Trading Intelligence (trading-intelligence)

Self-learning quant trading system with CFA-level research.

Overview

Quantitative trading system that combines CFA-level fundamental analysis with technical indicators. Features prediction tracking, paper trading, performance scoring, and adaptive learning from historical outcomes.

Stack

• Python

Quick Start

# Clone
git clone ssh://git@192.168.183.110:2222/pook/trading-intelligence.git
cd trading-intelligence

Backtesting

Install dependencies

cd trading-intelligence
uv venv .venv
uv pip install --python .venv/bin/python yfinance

Load historical data

.venv/bin/python scripts/backtest_data.py

Downloads 5 years of daily OHLCV for 27 tickers (equities, crypto, metals, indices) into trading_intelligence.db.

Run backtests

# Single strategy
.venv/bin/python scripts/backtest_engine.py --strategy momentum --start 2023-01-01 --end 2025-12-31
.venv/bin/python scripts/backtest_engine.py --strategy mean_reversion --start 2023-01-01 --end 2025-12-31
.venv/bin/python scripts/backtest_engine.py --strategy macd --start 2023-01-01 --end 2025-12-31
.venv/bin/python scripts/backtest_engine.py --strategy random --start 2023-01-01 --end 2025-12-31

# Compare all strategies
.venv/bin/python scripts/backtest_engine.py --compare all --start 2023-01-01

# Options: --tickers SPY BTC-USD --commission 0.001 --capital 100000 --max-positions 10 --json --no-store

Available strategies

Strategy	Description
momentum	Top-N momentum with lookback window
mean_reversion	RSI-based mean reversion (oversold/overbought)
macd	MACD histogram crossover
random	Random baseline (statistical control)

Status

In Development

License

Private — All rights reserved

Dry Run Usage

# Quick 2-year run (downloads data from yfinance):
python dry_run.py --start 2024-07-01

# 5-year backtest:
python dry_run.py --start 2021-06-01 --end 2026-06-30

# Today's signals only:
python dry_run.py --today

# Use existing database instead of download:
python dry_run.py --start 2024-01-01 --use-db

# Custom capital:
python dry_run.py --start 2024-01-01 --capital 50000

Output goes to dry_run_output/:

• trades.csv — every fill with strategy, side, strike, DTE, price, P&L
• equity_curve.csv — daily mark-to-market equity, cash, position value, and portfolio Greeks (net_delta, net_gamma, net_theta, net_vega)
• state.json — resumption state

Key Metrics (5-year backtest, 2021-06 to 2026-06, hardened config)

Metric	Strategy	SPY Buy&Hold
Total Return	+67.6%	+84.8%
Max Drawdown	-11.3%	-22.1%
Alpha vs SPY	-17.2%	—

After the risk-hardening milestones (M1–M5) the strategy runs roughly flat to SPY on absolute return while cutting peak drawdown ~49% (11.3% vs 22.1%). Raw return trails buy-and-hold in this bull sample; the edge is risk-adjusted, not absolute. (The original pre-hardening baseline was -13.5% MaxDD.)

Risk Controls (hardening milestones)

• M1 — IV-rank entry gate: sell-premium entries require IV percentile ≥ 0.40 (LEAPS buys ≤ 0.60); data/iv_rank.py.
• M2 — P&L roll triggers: take profit / cut the short leg by premium P&L, not just DTE; checked before the DTE roll gate.
• M3 — Portfolio drawdown kill switch: flatten all + cooldown at 10% drawdown; risk/portfolio_guard.py.
• M4 — Greeks + delta-aware allocation: per-position/portfolio delta/gamma/theta/vega (risk/greeks.py); new entries blocked when net portfolio delta exceeds max_net_delta (default 2.0 = 200 shares); strategies/allocation.py.
• M5 — Fast regime override: VIX-spike / SPY-gap floors the regime ahead of the slow classifier (regime/fast_override.py).

Not live-tradeable. Fills use Black-Scholes pricing with a VIX-derived IV proxy — real bid/ask spread, vol skew and liquidity are absent. A real options data feed (IBKR / Polygon / Tradier) is required before deploying capital.

Adding a New Strategy

1. Create strategies/my_strategy.py inheriting from Strategy base class
2. Implement generate_signal(ctx: MarketContext) -> Signal
3. Register in dry_run.py → self.strategies list
4. Defined-risk validation enforced automatically via _validate_risk()