TDAM Talks ETFs Podcast: Quantitative Investing - Navigating an Ocean of Data

ANNOUNCER: TD Asset Management welcomes you to this week's podcast. As a reminder, this podcast cannot be distributed without the prior written consent of TD Asset Management.

TOM GRANT: Hello and welcome back to TDAM Talks ETFs, a podcast about exchange traded funds in the market in Canada and TD's ETFs. My name is Tom Grant and I'm the ETF Capital Markets specialist here at TD Asset Management. In the podcast, we'll be discussing quantitative investment strategies. Today. We're joined by Julian Palardy who's a Managing Director here at TD Asset Management and the head of Quantitative Investments. We're also joined by Jonathan Needham, who is a VP & D here at TDAM and the Head ETF distribution. Jonathan and Julian, welcome.

JULIAN PALARDY: Thanks. Pleasure to be here.

JONATHAN NEEDHAM: Thanks. Tom!

TOM GRANT: Yeah, no problem. Now we're going to talk about quantitative investing and usually that invokes thoughts of algorithmic investing using models, predicting the future as such. A lot of this is computer based.

TOM GRANT: Sometimes I get you thinking about the Big Short, etc. in terms of from a movie background perspective. But I've got to say, Julian, I don't know if we've ever met in person, but I'm happy to learn that you are not a computer, although I've heard your IQ level is somewhere up near one. So, it's a pleasure to have you here. Exciting to have you in the office and here and join us in the in the studio. Now, this is the first time you've been on (the) TDAM Talks podcast series. Tell me a little bit about yourself and tell me a little bit about your team as well.

JULIAN PALARDY: Yeah, sure. So essentially, I joined TDAM 17 years ago now, and so it's been a long time, but I still do remember those days. I joined right the way on the Quant team. So, Jean Masson (previous head of Quantitative Investing) hired me back-to-back in those days and right out of school, essentially. And before that, I had a previous career in computer security - poking holes in in software and finding vulnerabilities. And now finding vulnerabilities in a market is essentially and I grew up with the Quant team even though the Quant team dates back from way before I joined it. So it's actually - the team dates back from 25-26 years ago approximately. So probably one of the first quantitative equity teams in the country. And I was very proud to join it. And I worked on Alpha Strategies, I worked on portfolio management and eventually took the lead on the quant portfolios and eventually ended up heading the group Jean retired... and the team, as I mentioned, very old-tenured team in the country, but also probably one of the largest as well. So, we are very ... I'm very proud of that. The individuals on the team are very qualified, extremely skilled individuals with very different backgrounds as well.

TOM GRANT: Well, we're very lucky to have you and in the team here at a TDAM and certainly your track record is one reason why now use the term quantitative investing. So let's just unpack that one just a little bit. Can you explain what quantitative investing is in the simplest terms possible, please? Yeah, it's fairly wide in terms of definition. It can go from ... all the way (to) your like fully active quant strategies doing factor timing or statistical arbitrage. We can go pretty deep and model construction to do an alpha, but we can go into things that are much lighter as well, such as factor investing. Next static factor investing, or even to some extent passive investing when you do not follow application but optimized replication. There are models that are involved in this, so all of this would fall into, in my opinion, and the grand family of quantum investing that the degree of touch will vary, and the degree of value add will vary as well and the objectives will vary. But at the end of the day, it's all about using models to to to make investment decisions. So instead of having an individual going over our numbers and deciding what stocks to buy or sell, it comes down to delegating, building a model that will do that job on a day-to-day basis, largely.

JONATHAN NEEDHAM: Awesome. Julian, again, pleasure to have you. I know you know, I'm a huge fan of evidence-based investing and quantitative strategies and you know, we talk about them often and whether it's value growth, momentum, liquidity, low-vol(atility), so on and so forth. But we also hear quite a bit of, let's say, myths and misconceptions out there in the marketplace. What are some of the ones that you hear most about and would like to squash, if you will, here today?

JULIAN PALARDY: Yeah, there's plenty of them unfortunately. So, the first one that I hear very often is that quantitative is “low-touch”, if you look at our team, you're going to see that it's slow-touch, we need people to build quant strategies and we need data as well. We need technology. So, it's a big investment. What is true, though is that quant tends to be fairly scalable. So, once you've set up, once you've put in the investment that's required to build a team and acquire the data and set up being the technology, then you can scale to much larger amounts of AUM than typical fundamental strategies that are high touch when it comes to individuals managing the money on a day-to-day basis. For quant, you can scale to larger numbers of funds and larger amounts of AUM as well. But overall, it's a big investment, so it shouldn't be seen as something that is necessarily cheaper to, to do than fundamental management. And the other thing that some people unfortunately tend to think is that it's easier to do possibly because when it comes to managing the portfolios day to day, no human beings are involved. But building those models is extremely complicated in the first place. So that's why at the end of the day, in the research team PhD’s that do this, and PhD’s are not cheap, as you can imagine, they come extremely qualified, and you need to compensate them consequently. So, the type of skill that's required is different, but it's nowhere easier - whenever it comes to generating alpha sustainable alpha and the markets, typically it’s tough. At the approach may differ, but it's always going to require some degree of skill. The fact also one thing I hear quite often is that quantitative investing is backward looking, while fundamental is forward looking. This is absolutely not true. And while it's true that human beings have the gift of imagination, quite often as we refer to it, so they can come up with ideas or thoughts about things that never happened.

JULIAN PALARDY: And the best machines are extremely good at taking the past and extrapolating into the future. We're generalizing into the future with as little as few biases as possible. And human beings tend to be really bad at doing this actually. So when it comes to managing money and extrapolating from the best without any biases, even when there's not too much data at play, humans are quite bad at making good decisions, even with few data points. So again, I would say that all techniques, human-based or model-based are typically about looking into the future, but quite often learning from the past.

JONATHAN NEEDHAM: Thanks for that, Julian. I mean, I can tell, right? It's not low touch, you got a big team, a very qualified team. It's not low cost - PhD’s - I call them propeller heads. Much smarter than this guy and a phenomenal team that you have. Yet we run our strategies at a very low-cost structure. So, I always like to talk to the investors about how these are actually giving you an active-like experience, but somewhere between passive and active cost. And so, the lower the cost hurdle, the more returns you keep. And so, another one of the reasons why I'm a huge fan of these strategies, because these factors tend to persist over time. You tend to get better risk adjusted returns and you have a low-cost hurdle to overcome in order to achieve those. Of course, in the ETF structure, being scalable is also music to my ears because obviously we want to be able to grow these and get more economies of scale and help more Canadians get better risk adjusted returns. And so again, why these are very suitable in an ETF structure. And so thanks for squashing some of those myths out there. Yeah, I think I think I'll hand it back over to you, Tom. I know you've got a couple other burning questions you'd like to ask.

TOM GRANT: Yeah, like, and it comes up a little bit, and sometimes I find that I just need a little help understanding some of these concepts.

TOM GRANT: So, you brought up the term “alpha” and ... how ... and I think people say and Jonathan, you talk about it sometimes you say how, how we capture this. So, can you just define a little bit for us what is alpha and how do we capture it?

JULIAN PALARDY: Yeah. So, there are multiple definitions of alpha, obviously, and one clear definition in my opinion is sustainable outperformance. And by sustainable, I mean that it cannot be the result of historical luck, which means that it cannot be also the result of a specific tilt that you maintained on a specific style. Because at some point people are going to find out if it worked in the past. People are going to find out that some point and they're going to try to replicate that. So alpha by definition cannot be easy to replicate, otherwise it's going to disappear. So that's part of that sustainable aspect of the definition. And it needs to come with scales, and it doesn't matter what was the format of the scale. Ultimately, it can be human beings, managing the portfolio is making stuff stock picks or building models. As long as there's some degree of scale there that's difficult to replicate. How is it captured? And that's a long discussion, a problem you are going to have for the rest of the podcast. But ultimately it comes down to building models that best extract information from historical data. And there's a lot of data on there that you can use to do this and is compressing this information or this data and to a small amount of information that you can use to extrapolate or generalize into the future. So quite often when people will ask about what quant is about, my best to word sentence would be data compression, which is literally it sounds like nonmathematical, more like computer and computer science stuff, but it's literally like everything that comes to Statistics, AI, it's all about data compression. It's turning a lot of data into a small subset of extremely useful information to have a view into the future.

TOM GRANT: it sounds like certainly math, science, all these fantastic elements that go into the construction of your portfolio and there's a lot of other terms that do come up just so that they're not confused, such as modeling, risk reduction strategies. And perhaps maybe what you can do is you can just talk as just through a little bit more about what the science is behind these strategies that you're incorporating.

JULIAN PALARDY: Yeah, if you if we want to simplify things a whole lot, I would say it's all based on the statistics that you learned in high school, but it would be kind of an understatement because ultimately we wouldn't be hiring PhD’s if that was the case, it would be hiring graduates from high school.

TOM GRANT: Yeah, for a second there at the future and in quantitative investing. But that got quashed pretty quickly because I didn't get my PhD - Jonathan, did you get one?

JONATHAN NEEDHAM: I have not.

TOM GRANT: No? Okay, I guess we'll leave it to the experts then.

JULIAN PALARDY: So the thing is, a lot of people build on those things, those basic statistical principles and obviously statistics itself is as much deeper than just what you learned in high school. But on top of this, people have built signal processing. On top of that, they've built econometrics that they've built machine learning and AI on top of that. So all of these things are fields from which we draw technical knowledge to build our strategies, and that's why we hire people with diverse backgrounds. Like my background is econometrics, because back then machine learning was not super popular, so not as much as today. So there was there was not a whole lot of people in that and that space back then we had more people with statistical science type backgrounds, lots of people in finance. But we're building more and more with people in computer science and machine learning as well today. So it draws from those various fields and the data compression example that - not even example, but generalization that I use before is the best way to represent things like we swim in an ocean of data and we need to compress that portion of data into something that fits in the pool in your backyard. That's why all the models ultimately end up doing this. So you can imagine, like I was talking about statistics: compute the average between ten points and you reduce those ten points to a single number. So this is what all models are doing, just a bit more complex models than this.

TOM GRANT: So, I guess we did talk of a backwards looking forwards looking. And what I want to ask is, you know, as we sort of think about the future and what might come next in the marketplace, like how do these strategies perform in an up versus down markets?

JULIAN PALARDY: It depends on the objective that you have. So, if your objective is to deliver a very strong asymmetry between upside and downside, typically what you're going to end up with is a strategy that is fairly low vol because these tend to deliver market like returns over the long run, but less volatile (inaudible), you end up with more upside, significantly more upside and downside capture, even though it's not always going to be the case. But in the long run, this is what we observe. Then you're going to have alpha strategies where you increase your upside, you increase your downside, the gap is going to decrease, but you're still going to have a gap between upside and downside capture, typically more than 100% upside capturing up markets and downside capture is going to be less than 100%. And then other types of strategies where you go into the high and both upside and downside space, these are pretty rare in the quants space, I would say like growth (inaudible), high volatility type of strategies where you still get do to beat the market on a risk adjusted basis. We don't see many of them because these would fight the against the low-vol anomaly, so not necessarily an easy thing to do. But generally speaking, there's going to be a quant strategy for every objective out there, to be honest. And they can all do well, both on a market relative basis or on an absolute basis. It depends. If you have this, you can use the same ingredients to deliver a different recipe for each investor.

TOM GRANT: Now, I know that when you were talking about artificial intelligence and machine learning a little bit earlier, John, eyes lit up. So, John, do you have something you want to ask on one?

JONATHAN NEEDHAM: Yeah. I mean, this is interesting because we're hearing more and more, Chat GPT and everything else that's out there in the marketplace. And so, we're hearing more about AI. In fact, I think we just released a podcast discussing it and in machine learning. So, is there any of this in your quant strategies that you're running?

JULIAN PALARDY: There is, but it's not as exciting as Chat GPT. And I say very early in our in our process, we add in our Alpha strategy to add signal processing techniques that are used today not in AI, but they use this in self-driving cars. There is in our risk modeling, we use something that we call the PCA – "principal component analysis." This is like the most basic version of pattern recognition that you can get. And it's it was used historically in the analysis of images and extracting features from images as well. Now there are much more advanced techniques that this uses, but this is a linear version of industry, and this is probably the simplest model feature selection as well. This this is the kind of stuff that we do in our Alpha models today. And this is these are things that are done in AI models as well and machine learning models, but 80% of our research pipelines. So, it's worth mentioning, is really machine learning and AI related with techniques that are much more advanced than these. And these were like the building blocks of machine learning back in the day is now we're advancing beyond this. A significant part of our research is on natural language processing. So, AI reading textual information and turning this into numerical insight that can eventually be used to beat the markets or reduce risk. And eventually I'm not going to stop there. We can open the door to any kind of unstructured data analysis using machine learning techniques. So, it's just at the end of the day, what we need are trade lists.

TOM GRANT: It's not like a script written by an AI, and we don't need generative AI that much. We need decision making type of AI and this stuff already existed way before Chat GPT.

JONATHAN NEEDHAM: This is cool stuff - I don’t know about you, Tom, it kind of gets me fascinated the type of information that you can gather. And so, let's talk a little bit about that. Like what about all the data right? There continues to be more and more obviously on a daily basis. How important is it? How do we collect it, how do we analyze it? Why is it important and how do you make sense of it?

JULIAN PALARDY: It's insanely important. In fact, I would say that it is a core of everything we do. A lot of people think that the modelling part is the most cool stuff, but actually not none of that happens without having the data. And the reason why we see, like all the training for Chat GPT is possible because of all this data that's available on the Internet today and its open source, largely, it’s like Wikipedia corpus, for example, people write an entire encyclopedia, and you get train models based on this stuff. If that didn't exist, it would be really tough for individuals to come up with things like large language models. So, and it's the same thing in finance. All our models rely on their being fed on data, and there's more, more and more of it. Fortunately, and quite often we refer to a notion of data to express the notion that there's an increasing amount of quantifiable data out there. And it's not all structured data, it's not all numerical data. A lot of that is sectional data, voice data, image data, videos, etc. So individuals are faced with an increasing amount of data to make their investment decisions and human beings tend to drown in this data. And while I like to think that quant models tend to swim in this in this data, and so when it comes to practical aspects, now most of the data that we use is not 100% public open source that's derived from public information, obviously, but it comes from vendors like most quant firms, and we have multiple vendors that we deal with. But even though it may sound easy like this, getting data from vendors is not a slam dunk. You need to ingest that data on a daily basis. You need to clean it; you need to match it. And people would be surprised to see how many errors there are and data that comes from vendors for which you pay big bucks, surprisingly. So, we have our own data team that's going through the data, not manually. Obviously, we build systematic processes to go through the data to identify anomalies that are likely to be errors, and we're actually extremely good at correcting errors in our data. And this is really the first step because if you do have errors in your data, you build a quant strategy that will be typically optimized. There's going to be an optimizer involved and a portfolio construction and for sure it's going to find this error and is going to maximize it. And your portfolio for sure you're going to trade on it. I can tell you that's like Murphy's Law. If you drop your toast, it's going to be on the side of the butter So same thing for the ... data errors. You're going to trade on it for sure. So having extremely clean data is very important. Having a long history of data that is point in time as well is extremely important. So, you want data unrevised as it was when it was released because you build strategies after that on this stuff. You don't want to have like forward looking information about the future that you shouldn't have known. So, a long history of clean point at that time. Data is like digital gold for us quants, so this is like invaluable, the stuff that we've built over the years. And then we feed that to our models. We can press that into models with lesser parameters and all that. So to forecast the future, but this is literally the cornerstone of all of our quant models.

TOM GRANT: Yeah. I wonder if you can, I guess, predict how many listeners we're going to get for this podcast using your model? Probably quite a few. John, I guess I'll just turn to you just for a second. Like we've got a ton of quant strategies here at TD and there's, there's some specifically inside of the TD ETF lineup. Can you discuss them?

JONATHAN NEEDHAM: Yeah, Thanks, Tom. Yeah, we're very fortunate. Julien runs and his team run seven different strategies here for us at TD management and so a bulk of them in four different categories. The first category is our low-vol strategy. For those that are looking for market returns with less risk, which I don't know why that wouldn't be everybody, you definitely want to look at our global strategies as a core component of your portfolio. So TCLV is the ticker for Canadian exposure to TILV for international exposure and TULV for U.S. equity exposure and of course, the second category is in our dividend strategies, where our quant team is looking for high quality companies that have dividend growth, return on equity and strong cash flow, and also great strategies for clients that are looking for income and growth. And so these strategies have done exceptionally well in the marketplace and Canadians continue to embrace them. And so TQCD is our dividend strategy for Canadian equity exposure and TQGD for global equity exposure, again, both with attractive dividend yields and high-quality bias. Our third bucket, if you will, is our multi-factor strategy, which is for clients that are looking for a global equity exposure and out- performance alpha, right? They're looking for better outcomes relative to the broader market, and that's TQGM And then the fourth bucket is systematic alpha strategy. And this is an area that I'm very passionate about that we find a lot of our advisor community embraces as well, and that's the U.S. small mid-cap space. So TQSM is the ticker on that one and that's an area of the market that's pretty inefficient. And so where you can employ a factor strategy from our quant team here, we've proven to provide much better outcomes in an area of the market that's very difficult to stock pick in. And so those are our seven strategies. Feel free to, of course, to look on our website at any point in time for any information and the performance of these strategies.

TOM GRANT: All right. Well, I'm going to make my prediction is that this is going to be our most popular podcast that we've hosted so far. John, Julian. Anything that you guys want to end on or anything you want to guys want to share that maybe we just didn't touch on today?

JULIAN PALARDY: On my end I would say if you want to buy into the AI hype I don't do it by buying overpriced stocks, just buy ETFs. That's going to give you a decent exposure through the at least process wise. I think otherwise I would cover that pretty well.

JONATHAN NEEDHAM: Yeah, I think for the listeners out there, you can tell that Julian's a true fiduciary, so he has your best interests in mind. And I think that's that's very sound advice to essentially, you know, outsource it to his team versus trying to pick stocks in a very hot sector at this particular time.

TOM GRANT: Very risky. Sounds good. All right. Well, for all of you listening, we thank you for tuning in and for considering TD ETFs for your hard earned savings and investment accounts. We at TDAM are fiduciaries as John mentioned, and as Julian has represented and of the best interests of Canadians in mind. We manage money for Canadians, by Canadians, and for the lens of the Canadian investor.

TOM GRANT: Not all ETFs are created equal, and we put a lot of thought and diligence in to bringing to market solutions for investors problems. Thank you, Julian, and thank you, John, for taking the time to dive into Quant investing today. Thank you for all we hope you enjoyed and have a great day.