Here, the final model will be used to create a flag for the investor/analyst warning him about the possibility of an impending crash. The investor will then make the decision to hedge his position against the possible fall. Therefore, it is very important that the model is able the predict all/most of the drastic falls. In other words, we want to choose a model with a better ability to have True Positives (better Recall), even if it comes with the cost of some False Positives (lower Precision). In other words, we do not want the model to miss the possibility of *‘sharp fall’. *We can afford to have some False Positive because if the model predicts that there will be a sharp fall, and the investor hedges his position, but the fall does not occur, the investor will lose opportunity cost of remaining invested or at most hedge cost (say if he buys out of money Put Options). This cost will be lower than the cost of false negative where the model predicts no ‘Sharp Fall’, but a massive fall does happen. We need to however keep a tab on the trade-offs in **Precision **and AUC.