High-Frequency Trading: A Grave Threat to the Markets and the Economy

Print Email

-By Eric Philo

High-frequency trading (HFT) is the extremely rapid trading of securities through the use of sophisticated hardware and software, enhanced by positioning this hardware in very close physical proximity to exchange servers for minimum latency.  HFT participants are the investment banks and market participants.  Trading generally occurs at millisecond (thousandth of second) and microsecond (millionth of a second) intervals.  Perhaps 60%-70% of all equity volume in the United States is HFT.

If you want to play the HFT game, you’d better find a way to get close to the action, and I mean very close – housing your equipment in the same facility as the exchanges, with direct, raw feeds, hopefully closer to exchange servers than anyone else.  This is meant to reduce latency to its lowest possible level.  To illustrate the importance of minimizing latency, and underscore the size of this business, Hibernia Atlantic, which provides transatlantic connectivity to a variety of customers, is laying a cable that will beat existing routes, offering 60 millisecond latency.  By comparison, it takes the human eye 350 milliseconds to blink.  This cable will be part of Hibernia’s Global Financial Network, essentially dedicated to HFT.

HFT provides liquidity, rapid price discovery (to put it mildly), and, via algorithms (“algos”) that instantly ingest stock technicals and other data and react immediately,  blindingly fast price adjustment.  One trader we spoke with said that HFT has created a “binary world” for human traders – either you’re invested prior to an event, or you have no opportunity to trade on it.  Stock movements in reaction to news or technical events (e.g. breaching a moving average) that might have taken minutes, hours or even days to unfold take place so quickly that “human trading” has become a very tough way to make a living.  This is arguably neither a good thing nor bad.  However, HFT brings with it a multitude of problems, and someday may help produce an enormous, confidence-shaking market crash.  The fact that HFT drives trading not just in stocks but in other markets such as commodities and currencies, often in linked fashion, means that a calamitous drop in one market could easily and quickly affect others.  HFT may be a ticking time bomb.

A Litany of Warnings

HFT and related algo trading have already exhibited disturbing instability.  The most infamous display of this instability was the “Flash Crash” of May 6, 2010 (Figure 1).  On that day, in the afternoon session, an algorithmic trade was place that sold E-Mini S&P 500 futures contracts as a hedge against long positions by a traditional fund manager (believed to be Waddell & Reed).   This particular trade targeted volume, ignoring price and time.  HFTs dominated the other side of the trade (that is, they bought what the institution was selling), and quickly accumulated the seller’s inventory.  Just as quickly, they dumped it, and a liquidity vacuum ensued.  All of this happened over a period of a few minutes, against an already negative backdrop in market fundamentals (driven by bad news on the European debt crisis that day).  This negative backdrop had already driven down buy-side liquidity for the E-Mini as well as the S&P 500 SPDR exchange traded fund (“SPY”), the two most active stock index instruments traded in electronic futures and equity markets.

Figure 1

 Image 1

The decline in these instruments triggered sell orders in individual stocks, generating transactions at irrational prices, from a penny a share to over $100,000 a share.  Accenture and 3M for examples, both traded at pennies.  The Rydex S&P 500 equal weight ETF went to zero.

At its low, the Dow was down 9.2% from the open – a drop of nearly a thousand points.

On March 2012, BATS withdrew its IPO in March 2012, following a software glitch that also caused a 9% decline in the price of Apple shares and paused the buying and selling of Apple for five minutes.  (By the way, BATS stands for “Better Alternative Trading System”.)

Facebook’s May 2012 IPO was marred by a computer “glitch” in NASDAQ’s pre-IPO auction process, resulting from a “race condition” whereby conditions triggered an infinite loop in the order matching software.  Obviously, NASDAQ’s testing was not thorough enough to catch the error prior to the IPO.  This software error prevented some investors from knowing whether they’d successfully bought or sold shares, leading to multiple order entries in some cases.  Swiss bank UBS is said to have suffered a loss of more than $350 million.

On August 1, 2012, Knight Capital Group, a market maker and provider of electronic execution services primarily as an equity wholesaler (executing orders for retail brokers off-exchange), lost nearly half a billion dollars due to a software glitch that created, in Knight’s words, “an erroneous trade position”.  Knight narrowly avoided extinction thanks to a capital infusion from a handful of financial firms that now control the company.  By definition, inadequate software testing is to blame.

Less spectacular “anomalies” take place all the time.   In early 2012, an interesting article was published entitled “Financial Black Swans Driven By Ultrafast Machine Ecology”[1].  The authors found analyzed 18,520 “events” – inexplicable, ultra-brief (650-950 millisecond) wide stock price swings (they termed them “black swans”) – from 2006-2011 – more than one per trading day, on average.  They note, somewhat troublingly, that “the ten stock with highest incidences of ultrafast [fluctuations] are all financial institutions…”

Not long ago, Nanex[2] observed that “On April 24th, 2012 at 15:51:44, the number of quotes for a single second in one stock set a new record: 47,138. The stock, PSS World Medical (symbol PSSI), is not active: only 1,992 trades (317,127 shares) traded the entire [prior] day…if HFT blasted just 10 stocks from one exchange at this rate, our national quotation system would collapse.”  According to Nanex, NASDAQ computer code that handles PSSI handles all shares with quotes between PC and SPZZZZ. That means all other symbols in that range were also affected by what happened in PSSI, including symbols like QCOM, QQQ and SIRI.  According to Nanex analysis, year to date through October 31st, there have been over 12,000 micro flash crash events, where very large individual stock prices fluctuated for brief moments.

On October 9, 2012 between 10:02:20 and 11:01:15, a large number of offers were made $0.444 off the then-prevailing market for 253 stocks, creating approximately 449 bad prints.  Retail orders taking advantage of this “fire sale” were canceled.  Nanex reasonably believes that an algorithm applied a $0.44 price improvement rather than an intended $.004.  (By the way, why were the orders canceled?)

 

Outright Manipulation

Of course, outright manipulation occurs.  The following is an example of “spoofing”: stuffing quotes into the order queue, intending to mislead other investors as to the BBO (best bid/offer).  These orders are entered and instantly cancelled at lightning speed.  First, a bona fide sell order is placed, putting the trader in a short position.  The trader simultaneously places non-bona fide buy orders intended to increase the market bid, and essentially induce the market to buy the stock offered for sale by the trader.  The non-bona fide buy orders are withdrawn, the bid side of the market declines, and the trader covers the short at a lower price, pocketing the difference.  It takes well less than a second for the entire transaction. This outright market manipulation was the subject of an SEC cease-and-desist order against Hold Brothers, in reference to trades done from January 2009 through September 2010.  Hold Brothers was fined $5.9 million.  The SEC action took place two years after the fact, on September 25, 2012.

What’s interesting is how long it took for the action to take place.  How far behind the curve is the SEC?  Do they have the resources to keep up with developments?  How many “Hold Brothers” are/have been out there?

The most disturbing aspect of the quote-stuffing outlined above is its potential to cause a major dislocation in the markets in the context of a fundamentally-driven major market selloff.  Millions of “flash orders” that come and go without execution create an enormous burden on the processing power of the exchanges.  In the presence of a large selloff, there is no telling what the impact on liquidity and price action such a potentially incendiary combination would create.  Refer back to our description of the cause of the Flash Crash of May 6, 2010.  HFTs stepped away from the buy side quickly, and the ensuing vacuum in liquidity caused wild gyrations in stock prices and of course the overall huge market decline.  (Again, by the way, we know of instances where orders placed to buy/sell stocks at the extremely attractive prices created by the flash crash were cancelled.  Why?)

The exchanges themselves have been fined for favoring HFTs over other market participants.  On September 14, 2012, the NYSE and its parent NYSE Euronext agreed to settle with the SEC and pay a $5 million penalty for providing unfair access to raw datafeeds, allowing certain HFTs with such datafeed access to get quotes ahead of the consolidated feed which the rest of the investing public relies on.  Supplying this information ahead of the consolidated feed allowed the HFTs closest to exchange servers to essentially front-run other investors.  Figure 2 offers an illustration of how this front-running can work.

Figure 2

Image 2

Plenty of Opportunity For Market Abuse

In 2011, the UK Government Office for Science[3] noted the following possible areas for market abuse in their report, An Ecological Perspective on the Future of Computer Trading[4].  The following is taken directly from the report.

Conglomerate effects:One obvious example is that a prop [proprietary] trading desk will be charged exchange fees that benefit from a volume rebate that is calculated at the level of the entire bank so that prop trading benefits from customer order flow generated at the brokerage unitIndeed, it is also well known that execution algos that are provided by brokerage desks to customers are often the old generation of algos initially developed by prop trading desks and then abandoned as improved versions were developed. Inevitably, prop desks at investment banks will have a good understanding of how client flow is being executed and of any deficiencies in these algos that might be exploited. Finally, consider that the huge investment necessary to provide brokerage services by investment banks can usually be leveraged by the prop trading desks of the same banks to create algorithmic trading (AT) operations at very little additional cost, giving them huge advantages relative to independent prop trading operations.

Vertical Integration: When the markets are owned and controlled by the same interests that drive the vast majority of trading, anticompetitive pressures will be at work. For example, Chi-X, which has the largest trading volume among all European exchanges, was established in 2007 by Instinet which is a subsidiary of Nomura Holdings and is now owned by a private consortium also including BNP Paribas, Citadel, Citigroup, Credit  Suisse, Fortis, GETCO Europe Ltd, Goldman Sachs, Merrill Lynch, Morgan Stanley, Optiver, Société Générale and UBS. Menkveld (2011) suggests a symbiotic relationship between Chi-X and a single large HFT market maker which we strongly suspect is also one of the owners. While Chi-X has effectively exerted competitive pressure on incumbent exchanges, we are not sure its business model would have been as successful had it not been owned by its own main customers. This example illustrates that while competition may exist between various dominant market participants, fair opportunities for outsiders to enter may be lacking.

Monopsony: A single HFT customer can be responsible for as much as half the business of an exchange (see e.g. the case of Chi-X above) suggesting monopsonistic pressures.  For example, the monopsonist can pressure exchanges to develop new products which benefit it at the expense of other customers.

Barriers to entry: There are large barriers to entry especially for HFT. Arnuk and Saluzzi  (2009) cite a TABB Group (www.tabbgroup.com) report according to which expenditures on colocation and facilities for fast access amount to $1.8 billion per year (it was not clear to us whether this number referred only to expenses in US equity markets or whether it is broader). In addition, markets are also spending huge sums and the NYSE alone is investing in facilities at a cost of $500 million. According to Price (2009), the cost of in-house solutions for competitive data feeds is of the order of $260,000 per month and a start-up cost of $270,000 per data center.

Incumbent advantage: In our experience it is very difficult to independently commercialize a good computer trading idea without passing ownership of the underlying intellectual property to an [incumbent] HFT firm.

Price discrimination: As an example, the Deutsche Börse was recently allowed to offer AT cheaper prices than other types of traders. Volume based discounts and loyalty   enhancing rebates for certain types of orders are extremely common. While the issue of   whether rebates should be treated as abusive per se is controversial, there is broad consensus about their potential anticompetitive effects (e.g. Motta 2009).

Waste and negative sums arms races: Some modes of competition such as competition in terms of latency among HFT are negative sums arms races, which can be expected to have a winner-takes-it-all outcome, suggesting that certain niches in the trading ecology may be dominated by a single, uncontestable player. We would like to emphasize that in our opinion the resources being wasted on trading activities in general and computer trading in particular are huge, especially in terms of human capital. One quantitative indication of this is the study of Phillippon and Reshef (2009) which shows a huge shift in US skilled labour – especially PhDs – towards financial services from other industries  since the deregulation of the 1980s.

Regulator handicaps: Regulators and academic researchers have several huge handicaps. They have incomparably fewer resources, personnel and incentives to collect and analyze market data than commercial traders, especially in HFT where strategies are quantitatively driven.

Testing?  We don’t need no stinkin’ testing!

In early October 2012, the SEC hosted a discussion with industry participants from exchanges, academia and elsewhere, the “Roundtable on Technology and Investing: Promoting Stability in Today’s Markets.”  This was not an encouraging discussion for those looking for assurance that HFT is a benign influence in markets.  For example, panel member Saro Jahani, The Chief Information Officer of Direct Edge (an electronic stock exchange), said the following:

Please imagine when you’re actually talking about testing, which we all want to do, it is absolutely impossible because it takes so much time.”

Here’s another useful quote from Lou Pastina, who manages the NYSE Cash Equity Market Operations Group, including the development schedule, the tactical operating plans, the NYSE trading floor, and the electronic trading component of the NYSE equity market:

“It’s amazing to me how many times software gets introduced and firms don’t test with you.  So whether we have test symbols in production or we run industry tests, it’s always the same firms that come in and test, and those are the firms that generally don’t have issues.  And then there’s a long list of firms that never show up…the 20 percent that show up represent most of the volume, which is good.  However, all you need is someone to come in with a bad message and ruin the whole system, and if you don’t have the proper defensive code in place, that message then can be propagated and because of the interconnectedness of our systems today, [and] be passed on to other marketplaces [e.g. commodities, FX].  And so I’m not advocating that everyone be required to come in and test.  I don’t think you can wait for every customer to come in before you put a new piece of software in place, but there should be some review about how firms actually take advantage of the testing opportunities that are afforded them.”

Finally, from Dr. Nancy Leveson, Professor of Aeronautics and Astronautics and also Professor of Engineering Systems at MIT; Professor Leveson conducts research on the topics of system safety, software safety, software and system engineering, and human-computer interaction:

“So sort of to summarize, I don’t want to sound like Chicken Little or a latter day Luddite.  I did not get a Ph.D. in computer science and spend 47 years working in the field just to try and convince everyone not to use computers.  But the bottom line is that there’s 100 percent certainty that you will have more upsets caused by the financial system software and probably in not too long a time, but it will occur, and it’s probably going to start occurring unless something is done more frequently because people are going to keep adding more functionality and more risk into the system unless it’s limited.  …all software contains errors.  I have not in all of that time ever come across any software that was not trivial in which no errors were found during operations.  The errors may not surface for a long time, but they’re lurking there and waiting until just the right conditions occur.”

Conclusion: HFT is a Runaway Train

So, we have seen:

1.Front-running

2.Quote-stuffing

3.Unfair latency advantages, aided by the exchanges themselves

4.Disturbing lack of testing

5.Computer experts raising red flags

6.Tens of thousands of micro flash-crashes that continue to this day

7.Multiple risks for market abuse by participants and exchanges

8.An out-gunned SEC, without the resources to police HFT, already over-burdened by policing Dodd-Frank, and trying to keep up with expanding financial markets overall

9.Linked HFT trading across asset classes, but policed by government agencies that focus on individual asset classes only

I fear the day when the market enters a broad, deep, fundamentally-driven selloff, and the thousands of HFT algorithms busy crunching and reacting respond in ways we cannot possibly predict – perhaps a liquidity vacuum as in the Flash Crash, but one that doesn’t correct in a few minutes.  And instead of happening in the middle of the afternoon, when the rest of the world’s markets are closed, it happens in the morning, when Europe is still trading.  In a worst-case scenario, trillions of dollars in wealth could be destroyed, public confidence in the markets shattered, investment capital for growth made virtually unavailable, and major deposit-taking institutions destabilized (prop desks are still part of banks, thanks to the 1999 repeal of Glass-Steagall and the slow implementation of the Volcker Rule[5].  This would represent a devastating blow to an already weak economy.  Sounds hyperbolic, but I wouldn’t bet against it.


[1] Financial Black Swans Driven By Ultrafast Machine Ecology. Neil Johnson (Physics Department, University of Miami), Guannan Zhao (Physics Department, University of Miami) Eric Hunsader (Nanex LLC, Evanston, Illinois), Jing Meng (Physics Department, University of Miami), Amith Ravindar (Physics Department, University of Miami),  Spencer Carran (Physics Department, University of Miami), and Brian Tivnan (The MITRE Corporation, McLean, VA; Complex Systems Center, University of Vermont).  Working Paper submitted Cornell University’s arXiv. February 7, 2012.
[2] A provider of historical and real-time market data.
[3] The Government Office for Science is the home of science and engineering across government and exists to support the Government Chief Scientific Adviser, Sir John Beddington. The key role of the GCSA and GO-Science is to ensure that all levels of government, including the Prime Minister and Cabinet, receive the best scientific advice possible, and to enable the many science-using departments across government to create policies that are supported by strong evidence and robust arguments.

[4] An Ecological Perspective on the Future of Computer Trading. J Doyne Farmer (Co-Director, Complexity Economics, Institute for New Economic Thinking at the Oxford Martin School, Oxford University), Spyros Skouras (Associate Professor, Athens University of Economics and Business).  August 21, 2011. See article for full citations.  http://www.bis.gov.uk/assets/foresight/docs/computer-trading/11-1225-dr6-ecological-perspective-on-future-of-computer-trading.pdf

[5] The latest development:  On November 29th, House Republicans Spencer Bachus of Alabama and Jeb Hensarling of Texas sent a letter to the Federal Reserve Board asking that the rule’s effective date be postponed two years after the final version is issued.  The current proposal would ban banks from proprietary trading, with exemptions for trades tied to market-making activities or hedging risk. It also limits banks’ investments in private equity and hedge funds.