Awesome-LLM-Trading

🔥 Large Language Models(LLM) have significantly improved the efficiency and quality of financial analysis and provided new solutions for automated trading. This repository curates academic research and trending projects related to LLMs in the fields of financial analysis and trading.

Trending Projects

RockAlpha

Live AI trading. Explore arenas. Copy-trade best models.
AI-Trader: Benchmarking Autonomous Agents in Real-Time Financial Markets

A fully-automated, live, and data-uncontaminated benchmark (AI-Trader) evaluating LLM agents across U.S. stocks, A-shares, and cryptocurrencies, revealing that general intelligence does not guarantee effective trading performance.

Survey

The New Quant: A Survey of Large Language Models in Financial Prediction and Trading

This survey synthesizes research on LLMs for equity return prediction and portfolio construction, proposing a task-centered taxonomy spanning RAG, agents, and multimodal understanding while outlining best practices for time-safe evaluation, leakage control, and deployment economics in the "New Quant" paradigm.
Integrating Large Language Models in Financial Investments and Market Analysis: A Survey

A structured survey categorizing financial LLM research into frameworks, hybrid integration, fine-tuning, and agent-based architectures, covering applications in stock selection, risk assessment, and trading.
Large Language Models for Financial and Investment Management: Applications and Benchmarks

A comprehensive survey reviewing the applications of LLMs in financial investment management, categorizing studies into predictive modeling and decision-making, while analyzing existing benchmarks and future directions in multimodal financial foundation models.
A Survey of Financial AI: Architectures, Advances and Open Challenges

A systematic taxonomy of Financial AI spanning predictive models, decision-making, and knowledge augmentation, analyzing trade-offs between architectural innovations and industrial deployment challenges.

Financial LLMs

BloombergGPT: A Large Language Model for Finance

A 50-billion parameter decoder-only model trained on a mixed dataset of FINPILE (proprietary financial documents) and public corpora, achieving SOTA performance on financial benchmarks like FLUE and internal sentiment tasks without sacrificing general abilities.
Can ChatGPT Forecast Stock Price Movements?

Documents the capability of LLMs like ChatGPT to predict stock market reactions from news headlines without direct financial training.
TwinMarket: A Scalable Behavioral and Social Simulation for Financial Markets

A scalable multi-agent simulation framework using LLMs and Belief-Desire-Intention (BDI) cognitive architectures to model the Chinese A-share market, demonstrating how micro-level social interactions drive emergent macro-level financial phenomena.
LLMFactor: Extracting Profitable Factors through Prompts for Explainable Stock Movement Prediction

A framework employing Sequential Knowledge-Guided Prompting (SKGP) to extract interpretable factors from news for stock movement prediction, outperforming baselines on US and Chinese datasets.
BBT-Fin: Comprehensive Construction of Chinese Financial Domain Pre-trained Language Model, Corpus and Benchmark

A comprehensive Chinese financial NLP resource featuring the BBT-FinT5 model, a 300GB financial corpus (BBT-FinCorpus), and a 6-task evaluation benchmark (BBT-CFLEB).
PIXIU: A Large Language Model, Instruction Data and Evaluation Benchmark for Finance

A comprehensive framework featuring FinMA (a LLaMA-based financial LLM), a 136K instruction dataset, and the FLARE benchmark for evaluating financial NLP and prediction tasks.
Data-centric FinGPT: Democratizing Internet-scale Data for Financial Large Language Models

An open-source framework (FinGPT) that democratizes financial data for LLMs and introduces Reinforcement Learning with Stock Prices (RLSP) for aligning models with market signals.
FinLlama: LLM-Based Financial Sentiment Analysis for Algorithmic Trading

A Llama-2 based framework fine-tuned for financial sentiment analysis that quantifies sentiment strength to enhance algorithmic trading strategies on the S&P 500.
LLMs for Financial Advisement: A Fairness and Efficacy Study in Personal Decision Making

A study evaluating ChatGPT and Bard against a rule-based system (SafeFinance) on personal finance tasks, revealing critical gaps in accuracy, fairness, and consistency for banking product advice.
Extracting Structured Insights from Financial News: An Augmented LLM Driven Approach

An augmented LLM-driven system that extracts structured insights like tickers and sentiment from unstructured financial news, achieving superior ticker identification coverage compared to existing commercial data providers.
Evaluating LLMs in Financial Tasks - Code Generation in Trading Strategies

Benchmarks various LLMs (e.g., GPT-4, Llama 2) on generating Python code for technical indicators, comparing their accuracy against standard libraries like TA-Lib to assess reliability in trading strategy implementation.
FinTral: A Family of GPT-4 Level Multimodal Financial Large Language Models

A multimodal financial LLM suite based on Mistral-7B that integrates text, tabular, and visual data, introducing the FinSet benchmark to achieve GPT-4 level performance in sentiment analysis, stock prediction, and hallucination mitigation.
RiskLabs: Predicting Financial Risk Using Large Language Model based on Multimodal and Multi-Sources Data

A multimodal framework leveraging LLMs to analyze earnings conference calls (audio and transcripts) and news to forecast financial risk metrics like volatility and VaR, outperforming traditional deep learning baselines.
Large Language Model in Financial Regulatory Interpretation

Explores using LLMs (specifically GPT-4) to distill complex Basel III regulations into mathematical frameworks and actionable code for risk management.
ChatGPT Informed Graph Neural Network for Stock Movement Prediction

A framework leveraging ChatGPT to infer evolving inter-stock dynamic networks from financial news to enhance Graph Neural Networks (GNN) for stock movement prediction and portfolio optimization.
Large Language Models Are Zero-Shot Time Series Forecasters

A zero-shot framework (LLMTime) that encodes time series as numerical digit strings, enabling LLMs to extrapolate trends in financial data (e.g., exchange rates) without fine-tuning, comparable to purpose-built models.
CFGPT: Chinese Financial Assistant with Large Language Model

Presents CFGPT, an open-source framework featuring CFData (141B tokens of Chinese financial text), CFLLM (a fine-tuned InternLM-7B model), and CFAPP for handling tasks like sentiment analysis and stock movement prediction.
Ploutos: Towards interpretable stock movement prediction with financial large language model

A framework comprising expert pipelines and PloutosGPT that utilizes rearview-mirror prompting and dynamic token weighting for interpretable stock movement prediction.
Learning to Generate Explainable Stock Predictions using Self-Reflective Large Language Models

Proposes the Summarize-Explain-Predict (SEP) framework using self-reflective agents and PPO to enable LLMs to generate explainable stock predictions and optimize portfolios autonomously.

LLM Trading Agents

Can ChatGPT Forecast Stock Price Movements?

Documents the capability of LLMs like ChatGPT to predict stock market reactions from news headlines without direct financial training.
TradingGPT: Multi-Agent System with Layered Memory and Distinct Characters for Enhanced Financial Trading Performance

A multi-agent framework featuring layered memory streams and distinct character profiles (e.g., risk-averse) that engage in inter-agent debates to optimize stock trading decisions.
TradingAgents: Multi-Agents LLM Financial Trading Framework

A multi-agent framework mimicking a trading firm with specialized roles (e.g., Bull/Bear researchers, Risk Managers) that collaborate and debate to optimize stock trading decisions, showing superior performance on US stocks.
TwinMarket: A Scalable Behavioral and Social Simulation for Financial Markets

A scalable multi-agent simulation framework using LLMs and Belief-Desire-Intention (BDI) cognitive architectures to model the Chinese A-share market, demonstrating how micro-level social interactions drive emergent macro-level financial phenomena.
Alpha-GPT: Human-AI Interactive Alpha Mining for Quantitative Investment

Introduces an interactive agentic workflow featuring specialized agents—Trading Idea Polisher, Quant Developer, and Analyst—to iteratively generate, backtest, and refine trading signals, achieving top-10 ranking in the WorldQuant International Quant Championship.
Alpha-GPT 2.0: Human-in-the-Loop AI for Quantitative Investment

A human-in-the-loop quantitative investment framework that integrates human expertise with LLM-powered agents for interactive alpha mining, modeling, and analysis.
FinAgent: A Multimodal Foundation Agent for Financial Trading: Tool-Augmented, Diversified, and Generalist

A multimodal generalist agent (FinAgent) integrating textual, numerical, and visual data with tool augmentation to outperform baselines in stock and crypto trading tasks.
FINCON: A Synthesized LLM Multi-Agent System with Conceptual Verbal Reinforcement for Enhanced Financial Decision Making

A multi-agent framework employing a manager-analyst hierarchy and conceptual verbal reinforcement to enhance decision-making in stock and crypto trading tasks.
FINMEM: A Performance-Enhanced LLM Trading Agent with Layered Memory and Character Design

A trading agent framework featuring layered memory (working/long-term) and character profiling that mimics human cognitive processes to enhance decision-making in US stock trading.
Data-centric FinGPT: Democratizing Internet-scale Data for Financial Large Language Models

An open-source framework (FinGPT) that democratizes financial data for LLMs and introduces Reinforcement Learning with Stock Prices (RLSP) for aligning models with market signals.
FinArena: A Human-Agent Collaboration Framework for Financial Market Analysis and Forecasting

A human-agent collaboration framework utilizing a multi-agent MoE architecture and adaptive RAG to integrate multimodal data (news, prices, reports) for personalized stock trend prediction and trading.
FinLlama: LLM-Based Financial Sentiment Analysis for Algorithmic Trading

A Llama-2 based framework fine-tuned for financial sentiment analysis that quantifies sentiment strength to enhance algorithmic trading strategies on the S&P 500.
QuantAgent: Price-Driven Multi-Agent LLMs for High-Frequency Trading

A multi-agent framework decomposing high-frequency trading into Indicator, Pattern, Trend, and Risk agents to align LLM reasoning with structured OHLC signals, demonstrating robust performance across crypto and equity markets.
R&D-Agent-Quant: A Multi-Agent Framework for Data-Centric Factors and Model Joint Optimization

A data-centric multi-agent framework that automates the full quantitative R&D pipeline by co-optimizing factors and models via a feedback-driven loop, achieving superior returns on CSI 300 and NASDAQ 100 compared to SOTA baselines.
CryptoTrade: A Reflective LLM-based Agent to Guide Zero-shot Cryptocurrency Trading

A reflective LLM-based agent for zero-shot cryptocurrency trading that integrates on-chain data and off-chain news, utilizing a self-reflection mechanism to refine decision-making and outperform time-series baselines.
AlphaAgent: LLM-Driven Alpha Mining with Regularized Exploration to Counteract Alpha Decay

An autonomous framework integrating LLM agents with regularization mechanisms (originality, alignment, complexity) to mine decay-resistant alpha factors, demonstrating superior performance in CSI 500 and S&P 500 markets.
Learning to Generate Explainable Stock Predictions using Self-Reflective Large Language Models

A Summarize-Explain-Predict (SEP) framework employing a verbal self-reflective agent and PPO to generate explainable stock predictions and portfolio weights, achieving superior performance in balancing prediction accuracy with explainability.
Can Large Language Models Beat Wall Street? Unveiling the Potential of AI in Stock Selection

MarketSenseAI leverages GPT-4 with Chain-of-Thought and In-Context Learning for explainable stock selection, analyzing multimodal data to achieve 10-30% excess alpha and 72% cumulative returns on S&P 100 stocks.
AlphaFin: Benchmarking Financial Analysis with Retrieval-Augmented Stock-Chain Framework

Implements a monthly rolling investment strategy using Stock-Chain to select A-share stocks (SCI, CSI 300 targets), achieving a state-of-the-art 30.8% annualized rate of return (ARR) through professional CoT-based reasoning.
When AI Meets Finance (StockAgent): Large Language Model-based Stock Trading in Simulated Real-world Environments

A multi-agent system driven by LLMs to simulate investor trading behaviors and analyze the impact of external factors like financial reports and BBS discussions within a realistic US stock market simulation.
FinVis-GPT: A Multimodal Large Language Model for Financial Chart Analysis

Proposed FinVis-GPT and a financial chart instruction-tuning dataset to enable multimodal LLMs to analyze charts, answer questions, and predict future trends.
Exploring LLM Cryptocurrency Trading Through Fact-Subjectivity Aware Reasoning

Proposes FS-ReasoningAgent, a multi-agent framework separating factual and subjective reasoning to enhance LLM cryptocurrency trading, improving profits for BTC, ETH, and SOL.
Enhancing LLM Trading Performance with Fact-Subjectivity Aware Reasoning

Proposes FS-ReasoningAgent, a multi-agent framework separating factual and subjective reasoning to improve cryptocurrency trading performance, especially for advanced LLMs.
Automate Strategy Finding with LLM in Quant Investment

A multi-agent framework leveraging LLMs to automate alpha factor mining and dynamic portfolio optimization, outperforming benchmarks in Chinese and US stock markets.
MarketSenseAI 2.0: Enhancing Stock Analysis through LLM Agents

A framework combining RAG and multi-role LLM agents to analyze financial reports and news for stock selection, outperforming S&P 100 benchmarks.
LLM Agents Do Not Replicate Human Market Traders: Evidence From Experimental Finance

Reveals that LLM agents exhibit textbook-rational behavior and fail to replicate human-like bubbles and biases in experimental asset markets.
FinVision: A Multi-Agent Framework for Stock Market Prediction

Proposes a multi-modal multi-agent framework with a visual reflection mechanism that integrates news and charts for stock trading, outperforming RL baselines on US tech stocks.
FinRobot: An Open-Source AI Agent Platform for Financial Applications using Large Language Models

Introduces FinRobot, an open-source platform with a four-layer architecture (LLM, Tool, Workflow, Agent) to democratize the development of specialized AI agents for financial analysis and trading.

Financial Benchmarks

FinBen: A Holistic Financial Benchmark for Large Language Models

A comprehensive open-source benchmark covering 24 financial tasks across 8 aspects, including stock trading, to systematically evaluate the capabilities of 21 representative LLMs.
AI-Trader: Benchmarking Autonomous Agents in Real-Time Financial Markets

A fully-automated, live, and data-uncontaminated benchmark (AI-Trader) evaluating LLM agents across U.S. stocks, A-shares, and cryptocurrencies, revealing that general intelligence does not guarantee effective trading performance.
PIXIU: A Large Language Model, Instruction Data and Evaluation Benchmark for Finance

A comprehensive framework featuring FinMA (a LLaMA-based financial LLM), a 136K instruction dataset, and the FLARE benchmark for evaluating financial NLP and prediction tasks.
Open FinLLM Leaderboard: Towards Financial AI Readiness

An open evaluation platform co-hosted with Linux Foundation and Hugging Face that benchmarks FinLLMs across 7 diverse tasks, including stock movement prediction and regulatory compliance.
StockBench: Can LLM Agents Trade Stocks Profitably in Real-World Markets?

A contamination-free benchmark designed to evaluate LLM agents on profitability and risk management in realistic, multi-month US stock trading environments using daily market signals.
FinQA: A Dataset of Numerical Reasoning over Financial Data

A large-scale expert-annotated dataset containing 8,281 QA pairs derived from S&P 500 earnings reports, designed to benchmark deep numerical reasoning capabilities of models on heterogeneous financial data.
FinanceBench: A New Benchmark for Financial Question Answering

A first-of-its-kind test suite evaluating LLMs on open-book financial QA with 10,231 questions derived from public filings of 40 US companies, revealing critical limitations in current models' retrieval and reasoning capabilities.
Agent Trading Arena: A Study on Numerical Understanding in LLM-Based Agents

Introduces Agent Trading Arena, a dynamic stock market simulation, revealing that LLMs perform better with visual charts than textual data in financial trading.
StockSim: A Dual-Mode Order-Level Simulator for Evaluating Multi-Agent LLMs in Financial Markets

Presents StockSim, an open-source order-level simulator modeling realistic market dynamics like latency and slippage to evaluate multi-agent LLM trading strategies and coordination.
When Agents Trade: Live Multi-Market Trading Benchmark for LLM Agents

Introduces Agent Market Arena (AMA), a lifelong, real-time benchmark for evaluating LLM-based trading agents across crypto and US stock markets.
MULTIFINBEN: Benchmarking Large Language Models for Multilingual and Multimodal Financial Application

Presents MULTIFINBEN, the first expert-annotated multilingual and multimodal benchmark for evaluating LLMs on diverse financial tasks involving text, vision, and audio.
INVESTORBENCH: A Benchmark for Financial Decision-Making Tasks with LLM-based Agent

Introduces INVESTORBENCH, a benchmark designed to evaluate LLM-based agents' reasoning and decision-making across stocks, cryptocurrencies, and ETFs in diverse market environments.
TradeTrap: Are LLM-based Trading Agents Truly Reliable and Faithful?

Proposes TradeTrap, a unified evaluation framework to stress-test the reliability and robustness of LLM-based trading agents in U.S. equity markets under system-level perturbations.