Views from the buyside
Algorithms are more popular than ever in FX. That statement isn’t a surprise to market participants, especially on the buy side. The Bank of International Settlements December 2022 Quarterly review highlighted growing market participants’ use of algorithms to navigate the fragmented FX markets.
While slicing orders and distributing them across diverse venues is now common, less attention is paid to algo performance measurement. While algo usage is high, stakeholders are still grappling with the challenges of ensuring their algos are performing as expected and unlikely to deliver unexpected results.
Richard Turner, Senior Trader, Currency Solutions at Insight Investment, is well-versed in the subtleties of algorithmic trading performance. “The availability (to the practitioner) of child order data is the key operational requirement,” he says when quizzed about quantitative algo performance measurement. “A rules-based structure to aid algo choice from previous experiences is also vitally important.”
So how are buy-side firms approaching TCA when using algorithms, and what metrics are they tracking? Here’s a deeper look.
Data organisation and metrics are critical
Algos offer a significant advantage when speaking of execution. They bring immense visibility to execution workflows, aiding TCA. Turner remarks that transparency is critical and “..enables the market practitioner to analyse the child orders of the algo from many angles.”
“The availability (to the practitioner) of child order data is the key operational requirement. A rules-based structure to aid algo choice from previous experiences is also vitally important.”
Richard Turner
“TCA can be carried out to assess many aspects of execution including market impact, information leakage, and implementation shortfall,” he continues. “The market practitioner can ultimately benchmark the behaviour of the algorithm and assess its suitability for their target execution goal.”
Jan Grindrod, Global Head of FX Trading at Invesco, says that his firm’s algo performance evaluation measures have always been quantitative. “Prior to using algos, our FX trading workflows had already been heavily electronic,” he says, “with nearly all trades being executed electronically or allocated post-execution on multi-dealer platforms.”
Given this electronic baseline, what metrics does he track? “Important metrics we use in the evaluation of algos include the fills versus various benchmark rates, spread capture, fill venues, speed of execution, and revaluations post-execution,” Grindrod says.
Turner points out that while metrics play an important role, they’re the result of well-organised data. “Accurate and easily available post-trade TCA data with good time stamps is key to this process,” he says. He also points out that a robust data analysis framework is critical.
“While Excel does the trick for internal consumption, an independent source of TCA is essential when presenting and qualifying findings to clients,” Turner explains. Technological improvements are playing a role in changing the way teams are looking at performance evaluation.
For instance, Turner cites the rise of live TCA as a key enabler of in-flight algo adjustments that optimises outcomes based on underlying market conditions during execution windows.
“Without “live” TCA the algo user is left to monitor markets and use their experience to adjust algo parameters in order to maximise the overall outcome,” he explains. “This experience is largely based on historical outcomes from a suitably rich source of previous execution outcomes.”
While data organisation and metrics tracking play an important role, Turner and Grindrod stress that the review process adopts a broader view.
Leveraging analytics and examining objectives
Grindrod explains that the performance review process begins with examining the order’s objectives. “First, we need to have a strong understanding of the objectives of the FX order, as well as the available algos best suited to meet those objectives,” he says. “After this has been established, we use third-party pre-TCA tools to garner details on prevailing liquidity conditions and estimations of how an algo will handle the order based on different parameters.”
He explains that Invesco uses a combination of in-house and third-party analytics to arrive at an estimated risk transfer price. “Lastly, throughout the execution,” he continues, “we use a combination of bank-provided in-flight TCA and internal tools to help determine whether fills, execution speed, and market movements in the currency pair are in line with expectations.”
Turner notes that Insight scrutinises algo performance heavily and monitors its approved list of algos closely. “We have a quarterly algo review meeting where we take a deep dive into the performance of all our algos,” he says. “We study algo performance and make recommendations if any pertinent action is required. It goes without saying that if we observe any live underperformance, in an algo, we take immediate action.”
Underperformance results in the algo’s removal from the approved list until remedial action is taken or the firm determines that underperformance was not due to the algo’s behaviour.
Given algo trading’s data-centric nature, analytics play an important role in helping firms determine performance quantitatively. As Grindrod puts it, “Post-trade TCA allows us to complete the feedback loop of having past experiences influence future outcomes. We can evaluate our decision to use an algorithm to execute the order, the algo selected, the parameters used, and the resulting fills to determine whether the correct decisions were made and if our expectations of the algo were met.”
Turner adds that his team focuses on a few things to improve algo performance. “The price decay/reval curve before and after fill, LP behaviour, whether the algo’s behaviour matches expectations, and the underlying market conditions are important factors,” he says. Turner also adds that examining the effect of limit prices and input parameters on algo behaviour is also critical.
Build versus buy and dealing with common problems
“It would be great to see TCA evolve to show how the currency pair moved on a risk-adjusted basis against a basket of highly correlated cross-asset markets over the life of the algo to see if the pair moved more, less, or as expected during the execution.”
Jan Grindrod
Given its importance in trade execution workflows, deciding to build an algo TCA solution in-house or buy a third party solution is a perennial question. As with all software platform decisions, several factors play a role when arriving at a decision.
Grindrod is in the third-party camp. “We prefer to use a third-party solution for FX algo TCA and analytics for two reasons,” he says. “First, it is important to have an independent third party to evaluate trade execution. Additionally, we do not have a competitive advantage in the consumption, storage, and analytics of the market data needed to perform these analyses.”
He stresses that evaluating electronic RFQ execution depends on different factors, and Invesco leverages in-house and third-party TCA analytics there.
“Resource is a massive factor in the debate on build or buy,” Turner says. “Across the industry, resource efficiency is an area with increasing focus and competing factions.” He explains that the need for resource efficiency governs the build versus buy choice.
“For analytics shared with clients, independence is key. In this instance, it is imperative that, where possible, an external TCA vendor is utilised to supply output to clients.” Turner contrasts this with internal analytics use where the benefits analytics bring are the most important area of focus. “If utilising an external vendor to improve your outcomes, what are your benchmarks? Can you consistently demonstrate the benefits versus the costs?,” he says.
Turner explains that governance and algorithm onboarding is just as critical as deciding whether to build or buy. “At Insight, we have an Algorithmic Trading Policy which details the governance of the approval and use of algorithms. The policy ensures that the algos we utilise are fit for purpose and meet our execution needs.”
Does Insight trial these algos before use? “As we trade with various bank algos, it is not possible for us to trial before use, so a lot of time is spent assessing the readiness and suitability of algos for use,” he says. “When a new algo is approved for use, it’s first run is heavily scrutinised during and after execution.”
These policies are in place to protect firms from any adverse impact emanating from algo misbehaviour. However, spotting these instances is challenging.
“Given the current limited metrics within FX algo TCA, underperformance may be in the eye of the beholder,” Grindrod explains. “We believe the best way to protect oneself against perceived algo underperformance is to ensure the user fully understands the ins and outs of the algo with regards to details such as available liquidity pools, aggressiveness, and expected time to completion.”
Turner concurs. “We are very careful to separate algo underperformance and natural market movement,” he says. “It would be very easy to blame the algo for execution underperformance in a market that moves against you. We do our best to try and ensure that we separate algo performance and use the metrics mentioned previously to measure the performance of the algos.”
Desired enhancements and looking ahead
Grindrod’s point about limited TCA metrics points to possible enhancements the buy side would love to see. “They (FX algo TCA metrics) are one dimensional, limiting their usefulness,” he says. He explains that analytics currently break down currency pair execution tick data into various metrics (like starting mid, starting top of book bid/ask, TWAP mid, etc.) and compare fill against them.
“While this is helpful, it doesn’t tell you how the currency pair performed in the context of broader markets,” Grindrod says. “It would be great to see TCA evolve to show how the currency pair moved on a risk-adjusted basis against a basket of highly correlated cross-asset markets over the life of the algo to see if the pair moved more, less, or as expected during the execution. Just because a currency pair moved for or against you during an algo execution doesn’t necessarily mean the algo performed well or poorly.”
Turner and Grindrod both wish for more “in-flight” TCA availability. “We are also striving to get TCA in a standardised form,” Turner says. “More data is key with information such as Limits, Urgency settings, and changes by the trader being captured. As signatories of the FX Global Code, we would encourage market participants to utilise the disclosures cover sheets and Algo/TCA templates available on their website. We would also advocate completion of the Global Code compliant field in TCA.”
Time will tell how analytics platforms catch up and cater to these requirements. Either way, there is no doubt that the algo TCA space is set to witness tremendous growth and sophistication.
