The Gateway to Algorithmic and Automated Trading

Buy Side Flocks to FX

Published in Automated Trader Magazine Issue 08 Q1 2008

Auto and algo trading techniques are being deployed by a wide range of buy-side firms in the increasingly crowded, but not always transparent, waters of the FX market. Chris Hall reports.

Buy Side Flocks to FXThe shotgun wedding between quick-on-the-draw automated traders and the venerable but slow-moving institutions of the FX establishment was a short-lived and tempestuous affair. Plenty of 'Wham! Bam!', but 'Thank you, Ma'am' was an uncommon courtesy as the new century introduced new trading technologies to the FX markets. The term latency arbitrage was coined to describe the highly-profitable strategies employed by hedge funds to exploit differences between the prices a bank would quote on different market channels. Banks retaliated, first by shutting off liquidity to hedge funds that persistently used high-frequency trading techniques to skim risk-free returns, then by deploying algorithms of their own to react more swiftly to price movements and provide liquidity to multiple electronic platforms.

Rather than precipitating a Cold War-style stand-off, a sense of mutual dependence has nurtured a policy of accommodation and entente cordiale. High-frequency statistical arbitrage still plays a part in buy-side activity in the automated FX market, but it is augmented by a wider range of strategies and time frames. Banks have broadened their perspective too. The investment required to handle large volumes rapid-fired from automated trading systems has opened up new revenue streams, with traditional asset managers eager to use execution algorithms and other advanced trading tools to bring their FX execution capabilities up to the standards already achieved in equities.

The resultant leap in volumes and, therefore, liquidity stands to benefit all. Total FX turnover reached USD 3.2 trillion in April 2007, according to the Bank for International Settlements' Triennial Survey, a 71 per cent increase on 2004. Much of this is attributed to a huge rise in buy-side participation. Transactions between broker-dealers and buy-side financial institutions (hedge funds, mutual funds, pension funds and insurance companies) have more than doubled in three years. Trade volumes between brokers and buy-side institutions (40 per cent) are now almost equal to inter-bank transactions (43 per cent). The FX market now resembles less a warring marriage than an online speed-dating service, where the needs of a diverse range of parties can be matched.

Douglas Connor,Maia Institute
Douglas Connor,Maia Institute

"For us, a trading strategy is only good if it makes money pretty consistently over the last eight or ten years of data."

Understanding FX market dynamics

Although not a newcomer to the space, the Maia Institute is an example of the breadth of firms trading FX on an automated basis. The Monaco-based institute does no actual trading, rather it conducts scientific research aimed at predicting how intelligent agents interact within complex systems, including, but not exclusively, the FX market. The institute's research shapes the design of automated trading strategies deployed in the FX market by its associate, Rudyerd. "We regard the FX markets as complex systems which are primarily human driven and much of our work has been focused on understanding the market dynamics," says Douglas Connor, Maia's Director General, who founded the institute in 1991.

Rudyerd's trading strategies are informed by observation of the dynamics between buyers and sellers. "We do not try to exploit the instant reaction of the market to external events such as news or figures. The key is a deep understanding of the underlying forces which drive market behaviour during the absence of external influences," says Connor. The firm deploys multiple trading strategies that predict short-term market movements (anything from a few minutes to a matter of hours) in major currency pairs. Trading is 100 per cent automated from trading decision to execution; humans only have the authority to close a trade in the event of a market interruption. Trading strategies only trigger trades when price movements replicate the complex multidimensional patterns that the strategies are looking for. "If the markets are too unpredictable, the strategies shouldn't generate trades," says Connor.

Access to reliable data is critical to Maia's research-led approach, but the path has been long. Because non-banks' options were severely restricted in the early 1990s, Maia built its own database by manually collecting and inputting data from three voice brokers between 1995 and 1999 and was among the first non-banks to subscribe to EBS market data in 2004. "The market models that we built on that original 1990s data are still applicable today," says Connor. These large historical data sets - and the software tools created in-house to analyse them - are used by Maia in both the development of market models and the development and backtesting of trading strategies. "Some people may run strategies only as long as they work, then replace them with new ones fitted to the latest data. We aim to model the underlying mechanisms and consequent dynamic characteristics of the market and build strategies that stand the test of time. For us, a trading strategy is only good if it makes money pretty consistently over the last eight or ten years of data," says Connor.

Though satisfied that Maia has developed a rare level of understanding of the factors underlying the FX market, Connor still believes the exploitation of that expertise remains a work in progress. For example, the firm is looking to introduce an element of intelligent adaption to individual trading strategies as well as greater levels of collaboration with other financial market participants. "We're still scratching the surface in trading terms," he says.

Competition breeds diversity

Long-term players such as Maia are being joined by a range of buy-side firms that see the automated FX space as a simple-to-access source of uncorrelated alpha.

Dr Photios C Harmantzis, a member of currency manager FX Concepts' investment research team who specialises in multi-asset class models, says the unique characteristics of the FX market pose particular challenges for model developers and quantitative researchers/analysts. "In FX, the investment universe is relatively small compared with other asset classes. In US equities alone, you have liquidity in thousands of stocks, but in FX you might be limited to major crosses due to a lack of liquidity in emerging markets currencies," he says. "The benefits of diversification are not as great as in equities, and both experience and research shows that the FX market is harder to predict."

Established in 1998, FX Concepts manages currencies for institutional investors using overlay and absolute return strategies based on a combination of quantitative modelling and technical forecasting; around 90 per cent of its trading is conducted by systematic means and 10 per cent is discretionary. Dr Harmantzis, who is responsible for developing model-based investment strategies at the firm, says not all the automated trading models deployed by market participants are specific to the FX market. "Some strategies are also common to other asset classes, such as price momentum, trend-following and relative value, but others are more specific to FX, such as the carry trade based on interest-rate differentials between currencies," he says.

As in other asset classes, higher levels of participation are increasing competition. "FX might be the biggest and most liquid financial market, but the fact that more people are using similar types of strategies makes active FX management more challenging," says Dr Harmantzis. "High-frequency strategies that were profitable a few years ago are finding conditions more difficult these days. All the participants are catching up technologically and inefficiencies that were being exploited are no longer there." Strategy development is becoming an increasingly intensive exercise as competition reduces the effectiveness and therefore the shelf-life of trading models. "In the last few years, many firms were effectively looking to generate 'free money' via carry trades or selling volatility. But since August 2007, carry trades have been unwound and volatilities are trending upward," he says. "Simple strategies that have been popular are not as profitable as they were."

Casting a wider net

With statistical arbitrage opportunities increasingly restricted to all but the best equipped, Thomas Parry, Director of Algorithmic Trading at Plimsoll Capital, an FX-specialist alternative asset manager, says automated trading strategies are becoming more diverse. "The mean time needed to capture some arbitrage opportunities has reduced from 12 milliseconds to around eight in a matter of three or four months. It's getting a lot faster for the simpler models, but the reaction patterns suggest a lot more complexity in models being deployed overall," says Parry. "The simpler the better is a great principle for model building, but more complex models can be more robust as they're based on data collected over longer time frames, they have higher levels of significance statistically and use more parameters."

Capital allocation, says Dr Harmantzis, is the appropriate response. "There isn't just one particular time frame over which a model works, hence the use of portfolio models that work on different time frames and different styles," he says. "Although some model types, such as valuation strategies, don't really make sense intraday, others such as price momentum strategies, can work on monthly, weekly and intraday bases." Furthermore, by combining the characteristics of different strategy types, the circumstances in which models are effective can be increased. "People should be more open-minded in how they combine models," says Dr Harmantzis, "For example, using trend model with a volatility overlay enables firms to trade according to prevailing trends and levels of volatility."

Many firms are looking to harness today's greater computer power and data availability across different asset classes to develop new models, particularly in the statistical arbitrage space. Dr Harmantzis argues that cross-asset activity has been accelerated by the reaction to the US sub-prime credit crisis. "After one shock to the credit market, the resulting lack of liquidity very soon affected the equity, bond and FX markets. When you're building a model in a particular asset class, it makes sense to see how it relates to trends in other asset classes," he says.

Transparent differences between venues

The growing use of cross-asset models by buy-side firms in the auto FX market has already translated into new demands on trading venues. "There's growing interest in identifying correlations across markets; customers are writing models that take inputs from news events and/or other asset classes," says Steve Toland, Head of FX Sales EMEA, ICAP Electronic Broking. "This might be models that take account of the recent strong correlation between the S&P 500 and USD or models that generate FX trades based on news-based inputs."

Trading venues have long been important allies of buy-side firms in enabling first high-frequency and now increasingly sophisticated automated trading strategies. Although most trading venues now abide by the principle of price-time priority when matching bids to offers, there are a number of nuances that can make a significant difference to model and execution algorithm performance. For example, a number of platforms do not permit orders to be modified, so that if an order is reduced in size following initial placement, its place in the queue is lost. In addition, some platforms give priority to the non-displayed portion of an order, while others restrict the number of orders that can be placed within a specific time period or place limits on cancelling orders within ranges around the market price.

The choice of FX execution platforms expanded in 2007, with the launch of FXMarketSpace by Reuters and CME Group and FXall's Accelor, designed specifically for the anonymous, automated and algorithmic flow from banks and buy-side institutions. The launch of a venue driven by an anonymous central limit order book by a firm known for its relationship-based FX portal for asset managers and corporates reflects a significant shift in market behaviour, with FX alpha-generation almost exclusively pursued via anonymous venues by all types of market participant. Accelor is mainly used by hedge funds running automated trading strategies and banks sourcing liquidity to offer positions to clients. Mark Warms, Global Head of Marketing, FXall, says matching rules, speed and market data are the key differentiators for automated and algorithmic traders. "Some venues do not honour price-time priority. Others don't have the technology to ensure a low-latency environment, which adds risk for those running automated or algorithmic trading strategies," he says. "Accelor provides data on every order that is placed and removed from the book, but that's not necessarily true in other markets." Maia's Connor notes the rise of new trading venues with enthusiasm insofar as they have capacity to bring greater transparency and new pockets of liquidity to the market, but questions how far transparency should go. "Some trading venues are going for maximum transparency while others are allowing banks to put prices in without being seen. But too much transparency could hurt the market-makers and liquidity would dry up," he says. "Through participants' preferences for the venues' different approaches, the market will find a balance," he says. High volumes and increased competition is fuelling rapid functionality developments to grow venues' share of the automated and algorithmic FX market. EBS, the former inter-bank platform now owned by ICAP, has recently reduced latency both by establishing matching engines in each of the three major time zones and refining the algorithms used to match deals. Processing times for in-region trades have been reduced to five milliseconds (from the customer's decision to aggress to deal confirmation), while the global average deal time across its 'EBS Spot Ai' API is 65 milliseconds, with further improvements planned. In addition, automated trading clients can use EBS Lab, a testing environment, for testing prior to using models in the live market. "Programming and testing support are also becoming more important, because clients want models to work properly first time," says ICAP's Toland.

Thomas Parry, Plimsoll Capital
Thomas Parry, Plimsoll Capital

"We're now using strategies that simply weren't viable previously."

FX algos: build or buy?

The range of execution platforms vying for automated FX order flow is forcing many firms to place increased emphasis on advanced execution techniques to support alpha generation. US-based Plimsoll Capital, for example, has recently implemented new adaptive, order-routing execution algorithms that take into consideration a range of performance metrics from execution venues, including price, latency, fill rates, spreads, range of participants and speed of confirmations. "Ultimately, it's about maximising the probability that when your order reaches the destination the quotes and the prices are the same as when the order left your system," says Plimsoll's Parry. The task of developing the algorithms is complicated by the fact that Plimsoll's managed accounts use a range of different strategies and a combination of seven execution venues. "Two accounts may both use Currenex but one uses Hotspot and the other trades with a single-bank platform, so we might end up sharing portions of orders between platforms and have to be able to allocate the liquidity back correctly," says Parry.

Plimsoll currently deploys two order-routing execution algorithms, which are parameterised by currency pair and execution venue, with one geared more towards improving on arrival price, the other to minimising market impact. "There's no one catch-all solution," says Parry. "It's about defining how you need to access liquidity and thinking about how your strategies can benefit from that, rather than obtaining the best price that one time." Parry estimates the impact of the new execution algorithms as a 1.5 to 2.3 basis point improvement per trade. "Although we need to do more work to optimise their performance, the impact so far has been huge," he says. "We're now using strategies that simply weren't viable previously."

While some firms prefer to develop execution algorithms in-house to meet their own specific needs, banks' FX offerings are witnessing increased demand from a range of buy- and sell-side clients, according to Jonathan Wykes, Head of Advanced Execution Services FX, Europe at Credit Suisse. In the second half of 2007, the bank rolled out a range of execution algorithms to FX clients that leveraged its existing experience and knowledge within equities with the latest ideas and technology from FX. "We are developing FX-specific strategies, but many client requests in the FX market can be filled by adapting the algorithms already used by equities clients, based on our core 'headline' strategies," says Wykes.

For Credit Suisse, one key aspect of developing effective FX execution strategies was the ability to access multiple trading venues. Part of the order-routing process involves 'heatmapping' to identify which venue has most liquidity at a particular price and at a given time. "The algorithm continually measures the probability of execution across venues. It may be that one execution venue has the better price in cable, but for some reason Currenex is seeing more executions. So the algorithm might send the bid or offer to Currenex, but also post part of the order on the other execution venue. The order can be re-segmented very quickly in response to execution data," explains Wykes. This functionality means the algorithm can re-allocate rapidly between FX trading venues according to how orders are being filled.

Jonathan Wykes, Credit Suisse
Jonathan Wykes, Credit Suisse

"… many requests in the FX market can be filled by adapting the algorithms already used by equities clients, …"

Universal FX appeal

Use of automated and algorithmic trading tools is rapidly becoming standard practice for a wider range of FX market participants. A recent report by Tabb Group1 provided further evidence of increased demand for execution algorithms, from traditional asset managers. Senior Consultant Laurie Berke says the motivation stems largely from the need to reduce trading costs. "Global long-only equity asset managers are bringing trading and active management of FX exposures in-house. The path has been well-worn already by hedge funds looking for new and alternative, i.e. non-correlated, asset classes in which to generate alpha," says Berke, who estimates that 9-12 per cent of US-based global equity asset managers have begun to utilise sell-side provided FX algorithms.

Although large buy-side institutions are looking to replicate the advances in execution efficiency made in the equities market, their FX tool kit has yet to develop to the same extent. "A trader at a large US-based global asset manager told me, 'If I had the tools of the guy sitting next to me in equities, I would use every one of them in FX'," says Berke who nevertheless asserts that the pieces of the puzzle - including connectivity to multiple price sources, FX-capable trading technology, access to historical data, performance benchmarks - are now falling into place. Berke estimates that while only 15 per cent of third-party EMS (execution management system) vendors were able to support algorithms in 2005, more than two-thirds will have the capability in 2009. "But the middle and back office need similar levels of investment and upgrade in capabilities to manage, allocate, clear and settle these transactions," she says.

A wider range of buy-side firms are looking to automate alpha generation as well as execution. "A lot of small hedge funds, with perhaps EUR 500 million AUM and a three-to-four-year track record, are now interested in systematic trading as a way of broadening investor appeal, but don't necessarily know where to start," says Peter Klein, Head of Saxo Bank's London Office. Saxo Bank has also developed a systematic trading platform to put sophisticated trading and modelling tools in the hands of a diverse range of banks, hedge funds and proprietary trading operations. Soft-launched in 2007, TradeCommander allows users to programme, test and execute trading strategies in a single software package that includes front- and back-office functionality and risk management tools, as well as providing the historical and live data feed handling and warehousing capabilities required for effective automated trading.

At first, TradeCommander will only execute FX trades in the 166 crosses that Saxo offers, but users will begin to be able to automate trades in other asset classes throughout 2008. "The variety of FX trading strategies being automated is enormous, from simple ideas based on price comparisons between different bars or time points to more sophisticated strategies. Now any trader can automate a trading idea regardless of programming background," says Stephan Martinussen, Head of Global Solutions, Saxo Bank.