The proportion of trades now executed by algorithms is a movable feast depending on who is asked. A buy side desk may say it uses algorithms for 30% of its trading, counting only the orders it handles using direct market access or other electronic broker pipes. The other 70% of orders are called in to brokers, where the sell side trading desk will likely enter the flow into algorithms for execution. “Most US equity trading now uses some sort of algorithm,” says Scott Daspin, a managing director in the global execution group at ConvergEx. “The average block size continues to fall and appropriate execution requires a trading tool.”
Growing confidence in algorithms has encouraged buy side traders to exploit more complex routines. While some players still rely primarily on VWAP—quantitative shops doing mass optimisations of market neutral trades, for example—this long-time favourite has given way to implementation shortfall routines designed to minimize market impact. Traders specify the degree of urgency and the algorithm tries to optimise execution within that time frame. Based on measures of liquidity in the name and where the liquidity is concentrated, the algorithm will select the best routing among venues and decide whether and when to cross the spread to obtain a fill. “Clients are moving toward implementation shortfall as their primary benchmark,” says Daspin.
Average investment holding periods have come down in recent years, which has reinforced the focus on implementation shortfall. The shorter the time horizon, the more market impact costs affect the expected return on the trade. If a portfolio manager expects a 5%-10% uptick in price from a positive earnings announcement next week, the difference between 50 basis points (bps) and 150bps in market impact matters more than for a stock expected to rise 30% over two years.
For sensitive trades that are not urgent, traders may prefer a dark aggregator algorithm designed to tap liquidity only in dark pools where the footprint of a large order is harder to detect. Some dark pools are darker than others, and some admit participants whose activities may be toxic to large orders so traders can exclude certain venues or order types on a particular venue. Jeffrey Bacidore, head of algorithmic trading at ITG, has seen attitudes toward dark aggregators evolve, too. At first, traders would designate where they were and were not willing to trade, but now they take a more nuanced approach. “Shutting a dark pool out completely means there is absolutely no liquidity in there a trader ever wants to participate in,” he says. “That can’t be true.”
ITG and other providers have built more sophisticated algorithms that expose bigger size in clean pools but still show some interest, albeit with stronger safeguards against gaming, in more suspect pools. The buy side does not have the resources to monitor every venue in detail, which has led firms to lean on brokers to deliver an acceptable end result. “Brokers have to justify their decisions and provide good performance,” says Bacidore. “Clients find it hard to stay on top of the landscape. They have outsourced that to brokers and hold us accountable.”
The buy side learned long ago that while brokers always claim to put clients’ interests first a broker’s own interests will take priority if the client suffers no harm, at least in theory. In the spirit of “trust, but verify” the buy side is demanding more transparency about how algorithms route orders and where they are filled. Convergex has just opened its kimono through a Web portal on which clients who enter a ticker symbol and size can see a forecast of the expected market impact, how long it will take to complete the order, where the trade will route and where fills are expected. When clients enter a live order, they can see in real time where the order goes and the fills come from.
“When we demo this technology to people we don’t know, they fall off their chairs,” says Daspin. “We can practically see them reaching for the phone to ask their existing brokers how orders are routed.”
The degree of transparency ConvergEx offers allows buy side traders to tweak their execution strategy based on hard data about which venues give the best fills in a name. Sometimes it requires just a change in the parameters entered into the algorithm, but Convergex will customise the algorithm if need be. “The beauty of transparency is that people can make the algorithm exactly what they want, which is not the same thing for everyone,” says Gary Ardell, head of financial engineering and advanced trading solutions at ConvergEx.“Transparency helps clients get the right tool for the right job.”
The heightened transparency may tax the capacity of some buy side shops to make good use of it, however. Paul Daley, head of product development at SunGard’s Fox River Execution Solutions, says many clients struggle with the sheer volume of data generated in the full routing disclosure his firm provides and prefer to rely on monthly summaries instead. The snag? The summaries only includes trades done through Fox River, so users cannot compare the results with trades done through other brokers who do not offer similar transparency.
Clients who use the complete data dump can see where orders went, whether they were ever routed from one venue to another, how they were executed and whether they took or provided liquidity. “Over time, people are getting more into the logic of why a broker went to a particular venue, not just where it went,” says Daley. “People will use the tools and get their hands around the data.” He expects buy side trading desks to hire quantitative analysts with a grasp of trading who can use their programming skills to mine the data and suggest improvements in how the desk interacts with the market.
The buy side trader’s role continues to evolve from the jumped-up order clerk of yore toward equal partnership in the investment process. Traders don’t have to watch the market all the time any more; they can focus on higher value-added tasks like picking the best execution strategy and leave implementation to the machines. “A human does not have to look at the screen, see the bid move up a penny and decide whether to cancel and resubmit the order,” says Bacidore at ITG. “The algorithm has already done that if it makes sense. The trader looks at the objective and works more closely with the portfolio manager behind the trade.”
The nature of product development for algorithms has changed, too. Ten years ago, Bacidore says the main concern was to ensure the routines were robust and would not break down or go haywire. Today, those safeguards are a given and developers spend more time figuring out how to source liquidity as efficiently as possible. They also know other technology-savvy market participants like high frequency traders will try to reverse engineer or game their designs, a constant threat to buy side clients. “We have to have cutting edge technology,” says Bacidore. “We must be as smart and efficient as the best people in the market if we are to deliver good results to our clients.”
While technical improvements in single name algorithms will continue, the bigger challenge is to perfect algorithms that can handle baskets of stocks. It’s a daunting task: not only must the algorithm process data on all the individual names but the trading in one name affects how other names are traded. In a market neutral (equal dollar amounts to buy and sell) basket of 1,000 names, for example, the algorithm must maintain balance so that buys and sells don’t run too far out of whack. “The algorithm takes into account portfolio level objectives and constraints,” says Bacidore. “It comes at a cost, though—they can’t be too dynamic.” If a block showed up in an illiquid name on an institutional dark pool, a human trader might grab it but the algorithm might not because a large fill would unbalance the basket.
Another difficulty is the lack of industry consensus about how basket algorithms should work. The objective is clear: to minimize risk and maximize return on the trade—but opinions differ over what that means in practice. The uncertainty has hampered development efforts, according to Daley. Fox River could build an algorithm that made sense to its developers but if clients reject the logic it would be wasted effort. “We all agree what a VWAP algorithm is,” says Daley, “but we don’t necessarily agree what a basket algorithm is. There is a tremendous amount of unsatisfied demand in that space.”