How predictable are large earthquakes beyond clustering statistics?
In plain words
Earthquakes cluster in time, which allows rough probability forecasts. Whether any physics or data can do much better than that is unknown.
Precise statement
Quantify the maximum probability gain over ETAS achievable for $M \ge 6$ earthquakes in windows of about 30 days and 100 km, using any combination of seismic, geodetic, hydrological or electromagnetic data. An answer is either a prospectively tested forecast with gain significantly above 1 in a CSEP-type experiment (CSEP: a community testing framework for earthquake forecasts), or an information-theoretic bound showing gains are limited.
What would settle it
Multi-year prospective forecast experiments scored against ETAS with likelihood-based information gain.
Status in the literature
Unverified note
A 2026 review (arXiv:2607.26918) frames predictability in information-theoretic terms; no method has shown large prospective gain.