The application of neural network based methods to the extraction of knowledge from accounting reports

A new approach to the problem of extracting meaningful information from samples of accounting reports is developed, which effectively avoids the search of appropriate ratios by the analyst along with some other major drawbacks of the multivariate statistical modelling techniques used in accountancy.

Abstract

Develops a new approach to the problem of extracting meaningful information from samples of accounting reports. Neural network-like algorithms are capable of implementing this approach. Such tools are able to automatically build optimal structures similar to financial ratios. Some results are presented. They suggest that this approach effectively avoids the search of appropriate ratios by the analyst along with some other major drawbacks of the multivariate statistical modelling techniques used in accountancy. The organization of the neural network models also outlines internal features of accounting data, providing new insights into the relative importance of variables for modelling a particular relation. The paper also argues that, in the accounting and finance context, a major problem of neural networks, that of understandability of the resulting parameters, is minimised. Much of the internal operation of the networks involves the construction of generalisations of the ratio concepts with which accountants are familiar.>