Does incorporating non-linearity into discretionary accrual models improve their performance?

Using a large sample of firms that restated earnings, this study investigates whether incorporating non-linearity (conditional conservatism) into discretionary accrual models improves their performance in detecting earnings management. The findings of this study are important because discretionary accrual models play a prominent role in several streams of accounting research and the models' ability to isolate the discretionary (managed) component from the non-discretionary (unmanaged) component of total accruals is critical. If the conventional linear discretionary accrual models are mis-specified, it is likely to result in misleading inferences about earnings management behavior. The findings indicate that the non-linear specification improves the performance of most linear models. The findings also indicate that a more sophisticated linear model that incorporates a performance measure and a future growth measure outperforms other simple models.