Climate change is driving unprecedented change in ecosystems. Ecosystem managers rely on predictions to inform adaptive management, yet it is unclear how accurately we can forecast abundance in rapidly changing ecosystems. Here we compare forecasts of abundances for 49 reef species in a global warming hotspot. We assess how accuracy of out-of-sample forecasts relates to species traits, time-series models of different complexity, protected and unprotected regions, degree of environmental change and different time horizons. Warmer water species were less predictable than cooler water species. The best model for cooler water species accounted for density dependence in per-capita growth rate whereas, models all models performed equally well for warmer water species. Finally, fished sites were less predictable than sites in the no-take marine reserve. Our results indicate that long-term warming and fishing pressure are reducing the predictability of this reef system. Testing model forecasts supports theory development, for instance, we observed density dependence in cooler water species, but not in warmer water species that may be at the early stages of immigration and population growth. We conclude that human pressures (climate change fishing) make adaptive management more challenging, as we are less able to predict species abundances.