jackson-cmd/risk-aware-trading-agents
AI AGENTS FOR STOCK TRADING WITH NEWS BACKTESTING
A set of AI agents for trading US stocks with honest news backtesting: headlines replay at the minute of publication, orders fill at the next open. Includes a public LLM agent leaderboard on three frozen markets (2020, 2022, 2026) with a look-ahead check on every run.
// KEY FEATURES
- Minute-accurate news backtesting without look-ahead bias
- Public LLM agent leaderboard (Llama, GPT, Claude, Gemini) on 3 frozen markets
- Trading agents with market regimes (VIX, FOMC/CPI), 30% cash floor, verified news
- Paper/live trading via 10 brokers, MCP server, dashboard and CLI
// DEEP DIVE: HOW IT WORKS
The package works as both a research framework and a production tool. Price, VIX, and news data are frozen for three periods — the 2020 crash, the 2022 bear market, and 2026; each headline arrives at the minute it was published and orders fill at the next open, eliminating look-ahead bias. A built-in look-ahead check runs on every execution.
Inside are four agent types: trend, swing, dip-buyer, and cash, which factor in the macro calendar (FOMC, CPI, jobs), fear gauges (VIX, Fear & Greed), earnings, and cross-verified news. Ten brokers are supported (Alpaca, Interactive Brokers, etc.), plus an MCP server for clients like Claude Desktop, a Streamlit dashboard, and a full CLI. The rta-bench-1 leaderboard publishes results for Llama, GPT, Claude, Gemini — each with three seeds, scored separately before and after each model's training cutoff.
No keys needed for basic use: backtests and dashboard run out of the box. Live trading requires broker keys and optionally a news API (or custom CSV/SEC 8-K). Runs on Python 3.10+, with Docker images for x86-64 and ARM, and GitHub Codespaces. The 2020/2022/2026 news datasets are licensed and not distributed — replay uses only the sources you provide.
// USE CASES
Best for: Quants and traders testing LLM strategies on historical data without future leakage
- Run a strategy backtest on the frozen 2022 market with a look-ahead bias check
- Deploy paper-trading agents with VIX regimes and macro-event calendar
- Compare LLM agent quality (GPT, Claude, Llama) on the public rta-bench-1 leaderboard
// QUICK START
pip install "risk-aware-trading-agents[dashboard,mcp]"
riskaware dashboard
riskaware init
riskaware scan
riskaware backtest --compare buy_and_hold,ema_bounceCommands are taken from the repository README — check the current version on GitHub before running.
New repositories every 30 minutes
REDDYX AI scans GitHub 24/7 and ships the best AI/ML/Web3 projects to Telegram.
Join on Telegram