The next meeting will take place in Utrecht on Friday 13 June 2025.
Location: Minnaert 2.02, Science Park, Utrecht University
Registration: here
Schedule:
14:00 - 14:30 : Thomas Walther
14:30 - 15:00 : Shuaiqiang Liu
15:00 - 15:30 : Nicola Zaugg
15:30-16:00: Coffee Break
16:00 - 16:30 : Wouter Andringa
16:30 - 17:00 : Özge Şahin
17:00 - 17:30 : Bud Schiphorst
17:30-20:00: Closing and dinner
Speakets, Titles and Abstracts:
Wouter Andringua (UvA): Hedging path-dependent derivatives in incomplete markets by using the signature transform
Abstract: The signature of a path has been successfully applied in many different areas of finance. First, I will define the signature of a path and outline some of the important properties that distinguish it from other types of transforms. Specifically, I will argue why it is useful for modelling path-dependent non-Markovian dynamics and that many complicated option payoffs can be approximated by linear combinations of the path signature. Since the signature of a non-Markovian model yields an infinite-dimensional Markovian lift, it is an interesting alternative to infinite dimensional Markovian lifts that have been considered in the literature. Next, I will highlight some of the results in the literature for hedging derivatives by using the signature. I will also present some of my own preliminary findings on hedging path-dependent derivatives in an incomplete market setting by using the signature. This project is a joint work with Drona Kandhai, Asma Khedher and Michel Vellekoop.
Shuaiqiang Liu (TU Delft & ING): Controllable Generation of Implied Volatility Surfaces with Semi-supervised Variational Autoencoders
Abstract: Deep generative models have demonstrated considerable potential for synthesizing implied volatility surfaces (IVS), with applications in option pricing and risk management. However, these data-driven models often lack explicit control over critical IVS shape attributes—such as level, slope, curvature, and term structure—which are essential for meaningful financial interpretation and decision-making. To address this, we propose a controllable generative framework utilizing a semi-supervised Variational Autoencoder (VAE) in combination with a feature extraction algorithm for IVS. This approach explicitly quantifies these key IVS shape features and embeds them into the latent space of VAE, enabling disentangled representations and precise control over the generated surfaces, thus creating synthetic IVS data that reflect targeted market scenarios rather than generating arbitrary samples. Numerical results confirm that this generative model produces realistic IVS that adhere to user-specified characteristics, maintaining both interpretability and the flexibility of deep generative techniques.
Özge Şahin (TU Delft): Redundancy and robustness in ESG metrics: A statistical perspective
Abstract: ESG scores are widely used in quantitative investment and risk modeling, but their construction from hundreds of Key Performance Indicators (KPIs) raises two key statistical concerns. First, imputing missing values with zeros can artificially inflate correlations among KPIs. Second, many KPIs are redundant, adding little information while decreasing the ESG scores’ interpretability. In this talk, we present a mathematical framework to address both problems. We compare Pearson and rank-based association measures, showing theoretically and empirically that rank-based measures are more robust to the effects of missing data in the ESG score construction. We also introduce a stochastic optimization method, based on simulated annealing, to identify compact KPI subsets (typically 10–20 indicators) that approximate Refinitiv ESG pillar scores with high accuracy. Our findings show a multimodal solution space, meaning multiple near-optimal KPI combinations exist. This flexibility allows ESG metrics to be tailored to sector-specific relevance or investor preferences without losing performance. We conclude with open questions on optimal weighting and statistically sound imputation strategies for incomplete ESG disclosures.
Bud Schiphorst (UvA & Rabobank): Connected Clients: Modelling Default Contagion Risk in Credit Portfolios
Abstract: Popular credit portfolio risk models often assume that the default events of different obligors are conditionally independent given some (low-dimensional) common risk factor. This yields a tractable model that can capture important macroscopic concentration risks. Additionally however, credit portfolios may also contain default contagion effects, in which the default event of one obligor directly causes an increase in default risk of another obligor. Therefore, we consider an extended model that allows for additionally incorporating default contagion effects that propagate along a network. As an important contribution, we construct an algorithm that iteratively computes (conditional) default probabilities along singly connected networks for the purpose of computing risk measures. Additionally, we propose an estimation procedure to estimate default contagion parameters from historically observed creditworthiness information. Finally, it is illustrated with a numerical example that portfolio tail risk is captured relatively well via estimated contagion parameters. Similarly, it is illustrated that ignoring default contagion effects may lead to significant underestimation of portfolio tail risk.
Thomas Walther (UU): Certainly! Generative AI and its Impact on Academic Writing (in Finance)
Abstract: TBA
Nicola Zaugg (UU): Stretching Volatility Surfaces
Abstract: A fundamental task in modeling derivatives is to transform the noisy market data of option prices into a clean implied volatility (IV) surface. The IV surface contains relevant information on the market's expectations, is used by market makers to set option prices and acts as an input for calibrating stochastic models. A popular method to construct IV surfaces is to use IV parametrizations, which are parametric functions, often based on underlying stochastic model (the SVI method, SABR-type, or their variants). This ensures the absence of static arbitrage on the surface, but indirectly imposes a model-specific volatility structure on observable market quotes. When the market's volatility does not follow the parametric model regime, the calibration procedure may fail or lead to extreme parameters indicating inconsistency. To address this limitation, we introduce a generic framework for letting the parameters from the parametric implied volatility formula be random. The method enables widening the spectrum of permissible shapes of implied volatilities while preserving analyticity.