Forecastability — Interactive Knowledge Graph

This graph maps the forecastability literature, from the information-theoretic limits of prediction to applied out-of-sample practice. Works are coloured by school: Shannon information theory, algorithmic information, dynamical systems, statistical and machine learning, and applied domain work. Each edge is a typed intellectual relation, recording what a work defines, estimates, builds on, evaluates or reconsiders. Edges are not citation links.

 

Read the shape before the nodes. Most of the applied literature shares one idea: a series counts as forecastable when some model turns out to predict it well, so forecastability is discovered after the fact, one model and one dataset at a time. The information-theoretic thread runs the other way: it measures the predictive information in the series itself, through mutual information, entropy rate and related quantities, so the limits of prediction are known at each horizon before any model is chosen. That divide is what this taxonomy shows, and the second path is the ground The Knowable Future occupies. Works marked with a star are the programme's own.

​

Search by author, title or concept; drag nodes; zoom; click any node for its full citation and relations. The bibliography is this graph’s static counterpart, and the two are maintained together. 

​

Cite this page: Catt, P.M. (2026). The forecastability graph. The Knowable Future. theknowablefuture.com/forecastability-graph