Trading Insights: An intersection between quant and fundamental investing, with Matt Jones, Global PM at Fidelity

[Music]

Matt Jones: No current model is perfect. We all know that. No AI or algorithm is perfect, but if we can marry the two together, it makes them more efficient and better. In running an NLP over those research notes, we can run an NLP that says, does this read like a buy or a sell rating? What that does is help us remove the lower conviction buy and sell ratings out there. We want to get quantitative conviction, which we've got, that this stock should come into the fund. I also want qualitative conviction so I don't have to read 400 research notes.

Eloise Goulder: Hi. I'm Eloise Goulder, head of the Data Assets and Alpha Group here at J.P. Morgan. And today, I'm delighted to be joined by Matt Jones, Global Equity Portfolio Manager at Fidelity International, who's based in Sydney, although we're having this conversation today here in London. So Matt, thank you so much for joining us here today.

Matt Jones: Thank you, Eloise. Very, very excited and happy to be here.

Eloise Goulder: So Matt, could you start by introducing yourself and your background?

Matt Jones: Yeah. I'm currently a global equity portfolio manager at Fidelity. My career originally started back at Bankers Trust in the early days as an equity quant analyst within a bottom-up fundamental international stock-picking team. I left Bankers Trust to go to J.P. Morgan's sell side for 18 months, and that was working within the quantitative research team for Asia based out of Sydney. After leaving J.P. Morgan, I was lucky enough to get a job at Fidelity in 2005, where I originally met you, Eloise. And my role has progressed from there. From running European long-only funds, European long-short funds, to global long-only funds, to now our flagship global long-short absolute return strategy.

Eloise Goulder: So could you dive into your role and your investment process.

Matt Jones: The core of what we do hasn't changed, which is taking a quantitative output from a fundamental research team, marrying that with external data and other alpha and information signals, wrapping around all of that with a strong portfolio construction process and risk models. The core part hasn't changed, but the evolution of how we got there and how we do it much better and more efficiently has over the years. Again, it's quite simple. Everything we do in this process is purely based on our fundamental analysts and the research team and the output that we take from that. That will never change. It has a proven track record. If we do this in a robust way, we're able to generate strong and idiosyncratic, uncorrelated alpha. And I guess the difference between our process to some of our peers is that behind everything we do is a human being running a research note.

Eloise Goulder: It's really interesting to hear how your strategy really differs in theory from a more standard multifactor systematic equity manager. Your starting point is the fundamental research analyst and their inherent underlying alpha. A systematic quant portfolio manager's starting point may be other sources of data. Do you view your strategy to be very distinct? And does it play out in the performance, in terms of a less correlated return?

Matt Jones: Yeah, our strategy is distinct to our peers. I'm very lucky as an, inverted commas, the only quant in the fundamental house because the alpha source I have, you can't take away in arbitrage from Fidelity. So obviously, other portfolio managers can use that alpha source, but you're not having the rest of the market tapping into that quant model that's just been built. And that alpha gets eroded and disappears over time. No one can get access to our research, other than portfolio managers and investment professionals sitting inside Fidelity. So when you take that unique alpha source, and you use some other external signals to filter it down, and then you have strong portfolio construction, just focusing on that stock-specific risk and return profile, you are uncorrelated because you're not doing the same thing. So I'm very lucky to be able to have that.

Eloise Goulder: And just thinking about the nub of the alpha source, these 120 research analysts globally, this is a very difficult question, but where do you think that underlying alpha comes from? And how replicable do you think it is in a world of more and more textural information being available and available to analyze via GPT and more and more data sources being available to systematic managers?

Matt Jones: I don't think the way these analysts produce their research is replicable. They're not fixed into writing a research note in any one way, so they have to come up with a 1 to 5 rating. And they have all the resource to do that, so to replicate something that is not fixed in how they come up to that buy-versus-sell rating I think is the most important thing. Simply just reading some company financials, or you meet management and you record that information and run an NLP on that. That's all a small part of the alpha source that they're generating, but not the full bit. The other critical thing is I also compensated on getting the stock calls right, and I think it comes back to the two key things. It's not replicable because they have freedom to do whatever they want to do to generate that alpha, and they're given everything they need to do that. And they're tied to that outcome.

Eloise Goulder: That's incredible. And I guess it speaks to the huge benefits of scale that you have given that Fidelity employs these 120 global research analysts.

Matt Jones: Yeah, it's breadth and depth of research. And when you've got breadth and depth of such high-quality research, that's when you generate really significant idiosyncratic alpha. It's the depth of that research and the history of that research which is quite unique. So we've been around since 1969, Fidelity International. Obviously, Boston FMR has been around a lot longer than that. And we have access to that research system to the point that our analysts today could look up a research note, the IPO note for Microsoft or Apple. And that information, a quant might say, well, that's old. But there's periods like now in the market where you see such high volatility, such dramatic macro influences, and you might actually look at a company, go, jeez, it'd be interesting to see how that company was during that equivalent period back 20 or 30 years ago. What was the analyst saying? What were the impacts? Is there anything I've missed?

Eloise Goulder: So coming back to the data, you've spoken to the way the research analysts are so core as a fundamental input to everything you do. But when we think about data more generally, what are you leveraging on your process, and how has that evolved over time?

Matt Jones: When I started at Fidelity 19 years ago, the data we were taking was Excel spreadsheets from brokers, and writing and hacking in your own code to clean that up and get it in the format you want. I'd spend 2 to 3 hours of the morning taking this data in, collating it, cleaning it, sorting it, combining it, finding it's messy, having to clean some more bits up, before you actually got to are these the stocks I want to own and send trades on the fund. Over the years, we've tried to improve that. That data will directly come in. We don't have to write code to pull data from our internal systems going as research notes anymore. That's all automated. It comes through the cloud-based systems and fully-integrated databases. We've had the evolution of that IT and support infrastructure to understand the data that we're pulling in, to understand where it needs to be cleaned, and that's very important in what we do. And then getting that to the point where I come in every morning now, and everything's already updated and refreshed and ready to trade. I think the hard thing we've had to do is having to marry internal data with external data. And because it's unique to Fidelity, you can't just upload that into an external system. It was very easy if you've got IBIS data and you've got World Scope data. It's all coming from one of the big data providers. So it's been quite difficult over the years. And every time we might integrate a new model into what we do or another external quant signal we might be looking at, we have to have this process of doing it all ourselves and making sure everything is clean before it gets to the portfolio manager making the decision to send the trades.

Eloise Goulder: Yeah, it's incredible how much these processes have evolved. And how has it changed the nature of the job for you and for your teams? I mean, you mentioned that 20 years ago, you would have spent three hours in the morning pulling together these Excel files and hacking it together. You're not having to do that today, so how are you spending your mornings?

Matt Jones: Yeah. Essentially the process is largely seamless now, so the whole migration of data from external providers, internal data sources, going into a database that sits on a cloud, it all happens automatically. So that means when I come in the morning, I can spend all of my time making sure we have the right stocks in the portfolio. And again, the job sounds really simple. In that sense, it is. A lot of the hard work and lifting is already done, from a data cleaning point of view. The optimization process might have run in the morning to show you how the portfolio looks from a risk portfolio construction point of view. And my job then is just to see if there's been any changes, and how impactful are those changes. So we have 2 and 1/2-thousand-odd ratings. I can't go through 2 and 1/2 thousand ratings or research notes every day to see if there's been a change. So all of that means my job is to focus then on the key alpha source and owning the right names in the right fund. Not only has it made the efficiencies and the improvements in the way we consume the data and then produce the output on the other side for the portfolio manager, what it's done for Fidelity is made it more scalable in what we do. So if you go back to what I said earlier, in 2005, we were able to run a European long-short fund and maybe two other European long-only funds. And with the improvement and the development in the way we take and consume the data and the way we generate the signals, it means we can run global funds. So we started running global long-only funds. We run global long-short equity market neutral funds now. We have a whole group of index enhanced funds that broader team team I work in generates. So we've been able to take those efficiencies, make the portfolio managers' job easier in generating alpha for our clients. But we've also made it much easier for our company to solve problems and provide solutions to our clients.

Eloise Goulder: Absolutely. You're benefiting from the scale in having all of these research analysts globally who are heavily incentivized on the alpha of their calls. But then you're also benefiting from the scale of the technological processes to really aggregate this together globally, to provide all of these different solutions for your clients.

Matt Jones: Yeah, 100% because that evolution of the way our IT infrastructure and our data consumption has been progressively improved on the margin every year, it's not something that we could have done overnight. It has taken 19 years to get to this stage.

Eloise Goulder: Yeah, it's incredible that you've been at Fidelity for those entire 19 years, following it through.

Matt Jones: Yeah.

Eloise Goulder: So I always ask our guests, what's more important, the data or the model? You've spent quite a lot of time talking about your input data, Matt, so what are your thoughts on this question?

Matt Jones: I'm 100% data. We know we have a robust portfolio construction process. We know we have our quantitative signaling process, but the underlying thing behind all of that is the data. So when that goes wrong, the trades at the other end of that decision go wrong. What I think is unique about that and different to, say, some of my quantitative peers, I guess if you're running a 2,000 stock long book, potentially if there's a data error in there, it's a very, very small bet overall in your portfolio. Say there's an incorrect feed from the analyst research team, and I go short a stock that should be a buy rating, and that's a 2% or 3% position in the fund, that moves against us. That's detrimental to everybody. So the data-- it's been drilled into me from 2005 to now that data has to be correct.

Eloise Goulder: It's fascinating to hear your conviction, Matt, that data is so much more important in relative terms than the model. That's quite different to the answers we've had from other guests on this podcast series, and I guess it speaks to the difference in methodology that you have and the fact that your funds are that much more concentrated. And so the relative risks of having the wrong data, as you say, are that much greater.

Matt Jones: Yeah. If you talk about how do you construct a portfolio, we all know the different ways. You know, there's mean variance. You could equal weight. You could cut weight. In the nicest way, so what? That's kind of a fixed thing that doesn't change much. And I would say even the error term in a lot of that typically isn't that great. And if you're tracking errors out by 50 basis points because your model is not predicting it perfectly, in my world, that's not a big deal. You know, you've got a 5% or 6% vol fund, or you've got a 3% tracking error fund. And you know what? The model didn't predict it perfectly by 20 or 30 basis points. That's not going to drive the alpha of my fund. The alpha of my fund is going to be driven by the 90-plus-percent stock-specific risk of that fund. Now, what's that risk associated with? It's the analyst research rating. So I guess my focus in the data and different to peers is that exact problem. When you have breadth, that problem isn't as great because I don't have the breadth. And it comes to my concentration. The concentration is really stock specific, so you want to make sure that the data going into picking those stocks is perfect.

Eloise Goulder: When we think about today and the future, how do you think further refinements can really come through, and how and if could AI really help you in your processes?

Matt Jones: There's not going to be any big-step changes, and we don't want that because it's relatively stable and consistent. What we're trying to do is make things more efficient. And one of those things we have included recently is an NLP on our research system. All of that textual data is sitting in our systems-- 2,500 stocks over 50-odd years-- so we can run an NLP on that. So we've got quantitative conviction in my process, so it has to pass certain quantitative criteria to get in to, say, the long book, by rated top analyst bet in their model portfolio with plus some additional signals. We have a scoring and alpha system. The stock gets in. In running an NLP over those research notes, we can run an NLP that says, does this read like a buy or a sell rating? What that does is help us remove the lower conviction buy and sell ratings out of there. We want to get quantitative conviction, which we've got, that this stock should come into the fund. I also want qualitative conviction. So I don't have to read 400 research notes. Those efficiencies and margins I expect to continue through different uses of AI, not only just from us but from the research team. If they have to do a marketing presentation, you know, they don't have to spend three hours putting all this information together. They could use AI to help make that process a lot quicker for them so they can spend more time on getting the stock calls right. Equally, they don't have to go out to read the company reports, or maybe management's out there talking about what's happening in the next six months, 12 months on the stock, how they feel about that. They could get an AI to read into that and summarize it for them. Again, on the margin, the AI will improve efficiencies for the analyst to get the stock call right. Do I think the AI will replace analysts? No, not a fundamental analyst. Even for quants, I think it will improve efficiencies. Small improvements on the margin to make things more efficient so we can focus on doing what we do best.

Eloise Goulder: It's fascinating to hear how you're using AI to scan the analyst report and infer the rating and cross-check that with the analysts' own rating. Do you often observe differences between the two?

Matt Jones: We do. And obviously, we've got quantitative information around whether this works or not. I think the most important thing to understand in all of this is no quant model is perfect and no AI is perfect. And I've got examples of this, where we know running, say, a more typical quant backtest on the AI, and it removing names that don't read well from our research improves average research performance. But it's not perfect because it will remove names in my daily process that are strong buy rated that the analyst has huge amounts of conviction in. So just through a nuance in the way they're talking about the stock, it can easily get that wrong. And that's why we don't use it in a pure quantitative form. Because again, going back to the point, my data is not perfect coming into that. The AI algorithm is not perfect. I don't want to trade off imperfect information.

Eloise Goulder: Yes.

Matt Jones: So it's improving on the margin to make my job quicker. But would I use it in a purely quantitative sense? No because the algorithm is not perfect.

Eloise Goulder: Yeah. Well, we've spoken so much about the benefits of your processes and the inherent stock-specific, idiosyncratic alpha that you're really capturing systematically from those fundamental analysts at Fidelity. Can we turn to the challenges? I mean, what are the main challenges to this approach?

Matt Jones: Yeah. I guess the obvious thing, when you're a very fundamentally driven bottom-up process, the challenge always is when the market's not being driven by that. And that's not my worry because it is what it is, but my concern in the environment and maybe even the last few weeks and going into the end of this year, even, is you're going to get market activity driven by exogenous, big, dominating effects. You've got the, inverted commas, threat of the Fed cutting rates. These are all big macro dominating things and not really individually stock specific. And that's, I guess, the drawback, and where we would expect this process not to work really well, if it's not a fundamentally driven market. Maybe it's even more retail driven. So when that total risk is being driven, things that aren't stock specific and not fundamentally driven, that's a period where the strategy, you would expect it to struggle. But that's not a bad thing. I think the great thing about that is when it does struggle, I can explain to our clients as well. When it's more macro driven, it's not fundamentally driven, I would expect the fund not to work then.

Eloise Goulder: Yes, absolutely. Coming back to the longevity of your portfolios and your process, I guess you have the benefit of a significant track record of that bottom-up, idiosyncratic alpha really playing out, and that should give you the confidence to live through these periods of macro volatility.

Matt Jones: Yeah, exactly. And from a global long-only fund, we've been doing it for 11 years, so we know where it does and doesn't work.

Eloise Goulder: So finally, Matt, we've covered the past. We've covered the present. We've touched on AI. Are there any other observations from your side on what the future holds and what you'll really be focusing on?

Matt Jones: Again, the core to everything we do is the analyst research team and that fundamental input, so I would never expect that to change. And that's part of a repeatable, robust process. And I think I alluded to it before, things like AI and technology, I think, are going to be some of the biggest improvements in what we do, but not improvements in a step form. More improvements in the efficiency and on the margin, to make sure that we're able to generate alpha and focus on what we do best, but ultimately to provide solutions to our clients. So I think that's where the next step and growth will be, is using the AI and the technology to take this whole process that we've developed over the 19 years and make that able to solve more problems for clients in a more efficient way.

Eloise Goulder: Absolutely. Well, I think that's a brilliant place to wrap up, Matt. And it's been so interesting to hear about your processes, the way you're marrying that fundamental bottom-up, idiosyncratic alpha source from the global research department with your systematic quant portfolio optimization tool kits. And it's so interesting to hear how this differs from the more traditional, systematic portfolio managers who don't have the research analysts as the starting point. So thank you so much, Matt, for walking us through all of this today.

Matt Jones: You're welcome. Thank you.

Eloise Goulder: Thanks also to our listeners for tuning in to this bi-weekly podcast series from our group. If you'd like to hear more about Matt's work at Fidelity International, then please do look at the show notes. Otherwise, if you've got questions, or if you'd like to get in touch, then please do go to our team's website at jpmorgan.com/market-data-intelligence where you can always reach out via the Contact Us form. And with that, we'll close. Thank you.

Voiceover: Thanks for listening to Market Matters. If you've enjoyed this conversation, we hope you'll review, rate, and subscribe to J.P. Morgan's Making Sense to stay on top of the latest industry news and trends. Available on Apple podcasts, Spotify, Google podcasts, and YouTube. The views expressed in this podcast may not necessarily reflect the views of JPMorgan Chase & Co. and its affiliates, together J.P. Morgan. They are not the product of J.P. Morgan's research department and do not constitute a recommendation, advice, or an offer or a solicitation to buy or sell any security or financial instrument. This podcast is intended for institutional and professional investors only and is not intended for retail investor use. It is provided for information purposes only. Referenced products and services in this podcast may not be suitable for you and may not be available in all jurisdictions. J.P. Morgan may make markets and trade as principal in securities and other asset classes and financial products that may have been discussed. For additional disclaimers and regulatory disclosures, please visit www.jpmorgan.com/disclosures/salesandtradingdisclaimer. For the avoidance of doubt, opinions expressed by any external speakers are the personal views of those speakers and do not represent the views of J.P. Morgan.

[End of episode]