For understanding how tricky it is to translate “historical understanding” into forecasting accuracy, I recommend the elegantly threaded arguments of @ProfPaulPoast