by Kyle Pedrotty and Jay Hinton. 

One of the hottest topics for institutional traders in 2020 has been the impact of automation on trading workflows. Charles River has been actively working in this area as well, and we recently launched a new product development team focused on trade automation tasked with the goal of delivering the next generation of trading tools to front office users.

Before exploring what we’ve been working on, it’s worth articulating our thoughts on trade automation. We believe the goal of automation is not to replace the human trader. Rather, trade automation is emblematic of Cooperative Game Theory (CGT), a branch of economics where a coalition of players in a game work together within a set of agreements to achieve the best possible outcome for the coalition. Just like CGT, trade automation is based on a set of rules or instructions defined by trade desks to determine how traders and automation tools work together to access and provide liquidity most efficiently,  governed by deep historical data documenting past outcomes.

In 2005, I was implementing and supporting an order management system (OMS) at various hedge funds. Mid-year one of my clients went live on their first broker-provided equity algo suite.  Over the next few months, several other clients were to follow with their own broker algo installations.  The perception of these new automated trade scheduling and execution tools was not universally positive.  Not long after initial implementation, one head trader reported that his traders had begun using the volume-weighted average price (VWAP) algo for everything, without regard for suitability. Another head trader opined that execution algos were useless since his job was to beat the algos. While these anecdotes are not universal, they are also not uncommon.

In the fifteen years since, buy-side usage of equity algos has matured, along with the understanding of the role they should play in the ecosystem of trading tools.  As such, not all liquidity is sourced through an algo suite or dark pool.  The high-touch, hands-on approach is still required for many types of trades like accessing large blocks of liquidity in a bank’s central risk book, executing cross-asset multi-leg orders, and utilizing trading expertise in unfamiliar markets, to name a few.

Many buy-side firms we engage with have implemented some form of trade automation, most commonly in basic order routing, but occasionally through more advanced methods with multi-placement decision logic.  Many of those firms are now deciding what business rules will drive their next generation of automation processes.  Like the principles of CGT, the forward thinking buy-side desks we work with are focused on how to achieve the best possible outcome for every order by creating a set of rules that empower human traders to work harmoniously with trade automation tools as a coalition in the ecosystem.  They are not focused on a binary analysis of whether traders or algos are better at sourcing liquidity for a portfolio manager’s order.

Just like their early use of algo suites, buy-side trade desks are in the vanguard of adopting desktop automation tools embedded in their order and execution management systems (OEMS). The OEMS captures and centralizes a number of key data elements across the trade lifecycle, from portfolio manager intent, to funds, orders, trade routing and pricing. In contrast, stand-alone EMS platforms mainly capture only execution details.

Adopting automation on the trade desk begins with identifying the business rules traders follow, from both defined procedures and tacit knowledge, and transforming them into automation rules.  In general, the automation rules will evaluate what type of order each one is, whether similar orders exist, and if that order is eligible for automated placements, to ultimately determine which sets of placement logic will be used to fulfill an order.  These rules start in the pre-placement workflows as the trade desk receives orders from portfolio management, and carry on throughout the price discovery, liquidity sourcing, and execution phases.

The rules also set guardrails: they define what the automation is capable of and not capable of, how the system should behave when conditions change, and when traders must retake ownership of certain orders.

Subsequently, in many automation rules, the paths in a decision tree are designed to ask human traders for inputs along the way, an effective hybrid approach to automation.  By systematizing order and execution management workflows in this manner, traders are empowered to focus more time on the critical orders that need their attention, and less on the ones that don’t.

Moreover, implementing trade automation rules is just the first step in the process.  Buy-side firms need to continuously evaluate how their rules are performing, a prime application of machine learning methodologies[1] . Automation can range from a fixed set of (human specified) pre-defined rules implemented as a decision tree, as mentioned above, to artificially intelligent trade execution processes that have been trained to optimize price return using deep reinforcement learning techniques and the rich trade history captured in the OEMS. These types of automated approaches are widespread in the equity arena and are growing in the fixed income world. Fixed income markets are more fragmented, less transparent and more manual than equity markets, and there are many obstacles to overcome before enough high quality data is available to allow for the use of machine learning.

Now that this data is increasingly available, especially in equity markets, we can apply machine learning approaches to decide how to optimize order clustering, time trades for best pricing, judge the quality of indications of interest, completely manage basic trades, decide when to pass a trade to a human trader and so on. This is currently an exciting and active area of research, and Charles River is examining machine-learning use-cases for trade automation.  For example, our quant research team has been applying machine learning techniques to institutional asset management; in one of their recent projects the team used the GRU Algorithm, a technique for gating within recurrent neural networks, to create a suite of prepayment prediction models for agency mortgages.

In tandem with establishing the business rules of trade automation, automation tools must earn the trust of the trade desks using them.  In the next parts of this series, we will explore how to capture trader knowledge within the rule sets, what transparencies and controls are needed for human traders to trust the systems augmenting their workflows, and how to manage complexity across automation processes.

[1] Source:

As of November 17, 2020

Disclaimers and Important Risk Information

Charles River Development – A State Street Company is a wholly owned business of State Street Corporation (incorporated in Massachusetts).

This document and information herein (together, the “Content”) is subject to change without notice based on market and other conditions and may not reflect the views of State Street Corporation and its subsidiaries and affiliates (“State Street”).  The Content is provided only for general informational, illustrative, and/or marketing purposes, or in connection with exploratory conversations; it does not take into account any client or prospects particular investment or other financial objectives or strategies, nor any client’s legal, regulatory, tax or accounting status, nor does it purport to be comprehensive or intended to replace the exercise of a client or prospects own careful independent review regarding any corresponding investment or other financial decision. The Content does not constitute investment research or legal, regulatory, investment, tax or accounting advice and is not an offer or solicitation to buy or sell securities or any other product, nor is it intended to constitute any binding contractual arrangement or commitment by State Street of any kind. The Content provided was prepared and obtained from sources believed to be reliable at the time of preparation, however it is provided “as-is” and State Street makes no guarantee, representation, or warranty of any kind including, without limitation, as to its accuracy, suitability, timeliness, merchantability, fitness for a particular purpose, non-infringement of third-party rights, or otherwise. State Street disclaims all liability, whether arising in contract, tort or otherwise, for any claims, losses, liabilities, damages (including direct, indirect, special or consequential), expenses or costs arising from or connected with the Content. The Content is not intended for retail clients or for distribution to, and may not be relied upon by, any person or entity in any jurisdiction or country where such distribution or use would be contrary to applicable law or regulation. The Content provided may contain certain statements that could be deemed forward-looking statements; any such statements or forecasted information are not guarantees or reliable indicators for future performance and actual results or developments may differ materially from those depicted or projected. Past performance is no guarantee of future results. No permission is granted to reprint, sell, copy, distribute, or modify the Content in any form or by any means without the prior written consent of State Street.

The offer or sale of any of these products and services in your jurisdiction is subject to the receipt by State Street of such internal and external approvals as it deems necessary in its sole discretion. Please contact your sales representative for further information.