Yuktibhasa seminar: Empirical VaR Failure: Reconciling KS and Wasserstein Metrics for Risk Validation

Abstract. In quantitative risk validation, practitioners frequently compare a full revaluation loss or return distribution to an empirical proxy, checking if sample risk measures match. While both the Kolmogorov-Smirnov (KS) distance and the 1-Wasserstein (W_{1}) distance converge to zero between an empirical measure and its population law, it is possible to observe severe instability in the empirical Value at Risk (VaR). This talk provides a rigorous mathematical framework to explain these empirical discrepancies.

All are cordially invited to the seminar.