August 16 began like most mornings for George Sofianos, who arrived early at his office in Jersey City, New Jersey, for a typical day of monitoring the movement of trades through Goldman Sachs Group’s electronic network. As Goldman’s chief trading execution strategist, Sofianos is the designated house sage when it comes to algorithms and electronic trading.
Trading kicked off briskly with the opening bell, but that was merely a prelude to what would follow. Prodded by the subprime lending crisis, spooked traders joined a massive sell-off. Staring at the spiking volume numbers on his screen, Sofianos realized that algorithmic trading programs were facing an historic test. The flood of orders caused backups and delays across Wall Street. Goldman added servers to double capacity, as orders continued pouring in.
The New York Stock Exchange, which experienced its first 4-billion-share day just a few weeks earlier, saw a record 5.8 billion shares change hands on August 16. Despite delays that proved costly to some traders, systems stayed up and algorithms continued to execute orders -- albeit not as quickly as on a normal day. Sofianos breathed a sigh of relief, as did hundreds of hedge fund managers who rely on algorithms. Although the architecture was up to the task, trade results differed widely. Problems reflected not just the traffic congestion and overall slowdown but also the way traders selected or employed algorithms.
Execution algorithms are a standard part of most hedge funds’ trading practices. But relying too heavily on any one algorithm style without having the knowledge or ability to shift to a platform better suited to fast-changing market conditions can invite trouble. Traders need to know the differences between algorithms and the features that make them more or less attractive in differing markets. Traders who chose the wrong algorithms to deal with the summer’s volatility found themselves vulnerable. And in some cases traders who returned to the old-fashioned way of trading -- through a broker -- wound up with better results.
The other family of financial algorithms -- quantitative models that use mathematical formulas to analyze historical trends and identify trading opportunities -- showed deeper cracks. In several instances quant models misinterpreted the importance of the subprime lending crisis and the accompanying liquidity shortage. Hedge funds that were heavily leveraged found themselves forced into costly margin calls.
“Everybody trading quant models made lots of assumptions about historical correlations and valuations,” says Gregory van Kipnis, general partner in Tiedemann Investment Group, a New Yorkbased hedge fund firm. “Those models clearly broke down.”
Because quant models differ from one hedge fund to another, each fund must solve for itself any associated problems. The models are closely guarded secrets, so fund managers generally aren’t talking about their experiences. But they have been poring over data and financial results since August, trying to determine where their formulas worked and where they need adjustment, then backtesting new assumptions.
Execution algorithms, on the other hand, face a different set of challenges. More widely used, these algorithms are available through prime brokers and a variety of vendors. Westborough, Massachusettsbased research firm TABB Group estimates that execution algorithms account for roughly 20 percent of all buy-side trades today and will rise to 23 percent within the next few years. Buy-side algorithms, in turn, are a big part of the explosion in electronic execution, which now accounts for 60 percent of trading by all market participants.
The first algorithms to be developed were fairly simple, created to break large block trades into several small orders that don’t leave footprints in the market. The goal was to prevent others from spotting big trades and jumping in to buy or sell the same shares, pushing prices up or down to make a quick profit. Thanks partly to the widespread use of these algorithms, the average equity trade size on the NYSE, which stood at 1,500 shares in 1995, dipped to a low of 292 in August, before rising back to more than 300 in September, according to TABB Group.
Over the years, as more sophisticated algorithms emerged, brokers began offering their clients multiple choices. Hundreds of algorithms are available today, and most institutional brokers offer about a dozen. Despite the wide array, however, most algorithms are simply variations on a few core themes.
Beyond slicing and dicing, more sophisticated algorithms try to time orders to capture market price or volume averages, or they kick in near the open or close of the trading day. Most recently, algorithms that search for dark pools of liquidity (trading among institutional investors not done on the primary exchanges) to execute orders have been gaining popularity.
Choosing the right algorithm can be critical during unstable market periods. Basic, timed algorithms with slow, measured execution are not suited for volatile, fast-moving markets. Newer algorithms are increasingly flexible, allowing traders to adjust execution parameters and more subtly tweak how and when orders are executed.
Sofianos recently looked at thousands of trades that were executed through the Goldman system to examine how traders responded to the hectic summer market . He had expected that during the most volatile periods, traders would shift away from passive algorithms to more-aggressive ones or to broker-handled trades. Many traders reliant on slow execution algorithms froze like deer caught in headlights.
“We did not see a substantial shift toward more-aggressive algorithms,” says Sofianos, “and that was kind of surprising. Where there is high execution risk, you want to trade more aggressively.” Sofianos did find that although most algorithm users stuck with the versions they had in play before the market volatility, not everyone was complacent.
Adam Sussman is the author of “Modular Algorithms: The Growing Choice of Buy-Side Execution Strategies,” a report published by TABB Group in September that looks at the history, current usage and future direction of algorithms. While researching his report, Sussman interviewed several hedge fund traders and managers, many of whom told him that they did in fact shift to more-aggressive algorithms to deal with the market’s volatility. He found that the larger, more sophisticated hedge funds led the way -- but that even they ran into problems.
Despite shifting to more-aggressive algorithms, many fund managers still found their trades bogged down in August’s swamped broker networks, says Sussman. “The algorithms did what they were programmed to do, but brokers couldn’t handle the flow,” he says. “Execution reports began to slow down. Everything began to grind to a halt. The traders began to call brokers, saying, ‘What the heck is going on here?’”
The worst trading bottlenecks occurred during the early morning and late afternoon rush hours at exchanges. Traders tend to base decisions on the opening and closing prices of stocks; most algorithmic programs and models in use today are set to wake up at the opening and close of trading.
“Few models are based on intraday prices,” explains Ernest Chan, who runs Ernest P. Chan Consulting -- a New Yorkbased quantitative consulting firm. “Before the close these models start running, and if they are triggered by certain price levels, they will accelerate market trends -- strategies that manage billions of dollars.” Chan says many firms tend to rely unflinchingly on risk assessment models, so when such models interpreted the market’s August gyrations as a sign to sell, many fund managers immediately plugged large sell orders into their algorithms. But when prices rebounded, selling often turned out to be a costly mistake.
The trick to making money in August’s volatile market, posits Chan, was not just choosing the right execution algorithm but also correctly interpreting the warnings from risk models. “Most people tend to adhere to the risk model,” he says. “But I’m sure that some people overruled their risk models and chose not to sell.”
Execution delays had the greatest impact on high-volume traders who had big bets dependent on rapid order execution. Some firms bypassed the broker jams by using in-house electronic trading architectures to access exchanges and alternative markets directly. “The guys who built their own infrastructure and were using their own proprietary algorithms had an advantage,” asserts Sussman.
Quantlab Capital Management of Houston, an investment management firm known for executing hundreds of thousands of trades a day, had done exactly that. “We were able to deal with the extra volume because of our bandwidth and the robustness of our system,” explains John Huth, Quantlab’s director of portfolio management. Quantlab relies on a suite of execution algorithms and in-house trading architecture designed and built internally.
Huth says that one big help for active traders is algorithms with built-in “intelligence” to monitor market conditions and automatically correct execution methods. Quantlab’s algorithms appear to have functioned well in the recent difficult markets, he says, by making adjustments in execution methods as needed. “The main complaint people had was that their algorithms were going too slowly,” he says.
The high costs of building and running such systems, however, make them a luxury available to only the largest firms. TABB’s Sussman estimates that the construction cost of a basic execution algorithm architecture runs close to $1.6 million, while annual maintenance runs an additional $900,000. Still, enough firms have armed themselves with in-house electronic trading capacities to make brokers say they were a noticeable presence in August.
“We saw the more sophisticated groups executing directly into gyrating markets,” says Carl Carrie, global head of algorithmic products at JPMorgan Chase & Co. “They were more automated. They were using dark pools and other liquidity merchants to find liquidity.”
Still, smaller firms unable to write their own algorithms or build their own trading systems aren’t completely out of the game. Brokers and vendors offer guidance for choosing among available execution algorithms, and a handful of companies have even begun offering software that helps users identify the best algorithm for their needs and for different market situations.
Though some traders chose to completely shun algorithms during the volume crunch, most have since resumed reliance on them. Execution algorithms remain a highly efficient way to trade, and their usage and the share of overall trading they control are expected to continue to grow. There is even some evidence that their efficiency helped pull the market back smoothly and quickly when investors decided that the rapid price drop in equities had created bargains.
“The crisis in confidence wasn’t so much on execution, it was on investment decisions,” Sofianos says. “Fundamental issues caused the market turmoil, but executions themselves were not the cause.”
