Main Conference Day ThreeThursday, 20 November 2025 - GMT (Greenwich Mean Time, GMTZ)

The state of play in quantum computing

Hear from various institutions followed by an interactive Q&A panel.

Although recent advances have made it feasible to estimate dynamic covariance matrices in high dimensions, portfolio managers face significant model risk when selecting a single forecast approach. We propose several forecast combination methods to mitigate model uncertainty and the risk of relative underperformance. These include distance-based approaches that penalize models producing forecasts that diverge excessively from others, as well as two optimization-based methods, which balance variance reduction with portfolio stability relative to individual models. Using daily U.S. stock data from 1980 to 2022, we evaluate ten individual covariance forecasting models and several combination strategies, constructing long-only minimum variance portfolios for universes of up to 1000 stocks. Forecast combinations--particularly simple ones--reduce model risk and deliver realized volatility close to the best individual models. However, optimization-based combinations often introduce additional turnover, potentially offsetting their benefits after transaction costs. Our results highlight the trade-off between improved risk forecasts and higher turnover, suggesting turnover-aware forecast combination as a direction for future research.

We reveal a geometric structure underlying both hedging and investment products. The structure follows from a simple formula expressing investment risks in terms of returns. This informs optimal product designs. Optimal pure hedging (including cost-optimal products) and hybrid hedging (where a partial hedge is built into an optimal investment product) are considered. Duality between hedging and investment is demonstrated with applications to optimal risk recycling. A geometric interpretation of rationality is presented.

Constructing an efficient hedge portfolio for derivatives within risk limits, which are typically based on Greeks, is crucial from both trading and risk management perspectives. Therefore, we propose simple algorithms to optimize the hedge portfolio with complex Greeks, inspired by the proximal point method with the Sinkhorn algorithm. One algorithm focuses on optimizing cost, while the other optimizes both cost and CVaR (conditional value at risk) with constraints to limit the residual Greeks. Our algorithms consist of iterations involving only simple vector and matrix calculations, similar to the Sinkhorn algorithm, which is now widely used in machine learning; thus, linear programming solvers are not needed. In numerical examples, we explore two ways to improve the hedge portfolio for Bermudan swaptions, moving away from conventional anti-diagonal vega hedging with coterminal swaptions. The results of the optimized cost, CVaR, and hedge positions produced by our algorithms are consistent with insights on hedging under various constraints.

This is actually an increasingly acute issue since the COVID market crash, where the hedging of large quantities of short term Equity options, sold by banks to the retail market, has had a large and paradoxical impact on the equity market itself, with the consequence of reinforcing any market drawdowns. This phenomenon was before COVID limited to a few albeit recurrent market crashes and ONLY wholesale market, which was actually the focus of my Ph. D (2006) and academic and practitioners papers (2012,2016,2017,2019) and presentations of previous business and academic conferences (QuantMinds 2016, and award of the Best investment presentation at the US Society of Actuary Conference in 2018).

Most counterparties do not have traded CDS instruments. This poses a challenge for calculating CVA which relies on risk-neutral default probabilities. Here we present a new model for the estimation of credit spreads using equity market data. In contrast to traditional equity-credit models, we take an empirical approach in order to determine a simple functional relationship that can be used in practice for CVA risk management. We find that our approach out-performs models that rely solely on credit data as well as alternative equity-credit models in the literature.

Hear from various institutions followed by an interactive Q&A panel.

Reflecting on the quantum computing landscape

We propose a nonparametric data-driven methodology for hedging using generative models. In contrast with model-based hedging approaches relying on sensitivity analysis of model pricing functions, our approach uses a conditional generative model trained on market data to simulate realistic market scenarios given current market conditions and computes hedge ratios which minimize risk across these scenarios. The approach incorporates transaction costs, leads to an optimal selection of hedging instruments, and adapts to market conditions. We illustrate the effectiveness of this methodology for hedging option portfolios using VolGAN, a generative model for implied volatility surfaces. The out-of-sample performance of the method matches and improves over delta and delta-vega hedging, without retraining the model for more than 4 years after the training period.

This talk is based on Cont, R., Vuletić, M. Data-driven hedging with generative models. Ann Oper Res (2025). https://doi.org/10.1007/s10479-025-06867-3

• Tasche model
• Stochastic variance approach
• LGD driver model
• Calibration strategy

• “Next-Gen Risk Architecture” strategic leadership in evolving institutional risk frameworks.
• “Cross-Disciplinary AI” practical ML/AI applications
• “Dynamic Market Defense” real-world, high-frequency trading and systemic risk control aspects he oversees.
• “Model Integrity” model risk management, resilience, and continuous validation.

Forward MtM (FMtM) Pricers are crucial whenever we need to evaluate a future MtM inside a simulated market for the purpose of computing a final t0 price. For instance, XVA pricing has relied on either analytical pricers for simple payoffs or regression pricers for more complicated ones. Evaluating the performance of these pricers is of utmost importance whenever the final t0 price has a heavy dependence on the distribution of FMtMs. To that end, we propose a framework based on Monte Carlo Branching Simulation to compute the Squared Error between an FMtM that we would like to assess, and the true reference FMtM which we don’t necessarily know. The Squared Error has the advantage of accounting for differences across the entire distribution of FMtMs. It can be used both as a selection criteria from a pool of available FMtM pricers, and as a tool to measure the performance of a single FMtM pricer. While it is one of the best ways to achieve that, we recognize that in some applications such as XVA this could be too strict as only the average positive part of FMtMs is needed. We thus propose a moment based approach also using Branching Simulation to evaluate the ability of an FMtM pricer to accurately calculate the average of a function of FMtMs under certain conditions. This effectively loosens the constrains of testing the entire FMtM distribution via the Squared Error.

As a numerical application, we use Regression Pricers in the context of XVA.

In this session we look at the history of derivative pricing adjustments, including XVA, showing that the majority of well-known adjustments can be seen as manifestations of a single fundamental representation that unifies these approaches. We look at historic examples, ranging from a classic option pricing adjustment for vol, to funding value adjustments. We then consider the important problem of trying to estimate the impact of unmodelled risk factors, and consider how the representation can suggest ways to attack this problem, which can be a headache for everybody from traders, risk managers and quants to senior management.