| Abstract: | There was a significant increase in the frequency and magnitude of asset writedowns by US firms during the 1980s. Auditors, financial analysts and regulators have shown considerable interest in evaluating the writedown phenomenon. This paper reports on the use of inductive learning to discover knowledge in financial data structures and describes the development and testing of a prototype expert system, WDXPERT, which evaluates asset writedowns. Real-world data relating to writedown and non-writedown firms are used for rule induction. A set of training examples comprising 42 writedown and 25 non-writedown firms are used to generate the rules using IXL, a machine-learning program. A separate holdout sample containing 43 writedown and 25 non-writedown firms are used to validate the expert system that incorporates these rules. A second validation procedure is performed by comparing the performance of the expert system with a conventional discriminant analysis model and a logit model using the same data sets. The results indicate that the expert system, WDXPERT, is a useful classification tool to group firms into writedown and non-writedown classes. |
