market_maker_trading_system

please be cautious! this is a simplified version and just for testing, the real version is much more complicated and involving sensitive data. But this is great for testing and simulation.

market_maker_trading_system

A minute-level research pipeline for market/medium-frequency trading:

1. fetch 1-minute OHLCV data (Polygon),
2. engineer features (technical indicators + Kalman trend + filtering + KMeans/RBF expansion),
3. train/evaluate an ML model with realistic transaction cost simulation (IBKR commissions + slippage) and “sniper mode” confidence thresholding.

⚠️ This repository is currently best viewed as a research/backtesting pipeline, not a production trading bot.

---

What’s inside

1) data_fetch.py — Fetch 1-minute bars from Polygon

• Pulls 1-minute aggregates from Polygon’s API.
• Saves/updates a CSV to: ./data/minute_data/{TICKER}.US_minute.csv
• If the file already exists, it appends only new rows after the last timestamp. :contentReference[oaicite:1]{index=1}

2) data_engi.py — Feature engineering pipeline (v4)

Produces an engineered dataset from minute OHLCV with:

• Core return/range features
• Technical indicators (via ta)
• Kalman Filter trend extraction
• Cleaning pipeline: impute → outlier clipping → scaling
• Two-stage correlation filtering (pre/post RBF)
• Feature selection (top-k)
• KMeans + RBF feature expansion
• Outputs to: ./data/{TICKER}.US_minute_engineered_v4.csv :contentReference[oaicite:2]{index=2}

3) Market_maker_ML.py — Model training + evaluation + net-of-costs backtest

A minute-level model training & evaluation pipeline with:

• “Sniper mode”: only trade when predicted probability exceeds a confidence threshold
• Anti-overfitting regularization focus
• Net returns simulated after estimated IBKR commissions + slippage
• Saves plots/results into an output directory :contentReference[oaicite:3]{index=3}

4) medium_frequency_trading_system.ipynb

Notebook version / experiments (large). :contentReference[oaicite:4]{index=4}