Equity Algos Deep Dive Part I: The Resilience of VWAP

PETER HAYNES: Welcome to episode 57 of TD Cowen's podcast series Bid Out-- A Market Structure Perspective from North of 49. My name is Peter Haynes, and today we will dive deep into equity algorithms with the first of a two-part series. The equity algo space is one that I've wanted to cover for a long time on Bid Out. And I'm pleased to finally break down algos with my two guests Jenny Hadiaris and Robert Miller.

Jenny is very well known to most of our listeners as a leading voice in US market structure, and she also oversees the Equity Algorithmic Sales Coverage team at TD Cowen. As well, I'm joined today by Robert Miller, Global Head of Equity Execution Consulting for Vanguard. Robert and Jenny, thanks for joining the show.

JENNY HADIARIS: Thanks for having me.

ROBERT MILLER: Thanks for having us.

PETER HAYNES: OK. So Jenny, I'm going to start with you. When the TD Cowen merger was first announced, I realized pretty quickly that I might have been the only person in our entire capital markets area that actually knew someone at Cowen. And of course, that person is you. There was just literally no overlap between TD and Cowen, which is really the genesis of the merger.

I would run into you at the market structure circuit and at the conferences, and always respected you and really have enjoyed our first nine months working together. And I'm really glad it's finally got to the point where we found a topic to get you on the podcast.

But I'm guessing the first reaction you might have had on your side was to tell everyone at Cowen that knew nothing about TD or a Canadian bank that you had some experience previously working for a Canadian broker dealer. And you've probably told them, of course, Canadians are generally nice, which is what we are accused of regularly. Is that how it went? And tell us about your career at RBC, where you started, and how you ended up at Cowen.

JENNY HADIARIS: Yeah. Thanks, Peter. Yeah, is that how it went? Definitely. I was so thrilled to be working with you, honestly, and you have been just an absolutely incredible partner. You are generous with your time, relationships, your podcast, your conference. So I really feel like this is a great example of how the merger and working together is working well.

So yes, I did start at a Canadian firm. I started at RBC almost 20 years ago, or Royal, as you guys call it up there. I spent 10 years there. And that's really where I learned and built my experience in market structure. As much as I might have some Canadian experience in terms of the Canadian bank, I really only moonlight in Canadian market structure issues.

So I was really excited, when we heard about the merger, to be able to work with you and to pool our efforts. And I'm particularly excited about being able to take some of your research and build it into the algo logic, which we'll probably talk about a bit more.

And then as far as how I came to be here, I joined Cowen in 2016. And I think what drew me to that role, as opposed to some other roles, was that it was really integrally tied into the algo product. So it wasn't just a market structure research role.

But it was also helping to shape how our algos worked, which I think was a turning point in my career because taking the theoretical market structure issues and then building them into algo solutions so that you can present a problem in the research and then say, hey, we also have something built here that can help you with that, that's been really critical to our success.

And so I was happily working away on our US electronic offering. And then we bought Convergex about six months, after I started at Cowen, in 2017. And that really kicked off a multi-year process of building our algo logic onto Convergex's global execution and global connectivity.

And I think the key theme is that we've always tried to lead with market structure expertise in any region. So if we can know the ins and outs of how the market structure works and the particular problems for that region, we can really build better algos around that. So we expanded in Europe following the Convergex merger. And we hired James Baugh, who is the most incredible European market structure expert.

And so now, with our teams combined with you as well, I can confidently say that we have one of the leading experts in each of the regions that we touch. So I think the market structure team has been one of the biggest beneficiaries of the merger, and maybe APAC next.

PETER HAYNES: Yeah. Well, I completely agree with you. And in fact, one of your pit stops between RBC and Cowen was Deutsche Bank, I know. And it was at Deutsche Bank that you got to know Heidi Fischer, who's a very good friend of yours. And it was actually Heidi who recommended Robert as the speaker for this-- the external speaker for this podcast. So thank you, Heidi-- who's now at the TMX-- for recommending Robert.

And Robert, why don't I turn it to your attention here? You were appointed to the current role you have at Vanguard in 2022. But you had extensive experience with algos from your previous stints in the securities industry. Can you tell us about the key spots that you cut your teeth in at over the last 20 years?

ROBERT MILLER: I'd probably start even before I went into the industry-- so going all the way back to when I was at school. The subjects that I was specializing in was maths, physics, and computer science. And it was whilst I was studying those I did a project where I looked at how the life markets or the London futures market went from open outcry to electronic trading.

And ever since then, I was hooked. I was like, this is what I want to do. I just found it absolutely fascinating how everything worked. And so once I left education, I managed to get a job trading. I think I've been lucky enough to actually experience trading in many different roles. So I've worked with private clients. I've worked at big asset managers. I've worked for hedge funds. I was market making for about 10 years.

And it was there where I started looking at market structure algos and how if you don't get your transaction costs right with your algo deployment, it can really affect your overall strategy. So you're going to have something that's absolutely is a winner on paper. If you don't take the costs into consideration, then it might not be a profitable trading strategy. So that's where I further developed my interest on the TCA side.

I then went over to the sale side. And I spent some time looking at algos and seeing how we can actually improve our current offering. So we were looking at things like venues. We were looking at timing, all sorts of things to try and gain an edge there. And that's when I was offered the role at Vanguard.

So this is just coming up to about five years ago. So I originally started in Europe, and then I moved over to the US in 2022 to look after the global algo rule process and the analytics team on the equity side.

PETER HAYNES: And now that you're back over in London and settling back in, I guess getting the family re-acclimatized to the London market-- and you mentioned your market making. You used the term algos. I was having a side conversation with Jenny recently about how I found it funny on a panel that she was on-- internal TD event-- where there were four people up there from four different areas of TD all using the term algo, all meaning something different.

For instance, the market maker for municipal bonds referred to the algo. Well, his algo-- what he thinks of as algo is really what's pricing the engine that's changing all the prices of the million different QCIPs that he's worrying about in the municipal market. So it's just funny how everyone has a different use of that term algo.

And we're going to focus today on equity algos, which I would think, really, is the sort of center of the universe when we have the conversation around algorithms. And of course, I want to start Jenny on VWAP, which is a much-maligned but still very popular trading benchmark-- Volume Weighted Average Price. Jenny, tell me why does VWAP remain the primary execution strategy for equity traders?

JENNY HADIARIS: Yeah. I have a funny story from my old Canadian days. I had a bumper sticker years ago in my desk at Royal that said, VWAP is for wimps. [LAUGHS] And I think that's really a misconception. I think the perception that VWAP is sort of antiquated, that you advance beyond VWAP is because, really, it was sort of the first algorithm. It was the way to stretch orders over the day. It's the OG of algorithms.

And in a lot of regions where maybe they're getting into algorithms to start, VWAP is usually one of the first algorithms that they use. Maybe because dark isn't as prevalent, it's usually the first strategy.

And I do think that the idea that it is antiquated is somewhat of a misconception. There's this common belief that VWAP, or spreading it over the day represents a lack of conviction or that you're unwilling to pick spots. As we've seen a lot of the market move to more passive algos, VWAP remains-- it's really still an excellent option for passive execution, reducing your timing risk.

There's also a ton of order flow in the market-- and I'm sure Robert sees this as well-- that is systematic in nature, model-driven strategies, transition flow, where the midpoint between the bid-ask and arrival is an incredibly arbitrary benchmark.

Especially if it's around the open and there's a ton of volatility at the open, there's been a ton more research as well done on the buy side, evaluating how much alpha their flow has, how much perceived alpha may be versus how much actual alpha the order flow has. And they're using that analysis to determine how quickly they should be trading.

And so the end result-- and this isn't necessarily a new story-- is that a lot of people have slowed down their algos. They're targeting lower participation rates. And with that, we've seen a return to VWAP usage. And VWAP is the core benchmark for those strategies. I think we have a lot of clients that are using an over-the-day strategy, a slower strategy, and that they're still using some combination of Arrival or VWAP as the benchmark for the algos.

I think it's because there was an old way of trading VWAP and that was-- that you were really measuring your VWAPs based off of how much spread capture they were able to accomplish-- and versus the VWAP. And one of the easiest ways to beat the VWAP was to be on the passive side more often.

The first route for a lot of VWAP algos was to post out loud at the near touch. If we look at market structure and how much it's changed over the past several years, the amount of volume traded on exchange, particularly displayed on exchange, has just dropped year over year over year.

And so what we've found is that clients are seeing that the traditional VWAP algorithm-- while the benchmarks might still work, the traditional VWAP algorithm, posting out loud at near touch, causes a lot of slippage versus Arrival. You're really seeing a lot of signaling or information leakage on those orders.

And so people have started to use a combination of VWAP and Arrival for their over-the-day strategies because they're trying to capture some of that slippage that you're seeing or information leakage on the exchanges. And so a big trend these days is to switch to a dark-preferencing VWAP so you're still using the benchmark but really adjusting how the algorithm works, trying to prioritize midpoint or dark fills instead of lip fills to get the benefit of blending the order over the day but avoiding some of the information leakage that you're seeing from posting out loud.

And I guess the last thing I'll say on VWAP is I think it was also historically maligned, as you say, because there's a perception that you can't really run into natural liquidity if you're spreading orders into small pieces over the day. And in recent years, we've had a ton of technological solutions that help with that. You see trajectory crosses have been really revolutionary in solving for that problem. And we've seen a big uptick in their adoption in the US.

So you can be an over-the-day buyer and a seller and match up with natural liquidity and just follow the trajectory pattern or the volume throughout the day. And so my long answer is, I would say that my bumper sticker no longer applies, and VWAP is very trendy again.

PETER HAYNES: Yeah, that's amazing. What goes around comes around. I'm curious-- you mentioned that it's a benchmark, and naturals want to run into naturals. But if you're a liquidity-seeking trader or perhaps a short-term alpha, how often are you seeing it, Jenny, where they will trade off that short-term alpha to run into naturals by slowing down their order and coalescing around VWAP? Is that happening frequently?

JENNY HADIARIS: It depends. It really depends on their aggression, their alpha. Again, are they willing to slow down? Because oftentimes, what we're seeing is if you're in a liquidity-seeking algo and you pair off against a trajectory order in some of the venues that offer trajectory, it does really-- you could be running along at a good 40% of volume clip, and you pair off against someone in a trajectory venue or trajectory order type, and you, all of a sudden, slow the algo down to 5% if the contra side is targeting 5% of volumes. So it does feel like you've almost hit the brakes on the algo.

And that's one of the issues that we've seen in pairing off liquidity seeking order flow with some of the over-the-day or trajectory orders. And that's something that I think people are grappling with as we see some of this order flow out in the market because at the end of the day, it's really good liquidity. It doesn't move the name. You're pairing off against natural liquidity. But if you have a lot of conviction in the prices that you're currently trading at and you were happy being 40% of volume, you don't necessarily want to interact with that order flow.

PETER HAYNES: So Robert, you mentioned that you fell in love with TCA. I want to talk about TCA from the perspective of performance measurement around VWAP algos. How do you measure the performance of one broker's VWAP algo from the next broker's VWAP algo? Is it simply Arrival? Is it interval VWAP? And do you adjust for outliers or fast markets?

ROBERT MILLER: So I think there's many different ways that you can view your algo usage. One of them is, are you going to be using VWAP versus liquidity-seeking strategy, versus more of a passive IS strategy? And so depending on what you're trying to measure, I think you'll use multiple different benchmarks for that.

And so if you're trying to determine which strategy is better, you might use something like Arrival and lean on that a little bit more. If you're trying to compare different broker VWAP algorithms together, then you're looking at VWAP as a benchmark. You're looking at some of the other things that get the difference from between the two different brokers, so venue analytics, timing, whether it's passive, whether someone's posting liquidity or taking liquidity.

So all of those things we feed into our analytics systems to try and understand exactly how brokers are piecing together their VWAP strategies and whether it's the right thing for our strategy at the end of the day.

PETER HAYNES: In the sense of outliers, where you have an example of something that just completely skews everything-- I guess if you have enough data outliers kind of maybe just average out-- but I'm curious, do you have to worry about outliers?

ROBERT MILLER: I mean, I think everyone should observe or look at outliers. Again, if you have enough data points, then maybe it's not so much of an issue. But I think it's interesting looking at outliers on both sides because it can help you redefine your strategy. If something is an outlier and it's actually performing well, is that something that we need to look into? If it's performing negatively, is that something that we need to be aware of so we don't run into that sort of issue again?

PETER HAYNES: Right. So that makes perfect sense. And when you're thinking about participation rates-- I guess we'll shorten. The industry likes to call them part rates. But when we're talking about participation rates for a VWAP strategy, what is the rule of thumb in terms of what percentage of participation you can have for a strategy before it causes too much impact?

ROBERT MILLER: I think the way to look at it is you need to classify similar order types, similar security types, and then try and get the best participation rate for that cluster or for that kind of type of security. If you've got a bunch of orders that are below half a percent ADV, for example, you're going to have a much different participation rate than you would if an order had a higher urgency or was slightly bigger. That, again, is when you're going to start using different flavors of VWAP.

So you may use-- if you have more opportunistic style of trading-- something that is less sensitive to keeping up with the schedule and something that is maybe looking a little bit more in dark pools or alternative venues for bigger block liquidity.

PETER HAYNES: So Jenny, Robert mentioned ADD, Average Daily Volume. Algo participation rates for schedule-based algos, whether it be VWAP or percent of volume, rely on accurate assessments of average daily volume or tradable volume. I know you've done a lot of work in the US on the concept of what we'll call accessible volume. Can you explain the reason why this is an important topic? And how have you, at Cowen, incorporated that research into your algo execution logic?

JENNY HADIARIS: Yeah, absolutely. I think whether it's VWAP, liquidity-seeking algos, close algos, anything, almost all algo strategies that are built globally-- dark algos, anything-- are built around a target participation rate. Or they have tactics that key off of the percentage of volume that you're achieving. So unless the algo is solely spread-based, you have some sense of volume that it's trading. And so there's a volume clock to the algo is kind of how I think of it.

And that's where accessibility of the volume comes in. So when you were asking Robert, what is what percentage of volume is considered impactful? Well, the key is you have to have a good sense of how much of the volume is actually accessible in the individual security, and it really-- it varies. And that goes back to elementary math.

So ultimately, in anything with an underlying participation rate, you're looking at percentage of volume. So volume is your denominator. And if you don't have an accurate understanding of what that denominator is, it's going to throw off all of your other analytics. And if we're using percentage of volume as a sort of proxy for how aggressive or how much impact people are willing to have, then that's where this becomes really critical.

So we've all been saying for years, at every conference we've gone to, that volume does not equal liquidity. So if your denominator is volume but the volume numbers we see on Bloomberg, for example, are not an accurate representation of how much liquidity there truly is in the market, the net result is that 10% of volume can be hugely different impact in one stock versus another.

So I guess the best example we have-- and this is where we kind of wrote some notes that are-- and you can find them on the website. But if you look at some of the meme stocks, which were obviously hot the past several years, retail orders in some of those names can be 30%, 40% of the day's volume.

So let's take a meme stock name where retail orders are 30% of the day's volume. In the US, a large portion of retail orders are split off and traded with wholesalers. And so those orders may or may not be accessible to an institution. So an institutional algorithm targeting 10% of volume in a name where 30% of the volume is just going and getting paired off at wholesalers, 10% is actually more like 14% of the accessible volume in the ticker.

And that's a much different rate of trading and expected impact than 10%. I think a lot of people would tell you that 14%-- the sort of in between 10% and 20% is sometimes, like, the death zone. You're not trading enough volume to complete the order, but you're not trading slow enough to not have impact. So you're sort of in this median place, which is a really, really difficult place to trade.

And so the key here is that we really don't have a perfect way of measuring accessible liquidity in the US from one stock to another. I think a lot of algorithms try to take out block volume that occurred that they didn't interact with because you don't want to be chasing volume that traded on a block that you didn't have access to.

But we don't have a great metric for, like, how much of that retail wholesale activity happens or how much volume is happening in venues that you don't go to. Like, another broker central risk book would not be accessible to your algos, and so how do you take that out? And we did some research a couple of years ago to try to take how much volume trades off exchange but not in a dark pool. Because at the end of the day, most of our algorithms, by default, were going to exchanges in dark pools. So our denominator should only be exchanged in dark pool volumes.

And so we created an inaccessible liquidity index. And people use it differently. I think the idea was you could haircut your target participation rate, so you can haircut that 10% for how much of the volume actually is accessible to the algorithm. But a lot of people kind of use it differently, where they're selecting different algos based on the rates of inaccessible liquidity.

So I want to trade a meme stock name differently than I would trade a super liquid, very institutionally focused, highly accessible name. We've even seen a lot of people looking to adjust the strategies intraday because some of that inaccessible liquidity tends to be centered around the open.

And so do I want to do something different in the first 30 minutes of the day, maybe not chase volume, not track volume, take all minimum participation rates off the algorithm in the first 30 minutes of the day? And even though the client or the wheel strategy might have an underlying base minimum participation rate of 5%, 10%, I want to strip that out in the first 30 minutes of the day because I don't want to chase volume that I don't have access to.

People are using it differently. I think people are trying to be smarter around the volume that they're chasing so that they're not having unintended impact, especially as people slow down their strategies, like we've talked about.

And I think the next evolution of this research will be really looking at some of the volume around the close because I think the perception has long been that we've moved volume into and on the close because that's a lot of institutional participation, theoretically. It's all accessible.

But Peter, you and your team have done some exceptional research in Canada that shows how much volume on the close is actually paired off at 3:50 before you might even have a chance to really interact with it. And should we be looking at that? Should we be applying some of these same concepts of inaccessible liquidity or some of these same algo haircuts to that volume?

And also, could we expand that research to the US? Because oftentimes people want to be x percent of volume into and on the close, or they're targeting being 20% of the closing auction volume. And if a huge percentage of that volume was already paired off by the time you submitted your order, then you're really targeting a much higher percentage of the overall close volume than you possibly intended.

PETER HAYNES: And I agree. It is an important topic. We've been looking at it. And the market structures are different in Canada and the US and globally. But it is an important topic. We can't lose sight of the fact-- even if there is a significant amount of noise in that on-close percentage, the truth of the matter is there still is a lot happening at the close.

And I'm curious, how has Cowen changed its algo execution to reflect the shift of significant percentages of trading in the continuous markets happening in the last 15 minutes? Once the imbalances come out at 3:50, all of a sudden, there's a huge amount of interaction with those imbalances, but also in the continuous market.

I'm curious-- what do we talk about from a benchmark perspective on closing algos? And do you think, now, of your closing algo strategies separate than you do any other schedule-based strategies? Are they totally different?

JENNY HADIARIS: It's actually-- more and more people have asked. There were a lot of closing strategies that were separate algorithms to your over-the-day strategies. And the number one request, I would say, over the past three to five years has been to incorporate those opportunistic close optionalities strategies or tactics into the underlying over-the-day strategies, just because the close is, like you said, a phenomenal liquidity opportunity.

So if we're in a challenging liquidity environment, from 9:30 to 3:50, all of a sudden, at 3:50, we need to get everything done. We need to be watching everything. We need to be making sure that we're seeing if there's a large close in one of the illiquid names we've been trying to trade. We need to see if there's an imbalance in our favor. We need to be trading aggressively the last 30 seconds. We need to be growing if the close gets larger.

And when you think of that from a workflow perspective, from the institutional client's seat, it's incredibly hard to be watching all the names on their pad in this condensed period of time, reacting to everything that they're seeing on the screen.

And so that's where I think people have asked for these close tactics to be incorporated into the other strategies so I'm not having to go in as a buy-side trader and select a target close strategy in the final 30 seconds of the day in a name that might be trading a lot of volume. I need the algorithm to be smarter and switch into that.

And so I think the biggest change that we've seen is really time-of-day switching and switching into different tactics or underlying strategies by time of day and then incorporating some of those opportunistic close reactions into the algorithm because it is just an impossible workflow issue to be watching all the names on their pad and trying to react to the imbalances.

You're seeing more people asking for that. Like, I know I'm going to have a carryover situation in this name. I don't want to push the name. I don't want to put an underlying min participation rate of 40% or something on the algorithm in the final couple of minutes of the day. But if the liquidity shows up, I want to be there.

It's almost like the final 2 minutes, 30 seconds in the close are the new block or conditional period where if you miss that liquidity opportunity, it's like missing a huge natural block at 12:30 in the middle of the day. And so I want to make sure that I'm there and I'm represented if the liquidity shows up on the exchange or if it grows or if the close grows. But I also don't want to push the name unnecessarily. And so that's a lot of the requests that we've seen.

And then I think the benchmarks, like you said, for the close have really had to adapt. Because at the end of the day, if you're targeting 40%, 60% 20% of the volume, even, into the close, let's be honest. You are the close. So if your benchmark for the close strategy, which a lot of them were, is close, ultimately, it's very difficult if you're running at a huge percentage of volume to miss that because you're ultimately driving where the closing price is going to end up if you're a large percentage of volume.

And so incorporating some metric of how much slippage versus Arrival I saw, relative to how much volume I accomplished-- so a slippage versus Arrival as a measurement of percentage of volume is a lot of what people are doing. And it's really interesting because I think what we're seeing is these hyper-aggressive orders work for large percentage of ADD orders, and maybe not so well for small percentage of ADD orders, especially if you're completing your trading aggressively and completing 30 seconds prior to the close and then seeing a lot of reversion into the close.

It's about balancing how large the order is, how aggressive you want to be, and then merging the benchmarks of participation rate, size of the order, and how much slippage you saw.

PETER HAYNES: For a person that grew up in the index world, and obviously, Robert's firm at Vanguard knows what I'm talking about-- we used to think of the close, and it was all about the index and index flows. And this whole conversation we've had so far has been unrelated entirely to index rebalances. It's just daily activity where there's things going on.

That will do it for the first part of our two-part series. Thank you, Jenny and Robert. And I hope our listeners will tune in to part two, which will drop in a week or so, where we will shift attention to algo wheels, AI and algos, and the process for incorporating new algos into algo routing. We thank you for listening, and we'll be back at you soon.

[MUSIC PLAYING]

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.