United KingdomFinance and insurance2025 – ongoingMan Group
AlphaGPT agentic AI for quant signal discovery
Man Group's in-house AlphaGPT agent has produced several dozen trading signals cleared for live use
Problem
Quant researchers face a flood of data and a slow cycle of forming ideas, writing code and testing them.
Approach
Man Numeric, the quant arm of Man Group, built AlphaGPT in-house. The agentic system mines historical data, proposes rule-based trading signals, writes the code and backtests it, mimicking a human quant's research steps. A senior portfolio manager said it is not on autopilot and that checkpoints and guardrails stay in the investment process.
Results
Takeaway
Man kept human sign-off on every signal. The firm also said hallucinations and inconsistent outputs remain, and it aims for a system that can verify its own work.
Categories
- Industry
- Finance and insurance
- Business function
- Finance, risk and legalIT and R&D
- Innovation type
- Process innovation
- AI technology
- AI agentsGenerative AI
- Adoption
- Built in-house
- Evidence
- Company's own claim
Sources
Source summaries
Hedgeweek, citing a Bloomberg report, describes AlphaGPT, an internally built agentic system used by Man Numeric. It mines data, writes signal code and backtests it. The only figure is that several dozen signals were approved for live trading.
Full details are in the original sources.