Antish Manna: Leveraging algo strategies to ensure the best outcomes

Head of execution analytics, multi-asset, at Man Group, Antish Manna, sits down with The TRADE to explore how to leverage algorithms to achieve the best results, touching on machine learning, broker collaboration, and ensuring minimal bias. 

How has the way you determine what is an optimum algo strategy changed in recent years and what is driving this change?

I have to say that in some ways it hasn’t changed at all; first and foremost, our role is to understand the objective of the investment strategy we are trading for and ensure the algo strategy we pick matches that objective. That won’t shift. We do, however, have to stay abreast of and adapt to developments in the marketplace, for example, the steady growth of closing auction volumes and close facilities. On the latter, we work really closely with our brokers to adapt our algo strategies to benefit from those changes, where possible through strict A/B experimentation. 

At Man Group we have invested in our experimentation framework over recent years, which has improved both our ability and capacity to explore and optimise through experimentation. What has also evolved is the maturity of our execution analytics platform, which means we can more effectively monitor and analyse key execution metrics – it enables better, faster and more in-depth work. We think this leads us to faster insights and, ultimately, a stronger partnership with our portfolio managers who can also access and consume the same analytics. Given the significant market volatility over the past couple of years, we think the ability to have good and trustworthy analytics is imperative.

What metrics do you use to monitor algos and broker selection on algo wheels?

We use machine learning, and specifically reinforcement learning, to optimally route flow to broker algos within what we call panels (what others might refer to as algo wheels). We think there are multiple benefits to this model: it is a systematic process and devoid of human bias; it offers a statistical framework to balance exploration and exploitation (because panels allocation doesn’t get stale); it’s complementary to experimentation, and finally, it offers a clear incentive for brokers to improve.

How do you normalise your data to ensure you’re comparing apples to apples?

The ability to fairly access execution outcome is extremely important and we spend a lot of time and effort on this. Specifically, there are a few areas where we focus: firstly, where possible we construct panels which have a similar order flow, whether in terms of order characteristics, alpha signature or both, as a way to maximise homogeneity. We also use automation and probabilistic routing to minimise biases, and transaction cost (TC) models that factor in the key drivers of cost. Where appropriate, we also use simulation techniques to reduce noise. Finally, a combination of statistical analysis, segment analysis and trend analysis is utilised to look at execution through different angles.

How much of your algo development and execution analytics do you outsource?

We do both proprietary algo development and execution analytics in-house and that choice has been made for a few reasons. With regard to algo development, firstly we think the information we have on our alphas and strategies gives us an edge when we’re planning the optimal path and approach to execution. Another key factor here is that we can control the pace and focus of development, and finally, being actively involved in the process and deep into the details has improved our team’s DNA and understanding of markets. The same applies for execution analytics – we’re aiming for a best-in-class platform, so every component needs to shine. What we’ve built is best placed to service the diversified business of Man Group.