Preconference Day: Summits & WorkshopsMonday, 17 November 2025 - GMT (Greenwich Mean Time, GMTZ)

How infrastructure, computation, and AI are redefining the edge

Large Language Models (LLMs) exhibit remarkable capabilities across a spectrum of tasks in financial services, including report generation, chatbots, sentiment analysis, regulatory compliance, investment advisory, financial knowledge retrieval, and summarization. However, their intrinsic complexity and lack of transparency pose significant challenges, especially in the highly regulated financial sector, where interpretability, fairness, and accountability are critical. Our research presents the first application in the finance domain of understanding and utilizing the inner workings of LLMs through mechanistic interpretability, addressing the pressing need for transparency and control in AI systems. Mechanistic interpretability is the most intuitive and transparent way to understand LLM behavior by reverse-engineering their internal workings. By dissecting the activations and circuits within these models, it provides insights into how specific features or components influence predictions - making it possible not only to observe but also to modify model behavior. While not yet widely adopted, mechanistic interpretability is expected to become increasingly vital as adoption of LLMs increases. Advanced interpretability tools can ensure AI systems remain ethical, transparent, and aligned with evolving financial regulations. In this research, we have put special emphasis on how these techniques can help unlock interpretability requirements for regulatory and compliance purposes - addressing both current needs and anticipating future expectations from financial regulators globally.

Understanding how central banks make decisions, and the decisions, is important for many uses. Large Language Models (LLM) have been used by many groups for different financial predictions with different timescales. Here we consider central bank rate setting decisions and use an LLM to help understand how the decisions are made, and tackle LLM explainability. Specifically we look at speeches by rate-setting members, and econometric indicators, using the Taylor model as a strawman. We develop a series of models using the LLM and different datasets, checking information leakage, then validate by switching to regression approaches. Considering two major central banks we show that their decisions can be explained from economics, and that speeches add little. We note that LLM-based model development does not need to lead to models that use LLM.

This session introduces two cutting-edge applications of large language models (LLMs) in quantitative finance:

1. Sentiment Trading with LLMs (Finance Research Letters, 2024):
   This study demonstrates how financial sentiment extracted via transformer-based LLMs (e.g., FinBERT) can predict short-term return anomalies. The model aggregates media tone to form daily sentiment signals, which enhance directional trading strategies across equity markets. Presented at over 20 international conferences, including ACL, QES, and CFAR. You can find this paper online.
2. Sentiment-Augmented Reinforcement Learning for Portfolio Optimization (Accepted at ACL 2025, ICLR 2025, Wolfe Research Quant Conference; under NeurIPS 2025 review):
   This paper proposes SAPPO, a novel reinforcement learning framework that integrates asset-level sentiment—extracted using LLaMA 3.3 from Refinitiv financial news—directly into the PPO policy update. SAPPO significantly improves Sharpe ratios and annual returns (2.07 vs. 1.67; 83% vs. 57%) by aligning allocation decisions with evolving investor expectations. Empirical results confirm its robustness across different market regimes and sentiment models.

Together, these contributions demonstrate how LLMs can move beyond text analytics into core decision-making, enabling adaptive, sentiment-driven asset allocation in volatile markets.

Deploying models, not just training them

In finance, valuation must obey law; conventional AI risks economic hallucination, distorting pricing, hedging, and tail-risk. Rather than adapting generic AI and correcting arbitrage post hoc, we rebuild neural networks on asset pricing first principles. The result is the first finance-native neural network, where neurons encode market states and Markovian Activation Functions turn the First Fundamental Theorem of Finance into a binding computational constraint. Applied to real-world QIS portfolios, it enables XVA-style forward-looking strategy design, learning markets in 8 minutes and simulating 100,000 scenarios across 800 layers in 80 seconds on a MacBook Pro. Rooted in pricing axioms, this framework transforms AI from a black box into a transparent engine delivering coherent valuation, robust risk diagnostics, and strategy innovation — where every neuron obeys the immutable laws of finance.

This work studies the dynamic risk management of the risk-neutral value of the potential credit losses on a portfolio of derivatives. Sensitivities-based hedging of such liability is sub-optimal because of bid-ask costs, pricing models which cannot be completely realistic, and a discontinuity at default time. We leverage recent advances on risk-averse Reinforcement Learning developed specifically for option hedging with an ad hoc practice-aligned objective function aware of pathwise volatility, generalizing them to stochastic horizons. We formalize accurately the evolution of the hedger's portfolio stressing such aspects. We showcase the efficacy of our approach by a numerical study for a portfolio composed of a single FX forward contract.

Predictions across infrastructure, people, and process