Preventing the Two-Minute Meltdown
August 3, 2009
This time around, it took financial markets about two years to melt down. The next time it could be two minutes.
That's because success in trading is now measured in how few feet your electrons travel. With high-frequency trading your algorithm has to beat he other guy's algorithm to the punch by microseconds.
This is what Ralph Frankel, the chief technology officer of Solace Systems, dubs "latency arbitrage,'' in a guest commentary on our site, www.securitiesindustry.com.
The arbitrageurs with the aid of their chief information officers and chief operating officers are working to wring every possible microsecond out of an action. The most obvious tack: Co-locating servers at the same place an exchange locates its servers. That saves the 25 microseconds it takes for your instructions to travel from Manhattan to, say, Weehawken, N.J.
You also have to have fast switches, which start reading and routing packets of data as soon as they arrive. And functions that used to be embodied in software are once again being embedded in hardware, to save even more microseconds.
Eventually, you'll have high-frequency traders fighting over how many feet away from an exchange's servers that their equipment sits in the "co-location" facility.
This capitalization on speed will never stop. As is explicated in a new section called Talent, the complicated art of parallel programming a decade ago revolved around the concept of coordinating the actions of the brains of 64 different computers in a data center. Now, the task is to train the brains of high schoolers to figure out how to best program 64 different central processing units-when they all sit on the same microchip ("Fast Times at Brooklyn Tech," p. 22).
While they learn, the potential catastrophe that will hang over our heads is the two-minute meltdown.
In a comment to the SEC on June 30 the Lime brokerage tried to calculate the potential damage of one instance of an electronic "fat finger" incident (See "NYSE Slaps Fat Fingers," page 8).
If a computerized trading order goes awry, 120,000 orders could go into the market before the problem gets recognized two minutes later. At an average of 1,000 shares an order and $20 a share, $2.4 billion of faulty orders could be executed.
That's a rather tame example, however. On February 24, UBS generated an unintended $31 billion order, 100,000 times more than it intended, Lime noted.
What if an algorithm gets stuck in an errant loop and takes all IBM, Google and, heaven forbid, Goldman Sachs shares off the market? Who's going to cover the capital that's been sucked out - or, at least, placed in limbo?
Then, there's the chain reaction that would lead to the next financial crisis. If America's new great high-frequency trading can't be harnessed effectively, the ripples will be felt in London, Paris, Shanghai and elsewhere.
Bearing the most watching will be the general, public investor.
Coming so soon after the mortgage- and derivatives-based meltdown, the general public might more wisely put into bank accounts whatever funds they have left in what they think are open, level playing fields.
You couldn't blame them. As Frankel puts it, the playing field is level. If you're willing to invest in the same technology.
