Passing the test

Algorithmic trading has become increasingly important for dealers and investors in the equity markets, contributing to a strong rise in equity trading volumes over the past year. But how did these systems respond to the sell-off in equity markets in March? Clive Davidson reports

Until recently, trading algorithms had fairly calm seas in which to operate, and their generally positive performance encouraged a wide uptake on the buy and sell sides. Then came the volatility of February and March, with the New York Stock Exchange's (NYSE) new Hybrid Market trading system and the Dow Jones Industrial Average index both crashing under the strain. But reports from algorithm providers suggest their trading robots relished the turbulence, transacting record volumes while the market infrastructure creaked. Furthermore, the robots have taken to so-called dark pools of liquidity like ducks to water, sneaking in to find best prices and executing orders without leaving market ripples. Now, algorithmic trading models are migrating to fixed income and foreign exchange markets.

"Even in times of relative tranquillity in the markets, algorithms are used for stocks that are volatile," says Carl Carrie, head of product development in the electronic client solutions group at JP Morgan in New York. "So all the providers of algorithms have a lot of experience in dealing with heightened liquidity in the market-place."

Joel Steinmetz, managing director of execution services at Citi in New York, concedes that some users may not have been completely happy with the outcomes of their trading strategies, but says that was more to do with the way the markets were moving or inappropriate choices of algorithms, rather than any malfunction with the algorithms themselves: "In general, algorithms performed exactly as they were supposed to."

London-based Robert Boardman, head of algorithmic sales in Europe at New York-based agency brokerage ITG, which operates the Posit crossing network (dark pool), agrees. "Volatile markets are not necessarily a problem for algorithms, but a wise user will make a different set of choices in very volatile markets compared with passive or becalmed markets."

In choppy waters, investors can be far better off with algorithms, particularly the newer adaptive generation that are able to change tack in response to new conditions and take opportunities as they arise. "It used to be thought that algorithms should not be used in volatile markets, but that's no longer the case," says Carrie.

Indeed, JP Morgan transacted high volumes via its algorithms when global stock markets tumbled in late February. On February 27, the Shanghai Stock Exchange composite index plunged by nearly 9%, the S&P 500 dropped by 3.47% and the Dow Jones Euro Stoxx 50 fell by 2.62%. "When the markets are extremely turbulent, it is even more important to be efficient in trading. At those times, when lots of portfolio managers are rebalancing their portfolios to adjust to new market conditions, they are forced to use algorithms because it's too inefficient to do each of the individual orders manually," adds Carrie.

But while algorithms have been riding the recent volatility in the equity markets with some composure, changes to the market infrastructure itself, particularly in the US, are proving more challenging. The NYSE has introduced liquidity replenishment points with its Hybrid system, where the market is slowed to allow participants to enter new orders, while the Securities and Exchange Commission's new Regulation National Market System includes a trade-through rule that requires exchanges to route orders to the destination with the best price at the moment the order is entered. The overall effect is a lot more quote data. "So algorithms have to be much more responsive to the increase in quote levels, particularly for listed stocks in the US region," explains Carrie.

Meanwhile, Europe is gearing itself up for the Markets in Financial Instruments Directive (Mifid), a regulation that will create a single market for financial services in Europe, and which includes requirements for best execution. This will also increase the amount of data that algorithmic trading systems will have to respond to.

In parallel with these initiatives has been the rise of dark pools - trading venues where prices are not displayed. The US has at least 20, and although Europe has only two major dark pools at the moment - Posit and LiquidNet - Mifid is likely to spawn more.

"Dark pools, the NYSE going electronic with Hybrid and similar developments all support quicker interaction with the market, as well as the opportunity for the intelligence embodied in the algorithm to really show its value - and not just the fixed intelligence coded into the rules, but the learning intelligence that is now common with algorithms," says Steinmetz.

But dark pools don't all operate in the same way - in some the order matchings are scheduled, while in others they are continuous. Some are event-driven and others have negotiated matchings. Then there are differences in minimum order sizes, time periods in which orders might be effective, and so on. "This is great territory for machines," says Boardman. "They can know in milliseconds what part of an order to send to which particular venue. Human traders can't keep up."

But that is true only if the algorithms have been pre-programmed with knowledge of the individual characteristics of each market. "We have spent a lot of time trying to understand the benefits and efficiencies of each of the dark pools, and we take that into consideration as we route orders to them," says Carrie.

Knowledge of how markets operate is essential to algorithm developers. ITG can call on its experts that develop and operate Posit, while Citi and JP Morgan both own electronic trading platforms - Citi owns New York-based Lava Trading and JP Morgan owns Utah-based Neovest.

Another good reason for algorithm developers to be aware of the specific characteristics of dark pools is that human traders can try to use these to outsmart their automated counterparts. "Some dark pools have flow that is highly correlated with other dark pools, so if you are in one and get filled, an arbitrageur who crossed against you in the pool can sometimes guess that you are going to be in another dark pool immediately thereafter and trade in front of your order," says Carrie.

Sophisticated algorithm and dark pool providers are aware of the market gaming techniques that certain traders will attempt and have processes in place to counteract them. ITG has a team that monitors attempts at gaming in Posit, while JP Morgan and others use randomisation of venue choice and order size and price limits in their algorithms to outwit human gamers.

Market participants agree that gaming is an irritant rather than a major problem. But the ability of algorithms to hunt out profit opportunities across products and markets, coupled with growing cross-border trading, present new headaches for regulators. In simpler times, market surveillance involved the monitoring by an exchange of human trading activity conducted under its own roof. Now, the authorities must keep tabs on all algorithmic and human activity across all markets for which they are responsible. In order to do this, the UK's Financial Services Authority (FSA) has hired data analysis specialist Detica, based in Surrey in the UK, to build a £17 million electronic surveillance system.

Recognition of abuse

"The new FSA surveillance system has to have similar technology to that which traders use in order to identify and understand the profit and other opportunities that traders are trying to exploit, and to be able to recognise abuse," says Simone Asplen-Taylor, manager of market and regulatory services at Detica.

This means making use of real-time data monitoring and event alert technology that is fundamental to algorithmic trading. In addition, the surveillance system includes data mining and visualisation technology to help differentiate suspicious patterns of activity from the background noise of market activity, and a proprietary tool called NetReveal that highlights relationships between traders, products or other entities that might also signify market abuse. The first phase of the project is planned for November 1, to coincide with the introduction of Mifid.

Meanwhile, algorithms are spreading their wings. Citi, for example, is developing algorithms that can manage global portfolios. "The idea is that customers should have one single entry point to deal with global portfolios, where the algorithms themselves are intelligent enough to operate appropriately in the various markets," says Steinmetz.

For example, in managing a portfolio, the algorithm would use a different formula to execute orders for General Motors compared with DaimlerChrysler because of the different characteristics of the NYSE compared with Germany's Xetra market, explains Steinmetz. And algorithms are moving into fixed income and foreign exchange (see box). It's going to take more than a little rough water in the markets to knock algorithms off course.

THE MIGRATION FROM EQUITIES

Having been developed originally for the equity markets, algorithmic trading technology is now expanding into other areas such as fixed income and foreign exchange. Credit Suisse, for example, is employing innovative organisational methods and finding creative ways to exploit new and existing technologies and infrastructure.

"We are using really lightweight adaptive and nimble ways of getting things up and running quickly," says Ian Green, head of algorithmic and electronic trading for fixed income and forex at Credit Suisse. Forsaking conventional software development, which can involve groups of 30 or more people working to a formal process, Green uses small teams, often simply a trader and a programmer, to rapidly develop trading ideas into algorithms, and calls on other intellectual and technological resources in the bank as and when he needs them.

"It's about organisational efficiency and leveraging the assets that already exist," says Green, who pioneered the method himself when he returned from a year's sabbatical in 2006 and who in previous incarnations at the bank built state-of-the-art equities and derivatives trading infrastructures.

The first thing Green did on his return was sit down next to one of the bank's top fixed-income traders and explore the potential of relative-value algorithmic trading in the fixed-income markets. "We looked at what we had in equities and we thought how we could apply that," says Green. "We spent a long time working through the maths, through the specifics of the Eurex and Euronext.liffe exchanges, we looked at the transaction flow that we've got, and at various algorithmic techniques that we could apply, and came up with some things worth trying."

With the trading infrastructure already in place, including a trading model that guaranteed separation between client and internal trader order flows, it was a matter of back-testing the algorithms, trying them out and then augmenting the library of algorithmic strategies with new ideas from other fixed-income traders who lacked the platform to implement them themselves.

Some new technology was required, and the bank made judicious use of vendors, says Green. One product they introduced was real-time streaming data management software from Massachusetts-based complex event processing software vendor StreamBase Systems. Other critical resources were found inside the bank. The first is the expertise of seasoned traders. "The people that have the most sophisticated knowledge and the best instincts on how to enhance your execution are the people who have been doing it for 20 years," says Green.

And the bank cancelled lower priority projects to free resources. "We looked at projects and said 'if we really want to get excited about five things, is this going to be one of them? If not, let's stop doing it, and give ourselves some assets'," says Green.

Once he had the lean, rapid development prototype up and running on the fixed-income desk, Green moved on to the foreign exchange spot desk.

"People say that forex is like equities was five years ago," says Green. "That's partly true in that the trading venues for equities are faster, have better data infrastructure and are more naturally adaptive to algorithmic trading than forex venues." But this does not mean that forex trading is any less sophisticated. "The spreads on forex compared with equities are wafer thin, clients get very good customisation services and there are very sophisticated products. It also has the benefit of being an over-the-counter market. One of the challenges for us has been how to keep the relationships and service levels, while still being able to offer some of the efficiencies you get in a developed exchange-based model."

His time on the forex spot desk has been an extremely fertile source of ideas, says Green. In July 2006, Credit Suisse introduced algorithmic trading for its spot traders. It was enthusiastically received and the bank has now also rolled it out to clients. "Increasingly, our clients have an obligation to show that they dealt at the best price they could," says Green.

One traditional method of achieving this is for the client to use a multi-bank platform and to show that at a particular time they took the best price on offer. "But the price differences between a group of dealers at a particular instant in time are inconsequential compared with fluctuation of the currency over the course of the potential trading window," says Green.

An algorithm can be analysed to show where the execution fell within the trading range. "That's a more meaningful measure of price transparency," he says.

Credit Suisse now has a range of forex algorithms. "Everyone thinks algorithmic trading should come to forex and we've done it," says Green. And the bank has done it without hiring more staff - not for cost constraint reasons, but to "leverage the intellectual and technical assets we already have", says Green.