Better Forecasting Through Information Theory
Abstract: Probabilistic forecasts are fundamental tools for making decisions under uncertainty in a wide variety of fields, including short-term and seasonal-term weather and energy supply and demand.
Carlo Graziani will present a scheme whereby a base probabilistic forecasting system that is poorly calibrated may be recalibrated by incorporating past performance information to produce a new forecasting system that is superior to the original one, in that it produces probability distributions that demonstrably furnish more reliable decision support than the original forecast system. The recalibration procedure need know nothing about how the original forecasts were produced, and can therefore be applied to improve any probabilistic forecast system. The scheme is formulated in a framework that exploits the deep connections between information theory, forecasting, and betting.
Bio: Carlo Graziani is a computational scientist and physicist. He earned his Ph.D. at the University of Chicago in 1993. His research interests include mathematical statistics, proton radiographic image reconstruction, numerical solvers for hydrodynamics and magnetohydrodynamics, and related applications.