Smart order routing, a technology that debuted in the Nasdaq market in the late 1990s, is being overhauled and revved up for the Regulation NMS era. Brokerage firms are spending millions of dollars and countless man-hours to update a tried-and-true workhorse of automated trading for a more complex and data-intensive trading environment.
Smart order routing has been overshadowed in recent years by its sexy cousin-the algorithm, which slices orders into the market. Now, no algorithm can afford to go it alone. Algorithms must rely on the smart order routing that underpins their executions to get their orders into an increasingly fragmented marketplace.
“The smart router is being raised to a new level of importance,” Jana Hale, global head of algorithmic trading at Goldman Sachs, tells Traders Magazine. “It must decide where an order should be across many exchanges and markets, based on how executions are likely to play out in those markets over the near term.”
The rules of engagement have changed. There are now more trading venues, more order types, more competition and new rules dictating the routing priorities and procedures of trading centers. New Reg NMS-prompted exchange-execution rules and order types have led to tremendous uncertainty about how traders should execute orders. Simply routing an order to an exchange for execution is no longer a safe option.
“In the good old days, three months ago, it was much easier to algorithmically send trades down to the floor,” says Michael Rosen, a market structure expert and product manager at agency broker UNX. “An algo could just grab a quote and didn’t have to worry about little funky things like every exchange having different rules for executing an order. Everything is now much more complex.”
Behind this shift is Reg NMS’s new quote-protection rule, which requires that the best electronically available bid or offer in each market be protected. This means no exchange or broker can trade through-execute an order at a price inferior to-a protected quote, wherever that quote resides. The rule went into effect for exchanges on March 6 and will become operative for brokers, for a pilot group of 250 stocks, on July 9. Full implementation takes place in October.
Order Placement
The new world of smart order routing concerns what some describe as the “minutiae” of how orders are placed into the marketplace and how a broker tries to aggregate liquidity. Smart order routing is necessary for algos and direct market access. Most brokers have a single smart order router underlying their many algos.
The router determines the best way to get the executions the algorithm wants. Algorithms decide how to slice “parent” orders into the market to try to achieve a client’s execution benchmark. Based on mathematical formulas, algos decide how quickly to execute the order, how many shares to execute at a time, how often and at what price. The algo’s smart order router then decides how those “child” orders should be placed in the market and how to efficiently take or post liquidity.
This process has become far more complex and less linear as order types and trading venues mushroomed. Many more variables must now be taken into account.
Algorithms themselves won’t be turned upside down by Reg NMS changes. “Reg NMS will have a strong effect on the details of how orders are placed into the market. It’s a huge technological challenge,” says Robert Almgren, head of quantitative strategies for equities at Banc of America Securities and a managing director in the firm’s electronic trading services group.
But Almgren doesn’t expect algos themselves will need a top-down overhaul. “My gut feeling is that fundamentally Reg NMS won’t change much above the order-placement level,” Almgren says. “You’re still trading with people [who make decisions]. Exchanges are the means to get there, so whatever the rules on exchanges, you’re still waiting for when people are ready to take the other side.”
Big Investment
To compete in the new Reg NMS environment, brokers need hefty technological resources. They must establish connectivity to all trading centers, develop sophisticated smart routing technology and invest significant sums of money to expand and bolster their market data infrastructures.
These infrastructures must capture tick, quote and trade data. Continuously tracking and processing information about where liquidity resides permits smart routers to make split-second decisions about how to place and take liquidity.
Algorithmic providers will be judged over the next year by how well they succeed at this new task of intelligently placing orders into the market and extracting liquidity from a marketplace that is becoming more fragmented. As a result of incresed electronic trading of listed stocks and the quote-protection rule, liquidity already has begun to disperse across market centers.
A recent TABB Group report noted that brokers will spend $91 million between this year and next for applications that help them comply with Reg NMS. Most of that will be spent by bulge-bracket firms.
Eighty to 90 percent of the outlay will be for low-latency connectivity to markets, order routing and market data aggregation. These areas constitute the “core technology underpinning the trading infrastructure that brokers need to compete” in a Reg NMS environment, according to the report.
Credit Suisse, long considered a leading firm on the algo front, has had four people working full time for a year to study and understand the order-routing logic at various exchanges and electronic communications networks. This effort, which is separate from the firm’s algorithmic effort, got under way even before the New York Stock Exchange introduced its hybrid trading model and Reg NMS began its phase-in.
Sorting out the myriad execution rules and order types at market centers-and their impact on how liquidity is distributed across venues-is no walk in the park. But it’s critical for algos. Dan Mathisson, head of Advanced Execution Services, Credit Suisse’s suite of algorithmic products, notes that there are 10 exchanges, five ECNs and anywhere from 10 to 30 alternative trading systems, depending on how they’re counted. “That’s 25 to 45 destinations, if you count the private networks,” he says. “You must figure out all the order types they support, how they route out and under what circumstances you will use which types of orders.”
Rick Holway, CEO at execution management system vendor Firefly Capital, also stresses the complexity and hazards of operating in a landscape occupied by rapidly changing order types and execution rules.
In addition to recognizing top-of-book and exchanges’ various rules, a smart order router must know that NYSE Arca, for instance, permits various kinds of tracking, Holway says. An algo buying stock can have the exchange track its orders on the bid plus a penny or track the primary market, or always bid a penny below last sale. He adds that figuring out when to use which order type and how those decisions might affect other market participants’ behavior is a challenging puzzle.
Goldman Sachs has spent “millions of dollars” over the last few years on its effort to expand and tailor its order-routing logic in the face of a shifting trading environment, Hale says. She adds that this effort supports Goldman’s algorithmic offerings as well as the firm’s overall ability to trade in the markets.
Size Matters
In Hale’s view, being a large broker is now key to getting better executions. “You can only slice and dice orders in certain ways, and everyone can do that well,” she says. “At every level, knowing when, where and how to place an order in the marketplace is increasingly a competitive process, and having access to unique liquidity generated by strong market share is what separates people at this point.”
Commanding a larger market share increases the order and execution data a firm can analyze, enabling it to fine-tune routing decisions and potentially improve its executions. Hale notes that given the financial and technological investment now required, “I can’t see how a player that’s 10th or 15th in algorithmic executions can compete against the top three.”
Executions are becoming more complex for several reasons. The quote-protection rule ensures that brokers will establish fast, low-latency connectivity to every market with protected quotes. Latency itself is a new competitive front that is increasingly important (see sidebar).
In the new Reg NMS environment, liquidity will go to the best protected quote. The number of exchanges and ECNs has already increased as market players saw opportunities to compete with established trading centers on the basis of price rather than relationships, history and habit.
Under Reg NMS, if an exchange receives an order that would otherwise trade through protected quotes on other market centers, it must route out a sufficient quantity of the order to satisfy its quote-protection obligation. However, that routing to the best price adds a slight delay in the execution and subjects the broker to the exchange’s routing decisions.
Most brokers that value fast access to liquidity expect to control their own executions. They do not plan to rely on exchanges to route orders to better-priced markets. They will instead use smart order routing and a new order type called the intermarket sweep order, or ISO, to grab those protected quotes themselves.
Exchanges now make two new sets of decisions that could affect the quality of algorithmic executions. Exchanges dictate how they route orders to market centers with protected quotes when those protected quotes are at the same price. For example, an exchange could decide to route to a stock’s primary market, to the venue with the cheapest execution costs or to the market with the deepest liquidity. It could also split the order and send it to different destinations.
Exchanges must also decide how to execute orders once they take out protected quotes on other markets. An exchange could route to markets with the next-best-priced quotes to get the best prices for its customers. It could also just fill an order from its own book, which could potentially result in a lower average price.
UNX’s Rosen says his firm advises clients not to send market orders to the NYSE because the Big Board preferences its own liquidity after protected quotes on other markets. “If you send a market order to the New York, and it’s at the top with two other exchanges but its next quote is 3 or 4 cents away, you’ll get protected on the top but then the exchange will rip down through its book instead of going down other markets’ books in an orderly fashion,” he says.
More ISOs
Smart order routers that can shift orders around and maintain the best bid or offer in multiple locations are likely to find more liquidity. “Algos that have the capability of maintaining top-of-book in multiple markets will get a larger share of liquidity than those posting in a single destination,” says Doug Rivelli, managing director and head of electronic trading at Weeden & Co. The ability of routers to respond dynamically is also critical for algorithms that try to “ladder the book” by posting liquidity at various price increments in the market, he adds.
All of these rules and decisions affect customers’ execution quality and must be thought through carefully. “It may be better to be at the top of the book at a secondary location or regional than at the primary exchange at another price level,” says Goldman’s Hale. “That ensures that if the market is swept [by ISOs], the order will be executed.”
The ISO, a subset of immediate-or-cancel orders, is the trader’s new power tool. It tells a market center that the broker submitting that order has taken over the responsibility for quote protection. Using an ISO requires a broker to simultaneously route additional ISOs to all market centers with protected quotes that would otherwise be traded through. Every trading venue supports ISOs.
To take charge of their own order flow, most large brokers are instituting a wholesale shift from current market and limit orders to ISO and IOC orders. Most brokers have already begun using these order types to see how they work and to test execution quality across venues under different scenarios.
Some made the change early. Morgan Stanley switched to ISOs for “about 99 percent” of its orders on March 5, says Andrew Silverman, head of U.S. electronic trading distribution. That was the day the Securities and Exchange Commission’s new quote-protection rule became effective for exchanges and the NASD’s Alternative Display Facility participants. The broker “didn’t want to add latency to customers’ orders,” according to Silverman.
Credit Suisse sends out ISO and IOC orders almost exclusively, Mathisson says. He adds that March 5 was Reg NMS’s “big bang” for algorithms. However, most brokers expect the broader changes to occur starting in July, when they must officially comply with the order-protection rule.
Dark Orders
For algorithmic providers developing advanced smart order routing, the current Reg NMS-inspired routing complexity in the displayed markets is only exacerbated by the growth of dark liquidity venues and crossing opportunities. To get best execution for clients, brokers must access that liquidity. Dark liquidity refers to orders that are not publicly displayed in trading venues and that cross within the national best bid and offer.
Some two dozen brokers already have algorithms that go to dark pools and use dark order types on public markets to stealthily execute orders. An increasing number of firms enable all or most of their algos to access dark liquidity, whether it’s resting in dark pools and ATSs or in hidden orders on exchanges and ECNs.
Goldman says 20 percent of its electronic customer flow is now executed algorithmically through dark order types. Hale notes that almost every Goldman order launched by an algorithm attempts to execute within the NBBO on an exchange or ECN before taking displayed liquidity.
The recent growth of dark order types that cross orders within the inside market is a response to competition from dark pools and an effort to draw liquidity back to the public markets. However, identifying which markets are attracting undisplayed liquidity is hard. Smart order routers cannot simply look at a market’s depth of book and analyze tick-level behavior, since those data feeds do not include reserve and hidden orders resting in order books.
To ascertain which markets have executable liquidity, algo providers must work backwards off trade and quote data and try to match it up to instant-by-instant snapshots of various markets’ books. This must be done in real time, so that smart order routers can optimize their next-instant routing decisions across a large number of destinations.
Firms that gather and analyze these vast quantities of quote and trade information can search for signs that certain markets may have undisplayed liquidity. That enables them to route orders to those markets more effectively, yielding potentially better executions for customers.
What’s unclear is whether the combination of dark liquidity and smart order-routing practices, which are altering the way algorithms route orders to dark and displayed markets, will affect trading behavior more broadly. It is possible, some say, that the presence of dark liquidity, coupled with changes in order placement, could provide new incentives for liquidity to enter the market.
“People use algorithmic spadework to try to take block trades and smooth them out over the course of the day to get into the slipstream of the daily retail flow,” says UNX’s Rosen. “If people move more into the dark, that will affect market behavior.”
Dark liquidity has already changed certain fundamental aspects of algorithmic trading, according to many market participants. Credit Suisse’s Mathisson points out that it has altered the fill rates traders can get without displaying orders and has created new ways to trade.
“You can do things you couldn’t before, like break an order up and spray out 20 dark orders simultaneously on 20 different destinations,” Mathisson says. “Instead of displaying orders, an algo can create a giant hidden net to catch the contra side no matter where it’s been routed.”
This change has been significant enough that Credit Suisse’s many algorithms hardly ever display orders now. “It’s becoming a rare event when an AES algorithm displays a bid or an offer out loud,” Mathisson says. That shift occurred over the last six months as a result of anticipated Reg NMS changes and the explosion of dark venues and order types.
Clocking the Exchanges
Data and low latency are increasingly critical for algorithmic trading. And one
without the other is useless. Algorithms need vast amounts of data to try to discern where liquidity resides and where executions are occurring at every moment. Analyzing that information in real time also enables a smart order router to figure out which markets may have slightly slower data.
“We’re constantly judging the execution strategies we’re using in our algos and how they’re working,” says UNX’s Michael Rosen. “It comes down to which venues are giving us better execution quality and better response time.”
“We’re talking about nanoseconds,” he continues. “If a firm has latent data and sends out intermarket sweep orders, it will have horrible execution quality. It will think it just cleared out the book, but that was the book of 50 milliseconds ago, and the broker will have hurt its client.”
Identifying latencies, or delays, in market data feeds is therefore vital. Banc of America Securities’ Robert Almgren says his firm tracks the delays in direct exchange market data feeds in milliseconds. The firm takes market data from multiple direct exchange feeds, time-stamps it, measures the latencies and cross-correlates them to see how much later one feed was than another. BofA also looks at how market centers’ clocks are synchronized, down to tens of milliseconds.
All this is done for the benefit of algos. Based in part on that data, BofA’s smart order router decides when and how to submit orders into various trading venues.
Credit Suisse uses proprietary “heat map technology” to chart response times, fill times and fill rates across market conditions on a name-by-name basis, according to Dan Mathisson. That enables the firm to overweight “hot” destinations and underweight “cold” destinations in particular names.
Goldman Sachs’ Jana Hale says her firm’s smart router also captures millisecond-level latencies in market data feeds and in response times at exchanges and electronic communications networks.
“If a trading center isn’t responsive, within our routing we can change to a different destination in milliseconds. We track it monthly, weekly, daily and on a millisecond basis,” she adds.
-Nina Mehta
