Thursday, August 2, 2012

Knight trading debacle, redux

Everybody seems to have an idea about how to fix whatever it is that went wrong with Knight Capital's new software this week.

Says the New York Times: Errant Trades Reveal a Risk Few Expected:

The fiasco, the third stock trading debacle in the last five months, revived calls for bolder changes to a computer-driven market that has been hobbled by its own complexity and speed. Among the proposals that gained momentum were stringent testing of computer trading programs and a transaction tax that could reduce trading.

Really, though, I have no idea how the NYT could claim that these were risks that "few expected". Anybody who thought about this for even a few seconds would understand that these risks were present.

And, I really don't understand how they expect to accomplish this:

Some S.E.C. officials are pushing new measures that would force firms to fully test coding changes before their public debut, according to a government official who spoke on the condition of anonymity.

What does that even mean, "fully test coding changes". It just exhibits a complete absence of any familiarity with the world of complex system software.

To its credit, the article does offer a bit of a constructive suggestion later:

“When they put these things out in the world they are really being tried for the first time in a real-life test,” said David Leinweber, the head of the Center for Innovative Financial Technology at the Lawrence Berkeley National Laboratory. “For other complex systems we do offline simulation testing.”

I wonder what these "other complex systems" are. For example, is this the sort of complex systems testing that would help figure out why half of India lost electrical power this week? If ever there was a need for simulation testing to assess the robustness of a complex system, power transfer infrastructure certainly jumps to mind...

This is actually the first time I've ever heard of "the Center for Innovative Financial Technology at the Lawrence Berkeley National Laboratory". So I went to look it up, and it appears that here is the home page.

Frankly, it looks like one guy, who's written a book, and writes a blog called Nerds on Wall Street (which I'm now subscribed to), and has access to some government-funded supercomputer resources. Here's a paper he's written describing how he thinks supercomputer simulations can help stabilize the financial markets: Federal Market Information Technology in the Post Flash Crash Era: Roles for Supercomputing.

It's all interesting stuff, but it's hard to buy the argument that requiring algorithmic trading software to be certified by a government agency using supercomputer simulations is really the right answer.

But, what is the right answer?

No comments:

Post a Comment