ION has incorporated artificial intelligence into its XTP Risk JANUS platform to improve the accuracy and efficiency of pre-trade risk management for cleared derivatives. The enhancement targets a key challenge in margin estimation using CME Group’s SPAN2 Approximation model.
SPAN2 is a margin methodology developed by CME that assesses the worst-case losses of futures and options portfolios using a historical value at risk (HVAR) approach. While accurate, its complexity makes it too slow for real-time pre-trade applications. CME offers a simplified version, SPAN2 Approximation, for faster calculations, but this often results in discrepancies with the full model.
The AI-driven solution, developed by LIST – an ION company – bridges this gap. Integrated into XTP Risk JANUS and its Margin Engine, the enhancement improves the accuracy of SPAN2 Approximation while preserving its speed, enabling more precise and timely margin estimates.
“We wanted to replicate SPAN2 calculations in real time for pre-trade use, where speed is essential,” Riccardo Bernini, Head of Financial Engineering and AI at LIST, tells TradingTech Insight. “Although CME provides an approximation for this purpose, it isn’t always close to the actual SPAN2 output. So we set out to improve both the accuracy and performance. To achieve this, our team developed a transformer-based neural network, trained entirely in-house, that understands the structure of CME-traded products and generates a correction factor to refine the approximation. The result is a model that delivers much more accurate margin estimates, while still operating fast enough for pre-trade validation, typically within milliseconds.”
The system leverages CME’s deployable SPAN2 library for standard margin calculations and offers the optimised approximation model for use in time-sensitive trading environments. This allows traders to perform pre-trade order validation with greater confidence in margin accuracy, supporting more informed risk and trading decisions.
“It’s not just about improving the accuracy of the approximation, but about doing it in the right way,” adds Luca Papaleo, XTP Risk JANUS Product Owner at LIST. “When you’re targeting a specific margin value, your approximation can overshoot or undershoot. From a risk perspective, underestimating the margin compared to what CME calculates is risky, it could expose firms to insufficient coverage. So while it’s safer to overestimate, the goal is to stay as close as possible to the true value without overshooting significantly.”
The team is now looking at potentially applying this methodology to other exchanges. “What’s interesting is that other clearing houses, such as JSCC in Japan, are also moving towards similar models,” says Papaleo. “If we can identify a workable starting point – like CME’s SPAN2 Approximation – we could potentially apply the same AI-based correction method elsewhere. By combining an internal margin estimate based on a VaR framework with a trained neural network that adjusts it, we could bring the result much closer to the clearing house’s output. This makes the approach adaptable beyond CME.”
Subscribe to our newsletter
