Algo Wheels: Best Execution, Workflow Solution, or Both?

While trading analytics tools have assisted buy-side traders for decades, money managers increasingly are turning to algo wheels not only to help traders measure and manage algorithmic performance, but also to streamline their workflow. TABB Group founder and research chairman Larry Tabb examines the use of algo wheels on the buy side and the drivers behind their accelerating adoption, as well as their impact on brokers' businesses.

This article originally was published Oct. 15, 2019.

Trading algorithms drive today’s markets. Not only are the most sophisticated market makers using algorithms to power their liquidity provisioning strategies, virtually all orders, whether automatically or manually generated, are routed and executed by machine, as the speed required to trade in our competitive and fragmented equities market makes it extremely challenging, if not impossible, to execute orders by hand.

While the majority of buy-side trading algorithms are implementing human-based decisions, it doesn’t mean that the execution algorithms are any less complex. Given the multitude of choices, the level of difficultly in matching an algorithm to a symbol, market conditions, trading strategy, and the level of separation between the money manager and the market, just aligning the product, order, and market liquidity takes skill.

While trading analytics tools have assisted buy-side traders for decades, money managers are increasingly turning to algo wheels not only to help traders measure and manage algorithmic performance, but also to streamline their workflow.

The use of algo wheels is still in its nascent phase, as only 13% of the 92 head traders we interviewed during the first quarter of 2019 were using an algo wheel in production; another 27%, or twice as many funds, are testing and/or evaluating the use of a wheel. Algo wheel usage was heavily skewed by size, as larger funds (>$150 billion AuM) were twice as likely to have a wheel in production than medium-size ($25 billion to $150 billion) funds and 5 times more likely than small funds (<$25 billion), (see Exhibit 1, below).

Exhibit 1: Usage of Algo Wheels in the US Equity Markets

Source: TABB Group

While data is spotty, many funds in our study developed their own proprietary solutions; the three external providers of algos wheels to our study sample were the Fidelity Service Bureau, Instinet and ITG. ITG was the leading outside provider (see Exhibit 2, below).

Exhibit 2: Number of Funds Using Algo Wheel Providers

Source: TABB Group

Why Wheels?

Algo wheels help buy-side firms automate their algorithmic workflow and rationalize their usage of broker algorithms. The wheel not only simplifies the difficulty of allocating order flow to a specific broker, it also helps the buy-side trader ascertain the quality of the broker’s algos given differing liquidity patterns and trading situations. Much like trading algorithms a decade ago, algo wheels can help automate the trading of smaller and less impactful orders. Wheels maximize the trading focus of skilled buy-side traders by automating the execution of less difficult orders, help traders normalize their brokers’ algorithms, and provide a framework to benchmark and measure brokers’ algorithms.

When used for best execution a broker wheel pits similar algorithms against each other using A/B-type testing strategies. How the fund leverages these tools and analytics, however, is firm-dependent. Some firms will reallocate flow based upon performance; others will leverage the results to pressure brokers to upgrade/customize their algos, and others leverage the information to streamline their workflows and validate their best execution methodologies.

Best Execution

Normalizing disparate algorithms to measure performance and ensure best execution are the algo wheel’s killer application. The complexity of aligning various vendors’ algorithms, randomizing their usage, and analyzing algorithmic performance is challenging, as different algorithms are better in certain situations. To start this process a trading desk needs to align the different brokers’ algorithms, determine their sweet spots and determine how each of the algo sectors are to be used. This enables traders to align the trade with a category of algorithm instead of a specific broker’s tool, as the algo wheel randomizes the usage of the different models within those categories. From there, the wheel routes orders to the algorithms, collects data, and measures the models’ performance.

While measuring algo performance isn’t like measuring snowflakes, each firm’s orders, size, impact, benchmarks, timing, and instructions are different, as even fairly standard algos such as VWAP can have different performance characteristics depending upon the timing of their use, the liquidity pools they traverse, and market conditions at the time. While this may make analyzing performance difficult, it doesn’t mean that traders should just throw up their hands. What it means is traders need a better way of measuring, analyzing and reinforcing algo performance. This is what the algo wheel does.

Just like content creators are moving toward A/B testing to better understand the impact of a title, picture, or broadcast timeliness, algo wheels provide a mechanism for measuring and standardizing the use, benchmarking, analysis, and comparison of trading algorithms. This in turn pushes brokers to embrace a continual improvement strategy, which measures performance, uses those analytics to improve performance, and generates more data, which again is used to hone performance even more.

But algo wheels don’t guarantee best execution by themselves. A trader’s perspective is critical in trading, especially for trading in difficult circumstances. While an algo wheel will help normalize broker algos, buy-side traders have many more tools in their trading bag than an algo or an algo wheel. As the old adage says, if you only have a hammer, the world looks like a nail. Algo wheels will ensure that you are using the best nail, but a wheel won’t help if the trader really needs a screw, a bolt, or a rivet – or capital, blocks, or plain old DMA.

Competing in an algo wheel world

But what does the shift toward algo wheels mean for brokers? Are all broker algos going to be pitted against each other in an increasingly battle-bot-style fight to the finish?

Well, yes and no. While the algo wheel pushes brokers to continuously improve their algorithmic suites, as noted above, algorithms aren’t the be-all and end-all of trading. In addition, wheels may help open the buy-side algo rotation for new and/or upgraded trading models.

Algos historically have been sold on a relationship basis. Firms with marketing dollars, a large sales force, and a significant technology budget have tended to dominate the algorithmic execution business, while smaller firms without big budgets traditionally have been given short shrift, no matter how good their technology. This makes penetrating a desk with a full complement of algos extremely difficult. To complicate matters, the more familiar the buy side becomes with the tools, the harder it is for a new provider to get on the desk. The number of algo suites on a trader’s desk increased to a peak of 12 in 2014 and has been on the decline, to approximately 9, through this past year (see Exhibit 3, below).

Exhibit 3: Average Number of Algorithms on the Buy-Side Trader’s Desk

Source: TABB Group

TABB Group believes that algo wheels will make the process of onboarding new algo brokers much easier for both the buy and sell sides.

The most effective way to leverage a wheel is very different than leveraging a traditional algo suite. While onboarding a traditional client means working with the trading desk to implement a significant number of customizations, algo wheels typically use algorithms that are similarly set and configured, as comparing two highly customized algorithms developed for two very different situations does not provide comparable performance characteristics. By using the brokers’ “best” versions of situational trading scenarios, the wheel can randomly pick between a number of brokers’ VWAP or IS or Dark strategies, to generate execution data that can be used to more effectively find the best performing algos within the overall execution strategy.

This means that the buy-side trader no longer needs to fully understand the nuances of each specific algorithm, as the trader can onboard 6 or 8 VWAP models that compete over time. While this won’t help the buy-side trader take advantage of specific situations, it will ensure he is using the best generic VWAP algo, as the data generated by the wheel will highlight the better performing algorithms. Many firms will then reweight the better performing algos as they drop off the lower performing models, which will then open the wheel to insert other brokers’ algos into the rotation.

Employing this strategy allows the buy side to cycle through a number of brokers’ algorithms, in search of the best-in-class algorithms for the situation. While the wheel will help the buy-side trader manage workflow for more highly liquid and simpler trades, however, it doesn’t alleviate the need for more tailored tools for more specific circumstances.

As algo wheels gain traction, TABB Group sees buy-side firms supporting two different types of tools. The first set includes more generic algorithms that fit into an algo wheel, which will be used for more traditional VWAP, TWAP, IS, or dark accumulation strategies; and the second set contains those more customized algos that will be tailored to specific situations.

Increasingly, the buy-side trader’s job will be focused on ascertaining the appropriateness of using a wheel, and those times when a more specific and customized tool will provide a better outcome.

Continual Performance Improvement

While wheels open up opportunities for brokers to get their models on the buy-side desk, they also mean that brokers must employ a continuous improvement process to monitor and tweak their tools to achieve superior performance. To accomplish this, the brokers will need to continuously measure, analyze and adjust their algorithmic performance not only to know how their algorithms natively perform, but also how well they perform when engaged by an algo wheel. This continuous cycle hopefully will demonstrate to the client not only a level of algorithmic performance, but also a level of client engagement that will provide both a better execution experience for the client and a more thorough understanding of how the client engages with these tools; in turn, that should be reflected in continual execution quality improvements.

While algo wheels theoretically take a lot of the human subjectivity out of the algorithmic selection process, brokers will have an opportunity to differentiate themselves in other ways. How a broker engages with the client and how it works to continuously improve its models may help the broker keep the algorithm on the wheel when the performance may dictate otherwise. More important, it will give the broker that can analyze the data, make adjustments, and work with the client an advantage over those that do not show the same level of dedication.

Creating this level of support will not be easy. The two types of firms that will excel will be firms with unlimited development budgets that can consistently measure, tweak, and re-measure and re-tweak, and those that utilize algorithmic providers such as Clearpool that have a newer and more open infrastructure where measurement, configuration, re-measurement, and re-configuration are done without significant technical cost. Firms with a more modern and flexible algorithmic infrastructure will benefit over those with older and more hard-coded infrastructures.

Conclusion

As the complexity of trading continues to increase, so does the pressure on buy-side funds to improve performance and reduce costs. We do not see these pressures relenting anytime soon. This calls for each and every financial intermediary – from pension funds and asset managers to market makers and brokers – to do more with less.

This can only be accomplished with technology, and nowhere is there a better example of this than on the trading desk, as measurement, analytics and trading have become virtually intertwined with the ubiquity of trading algorithms.

TABB Group believes that the next leap in trading technology will push buy-side firms toward algo wheels that help improve the workflow on the buy-side trading desk and amplify broker algorithmic competition.

While algo wheels are not a panacea or a solution for all execution problems, they will streamline workflows and enable the buy-side trader to continuously monitor, manage, and improve trading performance. In addition, algos wheels will enable a wider variety of brokers to get onto previously closed desks and demonstrate the performance of their models. While performance will drive the algo wheel, how the sell side continuously monitors and improves the performance of its strategies, and brokers support their clients will be the differentiating factors in determining who stays on the wheel and who gets dropped off. This new paradigm will favor those brokers with an agile algorithmic infrastructure.

While wheels are still new, and the technology is evolving, we firmly believe that these tools, like the development of trading algorithms themselves, will play an increasingly important role in the future of the buy-side trading desk and brokers’ businesses. Algo wheels not only will improve algorithmic performance, they also will push brokers to continuously innovate to keep their spots on the once-closed buy-side trading desk.

TabbFORUM is an open community that provides a platform for capital markets professionals to share their ideas and thought leadership with their peers. The views and opinions expressed are solely those of the author(s). They do not necessarily reflect the opinions of TABB Group, its analysts, TabbFORUM and its editors, or their employees, affiliates and partners.

Comments

  • profile image
  • profile image
  • profile image

Add a Comment

Your email address will not be published. Required fields are marked *