Using an earnings management model in which managers manipulate information when the firm's control system fails, I introduce a measure of earnings quality, based on the notion of integral precision, that has solid theoretical foundations. A trade-off between the frequency and the magnitude of overstatements is shown: overstatements are larger when misreporting is less likely. Overall, the model generates a distribution of earnings announcements similar to its empirical analogue and provides a structural method to identify the likelihood and magnitude of misreporting by exploiting information from the moments of the distribution of reported earnings.
