Detecting Earnings Management Using Neural Networks

Numerous studies have examined earnings management under various conditions. In most of these studies, it is assumed that earnings are managed through accounting accruals. Accordingly, models have been developed to detect earnings management based on accruals. However, some research has questioned the effectiveness of these models, arguing that their weak performance stems from relying on linear approaches, while evidence suggests the presence of nonlinear relationships in the data One proposed solution is the use of artificial neural networks. The aim of this study is to investigate the possibility of detecting earnings management using selected mathematical models and to compare their performance with neural network–based models. Two types of neural networks were employed: multilayer perceptron (MLP) and radial basis function (RBF) networks. The findings indicate that although neural networks outperform linear regression models, a definitive choice between the two approaches is not possible, as performance depends on modeling ability and the selected topology.

کلیدواژه ها:

Earnings management ، discretionary accruals ، neural networks