If political prognostication is usually so wrong, why do we keep listening? And what makes a good forecaster, anyway?