New AI Architecture Aims to Revolutionize Financial Modeling
NEW YORK – November 18, 2025 – A novel approach to artificial intelligence in finance, dubbed a “finance-native” neural network architecture, is challenging conventional wisdom about applying AI to dynamic markets. Developed by researcher Iabichino, the system prioritizes incorporating established financial principles directly into the AI’s structure, rather than attempting to force financial models to conform to the logic of existing AI frameworks. This could lead to more accurate pricing, improved risk management, and enhanced trading strategies for banks and investment firms.
The core innovation lies in a neural network design that connects neurons directly, rather than relying on traditional layer-by-layer connections, thereby constraining potential outputs to align with financial constraints. Iabichino argues that current AI applications in finance,such as deep hedging and models like Volga,are fundamentally flawed because they attempt to adapt AI designed for static problems – like image or text recognition – to the inherently dynamic nature of financial markets. “They try to bend finance to obey the logic of AI,” he stated.
This new architecture offers potential applications across various financial functions, including quantitative Investment Strategies (QIS), derivatives valuation adjustments (XVAs), and investment portfolio management. In the buy-side context, the system can incorporate investor views as constraints, comparing them with market-implied prices to generate trading signals, even where the no-arbitrage condition doesn’t apply.
The podcast details how the network’s interpretability is a key feature, allowing for a better understanding of the reasoning behind its outputs. Iabichino’s research also extends to XVA modeling, aiming to improve the accuracy and efficiency of these crucial calculations. The idea originated from a desire to build an AI that inherently understands and respects the underlying principles of finance.
Podcast Index:
* 00:00 introduction: what’s a finance-native AI architecture?
* 02:58 How to incorporate financial constraints into a neural network
* 08:22 Differences with other AI-based approaches
* 13:36 Interpretability
* 14:53 Uses for QIS,pricing and investment purposes
* 23:48 The genesis of the idea
* 24:55 Iabichino’s research on XVA
Listeners can access the full interview on Risk.net, Soundcloud (https://soundcloud.com/user-393208363/tracks), Spotify (https://open.spotify.com/show/2fKrhTrkJP0tdiT7RWkK8o),Amazon Music (https://music.amazon.com/podcasts/19067a46-6bb5-41bf-b7c8-9cd11d996af3/quantcast-%E2%80%93-a-risk-net-cutting-edge-podcast), or iTunes (https://podcasts.apple.com/gb/podcast/quantcast-a-risk-net-cutting-edge-podcast/id1340909285).Future podcasts in the Quantcast series will be uploaded to Risk.net.
