Integration of Neural Networks and Semantic Interpretation for Bankruptcy Prediction

Abstract

Authors: Yelena Y. Makeeva National Research University The Higher School of Economics len-makeeva@yandex.ru

Igor Vladimirovich Arshavsky National Research University The Higher School of Economics

For years, the prediction of corporate failure remained one of the controversial topics in the field of economics. Authors suggested a number of financial indicators which reflect quantitative information and in some way affect the probability of corporate failure. However sufficient part of the corporate information is stored in a qualitative form and is not reflected in various financial indicators. The quality of corporate governance and the degree of corporate disclosure are good examples of this type of information. As the corporate annual report contains important facts and indication of the company’s current and future performance, it is crucial to consider it as a source of nonfinancial information useful to predict corporate failure.This work demonstrates the methodology of corporate failure prediction based on semantic analysis of corporate annual reports and the use of neural network ensemble. The obtained results confirm the importance of textual information contained in annual reports and its positive effect on the predictive ability of the forecast model.