Detecting earnings management: a comparison of accrual and real earnings manipulation models

Abstract

PurposeThe use of models for detecting earnings management in the academic literature, using accrual and real manipulation, is commonplace. The purpose of the current study is to compare the power of these models in a United Kingdom (UK) sample of 19,424 firm-year observations during the period 1991–2018. The authors include artificially-induced manipulation of revenues and expenses between zero and ten percent of total assets to random samples of 500 firm-year observations within the full sample. The authors use two alternative samples, one with no reversal of manipulation (sample 1) and one with reversal in the following year (sample 2).Design/methodology/approachThe authors include artificially induced manipulation of revenues and expenses between zero and ten percent of total assets to random samples of 500 firm-year observations within the full sample.FindingsThe authors find that real earnings manipulation models have lower power than accrual earnings manipulation models, when manipulating discretionary expenses and revenues. Furthermore, the real earnings manipulation model to detect overproduction has high misspecification, resulting in artificially inflating the power of the model. The authors examine an alternative model to detect discretionary expense manipulation that generates higher power than the Roychowdhury (2006) model. Modified real manipulation models (Srivastava, 2019) are used as robustness and the authors find these to be more misspecified in some cases but less in others. The authors extend the analysis to a setting in which earnings management is known to occur, i.e. around benchmark-beating and find consistent evidence of accrual and some forms of real manipulation in this sample using all models examined.Research limitations/implicationsThis study contributes to the literature by providing evidence of misspecification of currently used models to detect real accounts manipulation.Practical implicationsBased on the findings, the authors recommend caution in interpreting any findings when using these models in future research.Originality/valueThe findings address the earnings management literature, guided by the agency theory.