Stock Market Consists of Computers Playing with Self

Anything that is not directly related to the game or its community.

Stock Market Consists of Computers Playing with Self

Postby elryacko » Mon Aug 03, 2009 6:00 am

;)

http://arstechnica.com/tech-policy/news/2009/08/nyse-builds-computer-trading-mothership-worries-abound.ars



With all of the scrutiny that high-frequency trading is now under in the media and in Congress, the New York Stock Exchange is probably none too thrilled that the Wall Street Journal has uncovered fresh details of NYSE's giant new datacenter, which the exchange is building in a former New Jersey quarry. The new datacenter will significantly advance the amount of computer-automated trading that already dominates global markets, housing as it will "several football fields of cutting-edge computing equipment for hedge funds and other firms that engage in high-frequency trading," according to the WSJ. So if you were recently shocked to learn that an estimated 70 percent of stock trading is just computers trading against one another, get ready for that number to go even higher.

The NYSE is reportedly already taking orders from firms that want to lease space in the datacenter so that they can co-locate their servers with those that run the exchange in order to execute trades more quickly. Actually building and running a datacenter of this size is new for the NYSE, which historically has rented space in others' datacenters. As for the impact this will have on the markets, the debate rages.

Not everyone is happy that the NYSE is poised to massively boost the already overwhelming amount of computer trading. At issue is not the simple fact of computers trading against one another over electronic networks—it's the speed with which they appear to be squeezing the humans out of the loop, and the potential instability and fragility that may result from increased automation of global markets.
Yes, there is actually something new here

For every technique or technology that comes under the heading of HFT, you can dig up an example of how people did this same thing on a much smaller scale without computers. Therefore, the argument goes, the relatively recent (see below) use of computers to do two orders of magnitude more of these activities in a given timeslice is "nothing new," despite the fact that the computers are now doing this among themselves without human intervention. This kind of reasoning keeps cropping up all over the Internet in the current HFT debates, in some instances coming from very smart people. (Tyler Cowen's response at Margina Revolution is the best statement of this view.)

Others, however, are concerned by HFT's rise to dominance. Quant guru Paul Wilmott, in a widely cited NYT op-ed this last week, says that it's exactly the potential systemic impact of the recent acceleration of the speed and volume of HFT activity that has him worried.

"There's nothing new in using all publicly available information to help you trade," Wilmott acknowledges, in an initial nod to the naysayers. But he continues: "What's novel is the quantity of data available, the lightning speed at which it is analyzed, and the short time that positions are held... The problem with the sudden popularity of high-frequency trading is that it may increasingly destabilize the market."

Michael Durbin, the man behind Citadel's high-frequency trading desk, echoed this warning to Reuters' Matthew Goldstein earlier this week:

"You have multiple HFT trading firms and sometimes their agendas are complementary and sometimes they're not," explains Durbin, director of HFT research with Blue Capital Group, a small Chicago-based options trading firm.

"There could be a time where these HFT programs unintentionally collaborate and you have a two- or three-minute period where the markets are going crazy. Then other traders respond to it and it simply gets out of control."

My own take on the question was summed up in a recent note to Felix Salmon:

It's quite remarkable to me that many of the econ and finance folks who insist that "HFT is the same thing we always did, just way faster" don't seem to realize that frequency and amplitude matter a whole lot, and that for any given phenomenon when you suddenly increase those two factors by an order of magnitude you typically end up with something very different than what you started with. This is true for isolated phenomena, and it's doubly true for complex systems, where you have to deal with systemic effects like feedback loops and synchronization/resonance.

What I've noticed anecdotally is that engineers and IT pros are more concerned about HFT than people who just handle money for a living. These guys have a keen sense for just how fragile and unpredictable these systems-of-systems are even under the best of conditions, and how when things go wrong they do so spectacularly and at very inconvenient moments (they get paid a lot of money to rush into the office to put out fires at 4am).

There's an analogy here with e-voting, which I did quite a bit of work on. In the e-voting fiasco, you had people who were specialists in elections but who had little IT experience greenlighting what they thought was an elections systems rollout, but in actuality they had signed on for a large IT deployment and they had no idea what they were getting into. To them, it was just voting, but with computers, y'know? They found out the hard way that networked computer systems are a force multiplier not just for human capabilities, but for human limitations, as well.

In sum, the growing concern with HFT is not that computers are doing something that people used to do; it's that they're doing a whole lot of it, very rapidly, so that the market as a system-of-systems now has starkly different frequency, amplitude, and connectedness characteristics that may (or may not) give it a new, currently unknown set of emergent properties. And if a fuse blows while the machines are driving the bus, we're all traveling so fast that we may hit a wall before the humans in the vehicle have time to react.
Reaction time, or what could possibly go wrong

If something does go wrong, the market moves so quickly now that by the time the humans intervene, the damage could already be too great to keep contagion and feedback effects from kicking in and pushing things further south. This is possible because HFT, by design, is capable of moving the markets much more rapidly than humans can possibly react to those moves. Again, speed matters, especially in activities where human judgment plays a critical role.

An example of what could go wrong happened on September 8, 2008, when a six-year-old news story about United Airlines' bankruptcy was somehow republished on Google News with a current timestamp. A source tells me that the algo traders kicked in at some point and started dumping United's stock, which eventually lost 75 percent of its value that day. About $1 billion of United's market cap evaporated in 12 minutes, before the humans figured out that they were looking not at a bankruptcy, but at the type of GIGO problem (garbage in, garbage out) that's well known in computer science circles. Much of the stock's value was restored by the next day, when the market had sorted itself out, but experts worry that something like this could happen on a much larger scale, especially in response to a genuine external shock.

A paper on the dangers of HFT by Lime Brokerage (uncovered by Zero Hedge) paints the following picture of just how rapidly things might go awry in the present environment:

Lime's familiarity with high speed trading allows us to benchmark some of the fastest computer traders on the planet, and we have seen [computerized day trading (CDT)] order placement rates easily exceed 1,000 orders per second. Should a CDT algorithm go awry, where a large amount of orders are placed erroneously or where the orders should not have passed order validation, the Sponsor will incur a substantial time-lag in addressing the issue.

From the moment the Sponsor's representative detects the problem until the time the problematic orders can be addressed by the Sponsor, at least two mintues will have passed. The Sponsor's only tools to control Sponsored Access flow are to log into the Trading Center's website (if available), place a phone call to the Trading Center, or call the Sponsee to disable trading and cancel these erroneous orders—all sub-optimal processes which require human intervention. With a two minute delay to cancel these erroneous orders, 120,000 orders could have gone into the market and been executed, even though an order validation problem was detected previously. At 1,000 shares per order and an average price of $20 per share, $2.4 billion of improper trades could be executed in this short timeframe. (emphasis added)

A final point: when there is a chorus of Wall Streeters and economists telling us in regard to some new, very profitable milestone in complexity and innovation, "don't worry, and the odds of anything going wrong here are passing slim," does this really reassure anyone in 2009? I think the burden of proof should be on people like Cowen to demonstrate that long-term "buy and hold" can peacefully coexist with the degree of opaque, computer-driven hyperspeculation the markets are barreling towards.
Addendum: the landscape has changed just this year

One of the things that perhaps wasn't clear enough from my previous article on high-frequency trading (HFT) is that, while people have been using computers to trade against one another for a decade or more, the bigger HFT picture has evolved rapidly in just the past few months.

This recent HedgeWeek article tells the typical story that you'll find on various trading blogs, and it goes something like the following:

Algo trading trading was on the rise going into the summer of 2007, when the correlations that the algo traders depend on suddenly began to break down. The August 2007 "quant quake" was a shakeout that had a catastrophic impact on number of the larger funds, and statistical arbitrage strategies continued to perform very poorly throughout 2008. As a result of this underperformance, stat arb was generally out of favor throughout 2008, until the strategy started working again earlier this year. In the past few months, algo trading has enjoyed a resurgence in popularity, with current market conditions apparently very favorable to stat arbs in particular.

But the fact that, throughout the first half of this year, everyone has been piling back onto the algorithmic trading bandwagon willy-nilly isn't the only major development in HFT. It's also the case that the 2008 crisis had the effect of radically shrinking the competitive landscape, leaving the field to fewer out-sized players whose HFT efforts are being scaled up; Lehman Bros., Bear Stearns, and Merrill Lynch, all had announced major HFT plans for 2008, and in 2009 these banks no longer exist. Then there's the flash order issue, which flared up this year and is right now creating a stir among the exchanges and lawmakers.

All of this together adds up to a significant shift in the high-frequency trading terrain just this year. And this is happening to markets that are still coping with the relatively recent rise of electronic communication networks (ECNs), which are largely (in terms of trading volume) a post-2000 phenomenon.

If we take a step back and look solely at the percentage of traditional "open outcry" trades (i.e., a crowd of traders, hollering and gesticulating at one another) vs. electronic trades on the major exchanges, we can get a sense for just how fast the market has changed in the last decade. In 2001, some 90 percent of trading volume on the NYSE and the Chicago Mercantile Exchange (CME) was carried out by humans on the trading floor. In 2009, the open outcry volume is under 10 percent on both exchanges.

Stock and commodity exchanges were based on the open outcry system for over 100 years, but in just nine years the ratio of open outcry to electronic trading on these two major exchanges swapped from about 90/10 to about 10/90. That's a lot of computerization in a little time; and as the NYSE datacenter news shows, the pace of change is accelerating. The markets are different now, and they're getting more different more rapidly. The post-crisis Wall Street of 2009 is, in effect, in the process of doubling down on the preceding 9 years of speed, automation, consolidation, and leverage.
Image

Don't Panic
User avatar
elryacko
 
Posts: 507
Joined: Tue Apr 07, 2009 6:40 am

Return to Off-topic

Who is online

Users browsing this forum: No registered users and 9 guests