FX execution analytics: A multidimensional framework

FX execution analytics has clearly moved beyond straightforward transaction cost analysis. However, one of the most important rules of any model – not to be tricked into thinking that outputs from a sophisticated model are inherently accurate – still applies.

Until relatively recently, TCA was largely a regulatory box-ticking exercise that was used to motivate counterparties to perform better. In this scenario, the accuracy of the data is not that important suggests Paul Lambert, CEO New Change FX.

“The thing about FX is that because it’s a market where the price you get depends on who you are and who you ask, you could argue there is no right price,” he says. “The basis of proper execution analysis is knowing what the right price is and to do that, you have to identify the benchmark price.”

This is what New Change FX has focused its efforts on, adds Lambert. “If you only take pricing from a particular segment of the market you are almost certainly going to be reflecting skew, which will work in the favour of your clients at some times and against them at others. So the conclusions you draw from comparing your outcomes to the data might be false signals.”

Increasingly sophisticated benchmarks

BestX, the State Street owned TCA firm, has always been focused on enabling clients to understand their trading performance and evaluate, record, demonstrate and achieve best execution, going beyond just looking at how they have done against market data into customisable benchmarks developed by practitioners that have traded FX or built algos.

The firm’s approach to execution analytics goes beyond a viewpoint against mid or WMR to really understanding trading performance and factors that go into decision making, explains Yusuf Nurbhai, head of BestX.

“We have always viewed the benefits of real-time, pre-trade and in-flight analytics as a continuous process,” he says. “You can have measures for outlier detection and exception reporting and fix issues as they arise rather than waiting until the month-end. All of that analysis can help drive pre-trade decisions around how to trade, where to trade and who to trade with.”

A static view on a monthly or quarterly or even yearly basis is no longer enough. Firms want to be able to have transparent and robust discussions with counterparties to improve performance and get the best outcome for underlying stakeholders.

As the only streaming benchmark for spot FX (and regulated by the FCA) and only provider of independent, granular FX forwards data across the curve, New Change FX takes data from across the market and has won various industry awards.

The reality is that the majority of transactions in the FX market are forwards, which account for around two-thirds of market turnover – and of that, two-thirds of transactions in the forward market are transactions into broken dates.

“The data that was available for broken date forwards was fundamentally wrong by design because it does not reflect all the granularity of the forward curve, for example driven by expectations of central banks changing interest rates at whatever date along the curve,” observes Lambert. “The data used by the vast majority of TCA providers to measure forward transactions incorporates maybe 10 or 15 of the 252 forward dates, joined together with straight lines.”

Using information carefully

The flaw in this process underlines the importance of thinking carefully about how data is used to drive execution decisions.

“If you are measuring your transactions against something which is incorrect by design and using the data from those measurements to drive your execution decisions, you are just embedding inaccuracy and have to ask whether you are doing a better job than when you had somebody on the desk using discretion,” says Lambert.

Yangling Li, head of analytics and quantitative research at BestX observes that buy-side firms may believe that adding more banks to their panel will improve spreads without considering the potential for information leakage.

“We can reliably show clients that there is an optimal number of banks to trade with, which might be six in the case of G10 currencies,” he says. “This selection will then be based on execution data rather than a feeling or existing relationships.”

On the question of how execution analytics has impacted algorithmic FX trading, Nurbhai says it is important to undertake a proper comparison of providers and their trading styles.

“We have mapped the algo names to a style so clients can look at their opportunistic or passive algos specifically against other opportunistic or passive algos,” he adds. “This is important because in some cases they will be really focused on arrival price whereas in other cases the focus will be on fill speed, or some other parameter that is bespoke to the strategy and the outcome they are going for.”

Banks in particular are developing their own single leader platforms that highlight the different liquidity conditions or volatility or depth of book that they are able to see.

Integrating alternative data

So how is alternative data – such as market microstructure signals – being integrated with next generation technologies like AI and machine learning to predict optimal execution strategies and detect hidden costs or patterns?

“The only way to achieve optimum execution is to combine data and analytics,” says Li. “To use a fuel analogy, if the raw material is not refined you won’t get a useable end product. So analytics based on incomplete data is not a solid basis from which to draw conclusions.”

He adds that the technology now exists for clients to design algos that optimise the signalling score for those that want to hide their market footprint.

“There is a lot of talk about machine learning and AI,” says Lambert. “But anyone that has ever built a model that had a P&L attached to it will tell you the first rule of every model is ‘garbage in, garbage out’. So we are completely focused on making sure that we are giving clients a data set against which they can accurately measure their transaction costs.”

Once firms have that accurate transaction cost analysis, they can employ a feedback mechanism between their transaction cost analysis and their trading decisions – and the most sophisticated way to do that would be by employing machine learning, by building models that analyse transactions in terms of what was the currency pair, what was the time of the day, what were the volatility conditions, what was the liquidity that was prevailing, what approach to market was used, did the client ask for a price, did they run an order, did they use an algo.

“The next thing we think is really fundamentally important is to take a step back and look at the decisions they have taken in terms of how they are going to manage their foreign exchange risk and how those decisions have affected outcomes,” continues Lambert.

Identifying the question

He explains that if a firm has decided that it is always going to trade its forwards into month-end and at a four o’clock fix and that the data used to measure that decision is standard tenors for forwards joined together with straight lines, then the outcome may appear fine but they are not reviewing the right question.

For instance, instead of asking ‘did I hit the fix?’ they should be asking ‘should I have used the fix and do I have the data to assess the benefit of trading to other days?’ so they are then able to decide whether this very rigid method of hedging or rolling forwards is the best option for them.

“One of the other things we are doing is giving our clients information on the granularity of the forward curve, so they are then able to decide whether this very rigid method of hedging or rolling forwards is the best option for them,” adds Lambert.

If they feel they would benefit from having more flexibility and knowledge around the shape of the curve they are faced with, they can take the next step into a model that not only has that information on the granularity of the shape of the forward curve but also knows what they are trying to achieve.

“When we have done this with our clients, what we find is that the impact on their outcomes is materially bigger than the impact of whether to use an algo or not in a spot transaction, for example, because they are fundamentally seeking a different price point on the curve,” he says. “They are not trying to improve the outcome of a price point they have already chosen.”

From an innovation perspective, Nurbhai notes that BestX has just introduced a service that enables clients to evaluate in real time whether they should do a particular trade as an RFS or an algo based on the statistical probability of outperformance. This service dynamically assesses each trade against size, currency, time of day and market conditions to give an expected outcome.

“It is still necessary to have human oversight to determine the urgency with which you want to trade and then go through the next stage to really evaluate algo providers,” he says. “But what it allows you to do at a macro level is have a viewpoint from which you can drill down in terms of how you want to execute in the future.”

This is invaluable for clients who need an audit trail but also have unique requirements and may not have the full suite of algo providers to choose from.

Delivering trading tools

The other development Nurbhai references is bringing component pieces of the industry together so that FX buy-side traders in particular have all the tools they need when trading. “Then there is an increased focus on forward points and the evolution over time in terms of value dates,” he adds.

His recommendation for firms looking for a FX execution analytics provider is to start by focusing on providers who work with market data that can be assessed against peers in the market and is not determined by what the sell-side are submitting.

“The other consideration is whether the analytics being displayed are provided by a firm that understands the industry and the nuances of FX,” adds Nurbhai. “We have seen other providers from an equity background, for example, trying to shoehorn analytics into an asset class that is unique.” “Clients need to be confident that their analytics provider understands what they are looking for, understands bespoke benchmarks, has the flexibility in terms of data and technology to deliver the analytics that they need and is independent,” he adds. It’s not about favouring a particular counterparty, platform or custodian – it’s about helping clients optimise their trading.”