Traders with a sharpshooter's eye for alpha are looking to beat the competition using the latest and greatest formulas-from customizable and forward-looking algos to those that can interpret the news. Consultancy Aite Group estimates that more than 50 percent of trades will be executed via algorithms by 2010, and in this pursuit, investment firms are pouring more research dollars into algorithmic trading. A back room full of mathematicians churning out formulas is no longer the sole recipe for success. Firms are looking ahead to the future of algorithmic trading.
Of course, investment firms still factor humans into the trading equation. One look at today's trading floors proves this, but it has become obvious that a human trader, no matter how over-caffeinated, cannot execute trades at the split-second intervals of a computer and a trading algorithm. A report from Kimsey Consulting, released in April, estimates that a staggering 12,000 trading positions will be lost by 2012, with one-fifth of those positions attributable to productivity gained from technology.
With uncertainty on the horizon, one thing is sure: Today's traders are relying on algorithms to execute trades and are now using them to pursue alpha, measure risk, and navigate through a myriad of asset classes and global trading venues. Waters examines the new trends that are shaping this red-hot space.
FIRMS BE NIMBLE
To be able to turn on a dime, some investment firms are offering customized algorithms that can be used to fine-tune a trader's strategy. According to Ari Rubenstein, managing partner for Global Trading Systems, an electronic trading buy-side firm, using vanilla algorithms from sell-side firms just doesn't cut it for today's traders.
"We are seeing a need for customized algos. A few years ago, it was fine for sell-side shops to have a few algorithms that were exciting for their clients. Now, what is going to help are algos that can be customized in some fashion so that you can customize on the fly," says Rubenstein. With customizable algorithms, buy-side clients are looking for variables they can modify and customize to their trading strategies needs.
The formerly off-the-shelf algorithms can be highly customized to the type of execution that is needed, explains Rubenstein. "How we customize our algos depends on the needs of the portfolio manager. Let's say there were time periods of the day where you had to do all of your execution because that is how your quant-based system was developed. If I were using a typical sell-side shop algorithm, I would need to be able to look at those time periods, at the history of names in the time periods and be able to build an algo from a standard piece they are offering. That would help me maximize my alpha in the time period I am working on."
Rubinstein says that customizing an algorithm doesn't take the weeks and months a buy-side firm would expect from its sell-side counterparty. "It's not a big deal for us because our infrastructure is built for that. If you were running the algorithmic trading sales desk for a large sell-side shop, ideally you would want your buy-side customers to easily be able to get info on how algos would work for them and potentially save them money. This is opposed to having conference calls and months later having the sell-side shop build something that the buy-side client might or might not need anymore," says Rubenstein. He adds that buy-side clients would prefer if the turnaround time on a customizable algorithm were quicker than the slower delivery times of a larger sell-side firm that provides the trading algorithms. "The time to market is much better when it is faster," he says.
Algorithm providers agree. "For firms that use algorithms to execute large trades through their brokers, the requirements remain the same: low cost of execution and achieving a benchmarked price," says John Bates, general manager for Progress Software's Apama, an algorithmic trading solutions provider. He adds that more firms are using algorithms for alpha-generation often in a high-frequency trading context. "The most important properties we hear from such firms are that customization is king and the algorithms must work with the minimal amount of end-to-end latency. Customization is important because firms need to differentiate their offerings and beat their competitors," says Bates.
Equities remain a key area for algorithms, but foreign exchange (FX) trading is a growing target for tomorrow's trading algorithms. Bates says FX is huge because market fragmentation is also an issue, and high-frequency and execution algorithms continue to grow in interest. "In fixed income, algorithmic trading on markets like eSpeed and Brokertec is growing. Algorithms are used for real-time bond pricing and hedging. In futures and options, algorithms, like spread trading, continue to garner interest. Rising oil prices is one factor driving interest in energy and commodities algorithms," says Bates. He cites cross-asset class algorithms, which combine two or more asset classes. "An example is cross-border statistical arbitrage, using an FX feed to normalize prices of instruments listed in multiple countries," he says.
Algorithmic trading will face challenges in the near future, according to industry experts. "Algorithms struggle when they are incorrectly applied. Selecting an algo is still like choosing from a range of options on a menu because not all will be appropriate for the given situation," says David Easthope, senior analyst for Celent, a market analysis firm. "For example, a basic volume-weighted average price (VWAP) algorithm in a very illiquid stock could be a bad idea. while crossing or internalization could be a better idea," he says.
"Algorithms have been rapidly adopted in equities and increasingly for options of course, but less so for futures, fixed income, and so on," says Easthope.
MORE ANALYSIS, PLEASE
Along with a desire for even faster trading speeds, traders and portfolio managers are looking for algorithms with more incisive analytical tools. According to Rubenstein, some customized trading algorithmic solutions could be fine-tuned to pricing from ECNs. "A lot of times you use an algorithm from a sell-side firm and often there is one price attached to it, but the bottom line is that the all-in rate the sell-side shop is giving you doesn't take into account the relative advantages the ECNs are giving on pricing. It would be interesting if the algorithms could pass on some of those advantages to the buy side," says Rubenstein. "That could be a point of differentiation among the sell-side shops."
Buy-side firms are looking for algorithms that go beyond simple benchmarking and move into the issue of market fragmentation, says Celent's Easthope. "Algorithms must be able to seek out liquidity wherever it may be, whether it is dark pools, hidden orders on ECNs, alternative trading systems (ATSes), public markets, and so on. A good example is EdgeTrade's Find and Nail algorithm or Credit Suisse's Pathfinder," he says.
This notion rings true inside HSBC. "We do not see a dramatic change in the type of product asked for, which is perhaps a reflection in some parts of 'product overload' from sell-side firms," says Kevin Bourne, the head of e-equities inside the global banking giant. "We do have regular requests for algorithms across emerging markets, as you would expect at HSBC."
Bourne predicts that the ubiquitous nature of algorithmic trading will lead to outsourcing of developing and managing algorithmic trading. "Algo product development is now much more a reflection of proven client demand. Three years ago, investment in new IT for such products was occurring at a substantial pace. I think some banks have now decided to take a closer look at the real cost of this technology and how they can obtain return on investment," he says. Bourne adds that outsourcing is becoming more common and that smart order routing is a candidate for outsourcing. On this basis, commercial advantage narrows quickly with universal suppliers, he says.
ALGOS IN THE NEWS CYCLES
Investment firms have been experimenting with using news algorithms, but they are still cautious, says an official from Dow Jones. The vendor offers a service that produces news, headlines and alerts in a format that can be easily attached to trading algorithms.
"At conferences, people talk about dark pools and how to modify your algorithms to take advantage of different pools of liquidity. A less discussed issue is the use of news in equity trading. This is a more complicated and difficult environment to implement but it has great potential value in generating alpha," says Alan Slomowitz, director of institutional product development for Dow Jones.
"We know implicitly that news impacts the markets. People read the news and it impacts the market. The question is how to take that implicit knowledge and make it explicit and quantify it and build it into a model. We have seen, through academic research, that when an event hits, it gets incorporated into the price and if you know how the model works, you can make money on those types of trades," says Slomowitz. For example, when Boeing lost a crucial military contract to build refueling aircraft for the US Air Force, the news had an impact on the aerospace community. "The challenge is getting the right news in the right time and in the right format," says Slomowitz.
While portfolio managers want sharper back-testing, they are also on the lookout for forward testing. Global Trading Systems has released a program called the Darwin Optimizer. Rubenstein explains: "It looks forward. Instead of back-testing, it forward-tests different algorithmic mutations on current trading strategies we have to identify potential trends in how an algorithm might work better." He says this can be filed under "artificial intelligence."
In the near future, Rubenstein predicts that the next catchphrase will be "algos that learn how to trade" as opposed to formulas that look to the past to see how they achieved best execution. "The best way is to learn in real time. This is going to be the trend in three years, which is like 1 million years in any other industry," says Rubenstein.
When it comes to processing news, the next generation of algorithms require sharper critical skills, says Slomowitz of Dow Jones. "Another area in its infancy is determining the sentiment of the story. Is it a positive or a negative news story? Of course, everyone has their own opinion of it-a news event might be positive for this one entity and negative for that one. This is certainly an area that is in its infancy and there is a lot more work that needs to be done so that people can incorporate it into models that they trust," says Slomowitz.
Algos that can critically analyze news are ripe for exploration, according to Slomowitz. "There are some products out today that do this, but I get a sense that people are still wary of the idea. They want to spend more time studying and understanding it. At some point in three to five years, we might have enough understanding and tools and applications that clients would want to use to incorporate the sentiment they are looking for into their quantitative models," says Slomowitz.
Apama's Bates foresees the rise of algorithms that can create newer algorithms. "These so-called 'genetic' algorithms can be used to dynamically synthesize new algorithms. This may be through an evolutionary technique of running thousands of permutations of different algorithms and seeing which is the most profitable, using it in live trading and growing its properties, while killing off the less profitable," says Bates.
Bates says he is a big believer that algorithmic trading is an intellectual property business and algorithms are only as good as what is encoded in them. "Yes, algorithms can replace certain activities done by traders but that is enabling the traders to become much more productive. It allows them to focus on the higher value tasks and to avoid the manual order entry of the past," he says.