In Wall Street 2.0, Computers Are King
SCOTT SIMON, HOST:
For most of the 20th century, trading stocks was a loud and pushy business. Today there's more hum than hubbub on the floor of the New York Stock Exchange. Computers print millions of shares of stock every second, moving wealth around the world. Earlier this month, something called an algorithm - we'll get to that in moment - by a company named Knight Capital, traded away $440 million in less than an hour, before human managers noticed and stopped the flow of money.
An algorithm still engineer most trades right now. Sean Gourley is a physicist and founder of the website quid.com. He joins us from KQED in San Francisco. Mr. Gourley, thanks very much for being with us.
SEAN GOURLEY: Thank you for having me.
SIMON: What is an algorithm?
GOURLEY: I think the best way to describe it is a series of instructions, you know, like a cake recipe that says, put cake into the oven, leave for 37 minutes, if brown, remove from oven and eat cake. So it's a series of instructions and those instructions have different pathways depending on the outcome of what we see in front of us.
SIMON: How could the algorithms at Knight Capital lose so much money in such a short time?
GOURLEY: I mean, what we've got here is an algorithm and the algorithm was doing something that was kind of the exact opposite of what you'd want something to do. It was buying high and selling low. And it was doing that hundreds of times every second and was losing literally $10 million a minute. And it took them about 30-odd minutes to track it down. And in that time they'd lost $400-odd million.
I mean, what they did, from the looks of things, is they released an algorithm out into the wild or onto the real world that was never supposed to be there, and it never had the right checks and balances in place that would shut it off once it started losing tens of millions of dollars.
SIMON: So an algorithm of the kind we're talking about here would instruct a computer to buy or sell stock when a certain set of circumstances occur.
GOURLEY: Right. To take a step back then, we can think of the price of Facebook, right? Google also has a stock price that at the end of every day has a dollar value attached to it. Now, if at the end of every day, they both seem to be higher or at the end every day they both seem to be lower, it can be said that these things are moving together.
And if they look like they diverge or it looks like, you know, they're different from each other, you know, the algorithm will say if stock prices between stock A and stock B are divergent, make a trade in the expectation that they'll come back together. So we've got a...
SIMON: But they not be moving together. I mean, they may be moving because of entirely different sets of circumstances.
GOURLEY: Right. And that kind of touches the heart of this. These algorithms are not looking for the change of the CEO or the fact that the new vice president of engineering is not a very good engineering hire. They are just simply looking at the price movements. So in that sense, they're quite naive. They're quite stupid in some ways.
SIMON: Is it dangerous?
GOURLEY: I mean, it's dangerous in...
SIMON: Or is it dangerous not to do it? I assume that's what the companies have decided.
GOURLEY: Well, one of the reasons that they're taking over is that the human mind operates at a certain speed and we can actually test that They stick you in an MRI machine. They project an image of a chess board and they measure the response time it takes you to figure out whether or not your king is in checkmate. And what they find is that the human mind can make that decision in about 650 milliseconds or about .65 of the second.
Algorithms can trade on the order of microseconds or millionths of a second so there's this whole world that emerges where human can't make decisions because the biological limitations of our brain don't allow us to think fast enough. Now, what that means is the algorithms can come in and say, hey, you know what, I can make decisions about whether or not to buy or sell a stock before the human mind can even think.
SIMON: So let's say - I'll play the human being for a moment. So I see a certain set of circumstances that suggest to me from my so-called wisdom and experience that we ought to make a trade. I see something and I go, look at that. Let's press this button. Let's make a trade. Whereas an algorithm goes (snaps fingers) like that.
GOURLEY: Even quicker than that. You wouldn't even hear the click of your fingers. You know, it happens literally in millionths of a second. And millionths of a second is something we can't even comprehend. To put that into perspective, Usain Bolt, when he wins the hundred meters, he reacts to the gun in .16 of a second. Even Usain Bolt is going to be beaten by an algorithm.
SIMON: Yeah, and he doesn't even have the best start, does he?
GOURLEY: Right. He doesn't have the best start. But he doesn't even have to make a decision. All he has to do is run. For you to actually process that information that, you know, Facebook has hired a bad engineer and then make a trade on it, you're probably going to be in the five to ten second range. But the algorithm can read that news article and make the trade before you've even comprehended.
SIMON: I mean this question with utter sincerity. Are these machines moving so fast we don't understand them, much less understand them well enough to legislate?
GOURLEY: I think that is probably a fair place to be. And it's not just the individual algorithms. It's when the algorithms interact with each other. And you can kind of think of it like an ecosystem. It's an ecosystem that we don't fully understand and we're trying to regulate to kind of inform the behavior and yet we don't fully know the ramifications of what's going to happen.
SIMON: Sean Gourley is a physicist and the founder of the website quid.com. He joins us from KQED in San Francisco. Mr. Gourley, thanks for one of the most chilling conversations I've had in a while.
NPR transcripts are created on a rush deadline by Verb8tm, Inc., an NPR contractor, and produced using a proprietary transcription process developed with NPR. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.