财务危机预警文献,仅供学习研究
Figure 1: ROC curve
A ROC curve always goes through two points (0,0 and 1,1). 0,0 is where the predictor finds no positives (detects no bankruptcy). In this case it always gets the negative (non-bankruptcy) cases right but it gets all positive cases (bankruptcy) wrong. The second point is 1,1 where every firm is classified as bankruptcy. So the predictor gets all bankruptcy cases right but it gets all non-bankruptcy wrong. A predictor that randomly guesses has ROC which lies somewhere along the diagonal line connecting 0,0 and 1,1 (Random predictor line in Figure 1). The average area under the ROC is a convenient way of comparing prediction models (Hayden 2002 . The greater the average area under curve, AUC, the better the predictability of model is. A random classifier (Random guessing line) has an area of 0.5, while and ideal one has an area of 1. We use U test of Mann-Whitney (1947) to examine if the average under curve of different models is significantly greater than 0.5.
