An Accruals Quality Model: A Group Method of Data Handling Approa

Abstract

Development of Earnings quality measures, especially Accruals quality measures, has been a critical line of research over more than three decades. Literature indicates that linear-regression-based measures are subject to (suffer from) significant estimation error in non-discretionary accruals estimation. Therefore, recent research used machine learning algorithms including multilayer perceptron and radial basis neural networks, in order to address the issue. However, being founded on Blackbox approach limits future development and applicability of these methods. So, to address the limitations, we have used Group Method of Handling Data (GMDH) approach, as a Whitebox approach, in order to estimate the accruals. Findings using data from 299 Tehran Securities Exchange listed companies during 1385 to 1397 suggests that GMDH-based models perform superior to regression models and multilayer perceptron neural networks in terms of estimation error measured by mean squared error. Moreover, Cash flow approach in total accruals calculation leads to less estimation error compared to balance sheet approach. As a result, the model developed in this article can be used by market participants such as regulators, analyst and auditors in order to detect probable financial reporting misstatements.

Keywords

• group method of handling data (gmdh)
• earnings quality
• neural networks
• accruals
• accruals quality