Feedback loops and marginal gains – using TCA to save costs and improve returns

“Without data you are just another person with an opinion”
W. Edwards Deming, Data Scientist

Continuous performance improvements, whereby all aspects of a process are examined with precision, are the hallmark of many leading teams and businesses. Seeking out such marginal gains, as exemplified by Sir Dave Brailsford with the GB cycling team, or the Formula 1 team of Mercedes McLaren, has now become commonplace, and in this article we explore how such approaches can be applied to a continuous refinement of best execution.

TCA and execution analytics add considerably more value than simply providing a framework to satisfy regulators, compliance teams, best execution committees and asset owners. Investing in such analytics purely to ‘tick a box’ represents a missed opportunity to save significant amounts of money. In order to achieve this, however, it is imperative that the execution process is approached with an open mind. There needs to be an environment whereby all aspects of the process are monitored, measured, questioned and tested on an ongoing basis. An ‘open feedback’ loop is required in order to allow lessons to be learnt, and the necessary changes to enable improvements to be made. The opposite of such a process, a ‘closed loop’, does not test assumptions, or measure the impact of changes, or indeed, learn from mistakes. In such an environment, the mantra ‘but we’ve always done it like this’ can often be heard quoted. 

Does this mean a world of decisions and actions taken by machines ? Of course not. The financial markets are too complex and dynamic to allow a purely automated execution process. Equally, however, the complexity means that it would be impossible for a human to be able to process all of the necessary inputs to arrive at an optimal decision alone and without any ‘help’ from analytics. Clearly, the answer is combining the best of both – the experience and intuition of humans with the processing power and objectivity of machines. Machine learning can add value, and is a topic of research at BestX, but is best deployed in the hands of an experienced trader.

There are many decisions taken every time a trade is conducted, some of which won’t matter in the overall scheme of things given, for example, the size of the trade. However, having the ability to at least monitor the impact of all dimensions, and checking whether decisions are having a material impact or not, seems a wise approach. The list of questions below, although not exhaustive, provides an idea of the range of decisions that now need to be taken:

  1. What time of day should I trade?

  2. What size of trade should I execute?

  3. Who should I execute with?

  4. Should I trade principal or request the counterparty to act as agent for my order?

  5. If I trade principal, should I trade via the phone or electronically?

  6. If I trade electronically, which platform should I trade on?

  7. Should I hit a streaming firm electronic price, or should I trade RFQ?

  8. If I RFQ how many quotes should I request?

  9. If I trade electronically, should I use an algo, and if so, which one?

  10. If I use an algo, which venues do I want it to execute on?

  11. If I use an algo, or order, should I employ passive order types or not?

  12. How quickly should I trade?

Clearly, such questions are not explicitly answered for every single trade. A desk may be executing thousands of tickets per day, and the process may be defined and automated for the majority of these. The larger trades may warrant more careful attention, and follow a decision making process which requires further insights and analysis. Either way, a comprehensive Best Execution framework should allow both the broad automated processes, and individual trade decision making, to be measured and monitored over time to check if the original assumptions are still valid.

For example, a best execution process may have defined that all funding trades of less than USD 50m notional are submitted to an RFQ process on a specific EMS, whereby 5 counterparties are requested to quote. This ‘rule’ may have been put in place several years ago. Does it still make sense in today’s market? Regular and ongoing evaluation of the execution performance and total costs are required to answer this question. Analysis of the costs for the entire book of business over a period of time may indicate that tickets of greater than USD 50m notional, that are not subjected to the RFQ rule, have started to systematically incur less cost, indicating that it may make sense to revise the rule.

The complexity of the financial system, however, means that it would be imprudent to simply make the change and hope for the best. There may be other factors at play that are resulting in the observed change in costs. Splitting the business into two and testing half of the portfolio with a revised RFQ rule for the following quarter would allow a more scientific approach to be taken. Such ‘controlled tests’ are widely used in all scientific disciplines and form the basis of any open feedback loop. Change an assumption, perform a controlled test and re-evaluate the results to check if the original hypothesis was correct. If yes, then incorporate the change into the best execution policy going forward with the benefit of having a quantitative and rigorous process behind the decision. If no, then go back to the original process and test a different assumption. For example, perhaps the majority of trades with notional of greater than USD 50m were executed by two counterparties that were different to those used for the RFQ business. In which case, maybe test the original RFQ rule but this time replace two of the counterparties with the two that perform well for trades greater than USD 50m. And so on.


Some of these changes may not result in cost savings. Some may result in marginal savings, and some may contribute significantly to the bottom line. However marginal though, in a world of either low or negative yields, every single basis point really does count. As an example, returning to the list of questions earlier, let’s just focus on the first one regarding timing. Using the BestX Fair Value Cost estimates, we analysed the costs of trading 50m AUDUSD every day for the period of January to May this year. If this trade had been executed at 9am London compared to 2am London, the total cost saving would have been approximately 33 basis points (AUD 167,000) over this period. For a US investor, if the trade had been executed at 9am London instead of 9pm London, the cost saving would have been a whopping 250 basis points (AUD 1.26m). Clearly, such simple analysis does not take into account factors such as opportunity cost, but the point is to illustrate that simple changes can result in considerable savings.

Over the course of 2016, we have analysed thousands of FX transactions at BestX from a wide and diverse array of FX market participants. It is clear from this analysis that there are many cost savings to be made for the majority of institutions, across many dimensions of the execution process. 

Best practice should never be to simply settle and assume that what I’m doing is the best I possibly can. After all, Dyson famously tested over 5,000 versions of his vacuum prototype before launching.

“I made 5,127 prototypes of my vacuum before I got it right. There were 5,126 failures. But I learned from each one. That’s how I came up with a solution.” - James Dyson

The dynamic nature of financial markets, especially in OTC markets at the moment as they continue to transform driven by the regulatory fallout from the financial crisis, require an ongoing evaluation of ‘best’ practices and ways of doing things. Learning by doing, including from both positive and negative results, in a measured, systematic and controlled way is one way to navigate this complexity. Indeed, employing such a feedback mechanism was explicitly mentioned in ESMA’s latest Q&A publication , where it is stressed that the results of ongoing execution monitoring are fed back into execution policies and arrangements to drive improvements in the firm’s processes.

FX is a simple product, but with a complex market, which is getting increasingly more complicated. The impact of the market structure changes, especially the drive towards a more order-driven market at a time when the traditional banking market makers are providing less inventory management in the system due to regulatory changes, is still to be fully determined. However, the recent flash crash in Sterling is a sign of things to come and such liquidity ‘air-pockets’ are becoming increasingly common. Fragmented markets, supported by less risk capital, with the majority of pricing and risk management processes managed electronically, all contribute to more volatile liquidity conditions for the foreseeable future. In such a market, the decision making around how, when and with whom to execute becomes increasingly difficult, especially when coupled with the need to justify such decision making. An interactive, rigorous and systematic approach to measuring and monitoring execution performance, and then using this information to continually enhance the process, is rapidly becoming an essential component of any best execution policy. Static, vanilla TCA reports of the past, produced and filed in a drawer for a rainy day if anyone asks, are no longer adequate.