Uniform accrual generating process grouping with self-organizing maps

Most earnings management and earnings quality studies rely on various types of discretionary accrual estimation models. Common assumptions when using these models is that the accrual generating process (AGP) is stable over time or that firms within the same industry have similar AGPs. These assumptions have, however, been challenged in a number of studies. Instead, it has been suggested that AGP is depicted by various accrual determinants and that firms should be grouped according to similarities in the AGP. The purpose of this study is to develop and assess the performance of a self-organizing map (SOM) local regression-based discretionary accrual estimation model. Overall, the results show that the SOM local regression model outperforms previously suggested discretionary accrual estimation models. For example, the detection rate of simulated earnings management for the SOM local regression model is almost twice the detection rate of the commonly used cross-sectional Jones model. In addition to outperforming previously suggested models, the SOM local regression model also gives a visual representation of the AGP of a specific firm in relation to other firms.