Can we detect fraudulent firms by automatically screening their financial reports? “It already works pretty well”, said Craig Lewis when visiting Frankfurt School this week. Lewis is the Madison S. Wiggington Professor of Finance at Owen Graduate School of Management at Vanderbilt University and the former Chief Economist and Director of the Division of Economic and Risk Analysis at the U.S. Securities and Exchange Commission. During his time at the SEC, Lewis was instrumental in developing what is called the Accounting Quality Model (AQM), or “Robocop”, as the financial press tends to call it. The system automatically creates a risk score for all registrants within 24 hours after their electronic filings. A higher risk score makes the enforcement staff aware of the fact that a filer might be worth looking at in closer detail. Thereby, the system makes the SEC’s inspections more efficient and effective.
How does the system function? The interesting fact from an academic perspective is that it strongly relies on models which have originally been proposed in the accounting literature on earnings management. For instance, the model estimates the magnitude of “discretionary accruals”. Accruals in general are neither good nor bad. They emerge because accounting rules generate a time series of net income figure which is, by construction, smoother than the underlying stream of cash flows. A simple example for this property to emerge is an investment of 1.000.000 in year 1. This investment negatively affects cash flow in year 1, but accountants will disregard this effect by recognizing the investment as a balance sheet item, depreciated, say, straight-line over 10 years, leading to smoothly expensing 100.000 per year. The idea behind measuring “discretionary” accruals is the following: Accruals emerge in the normal course of business. This particularly holds for working capital accruals, which include positions such as inventories, accounts receivable or accounts payable. If sales increase, we would naturally expect that this will have an impact on working capital accruals: The credit sale part will increase receivables, but there is also (more) inventory leaving stock. Identifying factors which explain these “normal” changes in accruals also allow us thinking about identifying a part of changes in accruals that is unexpected and thus likely to be explained by earnings management activities. Essentially, this is what Robocop does: It uses a sophisticated and calibrated earnings management model to identify firms with more aggressive accrual accounting.
The Robocop system likely will have some additional features in the future. Lewis’ own research contributes to this development. One of his recent working paper uses textual analysis to identify specific language patterns fraudsters tend to use in the MD&A (the management discussion and analysis) section of their financial reports. Findings include that fraudsters tend to under-disclose details regarding why the fraud firm achieved a certain performance level; also managers of fraud firms seem to disassociate themselves from some disclosures.
European enforcers so far do not seem to pursue own attempts to install a similar system. But that’s probably just one scandal to go…
Prof. Dr. Jörg Werner is the Academic Director for Frankfurt School’s newest Master of Finance concentration, Financial Accounting & Advisory.