Modular quant trading system for research, backtesting, and paper trading.

QuantForge consolidates four research projects — UARC (regime-adaptive RL portfolio), statistical arbitrage via Kalman filter pairs trading, HMM-based regime detection, and Monte Carlo simulation — into a single, testable system. The signal layer feeds into a portfolio agent, which is subject to a hard risk veto before any orders are sent to the broker. Each layer is defined by an abstract interface, making it straightforward to swap signals or allocation strategies via config.

Monthly rebalance, 5 bps transaction costs, 2020–2025.

Yahoo Finance ──→  HMM Regime     ──→  IQN RL Agent    ──→  Paper Broker
FRED macro    ──→  Kalman Pairs   ──→  Risk Manager    ──→  Alpaca (live)
Alpaca API    ──→  iTransformer   ──→  Position Sizer  ──→  Order Manager
SQLite cache  ──→  Momentum       ──→  MV Optimizer    ──→  Fill Simulator
              ──→  Monte Carlo    ──→  Rebalancer
              ──→  Ensemble

Each layer has a base.py defining the abstract interface. Concrete implementations are swappable via YAML strategy configs.

• HMM Regime Detector — Gaussian HMM over price/volume features, classifies market into N hidden states
• Kalman Pairs — Adaptive spread estimation for mean-reversion pairs trading
• iTransformer — Inverted-attention transformer encoder for multi-asset return forecasting
• Momentum — Cross-sectional momentum with configurable lookback and top-K selection
• Monte Carlo — Forward simulation of price paths for tail-risk estimation
• Ensemble — Weighted combination of the above with regime-conditioned blending

pip install -r requirements.txt

# Run a backtest (fetches data from Yahoo Finance, caches locally)
python scripts/run_backtest.py --universe SPY QQQ TLT GLD SHY --start 2020-01-01 --end 2025-12-31

# Suppress walk-forward for a faster single-pass run
python scripts/run_backtest.py --no-walk-forward

HTML report is written to data/results/backtest_report.html.

758 tests covering signals, portfolio engine, backtest framework, and integration pipelines.

pytest tests/ -v -m "not network"

Python 3.13 · PyTorch · hmmlearn · yfinance · scipy · plotly · pytest