Discretization, Simulation, and Swanson's (Inaccurate) Mean

Abstract

SummarySwanson's mean (SM) is heavily used within the oil and gas industry to approximate continuous probability distributions such as the log-normal. In this paper, we document the errors induced by this practice, which, as we show, has no theoretical justification for any distribution other than the normal. In parallel, we review methods to discretize continuous distributions and compare these methods to Monte Carlo (MC) simulation. We demonstrate that the best discretization methods have an accuracy equivalent to that of tens of thousands of MC trials.