United StatesPublic sector2017 – ongoingSecurities and Exchange Commission
Machine-learning prioritization of investment adviser examinations
SEC reported examination-prioritization models performed 5 times better than random in back-testing for potentially referable filing language
Problem
SEC examination teams had limited resources for identifying investment advisers that warranted closer scrutiny. Relevant signals were scattered across regulatory filings and examination histories.
Approach
DERA applied unsupervised learning to identify unusual reporting behavior in adviser filings. It combined those outputs with past examination outcomes in a supervised model that predicted adviser-specific risks. Expert staff reviewed the outputs when prioritizing examinations.
Results
Takeaway
The reported benchmark concerned back-testing for potentially referable language. A flagged filing could still have an innocent explanation.
Categories
- Industry
- Public sector
- Business function
- Finance, risk and legal
- Innovation type
- Process innovation
- AI technology
- Predictive analytics and ML
- Adoption
- Built in-house
- Evidence
- Company's own claim
Sources
Source summaries
Acting DERA director Scott Bauguess described models already influencing examination priorities. He reported a back-testing advantage over random selection and explained why experts reviewed false positives.
Full details are in the original sources.