trading ( Tumarkin and Whitelaw 2001 ) . As a line of work, these systems demonstrate the potential of LLM-based agents in finance, but their heavy reliance on textual inputs leaves them ill-suited for the structured, low-latency signals required in HFT scenarios. More recently, RD-Agent(Q) ( Li et al. 2025 ) takes a significant step forward by shifting to structured, data-centric signals and automating factor–model co-optimization.However, RD-Agent(Q) remains constrained to daily-resolution strategies and slower research-feedback cycles, making it less suitable for real-time decision-making in high-frequency contexts. Quantitative Trading Based on Indicators and Patterns. Prior to LLM-based agents, quantitative trading systems are predominantly built on technical indicators such as trends, volatility, and momentum for intraday decision-making. Early studies show that nonlinear price patterns can exhibit predictive power ( Lo et al. 2000 ) , but subsequent work highlight challenges including overfitting and researcher bias ( Chen and Chen 2016 ) . Momentum strategies ( Jegadeesh and Titman 1993 ; Moskowitz et al. 2012 ) are widely adopted to capture trend persistence, while heuristics such as Elliott wave theory ( Prechter 2005 ) and curated pattern libraries attempt to model higher-order market structures. Although these indicator- and pattern-based approaches are interpretable and computationally efficient, they often struggle in volatile or noisy environments, undermining their effectiveness in high-frequency settings. These limitations motivate us to design a framework that fuses structured price signals with LLM-based reasoning, enabling more adaptive and interpretable trading systems. 3 QuantAgent To bridge the gap between traditional high-frequency quantitative trading and recent advances in multi-agent LLM systems, we introduce QuantAgent, a collaborative framework for low-latency market decision-making. QuantAgent integrates classical technical analysis with prompt-structured LLM reasoning, enabling modular and interpretable financial intelligence. Built on LangGraph ( LangChain 2025 ) , the system simulates the workflow of institutional trading desks, where specialized agents execute distinct analytical roles to support

Read the complete text