Equity Algos Deep Dive Part II: Algo Wheels, AI and New Venue Analysis

PETER HAYNES: Welcome to episode 58 of TD Cowen's podcast series Bid Out, a market structure perspective from North of 49. My name is Peter Haynes, and today we are back with the second of two parts in our deep dive into equity algorithms with our guest Jenny Hadiaris, global head of market structure at TD Cowen, and Robert Miller, global head of equity execution consulting at Vanguard.

In the first part of this series, we focus mostly on VWAB and its continued relevance as a benchmark. And in today's episode, we will cover algo wheels, the sizzle around AI and algos, and the process for adding new venues to routing tables. And I'm sure there'll be a few other topics that we cover.

Robert, let's start with a breakdown on algo wheels. Now, I admit that term seems to mean different things to different people, and I feel like it's that buy side trader's badge of honor now to run a wheel, like the old cocktail conversations. Hey, do you own a hedge fund? Or, I bought an ETF last week.

What exactly is a wheel? What orders are included in the wheel? How many brokers are typically included in a wheel, and how do you measure performance?

ROBERT MILLER: They're all very good questions. So I think to start with, if you have a look what's happened over the last eight to five years on the buy side, especially some of the buy sides who are trading more heavily, we've definitely kind of moved to a more data analytical world.

And so the algo wheel is a tool which we can use that can help us achieve better data analytics. So by using an algo wheel, which is basically a randomization tool, what we're trying to do is we're trying to remove any trader bias that we have.

We're using it for apples-to-apples comparison. It's broker neutral. We're systematically ranking the brokers that are on the wheel or that feature on the wheel, which means it leads us to data-driven trading, which leads us to, ultimately, cost savings.

So we are reviewing like-for-like orders. We are classifying what type of orders should go to a wheel. And therefore, not necessarily the strategy, but like I mentioned before, if you take a certain percentage of ADV, or you might be doing some other sort of classification methods that are around at the moment, those orders should all be traded in a very similar fashion, and you're expecting to see a similar result at the end of it.

So what an algo wheel does is, obviously, because it's randomized, it removes certain biases, as I've already said. And then it gives us kind of nice, clean data that we can analyze, and therefore, we can then tilt that so we can increase the percentage of flow to the number one performing algo, and we can remove algos which are underperforming their peers.

PETER HAYNES: And so are the lists typically long? I mean, Vanguard would have one. But just in general on the street, do you find that-- is a wheel typically 10 brokers, five brokers? What seems to be the general consensus on the street on the number of brokers that are involved in wheels?

ROBERT MILLER: It's very different depending on what firm you go to. I've heard of firms that have lots and lots of different brokers on a wheel. If you have a lot of brokers, you're going to need a lot of data points to actually get to a statistically significant result, so that may take a longer process.

If that time period is dragged out too long, what happened a year ago might not necessarily be relevant now. So if you think back to last year, and you had some of the regional banking crisis in the US, that volatility was affecting algorithmic performance.

You then go to the summer period, where volatility is pretty much the lowest it's been in a long while. Then our different algo is performing well in these different types of environments. So many different firms set everything up in different ways.

If you're going to have quicker turnover of algorithms-- again, you're aiming to find the best set of algos out there-- you may have a reduced broker list. If you have a reduced broker list, you might be allocating more to the number one performer on that list.

Therefore, your underlying clients are actually benefiting from that because you're getting-- the majority of your flow is actually getting better performance, so more cost savings.

PETER HAYNES: I'm curious, Robert, when you're scoring your brokers and providing them feedback-- because we're going to do that all the time. We're, Hey, how are we doing? How are we ranking? et cetera. You know that we're going to be, as brokers, constantly asking for that feedback.

I'm curious. There's probably a trade-off in terms of the amount of data you want to provide your brokers that are in the wheels. How do you figure out how much to provide so that you avoid giving too much and therefore potentially leading to gaming?

ROBERT MILLER: So I think an important thing to say here is I think the buy side-sell side relationship has evolved over the last five, eight years or so. We're definitely working much more as a team, I think, if we have clear objectives, and we're both striving for the same thing.

So we want the sell side to outperform in performance. How can we help you guys get there? And I believe giving certain bits information will achieve that. However, like you said, if you give some information out, it does lead to gaming.

And so what we have on our side is we have a number of different checks. So we have pages and pages of analysis that comes out, and we can pick up on these things. So number one, we don't give away certain things which we know can be gamed, but we also try to give away or to give advice and feedback that you can use so you can actually tweak and improve algos.

So one of the last things that we want is for all brokers to have the exact same algo on the wheel. Therefore, we won't know ever again if anything is going to outperform this because everyone's doing the same thing.

And so if you're currently in the number one position, we're not going to go and tell all your competitors this is exactly what you need to do. What we're going to try and do is say, look, try focusing on this area of execution or performance.

And so I think there's a very fine line between what you-- or what information you give away and how much you're actually helping.

PETER HAYNES: And so, Jenny, you're on the other side of that conversation, and you're talking to clients like Robert, who are providing you feedback where Cowen is part of a wheel. In your experience, Jenny, have you found that when there's a smaller list on the wheel, that the relationship tends to be a little more collaborative with the buy side account?

And I'm curious, what do you find in terms of the frequency with clients, in terms of how frequently they're reviewing their wheels?

JENNIFER HADIARIS: It goes right back to what Robert said. It's always a balance, I think, and it is truly a partnership because, ultimately, the wheel is about helping to preserve alpha, right? If the brokers do well, the clients ultimately are going to see less slippage. That means gains for their investors.

This is truly a great example of where interests are aligned between the broker and the client. It's going to be a balance because I do think, like Robert said, if you give too much information about the benchmark, we are brokers. We will perform to the benchmark.

So it sort of goes back-- bringing it back to the VWAP conversation, if you say that your benchmark for VWAP is a rival, the first thing a broker is going to do is tilt the algo to the open. So take a VWAP and ultimately shift the trajectory.

So you're taking a VWAP, and you're manipulating the VWAP to a certain point to beat the benchmark because if it's a rival, it makes more sense to trade more around when you receive the order. You're also going to do things like it would be natural to do small blocks or "I woulds, if you could," if the algo is beating a rival right now or is at better prices than a rival.

That is a natural inclination because if you're saying the benchmark is a rival, then great. I will perform-- what the client has told you is that a rival price is important to us. We want to do better than the rival price. So if you can pick up liquidity at or better than a rival, then you should.

So communicating a benchmark is ultimately dictating algo behavior, and that's where I think clients are really, really careful as they set up wheels in talking about what the benchmark is, what the target participation rates are.

Another kind of common misstep or a common way that brokers may react to wheels is that a lot of people have target participation rates on their wheels. So I want to be between 5% and 10%. Well, if the benchmark is a rival, the best way to perform better versus a rival is to hug the lower end of that band.

So there has to be some-- I've seen a lot of clients also incorporate some sort of metric of opportunity cost, which is-- I know I said 5% to 10%. If you're hugging 5%, but I could have gotten a lot of liquidity at better prices if you had been up towards 10%, I need to incorporate some metric of opportunity costs for not trading at the high end of my band.

And that's just some examples of where the benchmarks drive behavior and how clients are evolving. I do think if you go back to the right number, it's not just about getting the benchmark right. It's also about making sure that you're right sized because you have to have that good sample set.

There isn't a right number of brokers, necessarily, for all clients. It really depends on how much flow you have, how much of your flow can be put into the wheel, because there is a lot of order flow that we see that isn't necessarily appropriate for wheels.

And so how much of your order flow can you put into a wheel? How thinly does adding each additional broker spread that liquidity around? Do you need to wait a longer period of time, as Robert said? And can you really get a true sample size with the existing volumes and number of brokers?

And then clients really have to think about how much feedback they're willing to provide, the periodicity of that, and that goes back to the sample size. So you can definitely see a lot of noise from month to month. Who does well in a volatile period might do really poorly in the doldrums of summer in the middle of July and August.

And so you want to make sure you're cleaning out some of that month-to-month noise. But also, I think there's understanding this human phenomenon of the squeaky wheel gets the grease. We do definitely see that clients that measure on a regular basis, provide feedback, you tend to hear brokers talking about their wheels more, making adjustments more, going back and tweaking them and trying to outperform that feedback.

And so if you provide feedback on an annual basis, then probably brokers are saying, well, we don't want to change something up without-- in the middle of the review period, not knowing if we could be number one, and if we change something up, it could move us to number 10.

And we're totally unaware of that because we don't even know where we are in the standings or if we're performing well. So providing regular feedback allows brokers to make regular adjustments to the algos, and that's where I think that partnership works between the buy and the sell side.

PETER HAYNES: Jenny, is that a combination of in person dialogue or all data?

JENNIFER HADIARIS: Yeah, it's in person and data. Oftentimes, it's a couple bullet points of, you outperformed on this, you underperformed on this. This is where we would make adjustments. We do see clients with-- it really does take a lot of order flow, but to provide full specs on broker comparisons.

You ranked two out of 15. You ranked two out of 12 in this, this, and this. This is where you should be making adjustments. That's quarterly. There are absolutely brokers who provide that, and I do think data will drive people making adjustments in the algos.

And so you don't necessarily want to see too many adjustments, but also you do want to see that people are continually striving to outperform and to kind of beat the benchmark and beat their peers.

And so if I were starting-- we have a lot of clients come to us and say, we grew in assets in the last year. We want to incorporate a wheel into our order flow. How would you do it? If I were building a starter pack wheel, I'd suggest doing dark and blocks as one wheel, low, 5% or lower, and then medium.

And there's a reason I didn't include high because I think high is a really hard-- a lot of order flow that you would designate as high urgency might not be appropriate for a wheel. And I'd set those as your three strategies and incorporate a couple of brokers.

And then really also think about how you are going to move off the initial people that you're including and how you're going to manage the rolling on and off process because you want to create opportunities for people who come up with new solutions or are adapting to market structure changes or the evolving market environment.

As volatility increases, you want to make sure that you're not always using the same sample set of brokers because there might be someone that's not on your wheel who has an excellent algo for this period of time.

And so some sort of experimental slot where people can get on and get a small portion of order flow, and you can see how they route and how they behave and how they perform in certain environments. And then they might be able to graduate to the big wheel, and you bump other people off, and that's a continual process-- so some sort of plan for how to move people on and off.

PETER HAYNES: Robert, I'm curious. A final question on wheels before we move on here, but I'm curious, Robert, in your experience, do you think there's a unique skill set necessary to be the wheel manager on the buy side?

And do you feel as though the wheel-- because typically it's taking away what we'll call mundane flow from the day-to-day pad of a buy side trader. Do you find that it could be, at the end of the day, a reduced trader headcount tool?

ROBERT MILLER: So for us, we have a very unique setup. So we have PM trader roles, so they're actually the same people. So this tool, it actually fits very well into our infrastructure. The way that we set it up is we have a number of different working groups which involve a number of the traders.

And so we're discussing the different proposals from different brokers, and we have this open forum of where the results go. And I think having that collaborative input works well. So I don't think it's going to reduce headcount. It's just a different or newer way of doing things.

PETER HAYNES: And so Jenny was talking about opportunity cost on part rates earlier. I want to get into that topic, actually, and we'll just focus on unfilled orders. It's an interesting topic.

So when you see a dealer that's overfilling good performance and underfilling orders that are underperforming at benchmark, is this a big problem? And what kind of a penalty would you impose on a broker that you were sort of able to identify was behaving that way?

ROBERT MILLER: The biggest thing is your intention of trading. Why are you looking to trade something? If you're being very opportunistic, and you don't need to execute the whole thing, are unfilled orders a good thing or a bad thing?

Obviously, if you're missing liquidity that is there, then there's obviously some sort of issue. When in the case of a wheel, obviously, we're trying to measure broker or algo performance. And so you've got to be very careful with, Is it something we're doing on our side? So has there been a limit price issue that we've implemented?

And so how do we kind of get around that so it doesn't affect your performance? And so I think that's one of the-- anyone who is starting a wheel, that is one of the things that I would mention, as well, just to make sure that if you're measuring someone else's performance, make sure that the things that you're doing aren't impacting someone else's results.

PETER HAYNES: So, Jenny, Cowen's been on the leading end or early end of experimenting with using some of the new venues in the US and new exchange order types, and these order types typically have been designed more recently for institutional orders.

So I'm thinking about IntelligentCross, OneChronos, IEX, conditional order type, and you mentioned earlier trajectory crosses. And there's the marketplace PureStream that offers that service.

How do you at Cowen go about studying these new venues before you feel comfortable using them for client orders? And what is the process internally prior to using a new venue? Does it have to be approved by the [INAUDIBLE] committee at Cowen?

JENNIFER HADIARIS: Yeah. I think first and foremost the reason why we support new venues is because I do think, for a long time, there was a lack of innovation on the venue space, right? Dark pools do have a very different regulatory structure and can do some different things, in terms of counterparty restrictions, order types, conditionals, all those sorts of things.

And so we really want to support innovation in the venues space, and that's why we really go in and try and test some of these new venues as they come out. I also think as we talk about inaccessible liquidity, these agency-accessible venues, some of these new dark pools that are not owned by broker dealers, are vital to the lifeblood of the market and making sure that there's accessible liquidity to everybody.

For a long time, we had the majority of dark pools in the US were owned by broker dealers, and dark pools are not fair access. So at any point in time, any broker could really tell you, you can't come in to X, Y, and Z pool.

And so having some of these independently-owned dark pools to compete with some of those venues, I think is really vital and provides some competition in the market space and is a really good evolution. And so we want to support that and test them out.

And so that's why we've tried to be really first to market with some of these venues. And then I think how we test them is really you do some internal testing. You take your order flow from the cash desk, Rhys and Andy's teams flow, where we can use their orders and test them to try the new venues.

Get a decent sample set. Put it in front of best X. See where the new venues mark out relative to the existing dark pools that we access or the existing exchanges. This is the same with exchange order types as well.

Test it in order flow where it makes sense. A lot of times, we will try it first in an algorithm that has an underlying minimum participation rate because the idea is if I could get a midpoint execution-- if I don't get a midpoint execution, and I don't maintain my min part rate, then ultimately I'm going to have to cross the spread anyway.

So that order flow tends to be a bit more open to trying out new venues. So usually that's where we'll test out some new venues first. And then we get the data, compare like to like in terms of the venues, and then make decisions based on that whether to roll out new places, by default, to clients and then make clients aware that we're making those changes.

PETER HAYNES: Robert, in the same vein, how closely are you monitoring brokers that are using these new venues? Do you still have to closely police the firms with their own dark pools from not over-preferencing their own venues?

ROBERT MILLER: We do a lot of venue analytics on our side. If a broker uses a new venue, then we will get alerted to that straight away. We have comparison metrics, so we can see where brokers are executing and how brokers are executing.

I think it's interesting doing venue analytics on the buy side because we don't have as much information as you would on the sales side. So we don't have fill probability and things like that because we don't see some of those where you've gone to a venue and you've not been executed.

But what I do think is interesting is we can see how each broker is interacting with the same venue. And so you can see the different types of fill sizes coming back. And you can see the different price improvements, potentially-- if someone's sweeping different venues, how that is kind of going through and affecting overall parent-level order performance.

And so from that end, I think that's very interesting. And then also, if someone starts using a new venue, we will probably see the performance sooner than some other brokers who are still experimenting or who haven't connected with that. And so we kind of get an early look at some data on new venues that are coming out.

PETER HAYNES: I was taken by a quote from [? Matt ?] [INAUDIBLE] at a conference recently that I attended. Someone asked him about a sort of similar question around experimenting with new venues. And the way he answered the question was-- he said, every single order we execute is an experiment. We don't know what the right venue is or the right marketplace or right strategy on any order.

So in that vein, I thought it was kind of an interesting way. And at the end of the day, it's performance that matters, and that's what you're measuring. So, Robert, not a day goes by without someone claiming AI is going to change the trading landscape forever, and I'm sure you've had brokers in pitching you daily on some new, fancy, new-fangled algo that incorporates AI and will be transformational.

I know I'm exaggerating this point a little bit, but what has been the real-world outcome of these so-called transformational trading solutions that have been pitched to you?

ROBERT MILLER: So I think AI is a very interesting one. There's been a lot of talk about generative AI at the moment. I think that is probably going to be quite exciting in the risk and compliance space. If you're having a look at actual execution and execution analytics, it's going to go back to machine learning.

It's going to go back to, How can we do a better job of actually analyzing data and then making a decision based off of that?

PETER HAYNES: Am I exaggerating? Are brokers in regularly, coming to you, saying, hey, we've done this, we've done that? It's fancy. It's cool. It's our new object. Is that happening all the time?

ROBERT MILLER: I mean, that's their job.

PETER HAYNES: Yeah, I know.

ROBERT MILLER: But I have to figure out when I'm being sales traded.

PETER HAYNES: Yeah. Well, probably a lot, I'm sure, is the answer to that. My AI story of the day I got a kick out of that I read-- there was a golfer on the DP World Tour, which is in Europe. And they've got a new process when you finish your round.

You go and sit in a room, and rather than have a bunch of reporters there, there's actually AI-generated questions. And so you sit in a comfortable chair, and it's a robot asking these questions. And what they found is that the golfers are a lot more comfortable answering questions, and they tend to be a lot more open in their answers.

And so the whole story was about how this golfer had played so badly and just went on a tirade about how he played. And of course, he wouldn't have done that if it was 20 reporters in the room, so I got a kick out of that.

But, Jenny, as we finish up, I want you to think back in your career to someone you might have dealt with early on that was totally opposed to using electronic tools, married to the old-school strategies, and set in his or her ways, and that person eventually changed.

So I'm not asking you to name names, but I'm guessing you've seen this play out somewhere in your career. What was the catalyst for this particular person to become more open minded to new execution strategies using electronic tools?

JENNIFER HADIARIS: Yes. I've been called Captain Cannibalizer in my career, so yes, I've experienced this. No, I think this really goes back to what Robert said about a lot of these tools are not about reducing headcount.

They're about allowing you to focus on the things you need to focus on, improving your workflow. Taking orders and putting them in a wheel is not about reducing the number of traders we need. It's about allowing the traders we have to focus on the orders that are more difficult, the orders that need more attention, improving outcomes for clients.

I think it's not an electronic versus high touch. It's about how we work together on these to accomplish the best outcomes. And so I think, yeah, I've experienced this. But I think the way that I deal with that is always, whether it's on the buy side or maybe someone who's more dedicated to high touch order flow or say we really only want to do high touch, I think it's really about helping them to solve their workflow problems.

So there are absolutely clients who, based on their wallet, their amount of flow, right now electronic doesn't make sense, and that is absolutely fine. It doesn't mean that we can't work with their sales trader to come up with a solution and put something on the sales trader's desktop that solves an issue that they're dealing with.

So we have situations where clients prefer to do high touch trading, and the sales trader has a customized button on their desktop that trades the way that client wants to trade, especially if they have a specialized subsector of names that are more difficult.

And so helping to solve-- I really try to come at it always at trying to help them solve a workflow problem. So that really is, Are you getting five million pop-ups every time there's a contra side in dark pools? Well, we can help you with that and do a conditional check. And that way you only have to hit this one button, and you can hit all the conditional venues.

Or are you having a problem, going back to our earlier conversation, around the close and watching imbalances and paired shares and those sorts of things? Well, we can build a tool that helps you with that. And so how can we make your life easier to get you to test something out?

And that's really been a tried-and-true solution in working with your internal partners and working with the client partners and making sure that you're working within the constraints that they have, because I do think for every client, they have different needs.

And so whether it's what capacity they trade in, whether they use wheels, whether they don't use wheels, whether they use algos or not. And so there's always something that I think we could do to improve their experience and solve a problem for them.

PETER HAYNES: And, Jenny, you mentioned the difficult environment, and I saw that a lot towards the end of 2023, where traditional buy side firms had been so quiet during the year, they were having trouble paying their research budgets-- their bundled research budgets.

And there's always been that mentality that those budgets have to be paid through high touch bundled commissions, where a portion is for research and a portion is for execution. Are you seeing a lot of traditional research budgets where accounts are actually now paying for that via electronic executions, Jenny? Or is it still that tends to be done on the high touch desk?

JENNIFER HADIARIS: I think MiFID II really allowed a ton of flexibility. If there's one good outcome from MiFID II-- I know there's a lot of headaches. Robert, I see you smiling. MiFID II created a lot of headaches for a lot of people, but I do think what it did was allowed people more flexibility with their execution versus their research budget.

It allowed them to divide the two to pay the way they wanted to pay, whether it's check for research, and then I can just go and get best execution on the trading side. And so it really varies. It's rare that I see clients these days saying, I have to pay for something in X way.

But certainly, in lower volume years, when people want to be relevant to the sell side, sometimes they're targeting a budget. And sometimes the easier way to get there is at higher commission avenues. And so sometimes that means that they can only use higher touch solutions.

But there's still a way that you can be innovative and use some new algo products, even if that's the avenue that you're choosing because our number-one client that uses our algorithms is our high touch desk.

PETER HAYNES: Yeah, I was going to say, they're absolutely the center of the universe, from the perspective of testing and starting out and using those products. So that's great. Robert, I'm going to give you the final question here.

There are some market structure experts that see the next generation of trading involving the buy side taking even more of a direct role in developing execution strategies through management of proprietary data sets, even finding ways to tweak algo logic directly.

Dan Viner at BlackRock, I think, referred to this as citizen development, where his traders can actually do some of their own programming. Obviously, it makes sense. This is only practical at the biggest shops, like Vanguard and other places like that.

But I'm curious if you agree with the notion about the future of the buy side trader, or do you have a different take on how this will-- how algo usage will look in the future?

ROBERT MILLER: Obviously, I've said a number of times that the buy side is now much more analytical. The buy side is also aware of the parent order, which the sell side doesn't necessarily know the whole order or the intent behind the order.

And so I think it makes sense that there will be more buy side involvement in algo tweaks and updates. However, I still see-- like, even at some of the mid-sized or smaller shops, the buy side-- the sell side still have the most data.

And so if you're looking for a way to execute a certain type of order, the sell side are going to have many more data points than we do because you have all your clients put together. And so I think it's still going to be-- it's going to be led more by the buy side. But I think some of the analytics is going to be still being driven from the sell side.

PETER HAYNES: Well, I accomplished my goal here today, and that was to learn because I'm so curious about the evolution of this space. And I think, Jenny, both you and Robert did an amazing job at demystifying some of the things that we hear a lot about wheels and other topics like that experimentation.

Thank you very much, both of you, for joining today. I'm sure this is just the start of a discussion on this topic, and hopefully our listeners will find it of interest. And who knows? Maybe you'll hear from them wanting to ask more questions.

So, Jenny and Robert, on behalf of TD Cowen, thank you very much for joining Bid Out.

JENNIFER HADIARIS: Thanks, Peter.

ROBERT MILLER: Thank you.

[THEME MUSIC]

This podcast should not be copied, distributed, published or reproduced, in whole or in part. The information contained in this recording was obtained from publicly available sources, has not been independently verified by TD Securities, may not be current, and TD Securities has no obligation to provide any updates or changes. All price references and market forecasts are as of the date of recording. The views and opinions expressed in this podcast are not necessarily those of TD Securities and may differ from the views and opinions of other departments or divisions of TD Securities and its affiliates. TD Securities is not providing any financial, economic, legal, accounting, or tax advice or recommendations in this podcast. The information contained in this podcast does not constitute investment advice or an offer to buy or sell securities or any other product and should not be relied upon to evaluate any potential transaction. Neither TD Securities nor any of its affiliates makes any representation or warranty, express or implied, as to the accuracy or completeness of the statements or any information contained in this podcast and any liability therefore (including in respect of direct, indirect or consequential loss or damage) is expressly disclaimed.