Forecastability: a formal definition

This note states, in one citable place, the definition of forecastability developed from the author's work on the subject since 2009 and formalised in Catt (2026, arXiv:2603.27074). Forecastability is defined as the amount of predictive information that a declared information set carries about a future value of a process at a stated horizon: F(h), the mutual information between the future observation and the declared information set, which under logarithmic loss equals the maximum achievable reduction in expected loss over the unconditional predictive distribution. The note sets out four properties of the resulting forecastability profile (conditional on the information set; resolved by horizon; a bound on every method under logarithmic loss; not raisable by processing), distinguishes forecastability from forecast accuracy and from exploitability, defines the exploitability ratio and exploitability gap, summarises estimation, and relates the definition to the coefficient of variation, entropy statistics and spectral measures. It is written for reference-work editors, package maintainers and encyclopaedia contributors, and contains no new results.