Training accuracy much higher than test accuracy - Performance degrades on new seasons - Complex models underperform simple ones