Neural networks as bond rating tools

Seven financial parameters identified as possibly effective predictors of bond ratings produced by recognised agencies were used to train small multilayer neural networks. The rating predictions of trained networks were compared with the results from linear discriminant models. Networks made superior predictions in evaluating the bonds of the operating companies resulting from the divestiture of AT&T. Although the comparison of the neural network and linear disciminant models was fair, it still remains to be proven that either neural networks or linear discriminant models can be relied upon to make predictions that can pass tests made on data patterns held out entirely during the model building process