Saying no to algos

A number of banks have launched algorithmic trading systems for foreign exchange, intended to provide more efficient execution for clients. But some question whether algorithmic models are actually needed in the highly liquid foreign exchange market. By John Ferry

Algorithmic trading has come in for plenty of criticism from US politicians, with many blaming the practice for the dramatic flash crash on May 6, which saw the Dow Jones Industrial Average drop by nearly 1,000 points before rebounding. But even though it has come under the microscope in the equity market, this hasn’t stopped banks from looking to extend their algorithmic trading services to other asset classes – specifically, foreign exchange.

A number of forex platforms have now been launched, with Barclays Capital, Citi, Credit Suisse and Deutsche Bank among those offering algorithmic trading capabilities to their clients. The approaches taken by these banks in some cases differ dramatically. Credit Suisse, for instance, has opted for an agency model, where the client is charged a commission and the order is filled at the lowest possible price at any one of a number of trading venues. Others have plumped for a principal model, where the client is charged a spread and trades directly with the bank.

However, some question whether algorithms are needed in the foreign exchange market at all. “It is not clear what the purpose of these algorithms is. I think the basic problem is a lot of the banks probably feel market pressure to offer algos but are not quite sure of the right way to do it,” says Sang Lee, Boston-based managing partner at consultancy firm Aite Group. The rush to offer forex algorithms is partly due to the fact certain clients, caught up in the hype of high-frequency trading, are asking for them, and partly because dealers don’t want to be seen to be falling behind their competitors, he adds.

One reason algorithmic trading may not be an obvious fit for foreign exchange is the liquidity of the market. Average daily trading volume in the global foreign exchange markets stood at approximately $3.7 trillion last year, according to figures compiled by Aite Group. “More than 90% of volume in the foreign exchange market is traded in 10 currency pairs. The level of liquidity in those currencies means clients are always getting good prices and liquidity,” says Mark Warms, London-based general manager of European operations at FXall, an electronic trading platform for foreign exchange.

Lee takes the same view: “When I talk to people actively trading foreign exchange, their attitude is they would never use an algorithm from a bank. The feedback is that the market is huge and very liquid, so why would they need an algo?”

Many participants say algorithmic trading is perfectly suited to the equity sector, which is significantly smaller and split across stocks, and where a large block trade can move the market. In the foreign exchange market, the case is less clear cut, they argue. “When these algorithmic trading models were developed in the equity market, a hedge fund trying to move its position could represent 30% or 40% of a day’s volume. In the forex market, if a fund is looking to move $1 billion against the euro, even though that’s a huge trade, it is a fraction of the day’s volume,” says Warms.

Dealers might argue their algorithms can enhance the execution of large trades – but this is a claim competitors might dispute. “The banks are interested in algorithmic trading so they can capture as much flow as possible. The pitch banks are making to customers is they can give them better execution. A competitor bank would say ‘why don’t you just call up my sales desk and we’ll offer you a far better rate and get you better execution’,” says Steve Toland, London-based head of forex sales at Icap, the interdealer broker.

So, what is the case for algorithmic trading in foreign exchange? The point of an algorithm is to enable clients to execute trades more efficiently. That could involve the slicing up of the trade into chunks and filling the order over time using a time-weighted average price (TWAP) algorithm, or executing a buy order as close as possible to the average price over a particular time horizon using a volume-weighted average price (VWAP) algorithm. TWAP and VWAP are designed to hide the activities of block traders, allowing them to transact without signalling their intention and moving prices.

That makes sense in the equities world, where trading primarily takes place on exchanges and a hefty block trade can easily get noticed by the rest of the market. In foreign exchange, however, the client typically trades directly with a bank. The dealer charges a spread and takes the other side of the trade and then transacts in other pools of liquidity to offset the risk. In this model, the bank is trading with the client on a principal basis.

And it is the principal versus agency model that is at the heart of the debate. In a nutshell, critics say any bank acting as principal has no reason to offer its clients algorithms. “If you have a very good relationship with your bank, then the bank shouldn’t be trying to push the market against you when you’re trying to do a larger transaction. So I think the algos the banks are offering are less important when the client is dealing with one bank as principal,” says David Hastings, London-based global head of sales at FlexTrade, a provider of algorithmic trading systems.

According to this argument, a bank should offer the tightest spreads possible when it acts as principal, so if it has an algorithm that improves on what it can provide on a stand-alone trade, then it is an admission it could be doing better in terms of liquidity provision and pricing in the first place, some claim. 

“If you’re trading with one bank specifically, then all you’re doing is trading on their price. So the bank has a responsibility to make the tightest spread possible for that institution, and what that bank does with its liquidity internally is down to them,” says Hastings.

Algorithm providers that trade on a principal basis with their clients beg to differ. They say their clients value TWAP, VWAP and other algorithms that slice into the bank’s own internal liquidity pool. So, a VWAP algorithm would be based on the bank’s own volume curves rather than the entire market.

Marek Robertson, global head of electronic distribution for Europe at Barclays Capital in London, insists it does make sense for dealers that see a large amount of flow to allow clients to access their liquidity pool using algorithms. He rejects the notion only banks that offer their clients direct market access to multiple liquidity pools should be offering algorithmic trading services.

“For those using a direct market access-type system, they may be told they can achieve the best deal by accessing many liquidity venues, but in practice more venues equals more impact than internalisation on a market-making engine such as the Barx electronic trading platform,” he argues. “When clients deal with us, they are dealing with probably the largest market-making engines on the Street, so we provide a unique liquidity offering for our clients.”

Barclays Capital offers a variety of algorithms through its PowerFill+ platform, including TWAP and VWAP, and does not intend to introduce a service that works on an agency basis. Others have taken a different approach. Citi, for instance, believes it has developed a platform that combines the two models.

“If you look at the spectrum of bank offerings from left to right, with agency execution out on the right and large banks offering execution tools that feed exclusively into the internal electronic market-making infrastructure on the left, then I like to think we sit somewhere in the middle. So we’re not militantly pursuing one strategy over another,” says James Dalton, Citi’s London-based director of foreign exchange algorithmic execution.

The bank’s CitiFX Intelligent Orders is designed to allow the client to take liquidity either from the wider market or from Citi’s own liquidity pool, depending on which algorithm they use. “The client can use a direct market access-type strategy, which is like an agency model, or they can use an internalisation-type strategy,” explains Dalton.

Its direct market access algorithm is called Silent Partner, which monitors the sector and uses historic data to work out how much liquidity can be consumed at any time without the market moving against the client. “Silent Partner responds to shifts in market liquidity in real time and uses each day’s data to refresh target buckets on a weekly basis. It is driven by a very granular matrix of market liquidity, which is refined according to the time of day or day of the week an order is running,” says Dalton.

Silent Partner uses smart order routing technology to execute foreign exchange transactions across Citi’s internal liquidity pool and other trading platforms, including EBS, Currenex, the Chicago Mercantile Exchange and Reuters. “Sometimes there is more liquidity going through our books than in the interbank market, but when the interbank market is firing and liquidity is at its peak, then there is an array of destinations for the client to clear risk,” says Dalton.

The algorithm creates a master order record and gradually begins to work passive orders into the interbank market. If target fill ratios start to lag, then it shifts into a more aggressive trading mode. Any crossing opportunities are filled at the market mid-price at the time the cross occurs, says Citi.

For those clients that prefer to trade with Citi directly there is the Ripple algorithm, which seeks to execute via liquidity drawn from the bank’s own client flows. Here, the bank internalises as much of the risk as possible and tries to beat the prices offered outside of Citi’s market. “Effectively, it works the order for the client inside the market bid-offer spread but without going to the external market,” says Dalton

Despite the criticism, some banks are convinced forex algorithms are here to stay. But there’s little agreement on the best approach, with some pursuing an agency model, most building on the bank acting as principal, and some offering a combination of the approaches. It’s unclear which, if any, will become most successful – ultimately, it will be up to clients to decide.

 

The rise in electronic trading and algos

Algorithmic trading appeared in the equity market around 20 years ago, when broker-dealers began offering institutional clients automated ways to transact large block trades. Computers were programmed to take sizeable orders and automatically cut them up into smaller pieces before sending them to different exchanges for processing – the aim being to stop the markets moving against the institution as the ticket was processed. This led to the development of algorithms such as volume-weighted average price and time-weighted average price.

Later, as electronic trading increased, algorithmic trading was augmented with smart order routing technology designed to monitor markets in real time and direct trades towards the markets where they should optimally execute in terms of parameters such as liquidity, price and latency.

Hedge funds and the proprietary trading desks within banks also developed algorithms with the aim of seeking out arbitrage opportunities in the market. A high-frequency hedge fund, for example, might use computers programmed to constantly monitor the markets and produce trade signals based on historical and real-time price and volume analysis. Others put in place latency arbitrage strategies that seek to profit from differences in the speed of price quotations.

This technology started to emerge in the foreign exchange market a few years ago, as electronic trading in the asset class reached something approaching critical mass. Consulting firm Aite Group estimates electronic trading comprised around 65% of all foreign exchange trading at the end of 2009, and reckons this will grow to more than 70% by the end of 2012. “Given that markets remain fragmented, the need to source multiple liquidity pools simultaneously has only strengthened the overall position of electronic trading,” Aite Group managing partner Sang Lee wrote in a report published in April. 

However, voice trading is still an important part of the client-to-dealer market – and increased during the latter parts of 2008 and 2009, as market uncertainty and wider spreads forced customers to trade directly with their dealers. At the end of last year, electronic trading represented 43% of the client-to-dealer market, according to Aite.

The consulting firm expects the adoption of bank-provided execution algorithms to increase over the next few years, reaching close to 15% of daily volume by the end of 2012 from a current estimated level of around 7%. But Lee believes algorithms have yet to bed down in the foreign exchange market. “It is still very early on to figure out how this is going to play out,” he says.