After taking the equities market by storm, algorithms are beginning to surface in the foreign exchange (FX) market. The opportunities for fast and effective electronic trading are myriad in the FX market, which turns over $2 trillion in trades each day. The spot FX market is traded 100 percent electronically among banks, and, now, the most-adventurous buy siders, dissatisfied with stale prices on bank-to-customer platforms, are firing algorithms at the dealers.
Currently, only about 3 percent of FX trading volume is executed using trading models, according to Larry Tabb, CEO and founder of industry research firm TABB Group. But that number is growing, he says, and it is getting a boost from developments at trading venues such as Electronic Broking Service (EBS) and Reuters.
Quantitative trading firms, hedge funds and currency overlay managers - whose investment objectives require them to trade FX frequently and to seek deep liquidity - now demand to see the same narrow spreads that the banks afford each other. They also want to participate in the interbank market, rather than depend on the banks' quotes. In order to be competitive with the banks, some of the funds have written algorithms that capitalize on minute changes in price direction. To serve its own need for greater liquidity, the interbank market has welcomed this development.
Hedge Funds Join In
EBS, the interbank dealing platform for spot FX trades, recently opened up its application programming interface (API), Spot AI, to hedge funds, allowing them to deploy trading models directly on the EBS platform. Historically, funds were required first to submit an order to an EBS member bank for placement on EBS, or to request quotes from a bank's inventory, a practice that often resulted in a price that no longer reflected the primary market. Additionally, EBS' major rival, Reuters, plans to offer algorithmic trading in FX later this year.
Hedge funds, which are typically highly leveraged - they trade mostly on credit afforded to them by banks and prime brokers - still trade using the credit relationships that their banks and brokers have established with other banks and brokers on the EBS platform. This allows them to see a greater range of prices than would be available to them if they entered the platform using only their own collateral. Using the banks' and brokers' names on the platform also affords hedge funds anonymity, so that their strategies are protected, explains Bill Moran, president of EBS.
"In December 2004, we commenced the pilot with hedge funds," Moran says. "We opened up to the general community in March, and we now have 15 hedge funds trading on the platform and another 30 in testing." So far, all of the customers trading and testing on the platform have built proprietary algorithms - none have used vendors, Moran relates.
From December 2004 to June 2005, EBS' daily spot volume traded through its Spot AI interface has risen from $5 billion to $12.3 billion, representing 4.5 percent and 9 percent of EBS' overall trading volume, respectively. Trading through Spot AI mostly is model-driven, according to Moran. In order to maintain orderly markets, the service allows trade sizes of no more than 5 million units of the base currency, and only 20 orders can be placed at once, he notes.
One of the largest algorithmic players on EBS' Spot AI is quantitative trading firm Financial Labs, a Cambridge, Mass.-based brain-child of five Harvard graduates who initiated a fund in June 2003, but now trade entirely for their own accounts, according to Aaron Sokasian, the firm's CEO. The firm uses self-built algorithms, run off of stripped-down, green-screen Unix and Linux shells.
"We develop software that runs devoid of any human intervention, which interacts with bank APIs, EBS and Reuters, and runs server-to-server between here, London and New York City for redundancy, with no GUI [graphical user interface]," Sokasian explains. "Banks usually had bulky overhead and recycled EBS prices with a spread. We wanted to be directly in the interbank market. Now, we can operate on prices that have underlying interest and aren't just recycled - almost like a real exchange. That had been missing in the FX marketplace for the buy-side community."
In essence, banks, until now, have offered their customers the same currency pairs that they trade with other banks on EBS, attaching commissions in order to cover their operating expenses. But by the time the prices were offered to the buy side, the market had already moved, leaving funds like Sokasian's one step behind. In order to get ahead, shops like Financial Labs needed a way to interact directly with the banks and to anticipate the market's next move, explains Sokasian. Now that EBS has opened up its platform to model-based trading, hedge funds and other professional traders can trade on a level playing field with the banks, he says.
Financial Labs, which specializes in Group of Seven currency pairs and crosses, trades about $2.3 billion on EBS alone and up to $4 billion per day across all of its bank and platform counterparties, according to Sokasian - all via algorithms.
Financial Labs trades about 1000 times to 3000 times a day, Sokasian says. The firm's models are developed based on sophisticated statistical analysis done internally and based on the firm's own IP. They are structured so as to insure that the cumulative position the firm builds up in any given currency never exceeds its collateral. As an additional safeguard, the Financial Labs' network engineers have built a dashboard that allows managers to see the performance of each fund and network status in real time at a single glance. Sokasian says that the firm's models have become so well known in the marketplace that it now consults with dealing banks on how they can reduce network latency and build quicker models themselves. He also emphasizes that the firm "does not arb banks' APIs and that ninety percent of the firm's volume is done on Reuters and EBS."
Forex Forges Forward
Reflecting another, more conservative camp in the FX world, currency overlay manager Richard Gluck, of Trilogy Advisors in New York, trades FX to counteract any difference in currency value that transpires in the time between the purchase and sale of international securities. Model-based trading in FX has begun to catch on for the buy side because outsized equity returns are no longer a sure thing, and because managers can no longer afford to "take a 25 percent bet against the dollar in their portfolios," Gluck says.
This means large amounts of currency must be purchased, and, in order to protect firms against the price of a currency increasing once the market sees demand, these amounts must be divided into smaller increments that can be executed separately, before the price moves unfavorably, according to Gluck. That's where algorithms come in.
Gluck, who manages about $1.25 billion, says there are a few standard models in the FX marketplace - "flow" models, based on custodial data on currency holdings; "pure trend" models based on previous price movements; and "carry" models, which dictate the purchase of currencies based on the best interest rate return for a given time horizon. Trilogy runs models, but trades only once or twice a day, and thus doesn't have much use for frequently deploying algorithms on a platform such as EBS, Gluck relates. Buy-side users have to trade a tremendous amount of volume to get decent spreads and to justify the use of algorithms on electronic platforms, he asserts.
"We've developed our own models using data links to typical data providers, but we are not high-frequency," Gluck says. "What I also find is that I often get better prices on the phone in principal markets."
Because most FX trades occur in a venue where banks have discretion over to whom they show the amount and price levels of the currencies they hold, depth and transparency are not hallmarks of FX trading. This is one of the reasons algorithms seem to benefit only the most-intense traders at this point, says Harrell Smith, securities and investments analyst at Celent in New York.
"EBS is considering adjustments to show its depth of book and study its effect on the market," Smith says. "The potential effects to the market in how depth of book is shown are not as apparent as in the equity markets. Although spot FX is 100 percent electronically traded, people still are trying to figure out the dynamics of how market adjustments will play out," he continues. "How would an algorithm really negotiate a changing marketplace? I think it will be wait and see for a while."
One manager unwilling to wait and see is Richard Olsen, manager of Olsen Invest, a Zurich, Switzerland-based investment advisory firm that manages about $100 million for hedge funds and high-net-worth investors. Olsen, who founded online currency broker Oanda, builds his own algorithms based on intraday tick data, paying particular attention to the increases and decreases in volatility around market opening times and closing times in New York, London and Tokyo.
"Very few people use high-frequency data," Olsen says. "By and large, they use daily data, and try to prime the entry and exit points of trades. This is where we are today," he continues. "I am very confident that in the next five years [the use of] high-frequency data - meaning millions of price ticks - to create a decision [will be prevalent]. Today, there are daily models using moving averages, over the last 10 or five days." In Olsen's view, algorithms relying on high-frequency tick data almost exclusively are the province of hedge funds and quant shops, and, at this time, even large banks are skeptical of model-based trading.
The world of FX algorithms still resembles the Wild West, and it will probably be some time before algorithms are used by traditional money managers, such as pension funds or mutual funds, that primarily manage equity or bond portfolios and use FX to settle international transactions, according to most industry sources.
Howard Tai, currency and derivatives specialist at American Century Investments in Kansas City, says that buy-side users of FX algorithms are a very small group, largely because of the bilateral relationships that typify FX trading and benefit the buy side with services beyond transactions. Just as entering the e-FX market in the first place represents a psychological hurdle for most buy-side firms, algorithmic trading would represent a further separation from the sell side's apron strings, he suggests.
"There is a two-stage evolution," Tai says. "You first have to get people to think electronically in the simplest sense - you get them to place limit orders and be comfortable doing it themselves. That is a big mental hurdle most buy siders do not overcome," he contends. "It is easier to lay it in someone else's lap; that way, you have an excuse when it goes wrong. So you have to have that first evolution of people willing to take [the responsibility of direct e-trading] on their shoulders before algorithms make sense."
Standard FX Models
In the FX market, most money managers develop their own algorithms. This is at odds with the common practice in the equities market, in which money managers rely on the sell side or technology vendors to create trading models for them. Below are some of the standard FX models.
- Flow models. These are based on custodial data on currency holdings.
- Pure trend models. These are based on previous price movements.
- Carry models. These dictate the purchase of currencies based on the best interest rate return for a given time horizon.