After Regulation NMS went info effect, the markets grew at a pace that no one could have predicted. Trading destinations were proliferating; fragmentation was rife. Algorithmic trading was considered too sophisticated and secretive to police. The chickens came home to roost after the May 6th 'flash crash', when algorithms essentially led the market on a wild roller-coaster ride that frightened investors and regulators alike.

The CFTC and the SEC scrambled to cobble together data to investigate the cause of the crash and realized how incredibly difficult that was. However, it should not have been. The CFTC even admitted that it still receives some data via fax machine. New CFTC Commissioner Scott O'Malia is very open about the shortcomings, and is convinced technology can be deployed intelligently to address future issues.

Using a system that can monitor full market depth and trades on all venues in real-time, get feeds from dark pools, and get OTC data electronically within a reasonable timeframe would have helped.

While such technology may not be able to predict a flash crash, it could certainly detect one early — due to volume and price behavior in specific symbols across venues. A crash could be stymied if a regulator could put all of the markets in "slow mode" and activate circuit breakers in specific symbols.

One of the problems exacerbating the crash was that certain markets behaved differently. It would be great if the regulator had a global circuit breaker for particular instruments when they detect concerning scenarios.

Using technology, a regulator would be able to see the radical price movements triggering stop losses, followed by mass shorting of instruments. This could appear on real-time dashboards, or heat maps, showing the "hotness" of particular symbols or potential worrying patterns.

Of course, there is always the potential for a 'black swan' event, but in most cases patterns repeat themselves. If you know how to respond to a certain crisis then you can encode automatic response processes or the system can seek human approval to take an automated action. Once the flash crash is understood in greater detail, scenarios can be encoded so that the detection, alerting and response to such a scenario can be automated. A surveillance system needs to 'learn' all the time — just like a human.



Previous 1 2 3 Next