Continuous-time risk analytics
that adapt when markets break.
The Fin-Tech division of Stokhos Labs builds Neural Stochastic Differential Equation models that forecast the U.S. Treasury yield curve across every market regime, with 64.3% lower forecast error when financial panic strikes.
Neural SDEs: a model
built for real markets.
The model evolves through a differential equation that responds instantaneously to new data, not once per day, but continuously.
The diffusion network learns state-dependent volatility from data. Uncertainty widens automatically in stress, with no manual recalibration.
Trained on all available FRED data, stress-tested on the March 2020 financial panic. ARX-GARCH MSE: 1.943. Neural SDE MSE: 0.692. A 64.3% reduction, zero retraining.
Go deeper.
Why ARX-GARCH and discrete-time models fail precisely when risk management matters most.
Read moreDrift networks, diffusion networks, and Euler-Maruyama integration: how the Neural SDE works.
Read moreYield curve forecast vs. actual, cumulative error, and regime classification: the live product interface.
View dashboardResearch notes on yield-curve risk, tail behavior, and where discrete-time volatility models break down.
Read insights