High-Frequency Trading and the Robot-Led Holocaust

Friday, July 31, 2009
By dreeves

Imagine evil glowing red eyes, and also that it's engaged in high-frequency stock market trading.

Michael Wellman launched a blog this week. Yesterday he made an excellent point about market efficiency. Tyler Cowen had pooh-poohed the problem of high-frequency trading. Mike set him straight and suggested an elegant solution to the problem, namely call markets. But what about the value of real-time prices? No problem, let the market clear, say, every second. After all, getting price information at subsecond granularity is of truly negligible value from a social perspective. And with a call market the incentives for traders are entirely changed. All the technology and brainpower devoted to high-frequency trading can be reallocated to more socially efficient uses.

Today Mike injected a dose of sanity into a different debate: the robot-led holocaust. Lance Fortnow also chimed in today on this topic, going a bit too far, in my opinion. Lance’s claim that humans are more than Turing machines, that we have “a self-awareness and reasoning beyond the capability of machines,” is frankly as hard for me to get my head around as I’m sure the opposite claim is for many layfolk. Not that I think the AI Singularity is imminent. In fact, a year ago I wagered a friend that no computer would pass the Turing Test (the first milestone Lance mentions) within ten years. I risked ten thousand dollars at 100:1 odds and am as convinced as ever that it’s tantamount to free money. (We piggybacked on the inaugural wager at longbets.org for the terms.) Nonetheless, thinking a little further into the future, I don’t think it’s an exaggeration to say that these questions are critical for the future of humanity.

Addendum (Low-Frequency Updates on High-Frequency Trading (HFT))

Tyler Cowen seems to still not get it.  MarginalRevolution commenter, “a student of economics” (who is that?), sets him straight eloquently:

1. There’s no real contradiction between asserting that at best, HFT creates essentially zero benefit (because the value of getting prices right a millisecond earlier is approximately zero) and that at worst, it actually increases volatility.

2. (a) “Even if the practice has no gain, I suspect the loss is small.” That depends on what you mean by small. If the practice has no gain, the social loss is easy to estimate: it’s the opportunity cost of the resources used. What’s the value of all those brilliant minds, computer capital, and other inputs? Probably in the billions. (b) “On an actual list of bad policies or practices in our world, would it be in the top million?” Well, you’re the one who’s blogging about it repeatedly. Is billions in social loss more important than the cost of a zoo killing a gorilla? Probably.

3. “There is no argument to date, and probably no argument period, that HFT can lead to financial insolvency or collapse on a major scale.” That’s setting the bar pretty low…

5. “We should still follow the rule of regulating practices shown to be harmful or likely to be harmful.” Actually, Michael Wellman’s proposal in the comments and on his blog seem like the best approach. Financial markets don’t just spring from the forehead of Adam Smith, they are consciously DESIGNED by humans seeking to achieve specific goals. If a small change to the mechanism for matching buyers and sellers drastically reduces rent dissipation, then that design should be implemented.

Specifically, Wellman’s suggestion of clearing markets once per second (or some similar discrete interval) would seem to do the trick by eliminating most of the incentive for microsecond improvements in trading algorithms, software and hardware. As long as we’re going to design mechanisms for trading, we may as well design and implement efficient ones.

Michael Wellman and Kevin Lochner have also followed up with new posts on this:  Wellman on the technology behind HFT and Lochner on Call Markets.

Tags: , , , , , ,

  • Munmun

    Interesting. But reaching the “AI Singularity” appears far from now. We can’t even solve simple block problems!

  • http://hunch.net/~jl John Langford

    I’d like to point out this earlier discussion on hunch.net, which is relevant. At it’s essence, the claim is that AI is extremely beneficial for interstellar travel, and interstellar travel seems critical for the longer term survival of civilization.

  • Vishal

    This is how Nasdaq describes the Flash order process around which much of the concern with HFT seems to be centered:
    It’s a two-page white paper and, to be honest, only mildly enlightening.

  • http://mblog.lib.umich.edu/strategic/ Michael Wellman

    Thanks for the plug, Dan. I was pleased to see that Felix Salmon picked up on my proposal for one-second call markets. It will be pretty hard to get there (discrete-time trading) from here (continuous-time trading), but I’m convinced it is worth hashing out the proposal.

    Besides, maybe working on designs for countering the trading bots will give us good practice for fending off the robots when the singularity hits…

  • michael vassar

    Chimps also can’t solve simple block problems, yet they are a trivial evolutionary distance from “Human singularity”.