Instinet won Best New Product (Micro Adaptive Sequencer) at the 2022 Markets Choice Awards.
Traders Magazine speaks with David Fellah, Head of International Quantitative Trading Strategy, Instinet to learn more.
What is your reaction to winning the award for “Best New Product” at the 2022 Markets Choice Awards?
We’re proud and delighted that Instinet’s Micro Adaptive Sequencer (MAS) has been recognized by Markets Media for “Best New Product”. At Instinet, we are wired to continuously look for ways to improve our clients’ efficiency and performance – to push our infrastructures and our tools toward the delivery of ever-expanding levels of execution quality. These initiatives can take a long time to implement, and can be incredibly technical and detailed. They are often hard to articulate in plain English. So, this response to our work on MAS has been very exciting. Many talented members of our global team have been deeply involved. On behalf of them and all of Instinet, thank you again for this recognition.
How did Micro Adaptive Sequencer come about? What gap in the marketplace does it aim to fill?
As we know, underneath every algo order, there is a series of child order decisions aimed at slicing up and attempting to reach the parent order’s goals. These tactical “sub strategies” are built to behave according to models – and these models operate on a schedule.
Schedules are informed by historical data that has been captured at the symbol, market, and venue level, among other data points. The schedule is what tells the child orders how – and when – to behave in order to meet its tactical goals. While most algos today also have the ability to react to atypical reference data (in real time), that reaction is usually rather “binary” – do I stop? Or do I go? But the child order’s underlying schedule still awaits, and tells the order what to do.
So, traditional schedule-based order management models have limitations and are highly prescriptive, determining pre-set directions between an order’s start and end points. We asked ourselves “what if you could make the schedule completely dynamic, rather than pre-set? Advances in deep learning and dynamic sequencing technology have enabled us to re-imagine new methods of order management and performance, offering continual optimization on the fly, as conditions change.
What I’m describing is like the difference between a static “map” and real-time GPS-based guidance technology. Operating like a GPS for orders, MAS steers away from sudden, unanticipated hazards of market impact, trading cost, or price dislocations, as they arise. And, it instantaneously calculates a corrected path targeting best performance, rather than being constrained by or forced to “catch up with” historical patterns or static target benchmarks. It’s a new way of approaching an old and thorny problem.
Please describe Micro Adaptive Sequencer — what are the primary capabilities and how is it unique?
MAS is a nuanced and nimble trading plan optimizer. It infuses both anticipatory and real-time responses to market “detours” and shifts. This innovation is like an advanced “wayfinder” for orders. It’s been engineered with world class tools, utilities and a minutely-detailed collection of signals to anticipate, as well as instantaneously adapt order behavior. All of the input models and trading tactics are carefully tuned to the dynamics of each specific market. MAS then uses all the information and models available to continuously reassess fair value models, then forecast and re-forecast optimal trading behavior throughout the life span of an order. The trading outcomes feed back into future predictions and model versions, leading to even greater performance improvement over time.
What were the key technology challenges to implementing this MAS solution?
People have been trying to untangle this Gordian knot of the “optimal schedule” for a long while. In order to do this properly, it was important to give an algorithm’s child orders a certain degree of discretion, so that they can move opportunistically. There are some strategies out there that propose to take discretion over an entire order, morph or switch as the algo sees fit, and manage its execution opportunistically. That “leap of faith”, however, is not easy and can sometimes run contrary to traders’ obligations to supervise the pursuit of best execution. Our clients generally don’t like to be surprised by their orders’ outcomes.
There are a few strategies out there that employ highly complex mathematical formulae and optimal control computations, such as the Hamilton-Jacobi-Bellman (“HJB”) equation, to continually recalculate the execution schedule while the order is in flight. Strategies using approaches such as these are definitely more dynamic, but the processing power and bandwidth required to execute these kinds of massive calculations can make this approach very expensive – not to mention rather less than green. Incidentally, an HJB formulation doesn’t easily permit constraints and is not readily extensible, not to mention the dreaded ‘curse of dimensionality’ problem referred to here. So that approach didn’t give us the nimbleness we needed.
Instead, we solved the exact problem with a different formulation – one that allows us to apply real-world instantaneous, cumulative and terminal constraints required in trading strategies.
Because we realized that what we actually needed was something that behaves like GPS navigation software for our orders. An intelligent utility that allows the trader to still drive, but offers a much more real time and anticipatory set of directions that continue to recalculate the schedule as things change during the life of the order, optimizing the “route”, until it ultimately reaches its destination – or in the case of an algo order, its trading objectives.
What are your long-term objectives and ambitions for Micro Adaptive Sequencer?
Going forward, MAS will serve as an underlying sequencer “utility” for orders managed by Instinet’s algo engine. While clients’ original order goals and parameters will still guide an order’s behavioral objectives, MAS will supplement and/or replace traditional schedule-based routing management with more insightful, dynamic and responsive management directions for child orders. The signals MAS can read are also continuously expanding. Currently, MAS is available globally via several of Instinet’s benchmark algorithms, including VWAP and Instinet’s Work strategy.
Our MAS framework can also be used in offline simulations for the calibration and systematic back-testing of settings. In this way, production MAS code can be used to validate settings before testing live orders. Pre-trade is thus consistent with intra and post-trade, resulting in a process of systematic algo development. As a next logical step in this framework’s evolution, the automation of this pre-trade tuning step suggests we could build an actual recommendation system for the initial parameters of an order. We’re continuing to explore this and will share more in the future.
