BMLL and Simudyne Train AI Market Models on Full Order-Book…
By Rick Steves
AI summary of the source article
BMLL and Simudyne have partnered to develop generative Large Market Models using BMLL's Level 2 and Level 3 historical order-book data. Designed for Simudyne's Pulse simulator, the models aim to enable reactive back-testing where simulated participants and market liquidity respond directly to test orders, overcoming the limitations of conventional static replay. BMLL provides harmonized, nanosecond-level venue data across markets, while Simudyne combines generative machine learning with its existing agent-based modeling. Developed through the BMLL Activate program, the technology is targeted at execution research, algorithm development, exchange rule testing, and clearing risk modeling across various trading regimes.
Why it matters
Conventional static back-tests cannot show how markets react to large orders, whereas reactive simulation allows analysts, algorithm developers, and exchanges to test price impact, slippage, and structural venue changes dynamically.
Key facts
- Simudyne is training Large Market Models on BMLL's normalized Level 2 and Level 3 order-book datasets.
- The models are built for the Pulse market simulator to provide reactive order-flow responses rather than static historical replay.
- The initiative is hosted in the BMLL Data Lab under the BMLL Activate program, which offers reduced upfront licensing costs.