The ability of earnings management models to detect and predict public discovery of accounting-fraud.

摘要

The purpose of this study is to analyze and compare: (1) the ability of competing aggregate accrual and frequency distribution models to detect extreme earnings management, i.e. accounting-fraud, and (2) the ability of a composite model to predict accounting-fraud using only prior period information. Studies have used various models to detect earnings management in circumstances in which, a priori, some management is likely to exist. Events with incentives to manage earnings analyzed include issuing securities, maintaining positive earnings or an upward earnings trend, increasing an earnings-based bonus, increasing subsidies during import relief investigations, or decreasing penalties during antitrust investigations. However, few studies have tested such models when there existed a virtual certainty about which firms managed earnings. Using the Securities and Exchange Commission's (SEC) Accounting and Auditing Enforcement Releases (AAERs) to denote accounting-fraud firms, I establish a format of analysis in which relative certainty exists.; Using that format, I test various aggregate accrual, frequency distribution, and composite earnings management models' ability to distinguish between accounting-fraud and non-accounting-fraud matched firms. Aggregate accrual model results show that total accruals, the simplest model, performs best in detecting accounting-fraud. Also, those models calculated from the statement of cash flows always outperforms those calculated from the balance sheet. Frequency distribution models show a surprising lack of ability to detect accounting-fraud. The power of the test is adversely affected by an apparent targeting bias for the SEC to investigate firms that miss earnings thresholds.; As expected, the data intensive composite model shows the greatest ability to identify accounting-fraud firms from ex ante data. The composite model only uses prior period variables to represent financial condition of the firm, income-increasing accounting choices, and potentially opportunistic behavior to distinguish an accounting-fraud firm-year from a matched non-fraud firm-year. Significant variables include total accruals, sales growth, cash sales growth, a proxy for age of firm, inventory valuation method, straight-line depreciation, and merger/acquisition activity.; Overall, aggregate accrual models calculated from the balance sheet and frequency distribution models appear to have minimal ability to detect extreme earnings management. Aggregate accrual models calculated from the statement of cash flows appear to be more useful to distinguish accounting-fraud firms, although they exhibit relatively low explanatory power. Composite model results represent a particularly useful contribution since only prior period information is used to predict future accounting-fraud firms. Additionally, the significance of certain variables representing managerial behavior and incentives provide strong insight for accountants and regulators concerning the prediction/detection of accounting-fraud.