The Use of Unsigned Earnings Quality Measures in Tests of Earnings Management

Abstract

Several recent papers employ unsigned measures of discretionary accruals to test hypotheses that predict general classes of firms will engage in earnings management, without specifying the direction or the time period in which the earnings management occurs. We argue that the research design choices using unsigned measures of earnings management heighten the threat of correlated omitted variables because these measures are correlated with stable firm characteristics. We provide evidence on the correlation between unsigned measures of discretionary accruals and firm characteristics such as market value of equity, total assets, sales growth, leverage, book-to-market ratios, cash from operations, volatility of sales, volatility of earnings, and volatility of cash flows. Our results show that unsigned measures have the highest correlations with volatility of sales, volatility of earnings, and volatility of cash flows, despite the fact that prior work often ignores these characteristics. We generate simulations to show that even modest correlation between these volatility metrics and the partitioning variable dramatically inflates the risk of incorrectly rejecting the null hypothesis of no earnings management. Finally, we replicate a recent study and we demonstrate that inferences change with the inclusion of these volatility measures in tests of earnings management based on unsigned discretionary accruals. Our results inform researchers of the potential hazards posed by correlated omitted variables in research designs that use unsigned measures of earnings management.

Keywords: Accruals, Earnings Management, Earnings Quality

JEL Classification: G30, M41, M43