Topic outline
-
All researchers (both junior and senior) of the research group are expected to approximately give one seminar talk per year. All talks should be an enjoyable learning experience to all members of the research group. In particular, a good motivating discussion is an essential ingredient. A talk does not have to be based on your own recent research paper, a presentation about methods or theory developed by other researchers can be a suitable topic for a seminar talk.
All researchers (both junior and senior) of the research group are expected to attend each talk of the seminar.
It is the responsibility of the host of an external speaker to explain the nature of the seminar to the speaker, in particular to explain what audience the speaker should expect. It is also the responsibility of the host of an external speaker to make sure that title and abstract are sent to the seminar organizer at least two weeks prior to the talk. Hosts of potential speaker should consult with the seminar organizer before inviting guests with the ambition to give a seminar talk. A standardized letter that can be submitted to external speakers, informing them about the matstat seminar can be found below (TBW). -
Title: From Risk Classification to Evolving Risk States: Predictive Analytics and Temporal Modeling with Player Account DataAbstract: This seminar presents two complementary approaches for characterizing gambling-related risk using player account data.
In Study 1, we develop a supervised learning pipeline to classify players into “low-risk” and “high-risk” groups. Emphasizing both predictive stability and chronological integrity, the pipeline integrates SHAP (SHapley Additive exPlanations) and Generalized Matrix Learning Vector Quantization (GMLVQ) for feature selection, and employs XGBoost (Gradient Boosting) within a nested forward-chaining cross-validation scheme to prevent information leakage. Hyperparameter optimization is conducted via Bayesian search strategy, and a temporal ablation study tests the pipeline’s stability under varying degrees of historical data availability.
In Study 2, we shift from static classification to modeling the evolution of gambling behaviors. We combine a Temporal Convolutional Network (TCN)-based Variational Autoencoder for feature extraction from sequences of player account activity with a Dynamic Bayesian Network to infer hidden “risk states” that offer a data-driven alternative to traditional definitions of problem gambling. In addition to the latent representations learned through the TCN-VAE, we incorporate select observable features—overlapping with those from Study 1—to capture critical behavioral signals. By merging both latent and observable dimensions, we hope to provide a more nuanced depiction of player risk trajectories than standard rule-based methods.
The overall aim is to demonstrate how predictive models and temporal representations can leverage player account data to advance the identification and understanding of gambling risks and the dynamics of problematic play.
-
Title: Stochastic Epidemic Models with Group Structures
Abstract: Social structures in communities are often formed in groups, such as households, workplaces, schools, ... Modern epidemic models try to capture such repeated type of contacts beside more random type contacts. In the first part of the talk we survey results for epidemic models having one group structure (households). In the second part we investigate what happens if a model neglects either the relative magnitude of group contacts, or the existence of some additional group group structure completely. For a wide class of epidemic models (but exceptions exist!) the meta theorem says that switching random contacts to new or existing group structures reduces the epidemic size. Properties of directed random networks is an important ingredient in the analysis.
-
-
Title: Self-normalized partial sums of heavy-tailed time series
Abstract: We study the joint limit behavior of sums, maxima and l^p-type moduli for samples taken from an R^d-valued regularly varying stationary sequence with infinite variance. As a consequence, we can determine the distributional limits for ratios of sums and maxima, studentized sums, and other self-normalized quantities in terms of hybrid characteristic-distribution functions and Laplace transforms. These transforms enable one to calculate moments of the limits and to characterize the differences between the iid and stationary cases in terms of indices which describe effects of extremal clustering on functionals acting on the dependent sequence.
Reference: https://arxiv.org/pdf/2303.17221 -
Title: Testing the mean-variance efficiency of a high-dimensional portfolio
Abstract: Optimal asset allocation is considered in a high-dimensional asymptotic regime when the number of assets and the sample size tend to infinity at the same rate. Using the techniques from random matrix theory, new inferential procedures based on the optimal shrinkage intensity for testing the mean-variance efficiency of a high-dimensional portfolio are developed and the asymptotic distributions of the proposed test statistics are derived. The practical advantage of the proposed procedures is demonstrated in an empirical study based on stocks included into the S&P 500 index. We found that there are periods of time where one can clearly reject the null hypothesis of mean-variance optimality of the equally weighted portfolio.
-
-
Title: Isotonic regression for variance estimation and its role in mean estimation and model validation
Abstract: We study isotonic regression which is a non-parametric rank-preserving regression technique. Under the assumption that the variance function of a response is monotone in its mean functional, we investigate a novel application of isotonic regression as an estimator of this variance function. Our proposal of variance estimation with isotonic regression is used in multiple classical regression problems focused on mean estimation and model validation. In a series of numerical examples, we (1) explore the power variance parameter of the variance function within Tweedie's family of distributions, (2) derive a semi-parametric bootstrap under heteroskedasticity, (3) provide a test for auto-calibration, (4) explore a quasi-likelihood approach to benefit from best-asymptotic estimation, (5) deal with several difficulties under lognormal assumptions. In all these problems we verify that the variance estimation with isotonic regression is essential for proper mean estimation and beneficial compared to traditional statistical techniques based on local polynomial smoothers.
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-