A UK Corporate Bond Rating Classification Using Probabilistic Neural Networks

摘要

The research attempts to classify bonds issued by the UK companies using Probabilistic Neural Network (PNN), which has not been the algorithm of choice for previous studies with small data sets. The study uses a total sample of 55 UK companies. The neural network was trained by means of five financial ratios. All experiments carried out showed that the PNN was able to classify corporate bonds. The degree of accuracy depends on the choice of a PNN model and the data used to train and test that particular model. When a PNN is trained, a non-linear function is created which is mapped on to the test pattern file. This test pattern file has its independent variables (ratios) evaluated by the different layers of the PNN and the predicted classification output is given based on this nonlinear function. The validation tests that there is no significant difference between the PNN output and random chance are applied. Two out of five models classifying corporate bonds managed to surpass the validation, the higher of these results achieving a 75.47% accuracy in classification (PNN jack-knife model predicting Aa and A bond ratings).