Improving and measuring the performance of a securities industry surveillance system

Recent disclosures of illegal trading practices in the securities industry have prompted the development of improved computer detection models. The authors designed for the NASD a surveillance system that employs statistical analysis to uncover unusual patterns of activity to better identify exceptions that should prompt investigation. Inferences concerning the likelihood that trading violations have occurred are made by examining deviations from historical trading patterns on a security-specific basis. Expectations are adjusted to reflect general market movement influences and to recognize the relationship between the timing of changes in security trading patterns and the dissemination of news related to the specific security. The initial design employs a set of indicators that were found to best identify exception conditions during preliminary testing. In addition, the system will accept suggested candidate indicators.