Josef Teichmann

Professor in the Department of Mathematics at ETH Zurich.

Title: The geometry of blurring – diffusion models under geometric constraints 

Abstract: We analyze generative technologies like diffusion models under global geometric constraints on the generated samples.

Josef Teichmann is an internationally renowned mathematician and Professor at ETH Zurich, one of Europe’s leading universities. His research spans stochastic analysis, mathematical finance, optimal transport, and machine learning, with a particular emphasis on uncovering deep mathematical structures and developing innovative methods at the intersection of these fields.

He is a co-author of the influential paper Deep Hedging (2019), which helped pioneer the use of deep learning and reinforcement learning in financial mathematics. This work demonstrated how neural networks can address complex portfolio hedging problems under realistic market conditions, including transaction costs and trading constraints. His broader research has contributed to the mathematical foundations of machine learning, stochastic modelling, and modern quantitative finance, bridging rigorous mathematical theory with challenging real-world applications.

His lectures are known for their mathematical depth, originality, and remarkable ability to reveal unexpected connections between seemingly distant areas of mathematics.

Ostap Okhrin

Professor, Vice Dean of the “Friedrich List” Faculty of Transport and Traffic Sciences, TUD Dresden University of Technology, Germany

Title:
Learning Low-Frequency Risk Functionals from High-Frequency Financial Data: From Heston Volatility to Skewness and Quantiles

Co-Authors: Haozhe Jiang (TUD) and Michael Rockinger (Uni Lausanne)

Abstract: High-frequency financial data contain rich information about the latent dynamics of returns, volatility, and tail risk. However, classical estimators of low-frequency risk characteristics often rely on nonlinear plug-in ratios, long historical panels, or stationarity assumptions that are difficult to justify in practice. In this talk, I discuss a simulation-based framework for estimating low-frequency distributional functionals from high-frequency observations.

The starting point is the Heston stochastic volatility model, where high-frequency data can be used to estimate spot volatility and the parameters of the latent variance process. This setting highlights both the potential and the difficulty of high-frequency inference, in particular under realistic parameter configurations in which the Feller condition may be violated. Building on these insights, we propose a High Frequency Network, a convolutional neural network architecture that maps intraday returns and their powers directly to monthly realized skewness. Instead of estimating separate moment components and forming unstable ratios, the network learns the low-frequency functional itself from simulated stochastic-volatility paths. Monte Carlo experiments show that this approach reduces finite-sample bias and remains robust under model misspecification, including jump-diffusion dynamics.

Finally, I outline ongoing work extending the same idea to the estimation of low-frequency quantiles. The broader aim is to combine stochastic-process models, high-frequency data, and machine-learning architectures to obtain stable estimators of distributional risk measures relevant for financial econometrics and risk management.

Ostap Okhrin is a distinguished statistician and Professor of Statistics and Econometrics at TU Dresden, where he also serves as Vice Dean of the Faculty of Transport and Traffic Sciences. His research lies at the intersection of mathematical statistics, stochastic modelling, econometrics, and artificial intelligence, combining sophisticated theoretical methods with applications to complex real-world problems.

His scientific contributions cover a broad range of topics, including dependence modelling, copula theory, financial econometrics, statistical inference, and machine learning. He is a co-author of several books on computational statistics and stochastic modelling, including Hierarchical Archimedean Copulas, and has published extensively in leading international journals.

One particularly exciting direction of his recent research is the application of deep reinforcement learning to intelligent transportation and autonomous systems. Together with his research group in Dresden, he develops mathematical and data-driven methods for modelling traffic dynamics, autonomous driving, and the navigation of self-driving vessels. This work illustrates how ideas from probability theory, statistical learning, and stochastic control can be transformed into algorithms capable of addressing challenging problems in modern transportation.

At the same time, Professor Okhrin maintains a strong research focus on financial econometrics and risk modelling. His recent work investigates how high-frequency financial observations can be used to infer stochastic volatility and other characteristics of financial risk, as well as how neural networks can improve statistical estimation when classical methods face substantial difficulties.

Anton Yurchenko-Tytarenko

Senior Market Analyst Statkraft Energi AS, Norway 

Title: Least squares Monte Carlo for pricing, hedging, and calibration in stochastic market models

Abstract: Modern financial markets exhibit complex phenomena that simple models cannot capture. Replicating these features requires sophisticated stochastic frameworks: rough volatility models, for instance, are widely used to reproduce the implied volatility smiles and skews observed in real markets. 

However, increased realism comes at a price. Complex models, even when they admit simulation via Euler-type discretization schemes, typically lack closed-form expressions for option prices or hedging quantities. This gap creates significant practical challenges. Without analytic pricing formulas, calibrating model parameters to observed option price curves becomes a difficult task. At the same time, classical Monte Carlo methods, while flexible, are computationally too slow for real-time applications such as dynamically updating a hedging portfolio during trading.

In this talk, we discuss the Least Squares Monte Carlo (LSMC) method as a flexible and efficient framework to address these challenges. The approach proceeds in two stages: one first simulates a large ensemble of model trajectories across a range of model parameters, and then applies machine learning techniques to learn how the quantity of interest depends on those parameters. The resulting surrogate is fast to evaluate and well-suited for tasks such as real-time model calibration against market data.

Anton Yurchenko-Titarenko completed a master’s degree in Statistics at Taras Shevchenko National University of Kyiv in 2019 and pursued postgraduate studies at the University of Oslo. He defended a dissertation on the topic “Stochastic Volterra volatility models” in 2022.

He is a specialist in stochastic processes, stochastic differential equations, fractional models, financial modeling and statistical estimation. He is an author of 14 research papers and a monograph Giulia di Nunno, Yuliya Mishura, Anton Yurchenko-Tytarenko "Volterra Volatility Models. Option pricing and hedging", Springer, 2026.

Alexander Kukush

Doctor of science, Professor, Leading researcher at Institute of Mathematics NAS of Ukraine

Title: Nominal-gain tontine fund with a heterogeneous cohort

Abstract: We investigate tontine funds with a heterogeneous cohort of participants. At time 0, each participant makes an initial investment 𝜋_𝑖 in the fund. At time 1, each surviving participant receives a part of the fund, proportional to the number of tontine shares (or protection units) 𝑓_𝑖 appointed to him at time 0. The tontine fund is self-financing in the sense that the total input in the fund (i.e., the total amount of initial investments) equals the total output from the fund (i.e., the total payments, also called compensations, to the survivor). Two versions of the tontine fund are considered: the case with an active administrator and the case with a passive one.

It is shown that under mild conditions the following results hold:
(a) Given the tontine shares 𝑓_𝑖 and given total amount of initial investments, there exists a unique actuarially fair tontine fund, i.e., a fund where the initial investment 𝜋_𝑖 of each participant 𝑖 is equal to his expected compensation.
(b) Given the initial investments 𝜋_𝑖 , there exists a unique choice of tontine shares 𝑓_𝑖 (up to a constant proportion factor) so that the tontine fund is actuarially fair.

The tontine fund considered so far is a so-called ‘nominal payoff’ tontine fund. We also consider ‘nominal gain’ tontine funds where the initial investments are used to pay every survivor his initial investment 𝜋_𝑖 back and where the investments of the deceased participants are proportionally distributed among alive participants, according to the tontine shares 𝑓_𝑖 appointed at time 0. Similar results as for nominal payoff tontine funds are presented for this class of for nominal gain tontine funds.

Further, we show that given the total amount that each participant spends to pay an initial investment in a tontine fund and a premium to buy insurance, combining a tontine fund with a standard insurance contract, one can achieve a less risky compensation (i.e., tontine compensation and insurance payment) compared with a ‘pure’ insurance case. Here, the riskiness of random losses is compared using the convex ordering.

Finally, we study tontine funds when the number of participants grows. This allows evaluation of the limiting compensations and investigation of the asymptotic actuarial fairness of tontine funds. A particular case of a tontine fund composed of several homogeneous tontine funds is studied in detail.

The results are co-authored with Prof. Dr. Jan Dhaene, K.U.Leuven, Belgium.

Presentation

Aleksandr Kukush is a renowned statistician and a leading research fellow in the Department of Theory of Random Processes at the Institute of Mathematics of the National Academy of Sciences of Ukraine. He also holds the posts of professor at Kyiv Academic University and senior research fellow at the National Scientific Centre for Radiation Medicine, and spent 25 years as a professor at the Faculty of Mechanics and Mathematics of Taras Shevchenko National University of Kyiv. His research focuses on mathematical statistics and the statistics of random processes, actuarial and financial mathematics, biostatistics, spatial statistics, and problems of dynamic system identification, combining sophisticated analytical methods with applications to the solution of practical problems. In 2006, he received the Shevchenko Prize from the Academic Council of Taras Shevchenko National University of Kyiv for a series of research papers on regression analysis. He is the author of a book on infinite-dimensional analysis, co-author of a monograph on the assessment of radiation risks, and co-author of several collections of problems published by Springer. He has published numerous papers in leading international journals, including first-quartile journals such as the Electronic Journal of Statistics, Numerische Mathematik, The International Journal of Biostatistics, Biostatistics and PLOS ONE.

One of the key areas of his long-standing research is the assessment of the risk of developing cancer as a result of exposure to radioactive iodine following the Chernobyl accident. Together with a research team at the Radiation Protection Laboratory, and in collaboration with specialists from the Cancer Institute (Washington, USA) and Texas A&M University (USA), he has developed original algorithms and programmes for estimating risk coefficients, taking into account both instrumental error in radiation doses and the Berkson error arising from data averaging. This work illustrates how ideas from the statistical theory of non-linear regression models can be used to address problems in modern radiation epidemiology.

At the same time, Professor Kukush devotes considerable attention to research in the field of actuarial mathematics. In his work, carried out in collaboration with Professor Jan Daene, head of the actuarial programme at the Catholic University of Leuven (Belgium), he investigates mechanisms for managing insurance risks and the interaction between the financial and insurance markets.

Vitalii Golomoziy

Associate Professor at Taras Shevchenko National University of Kyiv,
doctor of sciences in physics and mathematics

Title: The Mathematics of Generative AI in Image Generation

Abstract: Generative AI models can be broadly divided into natural language processing models and those working with images or spatial data. While the former are mostly engineering creations, generative models for images utilize non-trivial mathematics. In this talk, we will examine the main mathematical methods and their applications in this field.

Vitaliy Golomoziy’s research focuses on probability theory and stochastic processes. He is particularly interested in the stability and ergodicity of time-inhomogeneous Markov chains. His scientific contributions are centered on the development of coupling methods for the analysis of Markov chains, yielding important results on stability estimates, recurrence properties, convergence rates, and renewal processes.

Alongside his research activity, Vitaliy Golomoziy contributes actively to the advancement of modern mathematical education and interdisciplinary applications of probability and statistics. He teaches courses in machine learning, statistical learning, finance, and financial analysis. His work bridges classical probability theory with contemporary developments in machine learning and artificial intelligence, and he regularly presents his research at leading international conferences in stochastic analysis, probability, and applied mathematics.

Vladyslav Tomashyk

PhD, AIA C.Act IFoA Risk manager & actuary, Romania

Title: Valuation of Technical Provisions in Defined Contribution Schemes: The Case of Romania

Abstract: The Romanian pension scheme requires mandatory defined contribution (Pillar II) administrators to guarantee that exit payouts do not fall below net cumulative contributions. This presentation outlines this institutional setup, details the legal guarantee framework, and demonstrates the quantitative valuation of the required technical provision using the Cox-Ross-Rubinstein (CRR) binomial tree model to price the embedded option on the administrator’s balance sheet.

Vladyslav Tomashik is a graduate of the Faculty of Mechanics and Mathematics, holds a Doctor of Philosophy (PhD) and is a certified actuary. His professional work focuses on actuarial analysis, risk management and financial modelling in the fields of pensions and insurance.

He currently works as a risk manager at a Romanian private pension fund, where he is responsible for assessing and monitoring financial, actuarial and operational risks, analysing the long-term sustainability of pension schemes, and ensuring that the fund’s operations comply with regulatory requirements and international risk management standards.

Anna Panfilova

Deputy Chair of the Management Board, Chief Financial Officer, TAS life, Ukraine

Title: Actuarial Methods for Modeling Insurance Company Risks

Abstract: We look at how actuarial methods are used to model and assess risks in life insurance. Based on the experience of a Ukrainian insurance company, we show how actuarial calculations support business decisions, risk management, and financial stability in uncertain times

Anna Panfilova is Deputy Chair of the Management Board and a Chief Financial Officer of the insurance company TAS Life. In her role, she is responsible for formulating and implementing the company’s financial strategy, ensuring its financial stability, the effective management of assets and insurance reserves, as well as overseeing financial reporting and compliance with regulatory requirements.

Vitaliy Miroshnychenko

Ph.D. in Statistics, Assistant Professor at Taras Shevchenko National University of Kyiv

Title: ARIMA models for the analysis and forecasting of time series

Abstract: This presentation will examine the classical approach to time series modelling using ARIMA models. We will briefly discuss the key concepts underlying autoregressive models and moving average models, as well as their relationship to the stationarity of time series.

Particular attention will be paid to the practical application of ARIMA for forecasting. Using real-world time series from the M5 Forecasting competition as examples, the main stages of model development will be demonstrated: series analysis, parameter selection, forecast generation and quality assessment.

His research interests include mathematical statistics, mixture models, and machine learning. Current research focuses on mixture-of-experts models, their theoretical properties, parameter estimation, and applications to deep learning and generative models.

Vitaliy Miroshnychenko has 10 years of practical experience in data analysis and machine learning, mainly in retail, telecommunications, and legal document analysis. His work includes time series forecasting, classification, predictive modeling, and customer behavior analysis. He has experience with large datasets and the development of machine learning models for real-world business problems.

He also participates in machine learning competitions on Kaggle. His team achieved 7th place in an international competition “Handwritten to data” organized by the Ministry of Justice, focused on handwritten text recognition.

Vitaliy is also involved in the LCAM (Legal Case Law Analytics) research project, which focuses on the analysis of Ukrainian court decisions using statistical methods, natural language processing, and machine learning. His work includes developing a taxonomy for classifying court cases and building models to predict their progress.