Estimating the Amount of Estimation in Accruals

This paper examines the link between the amount of estimation needed during the accrual generating process and the persistence of the accruals portion of earnings. We measure the amount of estimation needed during the accrual generating process using the number of estimation-related linguistic cues in the notes to the financial statements and the critical accounting policies section of the management discussion and analysis. Consistent with the conjecture in Sloan [1996], we find that accruals that need more estimation are less predictive of future earnings. We also find that such accruals map less into the past, current, or future cash flows in the spirit of Dechow and Dichev [2002]. When we decompose the number of estimation linguistic cues into a component that is due to the existence of specific accruals accounts and a component that is due to the within-accounts variations, we find that our results are driven by both components. Lastly, we find mixed evidence as to whether the amount of estimation in accruals is systematically associated with the accrual anomaly. Overall, our results suggest that the estimation needed during the accrual generating process plays an important role in understanding the persistence of accruals. We thank Matt DeAngelis, Ilia Dichev, Scott Richardson, Lakshmanan Shivakumar, Irem Tuna, and the workshop participants at University of Buffalo, Florida State University, London Business School, University of Michigan, Michigan State University, Purdue University, and Rice University for their comments. We gratefully acknowledge the support of the Ross School of Business and the financial support from the Paton Accounting Fellowship and the Harry Jones Endowment. 1