In late March, the initial public offering on trading platform BATS (which stands for “Better Alternative Trading System”) was torpedoed by “systems issues” that slammed share prices down from $15.25 to a few cents in less than 900 milliseconds and caused a temporary halt in trading on Apple. 

Eric Hunsader looked at the trade-flow data and smelled an algorithm — something he’s gotten quite adept at as president of market data-feed firm Nanex.

“It’s not like somebody threw in a bad market order and it just worked its way to zero,” he says. “This was an algorithm that was waiting, and it waited and waited until the bid price posted lower, and then it hit that price — and hit it and hit it — 567 times.”

He says it’s not that difficult to see if you know where to look, and he assumes scores of traders have seen it, too.

“Whoever did this is a seasoned veteran,” he says. “The algorithm was designed with intent, and it used ISOs (intermarket sweep orders), which only people with sponsored access can use.”

In January, his company uncovered evidence of widespread spoofing (entering and then cancelling orders to trick the market) designed to manipulate the National Best Bid or Offer (NBBO), which is the reference price for securities across the United States. He says also that bid/offer spreads in equities actually widened after the 2007 introduction of Reg NMS (Regulation National Market System), which is the legislation that sparked the high-frequency trading (HFT) boom by requiring brokers to execute trades at the lowest price being offered on any platform.

All of this directly contradicts the generally held perception that HFT has made bid/offer spreads tighter and commissions lower than they’ve ever been — an achievement Hunsader attributes to the advent of automated trading in general and not to HFT in particular.

Still, HFT now accounts for 60% of U.S. stock volume and 40% of volume on U.S. futures, according to Rosenblatt Securities, which says the figures are lower on options in the United States and on all markets around the world, with high-flying Brazil and most of Asia hovering around 10% on equities.

The entire world is retooling for HFT, however, and established players are salivating over the fat spreads on developing exchanges as their own per-trade profits drop in the United States. 

Regulators, for their part, still are wrestling with a definition for HFT. In the United States, the Commodity Futures Trading Commission (CFTC) has set up a working group to come up with one, while the European Union is kicking around a working definition as part of its makeover of the Markets in Financial Instruments Directive (MiFID II). European Parliamentarian Marcus Ferber wants to define HFT as any trading that meets four of the following criteria: It uses co-location facilities; has a daily portfolio turnover of at least 50%; the ratio of orders to trades exceeds 4:1; 20% or more of orders are cancelled; the majority of positions are intraday and more than 50% of orders are placed on trading venues offering discounts or rebates for providing liquidity. 

Beyond seeking to define HFT, regulators on both sides of the Atlantic are asking the same question we examined one year ago in “High-frequency trading: Good, bad or just different?” — namely, whether HFT enhances liquidity in ways that longer-term algorithmic strategies don’t, or whether it disrupts trade.

The answer, increasingly, seems to be “both of the above,” and the question now is how to maximize liquidity and minimize disruption as engineers work to bring latency times down from five microseconds to three to less than two. For some, the answer appears to be: Just don’t go there. The Securities and Exchange Board of India (SEBI), for example, is considering a speed limit of, say, five microseconds on transactions. 

Technology challenge

Trading times have shrunk from seconds to milliseconds in the past decade, and some are talking about microseconds — which would require unheard of increases in technology.  Just one quote per microsecond on 1,000 symbols is 1 billion quotes per second. 

“That rate would saturate a terabit network, which is a future technology, and completely fill a terabyte drive in about 10 seconds,” says Hunsader.  

Indeed,  squeezing out those last few milliseconds is proving difficult.

“It’s more complex than many people realize,” says Daryan Dehghanpisheh, Global Director of Financial Services & Institutions for Intel. “These systems not only have to write fast code, but they have to write code fast, which means you’ll need lots of tools in multiple libraries just to get that microsecond.”

He adds, however, that plenty of trading houses are putting in the effort and, in the process, contributing to the development of new technologies that will have knock-on benefits for the entire financial system.

“We spend a lot of time with high-frequency traders because they’re forcing us to think about performance, latency and security,” he says. “These are the same issues that arise in payments for retail banking, where smart cards handle millions of transactions a second [around] the globe.”

In fact, he says, HFT has benefitted more from other sectors than it has contributed.

“I can chuckle when I hear traders talk about the overwhelming two billion messages a day, because Google, Facebook and Flickr are handling five billion posts a second,” he says. “High-frequency trading is not providing bleeding-edge technology, it’s consuming bleeding-edge technology. The markets are more of a beneficiary than a benefactor.”

Many traders, like Mark Reece, an eTrading Solutions Architect at HSBC, say the most important challenge now isn’t speed but monitoring capabilities.

“The traders I was working with were getting arbitraged on their exchange-traded funds business and assumed this was because they were taking too long to change the quotes and manage their orders onto the exchange,” he says. “It was only when I compared the time-stamped quotes being placed against the aggressive price taker orders that we discovered that … it was only very rarely that we got hit on a price that we were planning to change.”

He ran a check on his entire system using ITRS Geneos and found the problem wasn’t hardware or even software, but settings.

“Once we had the data, we usually found that it was something straightforward to fix — like the price construction parameter was set at a 10-second interval instead of a 100-millisecond interval,” he says.

The regulatory conundrum

HFT presents different problems and serves different purposes depending on whether it is being deployed in futures or in equities. On the equities front, it often links multiple platforms and fills a role previously occupied exclusively by registered market-makers, who were obligated to make two-way prices. In futures, that type of market-making was never the norm, and HFT generally is seen as simply a newer, faster version of the locals who always traded in their own best interest, with tight spreads being a by-product. Not everyone, however, is buying it.

“I hear from a lot of people in the financial sector how the cheetahs, the HFTs, are nothing more than a variation of the old day trader,” said CFTC Commissioner Bart Chilton at a March 29 meeting of the agency’s Technology Advisory Committee (TAC). “What we have found out, and I found out, is that it is naïve for us, in general, not to question how technology is changing markets.”

Academics like Joel Hasbrouck of New York University’s Stern School of Business are questioning it, but the answers are mixed. He referenced several academic studies at the TAC meeting, all but one of which contained data that was more than a year old (and, in fact, were included in our story last May). The most convincing pro-HFT argument remains Albert Menkveld’s “High-Frequency Trading and the New-Market Makers,” which followed one major HFT on Chi-X and demonstrated that his activities narrowed spreads across several markets. The most disturbing study remains “A Dysfunctional Role of High-Frequency Trading in Electronic Markets,” by Robert A. Jarrow and Philip Protter, who modeled the behavior of HFT strategies in a stable market and found they first created market aberrations — which they then capitalized on and corrected.

Hasbrouck’s own study identified peculiar bids entering a market in a way that might be designed to manipulate the NBBO in equities but that clearly would be benign in futures. The United Nations Convention on Trade and Development released a disturbing piece titled “The Synchronized and Long-Lasting Structural Change on Commodity Markets: Evidence From High-Frequency Data,” which says that HFT might be distorting commodity markets by creating synchronous price moves detached from fundamentals.

Most disturbing of all, however, may be the tendency of HFTs to follow similar strategies — raising the question of whether some confluence of factors could trigger a runaway market event as HFTs run amok. Most traders dismiss that notion — in part because HFTs are short-term in nature, but also because traders with a more intermediate time frame usually step in when markets move too far one way or the other — as they did in the Flash Crash. 

Hunsader, however, says that intermediate traders are backing away from the market at critical times lest they get tricked by HFTs (see “The race to the swift”), and that all the traffic being generated by HFTs gaming each other is raising the costs for those intermediate traders — accelerating their flight from the market.

“Our costs have gone up tenfold in the last few years on the equities side, while the number of trades and speed of trading have declined,” he says. “Last October, anyone processing stock quotes had to upgrade all their equipment from gigabit to 10 gigabit, which would be fine if this were due to an increase in meaningful data, but it’s not.”

That 10-fold increase in bandwidth, he says, translates into a 20-fold increase in cost, which could force intermediate-term traders out of the market.

“If this continues, we’ll end up with two groups of participants — those trading in milliseconds and the mom and pop investors,” he says. “That would mean a lot more volatility during the day, because you wouldn’t have diversity of opinion.” 

While Hunsader’s concerns focus on equities, futures markets have messaging issues of their own — primarily when HFTs start generating and then canceling thousands of quotes away from the market. ICE began to address the issue in March by charging lower fees for orders placed within one or two ticks of a market and higher fees for orders placed three, five or more ticks away. The pricing mechanism only applies to traders who generate 300 or more trades per second; it isn’t designed to cut down on message volume so much as to promote the kind of messaging affiliated with HFT that leads to tight spreads, while still encouraging intermediate-term traders to place orders a bit further away to provide depth of book. 

Zombie algos

Almost everyone we spoke to dismisses the idea of zombie algos that can run amok, at least in liquid futures contracts, but exchanges now require certification of algorithms before they can be deployed through co-location facilities. This means letting the exchange throw every weird market condition that they can think of at an algo and rejecting it if it behaves strangely. What’s more, if the exchange changes its code, the algos often have to go through the process again. That, however, hasn’t prevented a few funny things from happening.

In “Crude loop” (below), we see an aberration that began at 1:59:57 p.m. on Feb. 13, 2012, when quotes and trades on Nymex crude futures began queuing, only to be unleashed all at once at 2:00:35. Less than a minute later, at 2:01:08, it happened again — this time with quotes from the first 38 seconds and the intervening period. This repeated 12 times, each time with accurate timestamps that should have alerted algorithms to the fact that the quotes were old. But the algos reacted as if the quotes were real.

“Every blip led to an explosion of activity in USO, UCO and SCO — ETFs that mimic the price of oil,” Hunsader says. “If we had a 50¢ range in that loop, it would have been devastating.”

The CFTC is considering the imposition of mandatory pre-trade filters, price collars, trading limits, kill buttons and other measures proposed jointly by the FIA Principal Traders Group and the FIA European Principal Traders Association in March. Most futures exchanges say they’re already there, and both CME and ICE presented their risk-control procedures at the TAC meeting. Interestingly, neither exchange differentiates between algorithmic trading in general and HFT in particular when monitoring trades.

“We can be talking about something as simple as a spreadsheet to an auto spread to a sophisticated black box,” says Mark Wassersug, ICE’s vice president of operations. “All of them have direct market access. We do not look at subcategories beyond that.”

But they do monitor every order, every modification and every cancellation down to the millisecond and in real time. Both CME and ICE are able to reconstruct the entire order book over multiple markets and detect anomalies in real time, and both know an incredible amount about every single trade that comes in.

“When an order comes in to us, we get the clearing firm that is guaranteeing the trade, that trading firm executing it, the session ID, the account number, the country of origin,” says Dean Payton, CME Group’s managing director & deputy chief regulatory officer. “We also have the ATS order identifier, which we introduced last year, and the operator ID — or Tag 50 — which is the person who is entering the order into Globex. We have their names, we know who those individuals are. In the case of an automated trading system, you have a head trader, and you have the team of individuals who support that ATS. So you may have a risk manager, a secondary trader, a monitor — all of those folks would be registered with the exchange.”

Then there is price banding, which is designed to halt runaway moves. The CME had its Stop Logic function in place before the Flash Crash, and ICE implemented its Interval Price Limit (IPL) system after the crash. It’s a sophisticated limit-up or limit-down mechanism that puts a floor below and a ceiling above a market for a given time period. The floor and ceiling move slowly, and the market triggers a “hold” if the price hits either. During that hold, orders keep coming into the market, but bids are not allowed above the ceiling and offers are not allowed below it. The holds can last a few seconds or a couple of minutes, and then the bands begin moving again.

Such pauses provide an opportunity for living, breathing traders to think and react — and that, perhaps, always will be the ultimate safeguard against algos run amok.

The race to the swift

The depth of the order book on S&P E-mini contracts plunged dramatically from June through August in 2011 as these market studies from Nanex show. 

Every line in “Momma said there’d be days like this” (below) represents a trading day at one-minute intervals, color-coded by date. The higher the line, the deeper the order book. The purple lines are one year ago, the greens at the bottom are from August and the red lines are the most current.

Nanex believes this represents reduced activities as medium-term traders backed away from the market for fear of being stung by HFTs. Note the sharp plunge at 10 a.m. (when many economic reports are released) — reflecting an evaporation of liquidity on news, when it’s needed most.

Trading always has gone slow around news events, only to erupt seconds later, but this behavior has become more pronounced in the past year after intermediate-term traders found themselves getting stung by HFTs. “Liquidity gaps” (below) illustrates this phenomenon, showing how the order book depth now plunges from roughly 15,000 per side to less than 1,000 — leaving the market even more susceptible to wild moves on small reactions. The bottom chart shows the depth of book at each level and the top chart shows last trades. The color blue indicates less depth.