| The primary purpose of this study is to examine and compare the performances of two statistical approaches and two artificial neural networks approaches in the task of life insurer insolvency prediction. Multiple discriminant analysis (MDA) and logistic regression analysis were the first models to be used in insolvency studies. While they have been successful, there are some major limitations to these statistical models such as instability and linearity. Back-propagation (BP), an artificial neural networks algorithm frequently used in classification tasks, has been shown to be very good at classifying and predicting financially troubled companies. The second artificial neural networks technique, learning vector quantization (LVQ), is first attempted in this study for predicting financial failures.; The second purpose of this research is to investigate and compare the usefulness and effectiveness of several early warning systems, which includes the identification of significant variables in the early detection of financially troubled life insurers. The four early warning systems used in this study are the 22-variable model, IRIS (Insurance Regulatory Information System). FAST (Financial Analysis Solvency Tracking), and Texas EWIS (Early Warning Information System) model.; The third objective of this study is to examine another artificial neural networks technique, self-organizing feature map (SOM), as a possible method for a bankruptcy prediction in the hopes that this new model will offer enhanced visualization of the complex data.; We summarize the empirical results of this study as follows. First, the back-propagation (BP) and learning vector quantization (LVQ) outperform the traditional statistical approaches for all four data sets with a consistent superiority across the different evaluation criteria. Second, the results show that the 22-variable model and the Texas EWIS model are more efficient than the IRIS and the FAST model in most comparisons. Third, self-organizing feature map (SOM) supports the above findings by showing the distinct areas of bankrupt and non-bankrupt companies geographically. Due to its easier visual interpretation SOM can be used as a management tool by both insurance regulators and the companies themselves. |
