Re-visiting the efficacy of Neural Networks in Detecting Accrual and Real Earnings Management

Abstract

Earnings management practices by firms to conceal their actual financial performance have always been a matter of concern for industry and academia Therefore, over some time researchers have developed various models for the timely detection of earnings management practices However, the efficiency of such models has been frequently challenged in numerous studies One possible reason given in the literature for the ineffectiveness of these models is the assumption of the linear relationship between the variables used in the model Against this backdrop, the present study explores Neural Network based methods to model earnings management practices, which drops the assumption of linearity The objective of the current work is to investigate that, while detecting earnings management, do neural network-based models show better results as compared to traditional linear regression models in A Linear Regression (LR) model was compared with three neural networks i e Multi-layer perceptron (MLP), self-organizing map (SOM),