Evaluating New Earthquake Forecasting Models: Against a Strong ETAS Benchmark, Not an Underimplemented One

Abstract

New earthquake forecasting models are increasingly reported to outperform established models such as the epidemic-type aftershock sequence (ETAS) model. However, the scientific meaning of such improvements depends critically on how the benchmark is constructed. I use the term “under-implemented ETAS” for a benchmark that is unnecessarily restricted, weakly optimized, or otherwise disadvantaged, and propose the Parity of information, Optimization, Generality, and Assessment (POGA) test as a simple framework for assessing model comparisons. The purpose is not to protect ETAS from competition, but to ensure that improvements over established models are scientifically interpretable. A strong benchmark makes victory harder, but also reveals what a new model has genuinely learned that its competitor could not.