brain-decorrelator

Strategy-agnostic alpha decorrelation engine with plugin-based orthogonalization axes.

brain_decorrelator transforms correlated alpha expressions into orthogonal variants that pass WorldQuant BRAIN's correlation gate ($|\rho| < 0.70$) while preserving underlying economic Sharpe. It operates across all quantitative domains (Options, Sentiment, Risk Models) via an extensible plugin registry. For detailed mathematical formulations and before/after empirical benchmarks, see the Axes Guide.

Installation

pip install brain-decorrelator

Quickstart

from brain_decorrelator import DecorrelationEngine

engine = DecorrelationEngine()

# Automatically applies all 6 registered orthogonal axes
variants = engine.generate_orthogonal_variants(
    expression="rank(implied_volatility_call_30)",
    base_sharpe=1.65,
    max_variants=5,
)

for v in variants:
    print(f"[{v.axis_name}] -> {v.transformed_expression}")

The 6 Universal Orthogonal Axes

Extending with Custom Plugins

from brain_decorrelator import AxisPlugin, register_axis

@register_axis
class MoneynessInversionAxis(AxisPlugin):
    name = "moneyness_inversion"

    def apply(self, expression: str, context: dict) -> list[str]:
        return [f"-({expression})"]

Architecture

brain_decorrelator/
├── brain_decorrelator/
│   ├── __init__.py        # Public re-exports (DecorrelationEngine, register_axis, unregister_axis)
│   ├── plugin.py          # AxisPlugin ABC + thread-safe registry
│   ├── engine.py          # DecorrelationEngine — orchestrator & variant dedup
│   ├── py.typed           # PEP 561 typing marker
│   └── axes/
│       └── universal.py   # 6 domain-agnostic built-in orthogonal axes
├── tests/                 # Unit tests covering registry and all 6 axes
└── docs/
    ├── API.md             # Complete API specification
    └── AXES.md            # Mathematical formulations & worked examples

Testing

pytest tests/ -v

Security

Report vulnerabilities per SECURITY.md.

Contributing

See CONTRIBUTING.md for contribution guidelines.

License

Released under the MIT License. Maintained by Xtley001.