Applying asymmetric-stratified data envelopment analysis model for bankruptcy prediction

The high social costs associated with bankruptcy have spurred searches for better prediction capability. We propose a nonparametric approach for bankruptcy prediction, using data envelopment analysis (DEA) model to identify the boundaries of bankruptcy and non-bankruptcy. The benchmarks of non-bankruptcy and bankruptcy can construct two piecewise frontiers to dominate two convexity classes. Overlap might appear if companies dominated by both frontiers simultaneously, which results in the Type I and Type II errors. The proposed asymmetric-stratified DEA model was applied to find the sequential layers of frontiers. By minimizing the misclassification cost in accordance with different risks and costs of Type I and Type II errors, we identified the optimal layers of frontier to be as separating hyperplane. Based on a sample with equivalent costs of Type I and Type II errors, our approach perform high prediction accuracy with the hit-ratios 94.44% on training and 91.67% on hold-out samples. Iif the proportion of Type II to Type I cost is greater than 10, the results indicate hit-ratio and misclassification cost trade-off to each other. Not alike traditional cut-off value approach used in discriminant analysis, this asymmetric-layering approach can provide more than one alternatives of discriminant hyperplane for bankruptcy prediction in accordance with the institutions' risk attitude.