Spinning The Wheel
It is noteworthy that the term “algo wheel” has been coming up a lot. It was recently part of a panel discussion on “Next Generation Algo Trading” that QB participated in at the FIX Trading Community’s Boston conference. The development and use of these wheels has been prompted by the best execution requirements of MiFID II and increased scrutiny in many buy-side firms on which brokers are used and why. As a result, it has been growing in popularity with EMS providers as a new feature to attract or retain their clients. For the buy-side, it provides an automated approach to allocating trades across various brokers and algorithms within their EMS, something that many quantitative managers with proprietary systems have been doing for years.
ITG, Flextrade and REDI are some of the more high profile EMS providers who are offering algo wheels for futures flows. REDI describes their algo wheel as a “rules based order router” which makes sense. You can establish a set of rules for how to allocate flow across various brokers and algo strategies, and the router executes on that plan, automatically dividing up the flow to meet the intended goal. The great advantage of this is that the traders overseeing the order flow do not have to think about where each order should be sent, saving time and mental exertion while ensuring the appropriate allocation is followed.
At QB, we are supportive of a disciplined data-driven approach to allocating ones order flow, and we welcome the extension of the algo wheel concept across many EMS platforms. It enables those with manual execution flows to benefit from the automation of this aspect of individual order routing decisions, and partially catch up with other firms who have been doing this through proprietary systems for years. Given the nature of QB’s unique, premium service, we are typically subject to an extensive evaluation by our clients. The algo wheel can help set up an evaluation by fairly distributing orders between competing execution paths. The gathering and analysis of data on each path over time is the crucial part. You still need a good transaction cost analysis (TCA) solution, which may or may not be part of the algo wheel service. Understanding the TCA, and crucially, the unique characteristics of the order flow are important for making the right decisions.
Our recommendations are as follows:
Define the benchmark: The chosen benchmark determines which algo is most suitable, and consequently, which algos are relevant to be evaluated together. I.e. Comparing an implementation shortfall (IS/arrival price) algo with a VWAP will not provide an “apples to apples” comparison.
Define the order size: Randomly allocating orders is a statistically robust approach for ensuring the samples are not biased. That said, it is important to consider the trade size distribution and one may need to evaluate a more typical trade size band to ensure unbiased comparisons.
Define the strategy: Many firms have multiple underlying alpha strategies that generate trades. Tracking this is important as the benchmark and consequent choice of algo can be different. The fair allocation at the strategy level between different algos is also important to ensure that strategies with very different characteristics are evenly distributed across the algos being compared.
Keep it simple: Don’t have the algo wheel make overly complex decisions or amend order routing decisions once sent.
The automation that a wheel brings is great. But there is no magic bullet when it comes to the reallocation decisions to be made thereafter. We believe that the scope of the algo wheel should not be extended to the automatic reallocation of flow based on execution performance. A careful analysis needs to be undertaken, not an automated or cursory glance of the latest best slippage. Ideally, reallocation decisions should be undertaken by a thorough, multi-person human review, with full context of order flow characteristics and sufficient data. This means that making month to month or even quarter to quarter decisions may be unsuitable. Additionally, futures instruments are far more heterogeneous than equities and this variation requires significantly greater attention to market microstructure in the design of the algo wheel than one might do for equities.
TCA is challenging – it requires a lot of thoughtful set up, reliable data, contextual perspective, and honest interpretation. Machine Learning advances may help make this easier eventually, something QB is keeping a close eye on.
Quantitative Brokers
New York
December 10, 2018