Julien: I think there's nothing more complex than human beings in the first place, so there's nothing more black box, let's say, than a human being picking stocks, in my opinion....
Samuel: Hi everyone, welcome to Portfolio Manager Views. I'm joined today by Julian Palardy, head of Quantitative Investing, and Laurie-Anne Davison, head of Passive Investing. My name is Samuel Carrière and I'm an equity client portfolio manager. And today I decided to take two quants out of their offices to discuss everything we are seeing in terms of the improvements in AI and how that's affecting investing today.
I'm happy that you're joining us today really to tackle one important question. What happens when every investor has AI? Are humans still needed in the investment loop, or will machines and AI replace investing?
Julien: My view is that everybody having an AI is first of all, there's a cost to AI right? It's not free. Like, yeah, there's massive compute requirements to do this. So. And then after that, you mix that with the fact that in the quant space you need massive amounts of data so that cost is exploding even more. So, it's not at the reach of everybody out there to do this.
So, you need both the computing power, you need the right data, and then after that you also need expertise. So, it's not because you have AI that you don't need like scientists and machine learning engineers and all that, all the data team also do to ingest that data on a daily basis. So, there's going to be some types of jobs that maybe are going to be less required to manage money, but there's going to be other types of jobs that are going to become more critical.
And in the process of managing money I think is just going to shift the demand towards certain skill sets. But at the end of the day, there's still a cost. Like, even though AI is there, it's not like that cost of managing money using quant strategies is going to drop to zero, in fact. You know, apart from scale, I don't see what really can drop that cost to lower levels.
You need essentially to spend quite significant amounts to get things started. And after that, as you scale the team or as you scale the AUM, then the marginal cost is decreasing, but the startup cost is actually quite meaningful. You need the right data, you need the right people, you need the right infrastructure as well.
Samuel: Yeah, I think that's a good point. And when I'm in clients, the main discussion now we're having is that AI and machine learning is something new and I think that's a misconception. We've been doing machine learning in our investment processes and the quant world for quite some time. What really changed over the last few years is the sheer amount of compute that we're able to use to really enhance those models.
And with the appearance of LLM and Agentic AI, that really speeds up kind of quant research side of it. Because when you think about quant research, then its quantitative researchers look at models or risk models, alpha models. But then when you have Agentic AI that comes into play, you can really speed up the projects that you have in the pipeline.
Samuel: So, I don't know if you want to kind of share a little bit on that.
Julien: Yeah, So and we've seen that in recent years. But maybe to come back to your previous comment about the fact that we've been doing AI for a long time, we have. But I don't want to this to be reductionist in terms of the impact of what we've seen in the last few years, either like what we were doing 20 years ago, has nothing to do with what we're seeing today and that space like it was basic statistics that we were essentially applying to fairly large amounts of data and now it's literally entire processes that can be largely automated.
Julien: It's not just building models; it's cleaning the data. There's also figuring out what are the best approaches to doing certain things. To implement. A database, for example, what’s the best way to build certain models, automating the process of building and testing those models as well, documenting them. So, there's a whole lot of things that there's no way we could have thought about doing systematically and in an automated manner back in the days.
Now we can do so and this and the like.... We did see an acceleration in terms of how the team can operate. So, we've been pretty lean in terms of resources and how we, let's say, the things that we can accomplish with a limited number of people. So, let's say, the large language models that have been brought in our processes when it comes to building statistical models, while that has massively accelerated.
Julien: Like we have a guy on the team who built a dashboard over the weekend like that replaces ... I'm not going to name the previous vendor, but these, these guys probably are going to be out of business or with us at the very least. So literally that thing would have taken probably months to build. When I started 20 years ago.
So, there is clearly an impact. It's in different spaces than what we used to see. The models that we essentially apply on a day-to-day basis for forecasting stock returns or forecasting risk. They've evolved as well. But I think it is a testing of the models and the construction of the models that has accelerated because of the new models, let's say, the LLMs that are available to us and the acceleration in coding as well.
Samuel: Yeah, I think that's a good point because when you think about how AI impacts investing. Another misconception that at least I have is using AI is very different than implementing or implementing the model. Right? And Laurie-Anne, I think that's really what your team is, is doing at asset management is that you're implementing models in the market. And could you share a little bit about how really that that whole process happens?
Laurie-Anne: Sure. I'd be happy too. So, in terms of the model portfolios that are developed by the quant team, for us from a passive perspective that really functions as a benchmark that we manage to. It has a lot of the same characteristics that a benchmark would have. We have the constituents target weights. It's rebalanced on a specified cadence, whether it be quarterly or monthly, and we'll rebalance the portfolio or the ETF that a client is invested in alongside the rebalancing of the model or the benchmark.
And then we manage flows that come in day to day. You know, from a passive PM perspective, we're also managing things like corporate actions and trading and other things that really impact the performance of the benchmark and our ability to track the model as closely as we can.
Julien: Maybe the way Laurie-Anne talks about this looks really simple, but once you multiply this by like tens, maybe eventually hundreds of funds and massive amounts of AUM like that job becomes really complex very quickly, like a small grain of sand in the machine can derail everything. So, her team is actually managing an insane amount of workload extremely well.
Samuel: And I think you brought the word benchmark up. And I think this is a very important discussion point that I wanted to touch today is: are benchmarks still the best investment solution for clients? And when we think about the rise of AI, if I was to look at the S&P 500 at the moment of recording this podcast, semiconductors represented close to 19-20% of the index that has risen quite significantly from even five years ago, where IT exposure, which can be seen as more risky for an investor, was diversified between hardware, software and semis.
But now you're getting a lot more semiconductors within just the broad-based market benchmark. So, I don't know, Laurie-Anne, if you could touch on that and how the risk profiles of the benchmarks have changed in the last five years.
Laurie-Anne: Sure. So, when we think about passive investing, we think about MSCI, S&P. These are market cap weighted benchmarks. And what passive investing is really good at is giving clients broad market exposure in a pretty cost-effective way. We all have diversified holdings. As market leaders start to emerge, clients will benefit from that through the exposure and through increased exposure.
But what market cap benchmarks don't do particularly well is manage for concentration. And I think that's what you're getting at over the last few years. And quite frankly, beyond that period, we've seen an increase in concentration really driven by us mega-cap, really tech related names. So, if we think about an example, the MSCI Global Index, it is a diversified index.
It covers 23 developed countries. It is about just shy of 1300 holdings in it. But the influence of these mega caps has been so significant, we've seen the exposure to the US rise to 70% within that index, the top ten holdings comprise 25% of the exposure. So, if you think about that, if there's about 1280 names, less than 1% of the names in that index represent 25% of the weight in that portfolio.
So, you know, it certainly has implications for concentration and risk. And while we haven't seen really an increase in realized volatility, quite the opposite. We do have benchmarks now that are very sensitive to movement and in just a handful of stocks.
Samuel: Yeah, definitely. And something that's been a trend that we've been discussing a lot more is a term I call it thematic momentum, and that is when investors jump from theme to theme based on whatever is trending in the market. And if we look at these different themes throughout the last three or four years, I mean, we were jumping from defense.
We've even had a rally in quantum computing. Then it was semiconductors. More recently, we have gold and energy. And when we think about really these thematic rallies, quant investing or at least investing through alpha models or systematic solutions, have the ability to capture these inflection points in the markets a lot better. There are a lot more spread out and diversified.
And when we look at the data over the last three and five years in general, when we look at the median manager, quant managers have outperformed fundamental managers. And yeah, I don't know if you want to touch on that point as to reasons why this is occurring in the market.
Julien: I think there's probably two factors at play here. There's the I think you identify one of them, one of them is the fact that quant models are pretty good at identifying trends as they emerge in a market. But there's also the fact that across the spectrum of quant managers, and this is where passive also ends up in there. So, you're going to have strategies that are essentially close to zero or maybe even zero tracking error, like they won't deliver exactly what the market does.
And the market has been tough to beat in the last, let's say, five years because of this, because it's tough to pick up those trends quickly. And if you miss out on one of those strands and you're an active manager, that doesn't stick exactly with the index, you're going to trail probably. And when you imagine index, obviously you're going to deliver index returns.
But as you increase your tracking error, you're going to do it in a much more risk-controlled manner. If you're a quant manager, typically, then if you're a fundamental manager, you're going to have constraints on various dimensions of your portfolio. And as a consequence, this means that if there is something that, let's say a big trend that picks up into semiconductors and you have constraints around this, then the fact that you're much more risk controlled and your approach to investing and identifying trends in a market means also that you're much less likely to leave a lot of value on the table, like many fundamental managers have done in the last few years.
Julien: So, I think there's both factors at play here. One is more a question of risk management and how you manage your tracking versus the index. The other one is, is the systematic aspect of what we do. But then again, everybody builds their morals differently so that you can see differences from one quant manager to another.
Samuel: And just from experience discussing with clients. I think one very important point is to walk the client through the model and help them understand what is really happening behind kind of the alpha models and the investing that's happening because when you think about a fundamental manager versus a quant manager, a lot of clients are quite comfortable with stock picking, or at least you have a team that is responsible for picking stocks that are doing well or not well, and that ends up being the portfolio.
But when you think about more complex models or alpha solutions, clients don't always understand what is driving the stock selection and how these portfolios position in the market. So, I think, Julien, maybe could you share some of your experience meeting with clients and really dumbing it down or at least discuss how these models work?
Julien: I can tell you, first of all, there's something interesting in what you said. Like, I think there's nothing more complex than human beings in the first place. So, a stock picker will give you one stock pick on one day and then another day and the exact same market conditions, exact same economic conditions, maybe that stock pick will be entirely different just based on their mood.
So, there was a lot of input that comes into play into the decisions that a human being is making. And there was nothing more black boxy, let's say, than a human being picking stocks, in my opinion, because it's going to be very difficult for a person to tell you exactly what is their process that they went through to go from, let's say, the environment that we see to the exact stock pick and the exact way that the stock is going to have in the portfolio, I can actually do this quite perfectly with a quant model.
So, if I take all the inputs that are available to me today, I rerun that model another day in the same condition. I'm going to get probably pretty close to the same view again, unless there's a big error somewhere is going to give me the same input. You're going to get the same output. And on top of that, I'm going to be able to trace exactly how that decision was made.
Julien: So, when we go in front of clients, if it's, let's say a quarterly attribution or something and you're involved in those discussions as well, it's quite easy to actually explain how we generated performance based on our positions or how we came up with those positions. And Laurie-Anne also is involved in some of those discussions as well because her team is managing the QE ETFs on a day-to-day basis.
So, they take a lot of questions around the performance of the models and why we're positioned in a certain way, and they see those portfolios and how the models are built on a day-to-day basis as they may and they manage them. So, I think there's a lot more visibility, not necessarily for the end client, but once the PM, once a portfolio manager is able to ... give some transparency as to how the model operates, then things are much more open than people think.
It's much less of a black box than people would think.
Samuel: Yeah, for sure. And like when we think about our suite of solutions that goes from the dividend-oriented strategies, broad benchmark or even thematic type strategies that focus on just tech-oriented portfolios like Laurie-Anne, I think the most important question is how clients can implement that in their portfolios.
Laurie-Anne: Yeah, so we talked about, you know, how effective passive can be in terms of getting market exposure. And that really represents to me a core part of a client's portfolio. But there are shortcomings with paths of investing, despite it being cost efficient and intuitive, there is also concentration that creeps up and it's not necessarily well handled by typical benchmark providers.
And it really makes it, I think, important for investors to diversify across that exposure. So, you might have a core equity exposure in the S&P 500 or the global index or the TSX. But as concentration continues to increase and you know, there's not really anything to say that we would expect that to change meaningfully in the foreseeable future.
I think it becomes important to diversify that equity exposure or fixed income exposure away from these more concentrated positions.
Samuel: Yeah, definitely. And when I discuss with clients, I always bring them back to the risk profile they're comfortable with. And you need to monitor these very changes in the benchmarks, like you highlighted, the concentration because risk profiles and broad-based indices such as S&P 500 or MSCI World, they changed significantly through time. But as an investor, you don't really feel them or you don't really see them, if I could say so.
It's always questioning if the current product, you're invested in or whether your current portfolio aligns with your risk return profile or what risk you want to take in the market. I want to shift the conversation again to what we're seeing from clients, specifically in terms of style diversification. And over the last few years, we've heard a lot of clients ask, okay, I would like to diversify my growth tilted portfolio with a value tilted portfolio or at least be well exposed in terms of investment styles.
Samuel: But the conversation that we've been having now is that clients would want to diversify the investment processes within their portfolios so not only rely on a fundamental or concentrated manager that has 30 positions, they want to be invested through the whole space where you have one portfolio manager or one solution within your portfolio, which is a concentrated one.
And then you diversify that with a quant-based solution, which is more broad in nature and is less concentrated. So, Julien, I wanted to touch on the conversations we've been having.
Julien: Yeah, for sure. So, I think and I'm thinking more specifically in the institutional space and maybe after that I'll pass it to Laurie-Anne because she probably has a view on this as well in the institutional space, I think. And we need to draw a line here between, let's say, passive management and managing passively against the cap weighted index, because there's plenty of options outside of cap weighted indices out there.
And I think we're seeing some decent growth there. But also, among the clients that we talked to, it seems that there are really two camps. There's going to be the guys who essentially drag the index. They're being like, essentially you may have a board or an investment committee that that is being imposed, a specific benchmark for their strategy.
And if they underperform that benchmark, if the managers that they pick underperformed that benchmark by a specific amount, their jobs are going to be at risk again. And unfortunately, in many institutions, decisions are siloed. You're going to have asset making decisions that are being made in a certain way. And then after that there's going to be manager selection that's going to be made in a different way.
And those types of clients are going to be typically extremely sensitive to things like tracking error. And even though the market is going to be concentrated, they're going to be telling you things like, well, if the market is concentrated, it means that if you deviate from that quite a bit, then there's going to be a big risk for my job because you could underperform by a lot if those big names are rallying massively and then you would have another camp of clients where they do understand that on an absolute basis this is real risk, Right?
Julien: At the end of the day, you're the outcome that you go for that you try to do, to maximize protecting against downside risk or underperforming your liabilities as a pension plan, let’s say. And it's the same kind of thinking that you would find in the retail space as well, where they're much closer to the money of their clients than some institutions would be let’s say.
But to come back to the question of what kind of discussions we're having with clients on the I think on the passive side also, we're seeing a lot of growth in nontraditional benchmarks as well that are not cap weighted. And some of those concerns are around the concentration of the index as well. So, the fact that people worry about concentration doesn't exclude the option of using passive approach as just using different benchmarks.
And maybe Laurie-Anne, you could talk a bit about this, I think, because that's an interesting space where we've seen a lot of growth there as well.
Laurie-Anne: Yeah, absolutely. In an aggregate basis, passive, as you know, whether it's equity or fixed income, is a large portion of TD Asset Management's business. We continue to grow. We continue to see demand, particularly within the institutional space. You know, I think for a lot of clients where they feel markets are pretty efficient or, as Julien mentioned, in periods where markets are hard to beat, passive offers you a low-cost way to get that exposure.
But increasingly, to Julianne's point, we do see concerns around concentration and whether it's an alternative benchmark from a known provider like S&P, a capped index, for instance, or something that they've had more of a hand in creating. In terms of customization, we're starting to see more interest in that from our institutional clients.
Julien: And for Quant Solutions, it's worth mentioning as well, we have a large array of solutions that deal differently with tracking error, so some are going to be closer to the index there. There's been like a big resurgence in index-based type of strategies or indexing strategies at the end or at the other end of the spectrum you would have like fully risk management and absolute basis strategies like low-vol strategies.
And we're big in that space as well. And then in between you, you would have typical active strategies with active risk that is being managed using risk models or our dividend ETFs as well there in that space also. So, you have a broad spectrum of active quant strategies that just like you have a broad spectrum of benchmarks that you can manage against.
Also, if you want to manage your concentration risk.
Samuel: It definitely feels that everybody has their own take on the market and the rise of customization has really increased because some investors are not comfortable with. As you highlighted, Laurie, the concentration risk that we're seeing in the benchmarks, the high exposure to information technology securities. But instead, they're not reverting to other broad-based solutions. They might be requesting custom solutions based on their risk return profiles or how they manage their portfolios.
But I think in terms of one of the toughest questions I've had to answer when facing clients is, all right, you're using AI, you're implementing AI in the portfolios. It's a tool. It's speeding up the research. Isn't that going to lead to some sort of crowding effect where everybody's using the same models, everybody's kind of tailoring their own solutions in the same way, and leading to a kind of crowding effect in the market?
Julien: I would say it depends how you use them, but it's a reasonable concern, I would say. And so, if you just launch, let's say I'm not going to name names of specific LLMs, but if you just take an LLM and ask it to build models that that will optimize or maximize performance, let's say in sample, probably they're all going to end up with roughly the same thing.
Okay. But I don't think any reasonable quant shop would actually do this either. So, I think it's how you use it. And what part of your process is where things are probably all done differently. And on top of that, every firm has their own philosophy on how to do things. And depending on how you guide your models, let's say you automate or you do an agent, a research process, end to end.
You're still going to have to at some point dictate to those models how like what kind of what are your beliefs in terms of how to best build those models. It's really rare that you're going to find teams that will just ask a model to blindly optimize or maximize the performance of something and sample that would lead to probably disastrous outcomes in real life.
So, I'm not too worried about crowding with AI, I would say it's very similar to what we had back in the days when everybody was using the same statistical methods and regressions and things like this. But every quant manager out there had different beliefs on how to apply this, what worked, what didn't work. So, this is where you're going to get some differentiation in terms of outcomes between the various managers.
Samuel: For sure. And I think quant is not all built the same. It's becoming very popular with thinking about quant investing solutions. But what's very important is the research behind that, because I know we hear that a lot, but correlation is not causation and a factor, even though it might be looking through earnings report and then trying to find inferences between even a company call with management and how they position, how they talk and trying to build that as an investible factor that might not be persistent through time.
And that's really where I would say thorough research goes into building a product or building a process that is persistent through time.
Julien: I would say it's research, but it's also like if you just have good researchers, it's not going to go far enough. Like, yeah, you also need continuously to investigate new data sources. You need obviously the computing power that comes with the storage space that comes with this. So, you need the right infrastructure, you need the right tools in place.
Julien: So, there's quite a few factors that come into play. And this means that it is not just, let's say, academic research that would lead to generating alpha. I'd say it goes a bit beyond this. It's a machine. Like quant is really a business, very much like passive as well. Once you have the setup in place, it becomes a highly scalable machine.
Samuel: And Laurie-Anne, can you share a little bit on your side when you're implementing the portfolios? I've been, I've been reading articles that it's possible that AI would completely replace a trading desk, right? It could or at least has been it's been targeted that finance would be one of the industries where I could have a higher opportunity or replacement.
So, you could cover the impacts of AI and why and why not and might replace, you know, a passive investment desk, for example.
Laurie-Anne: Sure. So, you know, being part of the quant team, being part of the larger systematic team at TDAM, say we're all we were already very automated where we can be when we're where we can do it effectively, you know, longer term. Will AI replace bodies on the desk? Probably, but I'd say at this point we're quite a way from that.
I mean, as helpful as it's been in certain elements of our job and saving time and helping us gain even more scale. It's not at a place where it is accurate enough or sophisticated enough to deal with the issues that we encounter on a daily basis, whether that be custody tax, M&A related. There's actually a lot of noise that goes along with managing a portfolio to an index.
An index, when you think about it, is very theoretical. The implementation has a lot of noise, including fees and execution. So, I think it certainly has the ability to continue to evolve, to become increasingly automated. But where we stand right now, where we take advantage of it, where we can, which frankly is quite considerable, we're able to do more and get more scale than we had before.
But I think the place where it largely replaces human portfolio managers is still a way off.
Julien: And I would add to this that a large language model will never fear losing a job over a massive error. While Laurie-Anne or people on our team do so unfortunately. But you have the concept of ownership at the end of the day is really important. When you manage your portfolios, you're a fiduciary or you need to make sure that you don't make errors and environments that are really uncertain.
So, Laurie-Anne talks about noise. We have so many data inputs that are unfortunately unreliable. In a perfect world they probably would be, but in real life they're not. So, if you draw on a large language model like this and ask for managing portfolios automatically, I can guarantee you there's going to be big mess ups and there's going to be, sorry, I shouldn't have done this, but it's too late, you know?
Julien: So unfortunately, it's not going to be Open AI that's going to be payback our clients. Right. If something goes wrong, it's going to TDAM.
Samuel: So now, that's a good point. And when I think about our quantitative solutions, I always use the word purposely built because you could build quantitative solutions in any market targeting, for example, global equities, Canadian equities. But not everything works in the same place. And how we think about our market, when we do our own research is that we want to launch solutions that work in that specific market.
And I think I want to highlight the success that we've had with the Q dividend strategies, because that was an example, especially in Canada, where systematic strategies, it's more difficult, I would say, and there's less securities, there's less data. But when we did our approach, when we thought about the market exposing ourselves to dividend-oriented quality companies was the way to go, and that was persisting through time.
Samuel: So, I don't know if you want to share a little bit on how we thought about the Canadian market, for example, versus a U.S. market.
Julien: Yeah, I think you highlighted a few interesting points. The fact that it's a narrow market, less liquid as well in some spaces you have you cannot rely on... It's not going to be as easy to implement in the Canadian market as it is in the U.S., for example. But there are implications for this when you build the quant model like you have last breath we have in our databases more than 600 factors that we look at.
So, you cannot reliably estimate the fact that the return on those factors or select those factors when you have less stock is then factors. At some point there's a limit that you end up hitting when it comes to building those models. So, we decided to go with a simple approach limit. That's basically it was a dividend strategy so limited within the dividend paying universe, focusing on quality factors that help you pick stocks that will keep growing their dividend and avoid those that will cut those dividends going forward.
And it's a fairly simplistic model, but, you know, as they say, if even though it's simple, if it's not broken, then, you know, don't play with that too much or don't write to complexify things beyond what they need to be in terms of complexity. So, we decided to take an approach like this. And it turns out that this same approach is actually working really well also in the international space.
So, the U.S. market highly concentrated. Canada is concentrated for different reasons. It's not necessarily in specific names, is going to be more in a sector basis. But once you constrain those sectors, you actually have quite a bit of room to add value by picking stocks, overweighting and underweighting specific stocks in the international space, it's even there's even more breadth to implement this.
And it's really in the US dividend strategies have been it's been a bit more difficult for those types of strategies in the last few years. As most people know internationally, we haven't seen this actually dividend strategies have been adding massive amounts of value over the last few years and Laurie-Anne’s team is actually going to be implementing or launching it, I presume, in September.
Julien: So maybe around I'm not sure if you have comments on this as well.
Laurie-Anne: No, I agree. Well, we'll be implementing it as we do the other dividend focused ETFs. So, we continue to see a lot of investor demand and certainly strong performance. It's been actually a quite a winner for our quant team and our passive team overall.
Samuel: Yeah, that's a great point. And when I use again the word purposely built for our solutions is exactly what you're mentioning. So, we have a US small and mid-cap strategy and that market is completely different. Now you have a thousand stocks and that universe, and they have a lot of dispersion in the data between them. So, for that specific market, we're using our full machine learning capabilities, alpha models, in order to get kind to find the intricacies between these securities.
And then for Canada in international markets, we'll use some simpler models that target dividend-oriented strategies because securities have less dispersion in terms of the data that they have between them. So, I think this was a great conversation, and we really want to highlight why humans are still needed in the investment loop and why I think we still ... we still have some leeway for our jobs in the future.
I hope. I hope. But it was a great conversation. Thank you for joining me. And I'm sure we'll have the chance to talk in the future.
Julien: Thanks a lot, Sam.
Laurie-Anne: Thank you.
Disclosures
The statements and opinions contained herein are those of the participants and do not necessarily reflect the opinions of, and are not specifically endorsed by, TD Asset Management Inc. or its affiliates.
The information contained herein has been provided by TD Asset Management Inc. and is for information purposes only. The information has been drawn from sources believed to be reliable. The information does not provide financial, legal, tax or investment advice. Particular investment, tax, or trading strategies should be evaluated relative to each individual's objectives and risk tolerance. It is not an offer to buy or sell, or an endorsement, or recommendation or sponsorship of any entity or security discussed.
Investment strategies and current holdings are subject to change.
This document may contain forward-looking statements (“FLS”). FLS reflect current expectations and projections about future events and/or outcomes based on data currently available. Such expectations and projections may be incorrect in the future as events which were not anticipated or considered in their formulation may occur and lead to results that differ materially from those expressed or implied. FLS are not guarantees of future performance and reliance on FLS should be avoided.
TD Asset Management Inc. is a wholly owned subsidiary of The Toronto-Dominion Bank.
