This article was taken from the April issue of Wired magazine. Be the first to read Wired's articles in print before they're posted online, and get your hands on loads of additional content bysubscribing online
In 2002, after 12 years at Bear Stearns, investment banker Adam Afshar came to the conclusion that, for all their supposed expertise, most Wall Street traders really weren't very good at predicting anything. "I got the sense from my time there," says Afshar, "that market analysts have little insight and generally provide very little value to their firms." So Afshar set out to find market opportunities in, as he puts it, "a more systematic way". The result was Hyde Park Global Investments, a small trading firm based in Atlanta, Georgia, that doesn't hire market analysts or portfolio managers.
Instead, its employees are almost all physicists, computer scientists and mathematicians skilled in writing code and developing sophisticated algorithms inspired by evolutionary biology. The goal: to analyse historical financial data and spot and exploit fleeting opportunities in the market. "With algorithms and fast computing, even small firms can now buy and sell as fast as the biggest ones," Afshar says. "Technology lets you compete with firms having billions of dollars."
When Afshar says fast, he means very fast. What have become known as high-frequency trading (HFT) systems can execute transactions in milliseconds without human intervention -- basing their decisions on information they have received electronically. Joined by high-speed data links to the trading exchange, they draw on huge databases of historical data to test algorithms offline and then quickly use that knowledge to spot market opportunities and likely profitable trades.
They can execute those trades before anyone else has had a look in. In fact, to exploit split-second advantages, trading firms physically locate their servers as close as they can to the exchanges -- in some cases just a few metres away from the boxes where the trades are completed. The reason for all of this, says Joe Gawronski of New York brokerage Rosenblatt Securities, is that HFT requires extremely short "latency" to work -- a very small delay between sending an order and having it accepted, executed and receiving a confirmation, or, if it doesn't trade, confirmation of its being cancelled.
InfoReach, a New York-based company specialising in trading technology, has platforms capable of handling more than 10,000 orders per second with a sub-millisecond latency. Hyde Park Global relies on a trading platform that can execute up to 300 trades per second. It is currently relocating its servers so they run close to exchanges in New York, a practice known as "colocating". "By colocating in New York," Afshar says, "we're able to take 21 milliseconds off our trades. In the past, 21 milliseconds was a trivial matter. Now it's pivotal." The computerisation of financial markets isn't a new phenomenon. In the 70s the New York Stock Exchange introduced the Designated Order Turnaround system, which routed orders electronically.
Throughout the late 80s and 90s the traditional open-outcry system was abandoned in favour of electronic trading desks across the world -- most famously in London in 1986, when the deregulation of the financial markets prompted a huge shift to screen-based trading. Since then increased market liquidity and technological advances have created ideal conditions for the spread of high-frequency trading. It now accounts for about 70 per cent of US equity transactions and significant fractions of trading volumes in many other markets.
In 2009, estimates of the total profits of the several hundred companies involved in such automated trading (out of roughly 20,000 firms currently trading in US markets) ran as high as $20 billion. But as greater numbers of companies invest in the systems, industry watchdogs and regulators are beginning to ask questions about the possibly damaging effects that high-frequency trading might have on markets and the wider economy.
Most alarming is the parallel that's being drawn between the unregulated trade in derivatives -- which some economists thought would make markets more stable and efficient -- that triggered the collapse of 2008.
Could high-frequency trading trigger the next great financial catastrophe? HFT firms argue that their strategy is centuries old and beneficial to markets. This is particularly the case with high frequency "market-makers", such as Getco and Tradebot, which provide a buying-and selling service to others. Market-makers provide liquidity by ensuring that other participants can always find a buyer or seller for any product.
They make only a small profit on each trade, but execute hundreds of millions of trades every year. "High-frequency market-makers step up and make shares available for purchase or sale at specific prices on a scale never before seen in securities markets," says Justin Schack, a vice president at Rosenblatt Securities. "This flood of orders helps to make markets as fair and efficient as possible." But market-making is only one of a number of popular HFT practices.
A growing concern is that many firms use much more aggressive strategies to buy and sell at a profit. The recent market meltdown raises questions about the safety of all this activity.
Andrew Lo, a finance professor at the Massachusetts Institute of Technology, says that economists and financial experts in general have very little understanding of "systemic" market risks -- risks that can't be tied solely to the well-being of any one firm, but which emerge from the dense webs of interdependence among many firms. This behaviour, he argues, applies to HFT in the same way it does to mortgage-backed securities. "If we've learned only one lesson from the recent financial crisis," says Paul Wilmott of Willmott.com, a commentator on the use of mathematics in finance, "it is that people like to copy each other when they see a profitable idea.
The problem with the sudden popularity of high frequency trading is that copying behaviour may increasingly destabilise the market." Wilmott fears that volatile events of this kind could become more frequent as algorithmic trading, including HFT, comes to dominate markets. Wilmott has run simulations of markets to explore how algorithmic trading strategies evolve as the software effectively learns to play a game against itself.
His models may be highly simplified, but they show that the stability or instability of such a market -- its propensity to collapse suddenly -- depends to a high degree on whether the strategies are highly diverse or dangerously similar.
Too many firms following similar practices could precipitate a market crash. Indeed, research by Lo suggests that automated trading has already exacerbated market upheavals. In two dramatic episodes during the second week of August 2007, several prominent and successful US hedge funds suddenly suffered enormous losses in a few hours. The collapse - known as the "quant meltdown" -- has been traced by Lo and colleague Amir Khandani to a deadly feedback loop between hedge funds following very similar strategies.
These funds, it seems, were all heavily invested in similar assets and strategies. They were also highly leveraged -- borrowing heavily from banks and brokers to increase their positions and to amplify their gains if the market moved as they expected. When one hedge fund had to meet an unexpected margin call and pay back part of a loan to a bank, it naturally sold some of its assets to raise cash.
This drove down the assets' value, lowering the net worth of the other funds, so they also faced margin calls and also had to sell the same assets. The result was a death spiral of selling and margin calls.
This spiral, Lo points out, was driven by the decisions of human beings, not computers, but he and Khandani have found evidence that its violence may have been exacerbated by HFT activity -- or rather, the sudden lack of it.
Whereas some of the largest market-making firms are "exchange designated" -- and therefore obliged to continue making markets even in turbulent conditions -- others are not. It was their sudden flight from the market, Lo says, that appears to have made it more difficult for the hedge funds to sell assets, amplifying the collapse in those assets' values. "High-frequency trading may have increased the loss in liquidity during the second week of August 2007," Lo says.
Given these concerns, the US Securities and Exchange Commission is considering new regulation for high-frequency trading. It issued a document in January soliciting comment. "The equity markets in recent years have seen extraordinary changes," says SEC chairman Mary Schapiro. "We have to assess how changes in the market are affecting investors." Not before time, says Lo. "Today's rules weren't set up for high-frequency trading but the technology has gotten ahead of the rules."
Some commentators suggest an outright ban on HFT activity, or perhaps a limit that would prohibit trading taking place more quickly than, say, one second. The SEC is considering such ideas, and also exploring other options to make markets fair and safe. But their possible destabilisation isn't the only reason that high-frequency trading is under attack.
Traditional trading firms in the US and Europe argue that they're operating at a disadvantage because of the cosy relationship between high-frequency firms and the local exchanges. Many high-frequency traders pay exchanges such as the NYSE or Nasdaq for privileged access to fast data feeds or to colocate their computers in rooms at the exchanges.
Critics say these players have an unfair advantage as they have advance knowledge of other insiders' intentions and a split-second lead in trading mechanics. Joe Saluzzi of Themis Trading in Chatham, New Jersey, argues that special news feeds that exchanges sell to high-frequency traders -- such as the Itch, a direct data-feed interface for Nasdaq's options market -- create an unfair market in which some people, for a fee, are favoured over others. "Most Wall Street professionals assume that everyone sees the same price quotes and market data at the same time," Saluzzi says. "That was the tradition, but it is no longer the case." Others say opponents are acting like Luddites. "Objecting to [them] is like reasoning that it's unfair that a car has an advantage over the horse," says Allen Zaydlin, CEO of InfoReach. "A computer has an advantage over the telephone, sure, but there's nothing unfair about it. Anybody can do this."
One practice that the SEC is set to ban, however, is known as "flash ordering". By law, an exchange such as Nasdaq has to make the orders placed on its exchange -- by clients aiming to buy or sell certain amount of stock at a desired price, for example -- immediately available to all other exchanges. This is supposed to ensure that all market participants have access to the same information about pending orders. But some exchanges have begun to give their own high-frequency customers a glimpse of new orders -- lasting a few hundred milliseconds -- before making those orders truly public.
Critics say this gives those traders advance information on which they can profit, again in a matter of milliseconds, at the expense of other investors. One common outcome is that the client who placed the original order may have to buy the stock at a higher price, or sell at a lower price, than had there been no flash order.
Flash orders only account for an estimated two to three per cent of stock trading, but the practice has become more attractive as exchanges compete to attract high-frequency traders. "There are lots of good questions about whose seeing what, when, and what is fair," says economist Blake LeBaron, an academic who studies high-frequency trading at Brandeis University in Boston. "An exchange or other trading firms front-running their own clients is a really bad thing. It's really equivalent to stealing from them."
So why has the SEC, or the Financial Services Authority in the UK, not yet set out clear guidelines regarding high-frequency trading? The trouble is that there's no definitive science to prove that it makes markets dangerous. But neither is there data to prove it's safe. Jean-Phillipe Bouchaud, an expert in quantitative finance and cofounder of Paris-based hedge fund Capital Fund Management, notes that no one predicted the quant meltdown in advance, even though the mechanism by which it happened, in retrospect, is really quite simple. "The human mind," says Bouchaud, "just isn't very good at imagining the possibilities in advance."
Mark Buchanan is a science writer and former editor of Nature who specialises in network theory. His most recent book is The Social Atom(Bloomsbury).
This article was taken from the April issue of Wired magazine. Be the first to read Wired's articles in print before they're posted online, and get your hands on loads of additional content bysubscribing online
This article was originally published by WIRED UK
