On not evaluating economic models by forecast outcomes