Speaker 1: Matthias Linder, Leibniz University Hannover
Title: Elevating Trust in High-Stakes Decisions Using Glass-Box Models and Robust Feature Selection (Joint work with Judith Schneider and Brandon Schwab)
Abstract: We provide a unified framework that jointly leverages a robust selection procedure and a high-capacity, additive neural architecture in domains where transparency is crucial. We test this framework on a public and a large-scale proprietary insurance dataset. Our results prove that this approach allows the Neural Additive Model (NAM) to outperform traditional glass-box models in predictive accuracy and achieve superior economic outcomes in a competitive market simulation, establishing it as a robust and transparent alternative to black-box systems.
Speaker 2: Shihao Zhu, Ulm University
Title: Nonconcave Portfolio Choice under Smooth Ambiguity
Abstract: We study continuous-time portfolio choices with non-linear, option-style payoffs under smooth ambiguity and Bayesian learning. Our first contribution is a general non-concave dynamic asset allocation framework that admits arbitrary option-based payoffs and a wide class of utility functions and ambiguity attitude functions. We restore dynamic consistency via a robust representation that recasts the ambiguity-averse problem as an ambiguity-neutral one with distorted priors. This structure yields explicit trading rules by combining nonlinear filtering with the martingale approach and nests standard concave and linear-payoff benchmarks. As a leading application, delegated management with convex incentives shows that ambiguity aversion tilts beliefs toward pessimistic states, shrinks in-the-money regions, and lowers volatility through reduced risky exposure.