• Related Articles

• 

  Customising FX algorithms: fine tuning your order execution strategies

• 

  Build versus buy: choices for deploying advanced Algorithmic FX trading infrastructures

• 

  The next wave of FX Algorithms taking order execution strategies to another level

• 

  Identifying the optimal combination of FX algorithms

Harnessing more intelligent FX Algorithms for superior order management and strategy execution

By Nicholas Pratt

According to Paul Tolman, executive director and founder of Beta-Gamma Research, a UK-based developer of advanced trading models, there are a number of FX algorithms that are trying to take advantage of various complex FX market parameters such as momentum and velocity factors, moving averages, pricing ratios and news and even-driven behaviour. "The vast majority of algos may be looking at the strategies used in the equities market such as volume-weighted average price (VWAP) and then applying them to the FX market but there are also a huge number of momentum-based trading models out there.

"As for the news and event-driven algos, I think they are more prevalent in the speculative and prop trading parts of the market where traders are looking to take a punt on the direction of the market following a significant news event rather than the pure FX algos which are designed to trade on a more routine basis," says Tolman.

Changing role of the trader

In regard to these more complex algorithms, Tolman believes there is a core subset in the market that then move around the market as the various quants responsible for them move from firm to firm taking their intellectual property with them, so in some regard there are as many algorithms as there are quants in the market. The same core subset of models have always been prevalent as long as there have been FX algorithms but the value of these algorithms is clearly evident due to the changing role of the trader in today's electronically driven market.

"The job of a trader used to involve shouting orders down the phone as loud as possible in order to get some pips off the price but this has evolved now to the point where it is more about placing orders on computers. Now traders have to be able to work with algorithms," says Tolman.

There are two principle ways that traders are interacting with these algorithms says Tolman. "The first way is very direct in terms of using algos to place their orders and automatically split them up and find the best venues with the best prices. The second way that algorithms help traders is by enabling them to focus less on the operational side of each trade and concentrate more on profit generation rather than clearing every trade. It is a labour-saving approach in many ways with the algorithms carrying out the trades quicker and more accurately, giving the trader time to focus on other more strategic aspects."

As traders become more acquainted with working with algorithms and start to lend their own influence to the development of the algorithms, these benefits may become larger in number. "We get very good feedback from traders on how to improve the algorithms so that they are able to add more value and I think this will become an increasingly important part of their job," says Tolman.

Black boxes

Beta-Gamma offers black-box algorithms, an approach which may be seen as somewhat unfashionable in today's software market where transparency and customisation is king, however Tolman is clear that Beta-Gamma's algorithms stand up to the demands of today's customers. "Black boxes do have that image of impenetrable objects lacking transparency just doing random trades but our algos have very well defined instructions. There are very specific inputs and outputs so that the trading instructions and the circumstances in which to trade are very well-defined. I think this is very important, particularly for execution algorithms. We have some core strategies which are available to customers but we are now seeing more demand for customisation and the algorithms are increasingly modular and configurable so that, essentially, no two solutions are alike."

Paul Tolman

"The vast majority of algos may be looking at the strategies used in the equities market such as volume-weighted average price (VWAP) and then applying them to the FX market"

Aggregation has become a big factor in FX algorithmic trading due to the breadth of liquidity providers in the market but one consequence of the electronic nature of price discovery is the so-called liquidity mirage where a single price or offer can be seen on multiple venues. The result is that some aggregators may pick up prices that are no longer executable. "An awful lot of effort goes into aggregation and with the different latencies between price providers and others providing prices to multiple aggregators so the liquidity mirage is a big issue," says Tolman.

"There is a lot of interest in aggregation but I'm not sure if people can see the wood for the trees. If you buy into the whole idea of price aggregation then you have to deal with the mirage. Sometimes there is a gap between the promise that the technology offers and the reality. It makes sense to look at rates from multiple venues but you have to be clear about what venues you are looking at and be aware of the drawbacks."

Market impact is also a very important market issue for FX traders, says Tolman, and the algorithms that enable orders to be automatically sliced are simply mimicking what a real trader would do, only a much quicker and larger scale. However, says Tolman, it can be easy to get carried away with just how much resources are employed by the sell-side to second-guess the trading intentions of the buy-side and, consequently, how much energy and technology the buy-side should expend in
protecting their trading intentions. "I don't know if it is an area that will develop more," says Tolman. "People talk of a battle of the machines between buy and sell-side, like the Terminator. It is important to be aware of other machines but it can all get complicated and you have to sort out what is important."

Keeping pace with developments

One of the most important things is keeping pace with those market developments that are of real impact. "At the more sophisticated end of our algorithms, we have to cater for the changes in the markets. The market does change and you need to rethink the way the more complex algorithms will be most effective. With all the technology in this industry, it can be easily forgotten that there still has to be an idea, and this an area where the 'idea' comes into play."

Alongside the 'idea', there also has to be the proper application of the algo, says Tolman, which can only be improved by the ability to assess and evaluate just how effective each algorithm has been. As a result, the whole subject of post-trade analysis is attracting increasing interest, says Tolman. "We have invested a lot in post-trade analysis - looking at how profitable execution was based on different time-scales and so on. We offer a tool that does exactly that and we are looking at other tools that can marry the execution results against historical data."

The next potential area of development could be to provide the ability to benchmark algorithms against other algorithms, says Tolman. "Customers will see the value of that but it is very difficult to compare different algorithms because they could be doing different things. Some trades are easier to complete than others and there are different applications involved. The main thing is that we are very transparent about our algorithms but it is clear that there is a big industry developing in catering for all the changes in the FX market and ensuring that the algorithms can be made more effective."

OMSs

Despite the changes in the FX market, the development of algorithmic trading still lags behind other asset classes, notably the equities market, and the range of algorithms and related tools on offer is notably less in the FX market. For example, we are increasingly seeing equity traders link their execution algorithms to their order management systems (OMSs) or, in some cases, employing execution management systems. But this practice is less prevalent in the FX market, for now.

"Thus far, our experience with clients that use FX execution algos, has been that they deploy them separately from their OMSs," says John Miesner, global head of sales of Hotspot FX, the US-based ECN for institutional FX traders. "So we run our algos independently from OMSs for now but this may change in the future. The more you can streamline the execution process and the fewer steps there are to deploy, the better."

Currently Hotspot offers two FX execution algorithms to its buy-side clients - time-weighted average price (TWAP) and time slicing - which are effective when it comes to minimising market impact, says Miesner. Consequently, issues like momentum and velocity trading, moving averages, pricing ratios and news or event-driven pricing behaviour are not factored in as relates to the execution of an FX trade. "Essentially the timing of the execution is a heavily weighted factor in the two algorithms we offer."

Hotspot is a subsidiary of Knight Capital Group which provides trade execution services across multiple asset classes. Knight Direct, the group's electronic equities offering, provides over 15 execution algorithms to its clients. "I'm sure some of the factors behind the development of these algos will also become important in the FX market so we should see an increase in the number of FX algos we deploy," says Miesner. "Client demand is a large influence in what we develop but if it is easy to migrate certain algos from the equities market and if it makes sense, then we will definitely do that."

For now though, the focus for Hotspot is on developing execution algos that can focus on minimising market impact. "Minimising market impact while obtaining the weighted execution enables the traders to execute efficiently. This is the biggest factor that we see. A lot of our clients engage in large block trades and many of them are engaged in cross-asset trading where it is not so easy to hide your trades."

Randomisation

So what is the key to algorithmically slicing orders and allowing more flexible and covert execution and the absolute minimum of information leakage? "In a word - randomisation," says Miesner. "For example, if you are deploying a basic time slicing algorithm for trading $1m over 30 minutes, it will execute an equal amount every 90 seconds. But if this slicing is done on a random basis both in terms of time and value, then it will be very hard to track. I think Hotspot is a big enough exchange with enough liquidity that any FX trading firm that is randomly slicing its orders is unlikely to be detected by its counterparty."

John Miesner

"Minimising market impact while obtaining the weighted execution enables the traders to execute efficiently. This is the biggest factor that we see."

Just as Beta-Gamma's Tolman spoke of the importance of the post-trade industry and the ability to assess the effectiveness of execution algorithms, Miesner is also looking to develop this part of the trading process. "Transaction cost analysis (TCA) is a frequently talked about topic amongst asset managers and pension funds. As these tools are developed, we feel that sector of clients will gravitate towards more electronic execution. This will lead to the development of more sophisticated algos."

"For many of these buy-side firms, it is not so much about trading for profit but more about following a mandate and they welcome any tools that can affirm their execution. Right now it is still very early in terms of the development of TCA. There really needs to be more venues publishing their volumes for TCA to be effective. We have recently started publishing our volumes and we've had tremendous feedback from our clients who say they did not know we were doing so well. I can only assume that those venues which do not publish their volumes have something to hide."

Bloomberg's Tradebook is one market offering that features an FX execution platform that aggregates multiple venues and also integrates fully with clients' OMSs, according to company president Kim Bang. "Very few have found an elegant way to offer clients an average price execution and a single ticket for settlement when trading with multiple counterparties."
Currently Tradebook clients' look for two types of algorithms, says Bang. "One is for best execution and automation of the implementation process. This includes algorithms such as TWAP, Arrival Price, SCALE-in and Trailing Stops. These algos use our proprietary market data, liquidity and trading analytics to capture spread and to extract liquidity with minimal market impact." The second category concerns those algorithms that are designed for profit and to buy low and sell high. "For proprietary trading algos we enable clients to leverage Bloomberg market data, analytics and set their own parameters and triggers using our CEP contingent servers and ECO/NEWS-linked orders," says Bang.

Kim Bang

"Traders who take control of the execution process are able to stealth trade, to capture spread and extract block liquidity."

Adding value to the trading process

There are still many ways in which more intelligent algorithms can add value and efficiency to the trading process, says Bang. "There is plenty of room to add alpha by reducing the cost of trading using an electronic agency broker using DMA technologies and algorithmic execution strategies. Traders who take control of the execution process are able to stealth trade, to capture spread and extract block liquidity."

For example, the use of so-called 'stealth' algorithms and the development of other new trading strategies can help traders see through the aforementioned liquidity mirage that exists in the FX market. "Using next generation intelligent liquidity extraction - quoting in the most active venues and liquidity sweeps that look beyond displayed and into dark hidden pockets of blocks liquidity."

As with most other service providers, Bang is also seeing more interest in post-trade services such as TCA, which has been more of a fixture in the equities market rather than FX. "We are starting to see traditional equity transaction cost analytics firms offering FX TCA analysis. Their findings argue FX aught to be treated as an asset class and warrant front-office best execution processes and procedures." And if the whole area of benchmarking and transaction cost research develops to the right degree, Bang expects it to stimulate the creation of more intelligent FX algorithms and increasing client demand. "When fiduciaries realize the potential cost savings and alpha pick-up from taking control of the execution process they will quickly begin to ask for more sophisticated trading tools and algorithmic strategies similar to those available in the equity markets."

Right now, the focus for Tradebook is on the development of a number of cross-asset trading algorithms that will enable clients to auto-hedge or realise trading opportunities, says Bang. "We enable our clients to configure a 'Tradebuddy' - an alert-based database that functions as an electronic trading assistant."

Rich Gula

"Even more challenging is that the basic dynamics of the markets are changing faster than ever. This means that the average useful lifetime of a given algorithm can be shorter than ever."

Key to developing good algorithms

It is tempting to ask the likes of Rich Gula and Rich Moore, the two men behind US-based buy-side FX algo traders Argyle Trading Salgos (systems and algos), whether an electronic trading assistant would be a welcome addition. Or if they would already consider their algorithms to be electronic assistants of sorts. According to Gula and Moore, the current algo 'boom' reflects the increased emphasis on developing algorithmic-based strategies in all markets, including FX. "The stress on the world financial systems is making everyone examine how well their tools work and whether there isn't another tool out there somewhere that works better. Even more challenging is that the basic dynamics of the markets are changing faster than ever. This means that the average useful lifetime of a given algorithm can be shorter than ever. Trading algorithms, like any other model, use a set of key inputs and their relationship to each other to make predictions and there are hundreds or thousand of potential variables that could be used, says Gula.

But the key to developing good algorithms is to be selective. "Model builders simplify their models by picking and choosing the variables that they think are the most significant. Simplification is a positive thing to do with a model because it can reduce the cost and time it takes to run the algorithms; to improve the understanding and intelligence of the algorithm; and to reduce the implementation time involved. But for these benefits there is a cost. The simplified algorithm is less able to remain useful as the dynamics of the system measured changes. For FX algorithms this means less accuracy, less profits,"says Gula.

The growing number of people that want to use algorithms for trading are getting used to needing to change tactics, plans and tools more often than they ever have before he says, and "as a consequence, algorithms are being created, used when successful, and then discarded when they get stale. Therefore, there is a boom in demand for algorithms as each algorithm 'hit' is less able to satisfy the customer for as long as the last 'hit'."

Rich Moore

"Negative momentum termination models 'push' algos to execute into weakness, especially, where liquidity often improves as more emotional sellers panic sell."

"At Argyle, we believe that the constant moving on to the next best tool can be counter productive in the long run," says Moore. "Instead, each algorithm should be maintained and its success rate should continue to be evaluated and correlated to whatever larger attributes can be associated with market conditions. The 'best practice' end-case would evaluate market conditions, determine which algorithms are best for the conditions, and then spread investment activity across those algorithms based on their relative success rates," he says. With this approach, returns have the best chance to be maximized and risks reduced says Moore. "We have found, however, that market directional strategies add considerable value. Negative momentum termination models 'push' algos to execute into weakness, especially, where liquidity often improves as more emotional sellers panic sell. The opposite of course, works to some degree, but is more challenging to measure."

Cross-correlation models may help to highlight pairs for trading, with a reliable expectation that if your move them, liquidity will come, says Gula. "Algos that seek volatility may have better batting averages than those that ignore volatility, or see it as a necessary evil. Price rotation algos often show good performance after a sharp price move over a limited time, like two recent February events, the Chinese bank reserve announcement at 5am EST, or the US Fed Reserve rate uptick at 5 pm EST. Both produced very large magnitude moves in major currency pairs like the EURUSD and the so-called backwash became very tradeable.

"We wonder who got caught in these large and fast moves but we doubt there are reliable algos that will anticipate such announcements, although the old days of primary desk trading might have purported to have inside information on such moves and profited from them. Lacking such information access, volatility tends to increase at such announcements, and the cross-correlation models do their stuff to focus trading on the most volatile pair, assuming liquidity will follow. Ultimately, we feel that the most fruitful area for research and trading models emphasise market parameters like momentum, velocity, and futures trading factors derived from the equity futures markets where 'gaming' is always part of the fun," says Gula.