Addressing the confidence challenge of peer analytics in post trade TCA

Peer benchmarks are emerging as an important requirement for contextualizing trading performance in fixed income markets.

They allow traders to compare their absolute performances against the average performance obtained by their peers trading similar orders. Trading generates transaction costs and peer benchmarks provide the critical relative perspective that can put trading performance in context.

Below we detail how to use peer analytics for relative analysis in Bloomberg’s post trade TCA (BTCA), and highlight challenges to consider when using peer analytics.

Setting up peer analytics
There are important considerations to ensure confidence in the results. Firstly, the trading community is vast so the peer universe must be sizable to provide a fair representation. Secondly, an important component of any peer universe is to maintain the integrity and anonymity of the data while providing insights to customers.

Bloomberg has implemented performance comparison against respective peers at the group level for equities, fixed income and FX securities. The peer groups are designed to reflect key order characteristics that enable traders to accurately profile their order flow and derive meaningful insights from their analysis. For example, traders can profile fixed income orders by attributes such as: size, maturity, country of risk, rating, sector, and difficulty profile. Using appropriate groups and grouping combinations is essential for the quality analysis.

How peer benchmarks work in a post trade TCA

Analyze peer performance in BTCA

Traders can use BTCA to analyze performance in arrival price terms aggregated at the country of risk for a 3-month period. In this example, for orders in bonds where the associated country of risk is Great Britain, the performance against arrival price is -0.98 basis points. For the same period and country of risk, the performance among the peer community was on average -12.69 basis points. This shows that the firm, despite exhibiting absolute slippage, is outperforming its peers by almost 12 basis points.

Moving on to the Peer Rank, we can see that outperforming the peer group indicates the relative quality of the execution. For example, compare orders in Great Britain where the firm has outperformed its peers by almost 12 basis points against performance in Germany where outperformance is only 1.51 basis. However, the firm’s performance still ranks amongst the top 10% to 20%

amongst the peer universe. By comparison, in the United States, the firm has underperformed its peers by 2.91 basis points, resulting in a slip in its ranking amongst peers to 20% to 40%.

The analysis above helps traders build a narrative to help infer performance. However, it is important to note that generalizing at a single group level is not ideal because traders may not be comparing all relevant order characteristics.

For example, the group Great Britain has 221 orders and most certainly made of orders with different maturity, ratings, and liquidity profiles, such differences can meaningfully affect the quality of the analysis. Hence, further refinement in the peer data is key to achieving the correct results.

BTCA solves this issue by combining groups of characteristics in a waterfall like approach. Traders can filter high yield and investment grade, then sector and then add additional layers of groups like size, maturity ordinal and country of risk. For instance, analyzing high yield bonds, in the technology sector with a maturity order of five to ten years, using a refined peer dataset will yield a more relevant signal than analyzing at one group level, such as all high yield orders. It is therefore critical to ensure the comparison is made at a level of granularity that can provide a meaningful result. Too general and it will lack precision. So how granular should traders go?

 

The industry challenge
The more granular the analysis, meaning the more groups that are combined, the fewer observations are available. This means that the signal becomes weaker because the data becomes scarce and abnormally distributed around the mean. So if the quality of the peer data is critical for the accuracy of the analysis, then understanding and having transparency around the quality of the data is key to finding the correct balance in between granularity and data quality, and successfully using peer data in post trade TCA.

A recent study performed by Bloomberg’s trading analytics team attempted to identify the overall quality of the data across all available peer groupings and groupings combinations. The study assessed the data quality based on known statistical parameters such as size, distribution, skewness, variance, and standard deviation. It also analyzed the density and distribution of the performances in each group and all possible group combinations and classified them as excellent, good, fair and poor.

The results confirmed the hypothesis that the greater the number of groupings used in combination, the lower is the quality of the underlying data. The study also showed that at the first and second group levels most of the peer data is of high quality, and that issues with data quality arise when adding further groupings.

The chart displays data quality when using one, two and three groups in combination.

Analyzing the chart above, at the single group level the majority of the groups have at least a fair data quality, and most of the data is either excellent or good. The quantity of observations and their distribution are high quality and therefore produce a strong and clean signal.

At two group levels, such as sector and country of risk, the majority of the groupings combinations have at least a fair data quality, however there is an important decrease in the number of combinations that produce an excellent 

or good score and an important increase in the number of combinations that result in poor data quality.

At three group levels, for example sector country of risk, and maturity ordinal, about half of all groupings combinations have at least a fair data quality associated to them. However, there is an even greater decrease in the number of combinations that produced an excellent or good score and an important increase in the number of combinations that result in poor data quality. Finding the correct grouping combinations at the third level of grouping is challenging, roughly, 1 in 2 groups will produce a weak and noisy signal.

BTCA customers who want to start using peer analysis can by run BTCA <GO> and locate the FI Peer Benchmarks report under sample reports, right click and select run. The report will provide you with insights into your performance vs your peer’s performance over a rolling 3 month period.

Where do we go from here?
It is not possible to use peer data at a single bond level, as it will most certainly result in a weak signal due to data scarcity and abnormal distributions. Even with minimum sample constraints it could also introduce a potential for data leakage or signaling. If quality of the peer data is essential for the accuracy and quality of the analysis, then finding the correct balance in between granularity and data quality is a gap that TCA providers must bridge. The key to resolve this is to provide transparency about the quality of the data. Developing a way 

to highlight the quality of a peer result will allow users to make informed choices on how they construct their peer analysis and to what extent they should feel comfortable in the outputs. Challenging providers to provide transparency into the quality of their peer benchmarks should be the way forward for the post trade TCA industry for only with transparency will users be able to find the optimal balance between granularity and quality in their analysis.

BTCA traders who want to discuss BTCA’s peer universe in detail can contact their account manager or the Bloomberg Help Desk to schedule a meeting.

To learn more about BTCA, go to TCAI <GO> on the Terminal, or watch our video here.