财务危机预警文献,仅供学习研究
The estimated parameters and variable data are combined to yield estimated probability of bankruptcy for each holdout firm at age t. The estimated values are compared with the optimal cutoff scores that minimize the sum of type I and type II errors in the training sample (e.g., Begley, Ming and Watts 1996). A type I error (α(P)) occurs if the firm is bankrupt but is misclassified as non-bankrupt. A type II error (β(P)) occurs if the firm is non-bankrupt but is misclassified as bankrupt. 3.3.2. ROC (receiver operating characteristic
This study also refers to ROC receiver operating characteristic curve (e.g., to assess the quality of prediction models. When making Sobehart and Keenan 2001
prediction, the decision maker’s prediction can fall into one of the following four outcomes.
Predicted Bankruptcy status
Actual status
Bankruptcy I
III IV
Non-bankruptcy II
I, II, III, IV represent the number of firms falling into each category
Under a selected cut-off point C, the hit rate of bankruptcy is defined as HR C =I/ I II . The false alarm rate is FAR C =III/ III IV . A ROC curve of false alarm
rate versus hit rate is plotted while the cut-off C is varied, as depicted in Figure 1.