Discriminative Power
The first model behaved like buy-and-hold because it predicted almost all signals as up. A model with no discriminative power has no predictive power even if its apparent accuracy is acceptable. AUC-ROC represents discriminative ability on…
1 sources - 5 claims
The first model behaved like buy-and-hold because it predicted almost all signals as up. A model with no discriminative power has no predictive power even if its apparent accuracy is acceptable. AUC-ROC represents discriminative ability on a normalized scale from 0 to 1. The second model has a small but real AUC edge above random guessing. The second model's negative lag weights indicate a mean reversion dynamic.