PokieTicker addresses the challenge of understanding stock price movements for beginners by correlating news events directly with price action on candlestick charts. It offers a unique approach to developing event-driven thinking, enabling users to grasp the "why" behind market fluctuations. The tool benefits stock market novices, researchers, and power users by providing AI-driven explanations, trend predictions, and historical context for price changes.

How It Works

PokieTicker employs a sophisticated data pipeline that integrates news sentiment analysis with technical indicators. News articles are processed by Claude Haiku for sentiment scoring, key discussion summaries, and bullish/bearish reasons, then combined with OHLC price data into daily feature vectors. An XGBoost model predicts price direction (T+1, T+3, T+5 horizons) using these features. The system also identifies historically similar news patterns via cosine similarity on feature vectors, leveraging the principle that markets react predictably to similar information events and sentiment clusters.

Quick Start & Requirements

The project offers immediate usability with a pre-built SQLite database (pokieticker.db) and ML models, eliminating the need for API keys for initial setup.

• Primary install: Clone the repository, unpack pokieticker.db.gz and models.tar.gz.
• Backend: Python 3.10+, requires venv and pip install -r requirements.txt. Run with uvicorn backend.api.main:app --reload.
• Frontend: Node.js 18+, requires npm install in the frontend directory. Run with npm run dev.
• Prerequisites: Bundled database and models. Optional API keys for Polygon.io (free tier) and Anthropic (pay-as-you-go) are needed for updating data and running AI analysis.
• Demo: A live demo is available at mitrui.com/PokieTicker.

Highlighted Details

• News events are visualized as dots on candlestick charts, with details accessible via click.
• News can be filtered by impact type (Market, earnings, product, policy, competition, management).
• AI-powered explanations are available for selected date ranges to detail price rallies or drops.
• Trend prediction models forecast future price direction based on the past 30 days of news events.
• Historical periods with similar news patterns are identified, showing subsequent price behavior.
• Feature engineering includes 31 inputs from news sentiment (count, score, momentum) and technical indicators.

Maintenance & Community

No specific details regarding notable contributors, sponsorships, community channels (e.g., Discord, Slack), or a public roadmap were found in the provided README.

Licensing & Compatibility

• License: MIT License.
• Compatibility: The MIT license is permissive, generally allowing for commercial use and integration into closed-source projects without significant restrictions.

Limitations & Caveats

This project is explicitly stated as an "experimental tool for learning, not financial advice." The README acknowledges the complexity of markets and that no single model can capture all influencing factors. Full AI analysis and data updates require optional API keys and incur usage costs.