Do models of discretionary accruals detect actual cases of fraudulent and restated earnings? An empirical evaluation

We examine the association between the existence and the magnitude of a fraudulent event, non- fraudulent restatements of financial statements, and nine competing models of discretionary accruals, accrual estimation errors (Dechow and Dichev 2002 and McNichols 2002), and the Beneish (1999) M-score. We use the size of the downward earnings restatement following the discovery of the fraud to proxy for the degree of discretion exercised to perpetrate the fraud. We find that while total accruals are associated with the existence of fraud, discretionary accruals derived from the Jones model, the modified Jones model, and performance-matched models are not associated with fraud. Accrual estimation errors and M-score have explanatory power for fraud beyond total accruals. We also find that commonly used measures of discretionary accruals, accrual estimation errors, and the M-score are associated with the magnitude of the fraud. Only the accrual estimation errors are associated with non-fraud restatements.