#108 - David Wright: Quant Investing & AI Explained

0:05Welcome to the Insightful Investor Podcast, a weekly series that seeks to share industry, investment, and market insights. We define insights as concepts that are counterintuitive, widely misunderstood, or underappreciated. In other words, unique ideas that you probably won't hear elsewhere. I'm Alex Shahidi, the host of the podcast and co-CIO of Evoke Advisors, at leading investment advisory firm. Learn more about our show at insightfulinvestor.org.

0:38Today's guest is David Wright, head of quantitative investments at Pictet Asset Management, the asset management division of the Pictet Group, an independent wealth and asset manager that was founded 220 years ago in 1805 and is headquartered in Geneva, Switzerland. Pictet employs over 5 ,500 people and manages$926 billion as of the end of September. The firm launched its first ETFs in 2025, including the Pictet AI-enhanced international equity ETF, a symbol PQNT, which is managed by David's team. Welcome to the show, David.

1:15David Wright:All right. Thank you for having me, Alex. Let's go back a few years. What originally drew you to investing and why did you choose Quant as your home base? It actually starts before finance, really. So I was always and remain a huge sports fan, particularly English or European football. So on a Sunday, I would get the newspaper, I would read through all the results. I was very into the statistics around the sport. And as I slowly started reading more of the newspaper, I got into the business pages. I was interested in all the different stock prices. And again, it was kind of the data side of things that really attracted me.

1:55David Wright:So I'd always liked the idea of data and how people use data, what it meant about events, what it meant about potentially predictions. And then after university, the opportunity to join a firm that had a kind of strong quant offering, it really blended the idea of working in finance, which interested me, and having that kind of data -driven approach to investing as well. It's probably also easier to succeed in finance than it is trying to be a professional athlete. I think that's very true, yes. Especially English soccer. You spent 20 years at Barclays Global Investors and eventually BlackRock.

2:37What lessons from that period most shaped how you lead and build systematic strategies today?

2:43David Wright:I mean, I loved working at those firms. It was such a good grounding. I worked with so many great people. I mean, for me, while we see that that's two different firms and evolution through BGI to BlackRock, But really, that for me was within the same quant or systematic team almost the whole way through that time. But the cultures had quite clear differences within them. BGI was very, very collegiate, very, very collaborative. It had a very academic focus in the way that it did its research. And again, I think that really was why it had so much success building out passive businesses and quant equity businesses.

3:30David Wright:And then BlackRock, we were able to continue that as we became part of the BlackRock business, but you allied that with a very commercial, client-centered focus. And as I build out a business here at Pictet, I think we want to combine the best of those two things. It's about having that very collegiate, academically strong, innovative environment and investment team, but then really having a strong focus and what the client needs are and developing strategies and products that meet those needs and requirements. So you've been an investor for about a quarter century, but what are the biggest shifts you've seen and what feels genuinely new in the last few years with AI's broad adoption?

4:15David Wright:I mean, I often talk to clients about the evolution in the quant space. And again, I've had, as you highlight, the time that I've spent in the quant industry, I've been long enough in it now to see some really interesting phases. When I started my career, this was really when there was a huge growth in demand for quant strategies. There was a lot of success in what we would now view the more sort of vanilla traditional quant strategies, multi-factor approaches, emphasizing factors like momentum, value, and quality. We then went through a quite challenging time for a few years, the kind of quant crunch that happened across 2007, 2008.

5:01David Wright:And I think the biggest thing for me since those times, probably from sort of 2010 onwards, is how disparate and spread the quant industry has become. So you have seen plenty of quant firms become almost market makers. And we're going to know the types of names that I'm alluding to there. There are those that really stuck to their guns around quant investing. And maybe it's some of the well-known value type managers. People have investigated from those days alternative data. And right from back sort of 2010, we saw the first interest in machine learning, more advanced technologies to initially try and capture information from new types data sources.

5:52David Wright:And then as the years have gone on, use those tools to build models in differing ways. So really, I view machine learning AI as part of a kind of continuation of the evolution that's happened in the quant industry over the last 25 years. So always that demand for more data, more technology, but increasingly doing things in disparate ways because many of us who are in the industry over those years, sort of 2007, 2010, are somewhat scarred by those sort of crowded periods where a lot of people were doing very similar things. Right. And when those unwind, as they inevitably do, it can be painful.

6:35Exactly. Can you provide an overview of the quantitative investment franchise at Pictet Asset Management? and tell us about its origin and the team.

6:43David Wright:Yeah, so the team have been managing money in a quant way for over 20 years now. And what that has grown to today is a team of around 25 individuals based predominantly here in Geneva with actually someone based out of our London office as well. Today, we're managing assets of about 30 billion US dollars. The team does broadly three different things. So firstly, we run some multi-asset quant money in more solutions-based approaches. So really trying to solve certain problems that clients have. And then we have our quant equity approaches. Right back from the earliest days, the team have built more factor-based strategies in traditional quant modeling that particularly emphasize the quality defensive factors and we deploy those in both long-only and long-short strategies and then the final piece which has really been accelerated over the last three or four years has been our use of machine learning ai to build a a ai driven approach that we now deploy in long-only and long-short strategies.

8:04David Wright:And that is our fastest area of growth and now accounts for about$4 billion of our assets. You touched on the evolution of using technology in investing. Looking ahead, what do you see as the next wave of innovation? One of the most exciting things about Qwant is that there are just so many different options there and things that are available to us. So where it stands today or where it stands today for us, we are very focused on using machine learning AI in combining signals together, in combining the data sources in the most effective way. We know that a big area of growth for the quant industry has again been using more advanced technology to crack open the potential alpha from new data sources.

9:01David Wright:I think that is by no means finished at that stage. So for us as a team, we don't, for example, do a lot of work with language models at the moment. But I think that is something that will allow us to bring a broader group of data and differing data to our signal set that we can then use machine learning to combine into our models and our strategies. Yeah, it seems like we have better tools to analyze data. And so if you can add more data to your set, especially with unique data, then that could be an interesting evolution as well. Correct. So let's get into quantum investing a little bit more.

9:44And let's start at a high level. In everyday language, can you tell us what the computer is trying to figure out when it studies all that data? And how do you try to differentiate your approach from other models that maybe try to do the same thing?

9:58David Wright:Let's start and think about a more traditional quant approach that is trying to exploit some style factors or risk premia, where we can group certain signals, signals that are telling us something about a company or the stock that have an economically sound reason why they should work. So it might be a signal in the momentum grouping of signals, momentum being something that where you're picking up on a trend, maybe in a stock's performance and a trend that is going to continue in the future. That has a economically sound reason why it should work, generally justified by the fact that it's accepted that different parts of the market pick up on certain news driving that trend at different times.

10:53David Wright:Or maybe it's a signal from the quality grouping where there is something about the quality of a company's earnings or its management team or its business model that allows it to have an outsized result versus its peers. Again, there's a rationale for using that. In a traditional quantitative approach, a portfolio manager will combine these different signals together in a model that they build. Maybe they equally weight the signals. Maybe they test some ways to forecast which signals will work most effectively at different times. In our approach on the AI side, we go beyond that. We want to learn from historical data about which of these signals work most effectively with each other, how they can condition each other.

11:54David Wright:So rather than naively following a recent price trend in a stock, are there other signals that justify why that trend might continue? So again, providing the conditions on why certain signals should work. The relationships between these conditioning elements, there are thousands and thousands of them. They would be very hard to identify by an individual portfolio manager, but machine learning allows us to crack and understand these relationships from data. And I suppose some of those relationships may exist because many investors tend to have these behavioral biases that, well, when you analyze the data that closely with that much data and history, might reveal that an individual may not see.

12:46David Wright:Exactly. So I think an interesting example of an interaction that we can identify is something, but let's use a simple example of an analyst sentiment signal. So analyst sentiment, another kind of grouping of quant signals that's been used for decades in the industry. I think many of us see that if you use those in a traditional way, sort of naively following, for example, the more a stock is being upgraded, you follow and buy the stock, more downgraded it's being by the sell side, you should sell or short the stock. the actual efficacy and opportunity set in that has diminished potentially quite significantly in some markets over the last 5, 10, 15, 20 years.

13:34David Wright:However, if you can understand which other signals will condition that analyst sentiment signal very effectively, you can generate alpha that is not decaying in the same way. And one of the examples that we're able to identify from the data is that a lot of calendar information is very useful to condition those analyst sentiment signals, particularly calendar information at the company. So its reporting cycle, how close, for example, are you to reporting your official results? That can have a very strong relationship into the interaction with the analyst sentiment piece. And again, I think there is a behavioral element that is driving that, we can put a clear and obvious story on why that might work.

14:23David Wright:But there are thousands and thousands, tens of thousands, hundreds of thousands of relationships like these between different quant signals that you would really struggle to put a kind of story or a rationale on exactly why they should work. When people hear AI in investing, they might imagine a dramatic upheaval. Why is the reality more of a steady evolution, as you alluded to earlier? And what makes that gradual approach potentially more effective than a sudden revolution? I think for most people, the AI has felt like a big bang post the chat GPT of 2022. But in practice, you can go right back 50 or 60 years, the underpinning of the theory around a lot of AI.

15:16David Wright:And there's been peaks and troughs in the last 50 or 60 years of differing industries, whether it's finance or others, trying to make use of AI and in particular machine learning, a very large subset of AI. So I think within finance, some areas have probably been a little bit slow to see the potential use of AI. But in the quant industry or the quant subset of finance, because technology has always been alongside data, the kind of the core underpinning or foundation of the way our processes work, we were already very interested in how we could bring in new ideas, how we can evolve processes. And again, I think the experience that a lot of quants went through in that kind of 15 years, 16 years ago, and then the desire to look for new things, it has just fitted in very well to the improvements in the opportunity set, the improvements in machine learning, the availability, the ease of using these types of things are kind of really gone hand in hand.

16:29David Wright:If everyone can use AI now, where does the real edge come from? There's two different aspects and two different ways that I can think about that. So the first one is that where do we see our edge coming from in the way that we build and train AI models? So We don't spend a lot of time trying to find signals or features, as we would call them, using the machine learning jargon that we think are exclusive to us or unique to us. That's very hard for any quant team to do, to find a truly unique signal. I think that was one avenue that a lot of quants went 10, 15 years ago, and it is hard to do. Again, even some of the more interesting, harder to use data has become much more kind of democratized in recent times.

17:20David Wright:So we see the edge really as building a large set of understandable, usable signals, features. But then the real edge is spending many, many years refining a machine learning framework and structure that most effectively combines those different signals together and understands how they work together. So that's where we see our edge coming from. So how do we then potentially differ to others in that? So I think one of the key things here is that I've already talked about how the quant industry has become quite disparate. So varying horizons that people invest over, varying different focuses that they have, again, where they sort of span the spectrum in having a strong belief in economic rationale, their signals right through to the more kind of data-driven end of the market.

18:21David Wright:people are doing very different things now. So I think there have been many examples of last year as a good example, that there were several different periods where certain quant firms struggled, but others did particularly well within it. So I do think AI, machine learning, has actually created an environment where people use these tools in very different ways and are getting quite different outcomes from using them in a different way. So I don't think we need to be quite so worried about crowding, looking for that edge in a very crowded area in the same way the industry struggled with previously.

19:01In general, what do you feel humans do better and what do AI models do better? And how do you combine the two?

19:08David Wright:If we think about, again, our AI-driven process, what do the portfolio managers still do within that. So firstly, they're the ones that define the group of features that we train our model with. They either are developing those features themselves. But as I mentioned, I think the true edge there is getting harder and harder. So it's more about finding signals that have been developed more broadly, maybe refining those, finding large numbers of them that we can train the model with. I think then what the machinery is very good at doing is learning how those features work together and how it should combine them.

19:57David Wright:It's not exactly like this, but you can kind of think of it as how it should weight those signals and how that should evolve over time. And again, it's machine learning and a lot of data that allows you to do that. I think the machinery is very good at building the portfolio of understanding how it should combine return forecasts, risk forecasts, cost forecasts together. But then I still want portfolio managers at the end of the process to be checking the output to potential late-breaking information to have that additional risk control on top. So I think the start and the end need to be portfolio managers who are also our developers of our strategies.

20:40David Wright:But in the middle, those machine learning strategies that they've built, they really offer the benefit of, again, being able to choose and combine the most effective signals to use. And I suppose that's because that middle segment is where you have massive amounts of data and the machine that can analyze much more data than a human can and find patterns that a human would most likely not be able to identify. That's where the real edge comes in. Yeah, that's completely right. Are there specific decisions that you would never delegate to a model? The parameters around the way that we structure and train our models, I still think are important that we're testing those and refining those as humans, as the portfolio management team.

21:28David Wright:So the work that we did to build the structure that we then frequently retrained to produce our forecasting model, the different parameters that we use and tested by them were, again, suggested by the managers. You can do parameter searches using, again, the machine learning itself. But I think having a lot of structure around that and having a human element is important. And again, we could systematize that end final check. We could systematize a much shorter term view on maybe news flow to try and check that nothing has broken since our forecasts were made. But I still think there's value in having a portfolio manager doing that.

22:19David Wright:So I will always, within our approach, want to have the portfolio managers providing guide rails around the process. I guess you could think of it as a partnership, right? rather than just letting the computer take over entirely. There's a partnership between the two, particularly in the beginning and the very end segments. Very much a partnership. And another way of thinking about it is, it's a little bit of a folksy example, but it's a little bit like a plane. I mean, like a human is designing the plane, engineers are building it, but they're doing that with a huge amount of tools. The plane these days has, with AI, with machine learning, can fly itself.

23:02David Wright:But you need to have humans in that plane ready to take over if there are particular challenges, sometimes land if there's fog, for example. I view it much the same within our investment process. And if you just think about the 300 passengers on the plane, they probably feel better if there's a human at the front, just in case. I think that's exactly true. And we see that from our investor base as well. We are very transparent about how we train our models, the way that we build our process. But we do talk to clients about the human elements of it. They want to meet our portfolio managers often before investing.

23:43David Wright:And again, it's for exactly the comfort that you describe. If we talk about different investment horizons, short-term versus long-term, where do you feel machine learning potentially adds to the most value? We do find it is in the shorter end of the investment horizon. So to give you some examples on this, our models predominantly are trained with a 20 business day forecasting horizon. So about a one month forward look through, which actually for machine learning approaches, some people might even consider that actually quite long in horizon. And a lot of the machine learning spend is on intraday strategies, one day, five day strategies.

24:27David Wright:But we find 20 days is a kind of very nice, a very nice optimal point to use these types of approaches. We do also train for some of our hedge funds, some six month models. What we find is once we go much further out than that, maybe looking at building or training a 12 month model. If we compare a strategy that is built using a more traditional construction where portfolio managers are assigning weights to a model versus a model that is being trained on data, you actually don't see much uplift on using machine learning over that kind of 12-month horizon. That one month to six months, you definitely see a big differential of using a trained model in that.

25:19David Wright:So I think a big part of that is the higher frequency, the more data you've got. The more data you've got, the better model that you can train. Part of that is also pattern recognition of finding predictive elements within that data that a human would not be able to see because of just so much data. Exactly. And I also think there's an element as well of over differing horizons, different things are important to stock. So the models that we build are very focused on being factor neutral. So I've already mentioned the momentum factor, the value factor, the quality factor. We don't want our strategies to emphasize that.

26:02David Wright:In fact, we want them to be neutral to them. We want our strategies and our models to be good at forecasting the non-fact specific part of a stock's return. Now, if you go out to a 12-month horizon or maybe if I think about some of my colleagues and sister teams here at PICTA investing in a more traditional way over multiple year horizons, over those horizons, it's much more about the economic cycle or more specific company fundamentals. fundamentals. But once you get in a kind of month-to-month, week-to-week horizons, it's sort of thousands of different idiosyncratic things that are driving stock returns.

26:42David Wright:And to your description, these are very hard things for individuals to understand. It's really the machinery with the data that allows us to understand those things. And that also goes back to what you said earlier, which is it's hard to find a single data point that is predictive because you have so many different things that influence what happens over the next month or two. And so you need to take in all that data and find those patterns. Yes. Yeah. So let me ask you this. If you're just forecasting for one month ahead, how do you reconcile that with long-term investing for clients? Yeah, I don't want my clients just to invest with me for one month.

27:23David Wright:So I think the key point here is that it's that we've got to repeatably do that. So every single day, we are making a one-month forward view on every stock in our investable universe. Now, when we then combine that with risk modeling, cost forecasting, constraints on portfolio construction that help us maintain good diversification across the number of stocks, diversification across countries, industries, sectors, remove any style exposure. It does. And we also have a strong cost focus in that. It does slow our investment horizon or kind of think of it as holding period down to maybe three or four months, but even so much shorter than we would like our investors to think about an investment with us.

28:18David Wright:So the point being, we have to use the model's ability to regularly have an edge in forecasting the next month and then in a disciplined way, do that repeatedly quarter after quarter, year after year. And we see that in the performance that we're delivering in strategies like Pequant, the ETF that we've produced. So the sister strategy to that, that we launched first for investors here in Europe, that has been able to deliver alpha above 2 % above the benchmark in each of the last two calendar years. So a good consistent ability to generate alpha, but from that shorter term view. So when you talk about shorter term perspectives and factors that influence the price, do those tend to be more behavioral factors or are they more fundamental to the companies?

29:18David Wright:I mean, so I talked about the, particularly when you're out in kind of the medium term, maybe six, nine, 12, 18 months, company fundamentals are very important. What is really important over the next month is more like changes to those fundamentals. So if I think about the signal or feature set that we're training our model with, around 100 of the features are built from accounting data. So again, the types of data that you do build some fundamental value or quality signals from. But again, we're very focused on the change. The return on assets of a company is not interesting to us, but how that return on asset has changed over the last year is.