Forecastability: Insights from Physics, Graphical Decomposition, and Information Theory

Catt’s aim with this paper is to equip forecasters with some cross-disciplinary theory on forecastability and to provide practical techniques for assessing how forecastable a historical time series is. A time series is a sequence of values at equally spaced time intervals: days, weeks, months, quarters, or years. The historical time series can be viewed as an outcome (realization) of an underlying data generating process (DGP). Assessments of forecastability require an understanding of the DGP and its components. Copyright International Institute of Forecasters, 2009