Operating Asymmetries and Non-Linear Spline Correction in Discretionary Accrual Models

Earnings management research often uses discretionary accruals from Jones-type models. These models assume a linear relation between sales changes and accruals. However, we predict and find that sales changes have a non-linear asymmetric effect on accruals through managers' operating decisions. By forcing a linear specification on this non-linear effect, the modified Jones model overestimates (underestimates) discretionary accruals for moderate (extreme) sales changes. This non-linear bias causes excessive type-I error in tests of positive (negative) discretionary accruals for subsamples with moderate (extreme) sales growth. The literature often controls for non-linearities using performance matching (Kothari et al. [2005]). However, we show that this approach leads to false inferences due to matching on the dependent variable. We use a flexible non-linear spline specification to control for the non-linear operating effect of sales changes. Our method successfully mitigates the bias, improves type-I errors without sacrificing test power, and changes inferences about major prior findings