Title: The most predictable aspects of time series
Authors: Tommaso Proietti - University of Roma Tor Vergata (Italy) [presenting]
Abstract: The focus is on establishing the most predictable aspects of a univariate time series. Assuming the mutual information between past and future as the measure of predictability, we consider classes of transformations depending on a set of basis functions, that are combined linearly. The coefficients are chosen so as to maximize the mutual information between past and future.