HKUST Business Review

S&P Ratings Predict Fraud Earlier and More Accurately Specifically, we analysed 177 cases of accounting fraud identified through SEC Accounting and Auditing Enforcement Releases and settled securities class action lawsuits between 1999 and 2018, all of which were committed by companies that had been rated by both S&P and EJR. We started with a baseline fraud prediction model that incorporated commonly-used factors such as accrual-based F-scores (a standard financial metric), short interest, equity returns, analyst recommendations, and credit default swap spreads. Then, to quantify the value-add of CRA ratings, we incorporated ratings from both S&P and EJR into the model. And what did we find? Adding S&P’s negative rating actions to the baseline fraud detection model improved performance significantly, while adding EJR’s ratings had minimal impact. This is especially important because the baseline model was far from perfect: It correctly identified firms as fraudulent or non-fraudulent just 54% of the time, it had a false positive rate of 46% (that is, of the non-fraudulent firms, it incorrectly classified 46% as fraudulent), and it had a false negative rate of 38% (that is, of the fraudulent firms, it incorrectly classified 38% as non-fraudulent). Adding S&P’s negative rating actions increased correct classifications by 2.5% and reduced false positives and false negatives by 2.5% and 1.8% respectively, while incorporating EJR’s negative rating actions only increased correct classifications by 1.01% and reduced false positives by 1%, and it actually increased false negatives by 0.8%. We also found that S&P began to downgrade fraudulent firms much earlier than EJR did. S&P downgrades occurred up to a year before the fraud was revealed, while EJR did not downgrade until two quarters beforehand, and on average, S&P’s earliest negative rating actions preceded those of EJR by 83–121 days. Lastly, we also explored fraud prediction performance before and after the 2008 Financial Crisis, since the Dodd-Frank Act (passed in 2010 in response to the crisis) made it substantially easier both for the U.S. government to impose sanctions on CRAs and for investors to sue CRAs in response to misleading ratings, significantly increasing CRAs’ liability. Due to this increased scrutiny, S&P’s fraud detection effectiveness improved post-2008—but we saw no such improvement for EJR. Taken together, our findings suggest that despite the potential for conflicts of interest, S&P’s ability to leverage private information enables it to detect fraud earlier and more accurately than credit rating analysts who must rely more heavily on public data. Access to Private Data Boosts CRAs’ Fraud Detection Capabilities Diving deeper, our quantitative research coupled with several in-depth interviews highlighted a few key reasons for the improved fraud detection performance of CRAs like S&P. First, while the SEC’s Regulation Fair Disclosure rule prevents managers from selectively disclosing information to many intermediaries (such as financial analysts), CRAs are specifically exempt from this restriction. Next, CRAs that are paid by the firms they rate typically sign nondisclosure agreements with their customers, giving them access to important documents such as credit and acquisition agreements, private placement memoranda, budgets, forecasts, and detailed financial reports. These CRAs are also often privy to advance notice of significant events, and their analysts typically conduct in-depth site visits during initial ratings and annual reviews, in which they can observe company operations up close and meet face-to-face with top executives. Of course, fraudulent companies are unlikely to reveal evidence of their fraud directly, but these private meetings often allow analysts to pick up on vocal cues, facial expressions, and other forms of indirect communication that can facilitate fraud detection. Change in Fraud Detection Performance with S&P Versus EJR 41 HKUST Business Review