Model risk in American options
A new preprint—'If Not Now, Then When? Model Risk in the Optimal Exercise of American Options'—applies econometric model-risk analysis to early-exercise decisions, highlighting where standard optimal-exercise rules can fail. It's a solid hook for an options-focused quant project on robust exercise policies. (x.com)
arXiv:2603.19984 — "If Not Now, Then When? Model Risk in the Optimal Exercise of American Options" — was submitted on 20 March 2026 by Luna Rigby, Rüdiger Frey and Erik Schlögl. (arxiv.org) The arXiv HTML lists Luna Rigby with contact luna.rigby@wu.ac.at, indicating a WU affiliation. (arxiv.org) Rüdiger Frey is a full professor of Mathematics and Finance at Vienna University of Economics and Business and Erik Schlögl is Professor and Director of the Quantitative Finance Research Centre at University of Technology Sydney. (research.wu.ac.at) The paper assumes the true data-generating process follows the Heston stochastic-volatility model and contrasts optimal-exercise boundaries from a correctly specified Heston user with those computed under Black–Scholes and Dupire local-volatility misspecifications. (arxiv.org) Optimal-exercise boundaries are computed numerically using finite-difference methods in the authors' experiments. (arxiv.org) The study adopts the benchmark model-risk methodology introduced by Hull and Suo (2002) to mimic practitioner calibration and pricing workflows. (www-2.rotman.utoronto.ca) Primary result: stochastic-volatility dynamics and return–volatility correlation substantially alter optimal-exercise behaviour across models, producing model risk that is not transmitted to exercise strategies determined solely by models calibrated to European option prices. (arxiv.org) The authors report that frequent recalibration of a misspecified model does not eliminate this exercise-model risk. (arxiv.org) The arXiv entry classifies the manuscript under q-fin.MF (Mathematical Finance) and q-fin.RM (Risk Management) and lists the initial submission package size as 19,832 KB. (arxiv.org)
