Applying the Pareto Principle to Best Execution

If you're Noah, and your ark is about to sink, look for the elephants first, because you can throw over a bunch of cats, dogs, squirrels, and everything else that is just a small animal and your ark will keep sinking. But if you can find one elephant to get overboard, you're in much better shape.
Vilfredo Pareto

BestX launched its first product last September, delivering a comprehensive set of analytics and reporting for post-trade best execution in FX. The software was designed to satisfy internal and external best ex requirements, including regulatory reporting requirements, often referred to colloquially as ‘box ticking’. Perhaps unfairly, given this is a vital component of the fiduciary responsibility of asset managers to asset owners, and more broadly, of increasing importance to all FX market participants given the Global Code of Conduct and other initiatives. However, this article seeks to explore the value that the software can bring over and above the core ‘box ticking’ requirements, which we have explored in previous articles, expanding upon the article published last October on improving the execution process . We return to this subject and expand upon it using the experience we have gathered over the last few months witnessing how our clients are using the BestX product. Fortunately, we designed the software to be very flexible which has proven essential given the creativity of our clients in employing the product to make tangible cost savings. From very first principles, the philosophy behind the software design was to empower clients to 'make informed decisions’, and to give clients the ability to use the software to make tangible cost savings. We explore some of this practical use cases further in this article.
As previously discussed in the October 2016 article, there are many different factors to the execution process, each of which can be refined to provide further improvements in efficiency and cost. We’ll work through some examples below.

1.    Execution Method Selection

One of the most significant structural changes to the FX industry is the increased range of execution methods available to clients, from risk transfer over the phone, to RFQ or RFS on multi-dealer EMS platforms, single dealer platforms, algo execution and direct market access. Our philosophy is that best execution warrants having a menu of options available to allow different choices to be made depending on the prevailing liquidity conditions and trading objectives. However, at the core of the process should be the ‘go to’ method for the majority of flow, at least in normal market conditions. How do you decide upon this method? Clearly, measuring costs and execution performance rigorously and accurately, whilst performing tests to see if different approaches add value, is one way. We have found clients using BestX to compare execution methods, e.g. multi-dealer EMS vs single liquidity providers, and finding considerable cost savings (of the order of several millions of USD per year). There can be a general perception that FX spreads are generally so tight that if a client is already executing in a very sophisticated fashion, then surely there can only be fractions of basis points to be saved? Perhaps, although annual turnover does not need to be particularly large for such savings to be make a significant impact on the bottom line.

2.    Counterparty Selection

Selection of counterparties is an obvious area for performance improvement and many clients have used either in-house analytics or external providers to help with this for some time. Different liquidity providers have different strengths, for example, in different geographies or through different client franchises or technological advances. Identifying these strengths and allocating business accordingly is where we see the majority of the BestX use cases in this field. Traditional cost analysis, however, only gets you so far and may result in potentially erroneous conclusions. The BestX Expected Cost analytics have proven particularly valuable here, by allowing clients to compare costs across a consistent and level playing field, taking into account the relative difficulty of the business that each counterparty has executed. 
A simple example is illustrated below, where in Figure 1 we display the average spread costs for 4 different Liquidity Providers. A cursory inspection of the results may lead to the conclusion that Liquidity Provider 3 has performed the worst over this period as their Actual Spread Cost incurred is the highest.

Figure 1: Example Actual Spread Costs by Liquidity Provider

Figure 1: Example Actual Spread Costs by Liquidity Provider

However, lets now compute the Expected Costs, or ‘fair value costs’ of the trades that each counterparty executed to allow a fairer comparison. We add these results to the chart plotted in Figure 2, and we now see a very different result. It would appear that Liquidity Provider 3 executed the most difficult, or expensive, business as measured by the BestX Expected Cost measure. When taking this into account, and measuring on which counterparty outperformed the Expected Cost measure, then Liquidity Provider 3 actually delivered the best execution performance and Liquidity Provider 1 underperformed.

Figure 2: Example Actual vs Expected Spread Costs by Liquidity Provider

Figure 2: Example Actual vs Expected Spread Costs by Liquidity Provider

Businesses are dynamic, for example, franchises change, technology improves, staff turnover, so it is essential to monitor such performance over time. This is obviously a very simple example, but we have seen use cases where clients have drilled into performance by currency pair to help determine which liquidity providers consistently excel in specific currency pairs such as Scandis or EM. We’ve also seen examples where performance by time zone has been assessed to see if any changes should be made to further enhance performance. To allow such bespoke and timely analysis, it was essential that we delivered a flexible user interface, providing clients with self-sufficiency in terms of analysis and reporting.

3.    Channel Selection

Channel refers to the system by which trades are executed. For example, clients may execute via a single dealer platform, a multi-dealer execution management system (EMS), direct APIs with either single liquidity providers or via aggregators, and so on. We have already found that different channels can result in different execution performance, not only in terms of actual costs, but also with regards to implicit cost measures such as impact cost resulting from, for example, information leakage and signalling risk. It is obviously important to compare apples and apples in such analysis, as we have also seen examples where clients may use two EMS, but one of these tends to be used for the less liquid business, so comparing actual cost alone can be misleading and result in erroneous conclusions. The BestX Expected Cost metrics add value here as discussed above in the section on Counterparty Selection.

4.    Venue Selection

An increasingly popular area for analysis is in the selection of execution venues for those clients that are using orders and algos. The Last Look debate is a likely catalyst for this increased attention, and we are seeing more clients wanting to get a quantitative understanding of the impact of executing on different venues with different protocols. It does not appear likely that Last Look is going to be banned in the foreseeable future, although the Global Code of Conduct may help increase both the transparency and standardisation of practices employed in the market. With sufficient transparency, and measurement, it then boils down to choice once again. However, to make an informed choice, it is clearly important to have analytics to measure the costs, including those experienced at the point of trade but also the impact pre- and post-execution. In our experience, not many clients are provided with all of the sufficient order data to compute the true cost arising from the opportunity cost of Last Look, e.g. clients generally don’t receive reject data post execution. We have seen some liquidity providers and ECNs supply this, although the market is still very inconsistent which makes complete analysis across entire portfolios of trades difficult. We have seen clients tagging venues as Last Look or Non Last Look, and then using the BestX analytics to compare spread and impact cost, and also pre- and post-trade revaluation data. We expect this type of use case to increase over time as complete order data availability increases and becomes more standardised.

5.    Algo Selection

Not all algos are created equal. We have seen considerable differences in algo performance over the last few months, depending on the currency pair, time of day, notional size and trading objective. Such performance cannot be measured on any one trade – large samples of trades are required, and then measured on a totally consistent basis, to allow statistically significant conclusions to be drawn. The different charging structures prevalent in the market, coupled with a disparate performance, means that improved algo selection can also result in considerable cost savings in cash terms over the course of a year. To illustrate this, we have conducted some empirical research based on the data set of algo trades we have analysed over the last year.
Using the BestX analytics, we computed the average values for spread cost, impact cost and benchmark performance across the entire universe of algo trades analysed to date, which represents a statistically significant sample size. Looking at EURUSD specifically, where a significant volume has obviously been executed, we found some interesting results in terms of the range of cost and performance experienced. In an attempt to ensure we were using a homogeneous sample, we filtered the trades to only include those executed within the most liquid window (8am-4pm GMT), and also stratified the sample by notional size. Summary results are provided in Table 1 below:

Table 1: Standard Deviation of Costs and TWAP Performance (bps) by Notional Size (USD) for EURUSD Algos executed between 8am-4pm GMT

Table 1: Standard Deviation of Costs and TWAP Performance (bps) by Notional Size (USD) for EURUSD Algos executed between 8am-4pm GMT

So, what do these numbers actually mean? By showing the standard deviation of the average costs and performance, we are illustrating how the impact of algo selection can have a significant impact to the bottom line. For example, for trades with notional sizes of 50-100m USD, the standard deviation of the average spread cost is 1.5bps. If you are trading USD 10bn notional of algos per year, and if you were using algos randomly from the sample we analysed, there is a 68.3% probability that your total costs could have fluctuated in the range of plus or minus USD 1.5m. 

6.    3rd Party vs In-House Execution

Another trend we are seeing is the evaluation of outsourcing FX execution to a 3rd party vs bringing execution in house. There are many factors involved in such a decision, and we won’t provide an exhaustive discussion around them all in this article, but suffice to say, a key factor is the comparison of cost. Even when a cost comparison has been performed, however, a number of other variables need to be taken into account, including the provision of other ancillary services such as post-trade reporting, provision of research etc. We have seen a number of clients using the software to compare costs to help with such decision making. Clearly, moving in-house has other associated costs as well as replacing ancillary services, including technology and human resources. It is  a complex decision, but a number of asset managers are finding the BestX execution cost analysis valuable, including use cases where we have seen the analysis justifying the cost of 3rd party execution.

7.    Streamlining the Order Management Process

As with most large institutions, many asset managers have complex technology architectures and operational processes that have evolved over time. Such constraints can result in inefficiencies in the lifecycle of an order, whereby considerable risk can be run from the time an order is first originated to when it is finally executed. Over time, and over large samples of trades, you would expect this risk to potentially net out as clearly the market could move for or against you whilst the order is finding its way through the process, but it is a risk, and one that is uncompensated. Generally speaking, minimising the time it takes to get the order from the portfolio manager to the execution desk is best practise for any best execution process. However, deciding whether to spend scarce investment budget and technology resources on projects to streamline this process is not straightforward and requires a proper cost-benefit analysis to make an informed decision. We have seen clients using the BestX software to help quantify the risk they are running through inefficient order routing processes, and thereby determine whether the cost is worth incurring.


There is sometimes the misconception that the FX market is so deep and liquid that trying to improve the execution process may only result in relatively small cost savings, and therefore not worth the focus of time and resources. In our experience, we have found the opposite. More informed decisions around different aspects of the execution process, some of which have been illustrated in this article, can result in very significant cost savings, especially when viewed in actual cash terms. Yes, a handful of basis points may sound like a relatively small number, especially in relation to the commissions and bid offer spreads available in equities and fixed income. However, if you are turning over USD 100bn of FX per year, and through improved selection of counterparty, channel, venue, algo etc, then a combined saving of, for example, only 1.5bps would result in a cost saving of USD 15m per annum. Such a number is very realistic based on our experience to date.
There is a lot of focus in the FX market at the moment on issues such as Last Look, and the quantification of this. We fully support these important initiatives in the spirit of rigorous, consistent quantification and transparency. However, we also feel that many participants in the FX markets can make more significant improvements to their execution performance and costs by taking a step back and applying a more systematic approach to their process at a macro level. Obviously, some institutions already have an informed selection process in place, and can afford to start looking for more marginal gains. The 80:20 rule applies – get the 80% right before sweating the small stuff.