Hyper-Network-Based Neural Approximators for Lifecycle Derivative Pricing on Low-Dimensional Manifolds
Overview
As derivative markets grow more complex, firms face increasing pressure to deliver fast, accurate pricing and risk analytics across the entire product lifecycle—from structuring and issuance to hedging and risk management. This is particularly evident in products like autocallables, where path dependency and high dimensionality make efficient, consistent modeling especially challenging.
This paper introduces a novel machine learning framework based on hyper-network-driven neural approximators—an approach in which one neural network dynamically generates the parameters of another, enabling fast and flexible approximation of complex pricing models across a wide range of market conditions. The framework is designed to efficiently model high-dimensional derivative pricing problems while maintaining flexibility as market conditions evolve.
By combining advances in neural networks with financial engineering principles, hyper-network-driven neural approximators offer a scalable path toward real-time analytics without sacrificing model fidelity.
Access the Research
Explore the full research to understand the methodology, implementation, and performance results of hyper-network-driven neural approximators.
Overview
As derivative markets grow more complex, firms face increasing pressure to deliver fast, accurate pricing and risk analytics across the entire product lifecycle—from structuring and issuance to hedging and risk management. This is particularly evident in products like autocallables, where path dependency and high dimensionality make efficient, consistent modeling especially challenging.
This paper introduces a novel machine learning framework based on hyper-network-driven neural approximators—an approach in which one neural network dynamically generates the parameters of another, enabling fast and flexible approximation of complex pricing models across a wide range of market conditions. The framework is designed to efficiently model high-dimensional derivative pricing problems while maintaining flexibility as market conditions evolve.
By combining advances in neural networks with financial engineering principles, hyper-network-driven neural approximators offer a scalable path toward real-time analytics without sacrificing model fidelity.
Access the Research
Explore the full research to understand the methodology, implementation, and performance results of hyper-network-driven neural approximators.
Authors
Jonathan Rosen, Ph. D.
Jonathan Rosen specializes in applying advanced quantitative methods and artificial intelligence to financial modeling, with a focus on equity derivatives and volatility modeling. His work includes the optimization of equity volatility surfaces, calibration of local volatility models under realistic market constraints such as bid/ask spreads, and the acceleration of quantitative models using GPU-based techniques. He holds a PhD in physics from the University of British Columbia, where he developed deep expertise in mathematical modeling and computational methods. Dr. Rosen's work has also contributed to industry efforts around the LIBOR transition, including research on SOFR futures convexity adjustments analogous to Eurodollar futures, published on SSRN.
Andrew McClelland, Ph.D.
Andrew McClelland’s quantitative research at Numerix focuses on XVA pricing and hedging, generating counterparty credit risk metrics for structured products, and estimating risk model parameters via time-series estimation. He earned his PhD in finance at the Queensland University of Technology for a thesis on financial econometrics. He considered markets exhibiting crash feedback, option pricing for such markets, and parameter estimation for such markets using particle filtering methods. Dr. McClelland’s work has been published in the Journal of Banking and Finance, the Journal of Econometrics, and the Journal of Business and Economic Statistics.
Serguei Issakov, Ph.D.
Dr. Issakov, as Chief Quantitative Officer and Senior Vice President of Global Quantitative Research and Development, oversees the company’s quantitative research globally, including the research of pricing models at Numerix. Since joining Numerix in 1999, his earlier roles at Numerix included Vice President of Financial Applications, Head of Engine Development (the forerunner to Numerix 7) and Head of Risk Analytics.
Prior to joining Numerix, Dr. Issakov held research positions in theoretical physics at the Nordic Institute for Theoretical Physics in Copenhagen, the University of Paris (Laboratory of Theoretical Physics and Statistical Models), the University of Oslo and the Center for Advanced Study in Oslo. Before that, he led research on models of brain rhythms at the Medical Radiological Center in Obninsk Russia.
Dr. Issakov has published over 40 papers in mathematics and theoretical physics. He is a co-author of the Issakov-Ouvry-Wu equations in fundamental quantum statistical mechanics. He has received numerous fellowships and research grants, including a NATO Visiting Professorship and grants from the Russian Foundation for Basic Research. He holds PhD in Theoretical and Mathematical Physics from Moscow Institute of Physics and Technology, from the Theory Group led by Physics Nobel Laureate Vitaly Ginzburg.
