AMF 2025 – Program

May 21–23, 2025

Location: FRIAS Institute, Lecture Hall (Hörsaal Anatomie) Albertstraße 19, Freiburg im
Breisgau, Germany

Online: The zoom link has been sent to you by mail.

Abstracts of the Talks

Poster Presentation

Search:

Name	University or Company of Origin	Title	Abstract
Atchadé, Mintodê Nicodème	Carl von Ossietzky Universität Oldenburg	Modeling economics and finance data with the Arctan Marshall-Olkin Weibull distribution	The Arctan Marshall-Olkin family is introduced as an innovative and adaptable class of heavy-tailed distributions for modeling extreme events in financial and economic sciences. This family emerges from the combination of the Marshall-Olkin framework with the Arctan-X approach, which leverages the arctangent inverse trigonometric function. A particular case, the Arctan-Marshall-Olkin-Weibull (ATMOW) distribution, is explored in detail. Unlike the conventional two-parameter Weibull distribution, ATMOW incorporates an additional parameter, enhancing its adaptability to heavy-tailed data. Monte Carlo simulations confirm the efficiency of the maximum likelihood estimation for parameter inference. Furthermore, closed-form expressions for key actuarial risk measures, including value at risk and tail value at risk, are derived to assess extreme financial risks. Empirical applications to real financial and economic datasets demonstrate better performance of ATMOWcompared to several competing multiparameter distributions.
Brutsche, Johannes	University of Freiburg, Math.Department	The level of self-organized criticality in oscillating Brownian motion: n-consistency and stable Poisson-type convergence of the MLE	For some discretely observed path of oscillating Brownian motion with level of self-organized criticality ρ0, we prove in the infill asymptotics that the MLE is n-consistent, where n denotes the sample size, and derive its limit distribution with respect to stable convergence. As the transition density of this homogeneous Markov process is not even continuous in ρ0, the analysis is highly non-standard. Therefore, interesting and somewhat unexpected phenomena occur: The likelihood function splits into several components, each of them contributing very differently depending on how close the argument ρ is to ρ0. Correspondingly, the MLE is successively excluded to lay outside a compact set, a 1/√n-neighborhood and finally a 1/n-neigborhood of ρ0 asymptotically. The crucial argument to derive the stable convergence is to exploit the semimartingale structure of the sequential suitably rescaled local log-likelihood function (as a process in time). Both sequentially and as a process in ρ , it exhibits a bivariate Poissonian behavior in the stable limit with its intensity being a multiple of the local time at ρ0.
Gabin, Taibi	Bern University of Applied Sciences and University of Twente	Transformers for Financial Narrative Processing	This research introduces an innovative and holistic approach to financial narrative modeling, employing transformer architectures to explore the multifaceted nature of narratives. Our methodology goes beyond the conventional analysis of sentiment or topic trends by examining narratives through multiple dimensions, including their thematic depth, polarization, emotional consistency, and frequency. Leveraging a dataset spanning from 2010 to 2025 and comprising news articles from various financial newspapers, our approach integrates these diverse narrative elements, allowing for a richer, more detailed understanding of the dynamics that drive financial markets. This comprehensive analysis not only enhances our grasp of narrative impacts but also sets a new standard for financial data processing.
Knaust, Sven	University of Freiburg	Fast Bayesian calibration of option pricing models based on sequential Monte Carlo methods and deep learning.	Model calibration is a challenging yet fundamental task in financial engineering. Using sequential Monte Carlo methods, we reformulate the non-convex optimization problem as a Bayesian estimation task, constructing a sequence of distributions from the prior to the posterior. This allows us to compute any statistic of the estimated parameters, mitigating the strong dependence on starting points and avoiding the troublesome local minima, that often plague standard calibration methods. To accelerate computation, we incorporate Markov Chain Monte Carlo methods with delayed acceptance and a neural network-based option pricing approach that uses risk-neutral cumulants of log-returns as additional informative features. To demonstrate the strength of our approach, we calibrate an affine stochastic volatility model with price-volatility co-jumps to both simulated and real data. Our Bayesian algorithms significantly outperform the standard approach in terms of runtime, accuracy, and statistical fit.
Kormaz, Yaren	Yildiz Technical University	Markowitz portfolio optimisation with group constraints according to market trend	This study aims to grouping financial assets based on their Beta coefficients and develop a portfolio that investors can follow during market upturns and downturns. By integrating a group constraint into the traditional Markowitz mean-variance model, the objective is to devise a new portfolio model. The bi-objective portfolio optimization problem, which simultaneously minimizes portfolio risk and maximizes portfolio return under the established constraint, will be addressed using a heuristic algorithm. The study proposes a novel portfolio selection strategy for investors, utilizing the Beta coefficients of stocks to construct an optimal portfolio that ensures maximum return with minimum risk during both rising and falling market periods.
Kuissi Kamdem, Wilfried	University of Rwanda / AIMS Ghana	Linear pricing rule for power and Epstein-Zin utilities indifference valuation	In this paper we tackle the open problem on whether it is possible to obtain a linear pricing rule for utility indifference valuation in an incomplete market; meaning, the indifference price of λ units of the pay-off F is equal to λ times the price of one unit of the pay-off F. In a possibly non-Markovian setting, we obtain a linear pricing rule for Epstein-Zin utility as well as power utility. This is possible because we are able to provide an explicit solution to the fully coupled forward-backward stochastic differential equation derived by using a backward stochastic differential equation approach. We adopt a transformation new to problems associated to recursive utility. The pay-off can be a function of the entire path of both the traded assets and the non-traded financial indices on the time horizon. Our results apply to the pricing and hedging of European-type options. The indifference price obtained is an equilibrium price in the sense that the seller’s price and the buyer’s price coincide. Joint work with Olivier Menoukeu Pamen and Marcel Ndengo
Kunst, Philipp	Fraunhofer IAO	Transformer-based Deep Hedging (indicative)	tba (we are working on using transformers and PPO RL to solve hedging problems)
Massaria, Michele Domenico	Politecnico di Milano	The Additive Bachelier model with an application to the oil option market in the Covid period	In April 2020, the Chicago Mercantile Exchange (CME) temporarily switched the pricing formula for WTI options from the Black model to the Bachelier model. In this context, we introduce an Additive Bachelier model that provides a simple closed-form solution. This new Additive model exhibits several notable mathematical and financial properties. It ensures the no-arbitrage condition, a critical requirement in highly volatile markets, while also enabling a parsimonious synthesis of the volatility surface. The model features only three parameters, each one with a clear financial interpretation: the volatility term structure, vol-of-vol, and a parameter for modeling skew. The proposed model supports efficient pricing of path-dependent exotic options via Monte Carlo simulation, using a straightforward and computationally efficient approach. Its calibration process can follow a cascade calibration: first, it accurately replicates the term structures of forwards and At-The-Money (ATM) volatilities observed in the market; second, it fits the smile of the volatility surface. Overall this model provides a robust and parsimonious description of the oil option market during the exceptionally volatile first period of the Covid-19 pandemic.
Osei, Price	Bielefeld University	Fat-tailed distribution under the Smooth Ambiguity Model	The notion that asset returns are normally distributed has been increasingly questioned. We provide an alternative explanation for the fat tails observed in the return series using the smooth ambiguity model. Within this framework, we apply the exponential-power utility to capture the impact of ambiguity on asset return distributions. The inherent ambiguity in asset returns cannot be fully described by a single probability distribution. Instead, we consider a family of normal distributions with a known mean but an uncertain variance. This variance uncertainty disrupts the conditions required for the central limit theorem (CLT), leading to its failure. Our approach incorporates the gamma distribution to account for the uncertainty in variance. The model proposes the variance-gamma distribution as the unconditional law of asset returns.
Pavarana, Simone	University of Freiburg, Math.Department	Bayesian learning with set-valued observations	We introduce a novel Bayesian filtering framework for set-valued observation processes. Leveraging random set theory, we demonstrate that standard Bayesian formulas hold under mild conditions satisfied by a broad class of examples. Our approach is motivated by the need to incorporate expert opinions — represented as random sets, such as confidence intervals — into estimation problems. We illustrate potential applications in credit rating and propose a revised version of the Black-Litterman model.
Rennspies, Jasper	University of Freiburg	Efficient Sampling for Realized Variance Estimation in Time-Changed Diffusion Models	This paper analyzes the benefits of sampling intraday returns in intrinsic time for the realized variance (RV) estimator. We theoretically show in finite samples that depending on the permitted sampling information, the RV estimator is most efficient under either hitting time sampling that samples whenever the price changes by a pre-determined threshold, or under the new concept of realized business time that samples according to a combination of observed trades and estimated tick variance, or or hitting time. The analysis builds on the assumption that asset prices follow a diffusion that is time-changed with a jump process that separately models the transaction times, which allows for leverage specifications and Hawkes-type jump processes. This provides a flexible model that separately captures the empirically varying trading intensity and tick variance processes, which are particularly relevant for disentangling the driving forces of the sampling schemes. Extensive simulations confirm our theoretical results and show that low levels of noise, hitting time sampling remains superior while for increasing noise levels, realized business time becomes more efficient. This finding can be explained by the high noise-sensitivity of hitting time sampling that samples directly on the observed asset prices, whereas noise-robust intensity estimators can be employed for realized business time sampling. An application to stock data provides empirical evidence for the benefits of using these intrinsic sampling schemes to construct more efficient RV estimators as well as for an improved forecasting performance.
Richert, Ivo	Kiel University	Estimation of dynamically recalibrated affine models in finance	Dynamic recalibration of risk-neutral parameters in stochastic models to align with observed prices of financial derivatives is a widely used industry practice that, however, often lacks a tractable underlying mathematical framework. We address this gap by proposing a novel methodology wherein recalibrated parameters are treated as unobservable components embedded within a larger-scale affine model. The estimation of the dynamics of this background model then boils down to a two-step procedure, in which unobservable states are first calibrated to observed option prices using classic least-squares optimization techniques, and risk-neutral and physical model parameters are then jointly estimated to fit the trajectories of observed components alongside with the recalibrated states. We embed this joint estimation of both measures into the framework of Heyde-optimal estimating functions and establish weak consistency and asymptotic normality of the resulting estimators. Moreover, we derive explicitly computable expressions of the asymptotic estimator covariance matrix.
Rockel, Markus	Universität Freiburg	Closed-form Chatterjee’s xi for approximating copulas	We provide closed-form formulas for Chatterjee’s xi of multiple types of approximating copulas, which measure the strength of dependence between two random variables. In particular, we obtain lower bounds to the true Chatterjee’s xi value based on approximating checkerboard copulas. Lastly, we empirically study the performance of the resulting checkerboard based estimator for Chatterjee’s xi in a Gaussian setting.
Tambe Ndonfack, Felix Barrez	University of Freiburg, Math.Department	Filtering with Stochastic Discontinuities	In this work, we present a framework for filtering in stochastic systems with discontinuities. To address challenges arising from abrupt changes in observations, we model these phenomena using paired stochastic processes that represent unobserved signals and observable data streams. By leveraging a general random measure, our approach systematically characterizes discontinuities, offering a structured method for estimating hidden states in complex systems. Key theoretical contributions include the derivation of the Zakai and Kushner-Stratonovich equations, which enable precise computation of conditional distributions and enhance filtering accuracy in non-continuous dynamics. This framework has broad applications in domains where sudden state shifts are critical, such as financial risk assessment and advanced signal processing, providing a robust alternative to traditional continuous filtering models.
Valett, Henrik	Kiel University	Parameter estimation for polynomial models	Polynomial processes, which include affine processes as a subclass, are a class of Markov processes characterized by the property that their conditional polynomial moments can be computed in closed form. Due to their computational tractability, polynomial models are widely utilized in mathematical finance, with notable examples being the Heston model and Lévy-driven models. The aim of our research is to estimate the parameters of discretely observed polynomial models. In asymptotic statistics, the maximum likelihood estimator is highly desirable for its favorable properties, such as consistency, asymptotic normality, and minimal asymptotic error. However, in practice, the score function is often not available in closed form, which motivates the use of alternative approaches. We propose using martingale estimating functions in place of the score function, with a focus on the Heyde-optimal martingale estimating function, which minimizes the distance to the score function in an L2 sense within a specified class of estimating functions. Our framework constructs a specialized class of polynomial martingale estimating functions for general polynomial processes, requiring only the calculation of polynomial conditional moments. Specifically, we consider: where the maximum degree k is fixed, and the integrand λϑ , α(m), measurable with respect to Fm − 1, can be freely chosen. By applying ergodic theory for Markov processes, we establish both the consistency and asymptotic normality of these estimating functions. Additionally, we demonstrate how to explicitly compute the Heyde-optimal estimating function within this class.