tradingbot

A risk-first algorithmic trading lab. Strategies move through explicit lifecycle stages (local → paper → live) and must clear a pre-flight risk gauntlet — drawdown cap, leverage cap, position concentration cap — before any backtest or deployment runs. The point of this repo is not to maximize returns; it is to make unsafe configurations unrunnable.

Risk gauntlet

Every algorithm load is gated by a pre-flight check that validates the strategy config against three hardcoded caps:

• Max drawdown — refuse if config.max_drawdown_pct > 0.30.
• Leverage — refuse if config.leverage > 2.0.
• Position concentration — refuse if config.max_position_pct > 0.20.

Violations raise RiskGauntletError(cap_name, actual, allowed) before any order, indicator, or universe selection is touched. The cap values are deliberately hardcoded in the gauntlet — the gauntlet is the boundary, not a preference.

Lifecycle stages

Strategies are tagged with a stage in strategy.yaml:

Stage	Notional cap	Purpose
local	$0 (paper, local stub)	Develop logic, run pytest, replay black-swan fixtures.
paper	exchange paper account	Forward-test against live market data, no real money.
live	per-strategy notional_cap	Promoted only when paper-stage promotion criteria are met.

Promotion is not automatic. The strategy YAML records the criteria; the operator decides when to flip the stage.

Usage

This project expects QuantConnect's AlgorithmImports runtime. It is not a standalone script.

1. Create a QuantConnect project in the dashboard or with the lean CLI.
2. Copy these Python files into the project root.
3. Edit config.py for dates, cash, universe, and strategy flags.
4. Run the backtest.

Local logic tests use a small QuantConnect stub:

pip install pytest
pytest

License

See LICENSE.