Creating algos was once the purview of just a handful of banks, but client demand has meant more options coming onto the market. And there’s only one way to figure out which of them are worthwhile – transaction cost analysis (TCA).
At the moment, however, TCA in FX markets is not only in its infancy, but also far more complicated because there is no standard industry benchmark to measure against.
Sang Lee, managing partner at Aite Group, a consultancy, said it’s a ‘by-product’ of the regulatory focus on best execution, as well as banks leveraging electronic execution tools: “They need to be able to prove to their clients that if they use their algos, they are getting better execution.”
What are the knobs you can tune on those algorithms?
That’s a tall order, he added: “The bank may be providing some level of liquidity back to the clients and they know what they’re offering in terms of pricing, but one could argue that there’s no way even a big bank is 100% sure there are better prices elsewhere because the FX market is global in nature and decentralized.”
Larger asset management firms are now some of the longer-term persistent users of execution algorithms, and as such are starting to focus on measuring comparative performance.
Understanding the subtle differences between bank offerings is essential to a valid analysis, said James Dalton Global Head of Algorithmic Execution at CitiFX. Depending upon how they trade, the output from any metrics can be distorted – whether or not one bank’s algorithm is a little bit more aggressive than another’s, for example.
“There’s a lot of variables associated with measuring this stuff and they are coming to terms with how they can calculate precisely how well they perform when they choose one over another,” he said.
As one example, some firms may use a historical volume curve over a given time horizon – one week, one month, three months, etc – to calculate optimal execution horizons, while other firms might look at market volumes in real time.
“Both want the participation-type execution strategies but running a slightly different set of inputs, and being aware of the difference between the two explicitly is important when evaluating the strategies,” he said.
…can you show me what happened in the market in this time zone?
The important thing is to understand why the outcomes are different, Dalton explained: “An algo using a historical curve may be run during a period of time when the markets are completely quiet. This can potentially increase the participation rate above the average expected on an execution. Whether this has a positive or negative impact on a single execution is very hard to establish.”
Other complications include getting access to a complete market view.
Soren Haagensen, Managing Director at Integral Development, currently the only provider of end-to-end electronic FX trading platforms, said that clients doing their own TCA are limited by the data available to them: “There is a lot more focus now coming from clients, as well as banks alike, to go to platforms like us and ask: can you show me what happened in the market in this time zone?”
The larger the order, the more benefit a company could expect by heeding TCA outcomes, he added, though there are some important caveats. Even a very large order, at times up to say some $500 million-equivalent, executed within a couple of hours won’t tip anyone off in a liquid pair at the right time of day.
…now it’s putting the TCA behind it and looking at more advanced algorithms.
In terms of ticket sizes appropriate for algorithmic trading, Haagensen has seen them begin at some $10 million-equivalent. “50,000 euro-dollar for a corporate account, I don’t think it’s worth the time and effort to put in an algo, to sell 10,000 in two minutes – you end up paying it away in ticket costs.”
Moreover, it does depend on people’s risk profiles, he added, because some firms don’t want anybody to know what they’re doing.
Last Look and the Hard Problem
Vendors are certainly seeing the opportunity in figuring it all out. Portware’s FX trading platform, in partnership with ITG, is working on having TCA that takes post-trade analytics feeding into a pre-trade scenario.
“The decision can be made on how to better execute the next trade. Automating that process is one of the goals we have for 2016,” said Chris Matsko, Head of FX Trading Services.
Jacob Loveless, CEO at Lucera, a high performance infrastructure provider, refers to TCA as a “hard problem” requiring a great deal of data to build on. But the decision that traders have to make is simple: which would you rather sacrifice, time or price?
The decision can be made on how to better execute the next trade.
“What are the knobs you can tune on those algorithms? How long are you willing to let something work to get a better price, or what are your time constraints? We’ve always had that flexibility in our matching engine, and now it’s putting the TCA behind it and looking at more advanced algorithms,” he said.
Loveless pointed at foreign exchange’s ‘last look’ as a major factor in how TCA is developing because of the corresponding uncertainty in whether liquidity on the screen can be executed.
“Before you’ve even started on the math, before you’ve even started on the algo design or the TCA, you’ve got to think of that problem,” Loveless said. “If you can (calculate) the probability of the fill being true, then you are going to start to be able to tease out microstructure things that make sense.
“If you see $15 million bid in euro-dollar, $20 million offered on your aggregator across all your venues, and you remove all the nonsense, then you feel a little more confident – is it worth me crossing the spread and listing this offer? Or should I be patient and sit on the bid? If you can’t answer that question I don’t think you can answer anything.”