Data-free statistics

Abstract:

Extreme risks must be evaluated in such contexts as quarantine, terrorism and banking. Unfortunately, an extreme risk is one that hasn't happened yet, so directly relevant data is non-existent. The talk surveys what is done in the Basel II compliance regime in banking and in Australian quarantine risk analysis, where there are formal processes for using small data sets to keep expert opinion honest. The usefulness of Extreme Value Theory is considered. Extreme risks raise in acute form the "reference class problem", of how to decide what is the right class in which to take statistics to bear on an individual case. The views of philosophers and legal theorists on the reference class problem are canvassed.