The Future of Trading
10 Top CIOs on Wall Street | Why Mutation Will Matter | Buyside Traders: Not Taking Orders Any More | Calculating Cost First, Trading Second | Trading Desk, in One Box | Changing the Minds of Star Programmers | 2015: The Always On Desk
2015: The Always On Desk
November 4, 2010
It’s happening faster than you think.
The day when trading occurs 24 hours a day, at least six days a week, may in fact already be here, according to trading firms interviewed by Securities Technology Monitor, as well as hardware, software and network experts.
But it may be at least five years before a trader anywhere in the world can trade, around the clock, in any financial instrument in any market anywhere else in the world.
The obstacles are technical, regulatory and human. But, if the technology permits it, trading firms will find ways to man their desks at all times. Because competition will demand it.
FOLLOWING THE SUN
ING Investment Management on Park Avenue in New York is already operating 24 hours a day, six days a week, according to its head of U.S. equity trading, Nanette Buziak.
The New York-based trading team handles U.S. and International trading. While ING is globally coordinated, they are not globally integrated as yet.
As a result, the firm currently uses Instinet for overnight trading in Asia-Pacific and early European trading. The electronic, agency-only broker handles orders on their behalf from 5:30 p.m. to 7 a.m. Eastern time each night as well as on US holidays that are not global market holidays.
That may well change, as ING pursues its effort to be a “global boutique,’’ a specialized investment firm that provides personalized service and deals effectively with regional differences. A worldwide project, under way for a year, will create, for instance, a single common order management system that will be used in all time zones. As its business grows, replacing Instinet with its own team in Asia is a logical next step.
ING’s “follow the sun” model is not the only approach in wide use. Decentralized approaches constantly compete with centralized approaches, according to Dave Csiki, managing director of Indata. The “centralized” approach translates to working around the clock in one locale. Three shifts, like a manufacturing plant. Including one graveyard shift.
That does not sound like the relatively family-oriented work hours of the Wall Street tradition. There’s a reason that its trading begins at 9:30 a.m. and ends at 4 p.m., every weekday.
“Humans go home and have dinner with their family and they watch their kids play baseball and they need to close, they need to shut things off, they need to go home,’’ said Joe Ratterman, the chief executive officer of BATS Global Markets, which now operates two equities and one options exchange in the United States and one equities exchange in Europe. “The close is an incredibly invaluable part of the price formation process.”
But there are signs that this is changing. “Finding talent to work overnight shifts is increasingly not difficult,’’ said Scott DePetris, chief operating officer of Portware, which developed an algorithmically based system for basket, index and automated trading around the world of stocks, futures contracts, options and currencies.
And there are already precedents. Cotton, cocoa and coffee futures trade from 5 p.m. each Sunday to 4:15 p.m. each Friday, Chicago time, with only a 45-minute break before trading resumes at 5 p.m. each day. Similarly, currencies are traded 24 hours a day, from Sunday afternoon to Friday afternoon, thanks to sequential activity on markets in London and Frankfurt, for instance; followed by activity in New York; followed by activity in Sydney, Tokyo and Singapore.
But equities markets are exceptions. The same financial instruments do not trade in all markets at all times, around the world. Those traders who enjoy working the late-night shift in the Big Apple?
“They just might not be buying IBM at 4 a.m. in the morning,’’ said Ruth Colquiri, managing director and head of product management at UNX, an agency brokerage and supplier of trading technology.
There are practical problems, she says, such as manning support desks, not just trading desks, 24 x 5 or 24 x6. There are also timing issues, resolving in every trade just what time it is and what time zone to go by in executing and tracking trades. And the international dateline does not help.
But there is little question that technology is on its way to expand the types of high-speed connections that will make instantaneous trading around the globe practical and increasingly competitive. And that the math is arriving that makes it possible to enact any trading strategy through algorithms that not only execute what traders have been thinking about doing, but also start to read their minds, in effect.
A startup called Spread Networks this year opened up a 825-mile route between New York and Chicago that eschewed railroad lines and, if need be, bored right through mountains to get the most direct route for light to travel between the two major U.S. financial centers.
The route could complete roundtrips in 13.33 thousandths of a second. Spread’s chief executive, David Barksdale, said its route was only the latest, fastest route and would not be the last to hold that title.
“The speed of light is actually too slow,” said Paul Schoenauer, director of system engineering for Ciena, a supplier to Spread that specializes in high-performance networking gear.
At this point, he says, fiber itself can be getting in the way of the light taking digital instructions between markets. Right now, he figures that optical networks are only operating at two-thirds the speed of light.
Besides the straight-line approach of Spread Networks, which tries to remove curves and turns from the path of light pulses, the other obstacle that could get removed is the fiber itself. In a fashion.
Companies such as BlazePhotonics in the United Kingdom have been promising hollow-core fiber for years. The idea is to send light through air, to speed it up. No fiber drag. But, so far, it’s not clear that any company can deliver a hollow-core fiber that is any faster than fiber itself.
But there is little question that automation will continue to take over basic functions of the trading desk and make humans less a part of the process, on typical days of buying and selling securities of all types for portfolio managers.
Given is the fact that the trading process will be increasingly automated.
“The machines are getting smarter. We're outsourcing more of the nuts and bolts of the actual trading process,”said Jose Marques, managing director and global head of Electronic Equity Trading for the Markets division of Deutsche Bank Securities.
But, more critically, the algorithms that are used to effect trades will only get more sophisticated.
"We're moving toward algorithms where the machine can understand what's actually in the trader's head or what the portfolio manager's real objective is in executing that trade and be able to make the real-time tradeoff between liquidity and the cost of that liquidity,'' Marques said.
To date, algorithms have been fairly simplistic. The most widely used benchmark is what is called the Volume Weighted Average Price of a stock. The “VWAP” is calculated by multiplying the price against the shares in every transaction in a stock, then adding them all up, then dividing by the total number of shares traded that day. If the average price of a firm’s trades is better than the average weighted price, then it is considered a good trade.
What’s next? Algos that will already be inferring what’s inside a trader’s head, said Marques.
"Rather than giving mechanical instructions such as 'go buy 10,000 shares by time-slicing it every 15 seconds, 300 shares at a time,’ we may see a future situation where the trader is able to communicate to the machine and say, my portfolio objective is to minimize my implementation costs and given my horizon, I'd like to be relatively aggressive."
With the right formulas, a company’s trading systems will spot opportunities and act on them, with the prior approval of trading managers.
A basic example is pairs trading, said Jamil Nazarali, senior managing director and Global Head of the Electronic Trading Group for Knight Capital Group, which provides market access and trade execution services to buy- and sell-side firms.
In this tactic, traders find two securities, often rivals, where their daily price movements are very similar because they face very similar conditions. The trick is to act when the correlation weakens and one price moves up while the other moves down.
An algorithm, Nazarali notes, can fairly reliably be built that can make the correct buy and sell orders when the two stocks are moving apart and profiting from those moves by the time the prices converge again, to their traditional correlation.
Then there are “genetic” algorithms which evolve over time toward better solutions to trading in stocks or other financial instruments. These are often considered forms of artificial intelligence, where computer coding tries to replicate the thinking process of the human brain.
The approach has not made huge approaches onto Wall Street. But Adam Afshar, the president of Hyde Park Global Investments in Atlanta, employs no analysts, portfolio managers or traders to identify arbitrage opportunities and price discrepancies. He relies solely on a robotic trading system to carry out instructions – and figure out how to improve trading strategies in the process.
What’s more imminent are predictive algorithms, according to DePetris, the COO of Portware. Thease are formulas that take past history and project what while happen, at some point in the future.
They can also act as the kind of “recommendation engine,’’ that Amazon has popularized with its sales of books and other merchandise. Its algorithms set up possible choices that can be made in a given situation that it thinks is appropriate, but leaves the final choice up to the human on the other side of the screen.
A firm that started in Edmonton, Canada, has thrown its neural networking technology at the increasingly vast streams of information that flood into Wall Street trading desks.
The firm is looking for patterns in the streams as well as specific “sensory” imputs that model recurrent behavior in capital markets. In effect, how the herd instinct works.
It’s compiled a pile of ten years worth of tick-by-tick data from all U.S. equities exchanges, including the New York Stock Exchange, NYSE Arca, NYSE Amex, the Nasdaq Stock Market, BATS Exchange and the Direct Edge venues.
Then, it mixes in streams of economic indicators and news that shows sentiment about stocks in general or a stock specifically.
When it finds five or six instances of similar conditions that lead to similar results, it figures it can act on the pattern the next time the conditions occur.
This can apply to “irrational exuberance” in oil prices, as occurred in the summer of 2009, or the failure to foresee drops in housing prices or activity, as occurred in late 2007 and again in the second quarter of this year, when the Obama Administration’s homebuyer tax credit was endin
Broadly speaking, its “algorithm” works like this:
If we have [trend] then [behavioral event] then [overreaction], then you can [buy/sell] into it, during “euphoric,” “depression” or “greed” phase.
But such models are not certain to last forever, said Eric Davidson, vice president of research and trading for titan.
“One of the core questions in modeling is once you have a model that shows an edge, is it going to self-destruct?’’ he said. “Over time, the strategy can become ineffective.’’
DRAG, NO DROP
Traders will begin to configure their own algorithms, rather than rely on programmers or quants.
“Configurability” already is being built into formula-generating tools so that traders can establish how fast and how much of a stock that they want to trade in, over a given period of time, according to Ron Santella, the chief executive of Fox River Execution, a trading technology firm acquired this year by SunGard Data Systems. The most prominent example: Waddell & Reed, May 6, 2010.
Over time, portfolio managers will have their choices set up for them, in advance, with or without requiring human traders to carry them out.
Odyssey Financial Technologies, in the United Kingdom, for instance, allows users of its Investment Manager product to establish a new ratio for allocating assets and see what it will take to change it. If the manager asks for the fixed-income portion to be increased to 57 percent, from 53, the software shows a series of trades that adhere to company policy and known regulations. If the manager wants to go ahead, the trades get executed. Or they can be rejiggered, first.
In effect, the choices or “recommendations” will become clearer and the algorithms behind them less visible. Tim Mahoney, chief executive of BIDS Trading, sees an “aggressiveness” button as a not-unlikely outgrowth of the increasing injection of trading strategy into code.
Most markets will be electronic, Marques asserts, over the course of the next five years.
But, if nothing else, the “events of recent months” – particularly the May 6 crash -- shows that human direction of trading is still necessary, Mahoney contends. Trading firms still have to understand the mechanics of the systems they’re using. And they don’t want to cause conflagration, accidentally.
“Who wants to be Mrs. O’Leary’s cow?” Mahoney says.
If or when things go really wrong, algos may not understand what’s going on. "When something new and unusual happens, people are still much better than machines at dealing with that and making very quick decisions,'' said Marques.
Traders still need to understand the algos that they have to use, DePetris said. And this will become a career-maker. Traders who understand algos are the “big winners’’ when desks are always-on and moving at electronic speed, said Timo Pentner, managing director for the Americas at RTS Realtime Systems Group.
The process is likely to be helped along by the adoption of a messaging language that allows buy-side and sell-side traders exchange clearly defined algorithms and update them as they go along.
The Financial Information Exchange algorithmic trading definition language (FIXatdl) will allow algorithms to be released in a standard computer-readable tagging format. This will avoid the kinds of detailed documentation, coding and testing required by conventional programming of algorithms.
This should speed up the release and updating of algorithmic trading strategies, according to proponents.
Ultimately, companies are likely to have “chief algorithm officers,’’ suggests Bill Hilf, general manager for technical computing at Microsoft. But traders might also have programming tools of their own, akin to Visual Studio, where users plug in the languages and functions they need. Call it Wall Street Studio.
The types of humans found on the trading desk also will change, as the complexity of digitally driven trading increases.
A decade ago, one big bulge bracket firm on Wall Street had just 10 sector traders, each dedicated to stocks in one broad industry, on its desk. Now it has four chartered financial analysts, one quantitative analyst and one programmer. “We look more like our investors, than we look like a trading desk,’’ the head of the firm’s trading said early this month.
And not just any programmer will do, any more, said Jeff Birnbaum, chief technology architect for Bank of America Merrill Lynch.
“No matter what programming language you choose, and they are not all created equal, the skill of the programmer makes the difference,’’ he said.
With the number of venues exploding, market data following suit and expansion of time and markets worldwide all part of the environment going forward, “the lifeblood of Wall Street will depend on our ability to harness hundreds and hundreds of processors and simultaneous threads of calculation of calculation,’’ he said at a September conference on high-performance computing.
In the last credit crisis, he noted, banks were not able to adequately calculate the risks they faced. That should not happen again.
There are ways to justify human wisdom, even in the era of the automated trading desk. One is the concept of “At-Trade Transactional Cost Analysis.” This calculates not just all associated costs of a transaction, in advance, but expected return or performance.
This allows firms to turn savings almost into a revenue stream. One big Wall Street firm figures it generated $2 billion in savings from aggressive monitoring of its trading costs and results. And, used properly, such analysis allows traders to be judged on the execution prices and performance they consistently achieve, said Allen Zaydlin, chief executive of Inforeach, a supplier of trading management systems.
24 X 7: NOT
Getting to 24-hour trading, anywhere in the world, on the same equities, though, is not likely to happen in five years or quite possibly even 10.
A big sticking point is likely to be regulation, say Ted Myerson and Gary LaFever of FTEN, which develops risk management products for Wall Street firms. Twenty-four hour trading in stocks would require that regulatory schemes around the world be “synchronized,’’ they contend.
Helping out might be a securities industry “currency,’’ sort of like air miles. But that is not likely to happen either.
Currency exchange on equities trades will continue to just get handled in the settlement process, said Colaquiri of UNX.
Even that will require back office systems to get better, over the immediate future, as more and more markets try to establish cross-border trading of securities.
But even if regulation and back office systems become consistent, there’s still no guarantee that the same equities will ever get traded at all hours, at all points on the globe.
Mathematics, though, can produce substitutes. Already, machines are "replicating a basket of stocks that mathematically looks like IBM" and choosing, at any given instant, which to buy or sell, said Marques.
Twenty-four hour trading will reduce the liquidity of stocks and increase the volatility of their prices, Zaydlin contends. For most stocks, there is not enough interest to keep trading fluidly all day long.
It’s different with foreign exchange markets, where there is “huge liquidity,” said Colaquiri.
Indeed, the answer may be at the other end of the clock, said Drew Miyawaki, head of European and Asian trading for Liquidnet, a market that specializes in moving large blocks of stock.
He suggests a much more concentrated trading day, to establish more concrete prices on a given day. His time frame: Three-hour trading sessions, every day. This should reduce the range of price swings in a given day and make it easier to move shares a couple thousand shares or more at a time, rather than a couple hundred.
It’s not that far-fetched, said Joe Saluzzi, a co-founder of Themis Trading and critic of electronic trading practices. In recent years, stock markets have huge volumes in their first 90 minutes and last 90 minutes. Trading in between if largely calm, unless there are major midday business events.
A separate approach could be to set up a two-tiered marketplace. With the encouragement of new venues, the establishment of an alternative marketplace for institutional investors is not out of the question, Saluzzi suggests.
Just like slow food and slow money that invests in long-term revolutions are gaining ground in agricultural markets, the same might be warranted in securities trading.
David Weild IV, Capital Markets Advisor at Grant Thornton and former vice chairman of Nasdaq, suggests, for instance, an “opt-in capital market” which could feed the growth of new stocks. The alternative market would have rules “preserve the economics necessary to support quality research, liquidity (capital commitment) and sales support,’’ thus favoring investors over high-frequency or day traders.
This could lead to wider spreads, but those nickels and dimes could also afford more research. And technology that allows trading between institutions to take place safely and directly.
On October 8, the Nasdaq OMX Group even launched the first “price-size” exchange, that rewards traders who bring big orders to market, rather than traders who simply bring any size orders to market first. In its first day, the average trade was 578 shares, compared to the general markets average of 220 shares.
But the size of an average trade on Liquidnet in the last year fell 7 percent, to under 50,000 shares, and its overall trading volume fell 15 percent.
Size may again matter, by the end of the next five years. But, the need for speed will be unrelenting, as the trading day expands, as do the number of markets and the types of instruments that populate them.
And the algorithms that will almost certainly be the drivers of any strategy that matters.
"Humans take complex problems, reduce them to a few number of variables and then try to create inferences between those variables,’’ said Marques. “Machines deal with complexity in a very different way. They take it head on. They have no issues with having 10,000 degrees of freedom, with each of those variables updating on a millisecond time scale and then exploiting relationships between them.”