Regime detection, tournament-based strategy selection, and ML signal filtering — the infrastructure hedge funds spend years building, running 24/7 across five asset classes.
How It Works
Every signal FerroQuant generates is the output of a complete institutional-grade pipeline running in real time.
01
HMM continuously classifies market state. Strategies are selected for current conditions.
02
178 strategies compete on live data using an ASHA successive halving algorithm.
03
ONNX ensemble model — separate models for long and short — scores and filters every signal.
04
Entry, stop, and target levels delivered in real time. Paper trade to validate, then go live.
Research
The full theoretical framework — including the novel application of the ASHA algorithm to live trading strategy selection — is documented in the FerroQuant whitepaper. The research overview covers the five open scientific questions the platform is actively investigating.
Read the full whitepaper → Research overview →Built With
Signal engines in Rust for sub-millisecond latency. Data infrastructure in Go. ML training in Python. Seven years of market data in Parquet, queryable in milliseconds via DuckDB.
View on GitHub →