Failure Discrimination and Rating of Enterprises by Semi-Definite Programming

We propose a new method forfailure discrimination and rating of enterprises using financialdata compiled from their balance sheets. No particular distributional assumption is made on the underlying data. Our method automatically discriminates and rates many enterprises using mathematical programming methods. We separate multi-dimensional data byhyperplane and hyper-ellipsoid, so that we can interpretthe results of classification from the geometric point of view. Theproblem to be solved here is a linear programming problem orsemi-definite programming problem which can be solved efficiently byinterior point algorithms. Numerical simulations usingreal data show that hyper-ellipsoid separation generates a result which can be used for practical purposes.