Bias and Measurement Error in Discretionary Accrual Models

Abstract

This paper examines bias and measurement error in discretionary accrual models. This is important because the empirical evidence supporting the conclusions from many earnings management studies is based entirely on estimates from these models. Models of discretionary accruals are actually models of expected and unexpected accruals. Therefore for most earnings management studies, unexpected accruals that arise for reasons other than managerial discretion over financial reporting represent measurement error in discretionary accruals. This study proposes that unexpected accruals will arise from large structural changes such as acquisitions and divestitures. Since most earnings management studies examine managerial discretion over financial reporting these unexpected accruals represent measurement error. I find that structural changes are associated with the direction and magnitude of discretionary accrual estimates. Also, the measurement error associated with these structural changes is correlated with the level of earnings. Since the level of earnings is often used by researchers to partition earnings management incentives (or is correlated with the partitioning variable). This suggests that the empirical results from many earnings management studies are biased.