---

---

In this episode, Stephan Kessler discusses the transformative impact of AI on quantitative investment strategies, signal generation, and portfolio construction, highlighting how AI enhances productivity, signal uniqueness, and alpha generation.  Kessler is a Managing Director and Global Head of Quantitative Investment Strategies (QIS) Research at Morgan Stanley, focusing on research around signals, portfolio construction and risk management in systematic strategies. He holds a master’s in Economics and Business Administration from the University of Mannheim and a PhD in Economics from the University of St. Gallen, where he still lectures on financial modelling. Kessler is also the author of numerous academic-journal articles on financial markets.

 

Keywords

AI, Quantitative Strategies, Signal Generation, Portfolio Construction, Hedge Funds, Sentiment Analytics, Alpha, Machine Learning, Investment Strategies

 

Chapters

00:00 The Evolution of Quantitative Research and AI

06:21 Navigating Alpha Generation in a Competitive Landscape

08:51 Balancing Systematic and Discretionary Strategies

08:57 Comparing Systematic QIS Strategies with Hedge Funds

13:26 The Changing Perception of Quantitative Strategies

 

Transcript

S. Aneeqa Aqeel (00:00)
Hello everyone. My guest today is Stephan Kessler, Managing Director and Global Head of Quantitative Investment Strategies Research at Morgan Stanley. He’s joining us from London. Welcome to the HFR Podcast, Stephan.

Stephan (00:12)
Thank you very much for having me.

S. Aneeqa Aqeel (00:13)
So I did some research on you, and I found a quote from a conference last October where you said that a key use case of AI for quants is signal generation from sentiment analytics by analyzing news reports, and you advocated for quants to form a symbiotic relationship with AI. So I want to kick off with that. Can you elaborate for us on what sentiment analytics with AI have yielded particularly in this current year with  concurrent macro shocks.

Stephan (00:46)
Yeah, so I think the whole research process in the quant space has changed fundamentally over these past few years. I’m in this business for for twenty years now and if I look over the past years, there hasn’t been a time that had so much disruption, it’s had so much change in terms of what we can do and and how we how we go about it. So this whole AI hype, it’s in my view really a game changer.

And the way we use it is in many different ways. So maybe to firstly talk about the symbiotic part that you’ve mentioned and then we can go a little bit into the sentiment question. I think as we as we get more and more used to this new technology, we are starting to build a working flow where we don’t work anymore as individual researchers that come up with their idea and then they type something in their computers, some feedback. Rather than that, I think as researchers we need to start to work with the AI in mind. So it becomes more and more a conversation between the AI as your research assistant of sorts and oneself as the researcher. And as a matter of fact, it’s not just one research assistant, it’s many research assistants that can run at the same point in time.

So that means that we actually need to change the whole workflow in the sense of in the past we could only work linearly. We had a problem, we had a question, we worked on it, then we you know we researched literature, we coded it up, we got some answers, then we went to the next question, it’s the same thing. Now we can address several questions at the same time. We can let certain agents work with this information, with these questions, and then come up with the results.

As we do that and as we gain more and more experience there, I think we see it’s ever larger projects that these agents can do. Now, you know, you can essentially give a very high-level brief to an AI and you ask it to reconstruct a strategy or reconstruct an idea and backtest it.

That gives you an answer, but most likely that answer is actually not correct because the AI made choices, the AI used modeling assumptions that are not really holding true in reality. So what we do in terms of thinking how to how to use the AI is we sandbox it. We tell the AI, well, here’s a back tester. The back tester knows about the calendar of the business days, the back test knows about the transaction costs we use, then here’s a data source we want you to use.

That is a clean data source that’s properly tagged, that’s properly lagged, that’s point in time. Then we ask the AI to use a specific portfolio construction approach that’s also pre-coded. And then within this kind of several sandboxing tools and skills, or whatever you want to call it, you can then give the AI indications and guidance as to what it should actually do. And the more specific you can be and the more precise on which building blocks to use as it does this research, the better the output, right? And that’s what we spend a lot of time on at the moment, which is to think about how do we leave the AI still kind of working and being independent, but also give it enough guidance so it doesn’t fall into certain traps that can happen in research. And so you know what we then need to do in the symbiotic relationship is we on the one side need to provide the right guidance to the AI, we need to send the right questions, but we also as researchers need to change our workflow to kind of leverage this new technology as much as possible. And I would say we see already very large increases in productivity, in the speed we do research, but also in the quality that we can do research in, because the things such as literature review, for example, that’s now done much faster and actually much better by the AI because the AI can read an article much faster than we do, and it can give us a much more deep analysis of different articles quickly. We still need to verify that, right? I think we’re not yet in a space we can fully trust what comes out, but we can leverage it for certain steps very effectively right so and that’s simply a change of the workflow and equally you know I think if you go back a few months it was probably more of a resistance to move to that model. I think nowadays a lot of folks in the marketplace they realized if you work against the AI or if you try to exclude the AI you’re missing a trick right.

Now in terms of sentiment, right, to your next question, we see a lot of ability to do text analytics, to also quantify text that wasn’t there before. And that leads to the ability to generate new signals because you essentially can feed the AI an annual report, you can say this is a section I want you to analyse and then look for changes versus the previous quarter, evaluate those changes, give me a sentiment score for what those changes mean, and we can use that in our strategies, in our signals. We see a pickup on really through two ways. One is we don’t need as much pre-processed data from data providers, we can actually process the data much easier ourselves with these tools, and secondly, we see the ability to quantify and ask specific questions that you can then run in systematic backtests on the back of that, which also helps us with the alpha generation. So we find it quite powerful, and you know, when you think about the traditional quant, we had say your factors, your multiples, in equities, that is maybe quant 1.0 Then we start to do machine learning, which is like using more advanced statistics of sorts, that led to an improvement of the signals we get. Then we start to do natural language processing, through, you know, back of words at the start, but then a more advanced model as time went on. And nowadays with the AI we’ve got yet a fourth category, a fourth generation of signal generation that we use and that’s gonna be very impactful in alpha generation.

S. Aneeqa Aqeel (06:21)
Excellent. So just to touch on a couple of things there that you mentioned, in some sense though, AI obviously adoption might be staggered across the industry, but then wouldn’t it become a race to the bottom? So how do you identify true alpha generating signals that are unique to you? Why isn’t everyone then going to essentially arrive at the same point and thereby eliminate all of those premia? And then do you also think that there’s more noise now? Because there’s also more news generation then?

Stephan (06:55)
Yeah, it’s a very difficult question, and I don’t think we really know the answer where this will leave us yet. But if I were to draw a picture that I think could be the steady state, it’s the following. So I think first of all, the AI tends to interpolate, right? So it tends to look at what is already out there and kind of go in those steps that already were trodden. So I think to just read your average newspaper article or average academic article about new signal then implement it and earn a lot of alpha, those opportunities will become less. And here I think there’s a question also around are we talking about a high capacity signal, low capacity signal, or is it what’s called true alpha or risk premium where probably a risk premium has more legs to persist whereas an alpha that is grounded in illiquid markets that will be arbitraged away much faster right and so I think the lower hanging fruits in alpha generation those will be arbed away faster but at the same time the AI allows us to generate a lot more information a lot more signals a lot more data to process and I think in that analysis there are new signals that are relevant and those will generate alpha. So that’s why we need to use the AI. If we don’t use the AI, we don’t get to the level of efficiency I think in the midterm to really find the next alpha signal. And then it’s likely to just remain the way it always was, which is the alpha was always a game of cat and mouse between the people who first found it and produce very high alpha and then the people who followed and then, you know, deteriorated the alpha a little bit. And then I don’t think that will not change all that much. But the way we go about this will change and possibly the speed of information decay will change as well. Of signal decay.

S. Aneeqa Aqeel (08:45)
All right, so let’s switch from AI to the business of portfolio construction: I know you think about identifying high-alpha managers, decomposing hedge fund returns, so how do you compare the returns from purely systematic QIS strategies and a Morgan Stanley tracker fund?

Stephan (09:05)
Yeah, so I think what we tend to do is we characterize hedge fund returns and here it’s usually you want to kind of go a little bit further in terms of the level than the indices. You want to kind of look really at combinations of funds, for example. You characterize those through a factor lens. So there’s the Fung and Hsieh models from the early 2000s which essentially say, well, hedge fund returns can be characterized through trend following like factors, through value factors, through size factors and the like. And then once we have run these factor analysis, there is still alpha left, which is you know the kind of the valuable bit that hedge fund managers get compensated for the way they are. And so the factor analysis I think is very important as a starting point.

And what we do is very often is we do two things. So one thing is clients that we work with together have portfolios of hedge funds in their portfolios and they want to understand okay how can we complete these investments? So they already have their relative value credit manager, they already have their trend following manager in there, they already have possibly a leveraged loans manager in there. But they want to see how else can they make their portfolio more robust. And QIS strategies, so quantum investment strategies, which are systematic investment strategies that are usually classified in value, trend, or momentum, carry, and then you do that across asset classes, equities, commodities, fixed income, rates, and then also in the volatility space and non-linear strategies. You have then a menu of these QIS strategies.

And you can now look at what is already in a portfolio and what can one use out of that menu of quantitative investment strategies to build a more complete portfolio. This whole approach is at the moment quite popular and much discussed in the term of total portfolio allocation, which is about thinking about an overall outcome in the top portfolio and maybe breaking through traditional asset class allocation paradigms.

Think in a factor space about return drivers that we can combine together. And I think QIS is a great place to to achieve that. The other thing that we think QIS is used for and also quite successful is to get a baseline return profile in the portfolio. So it can combine together an FX carrier strategy, an FX value strategy, an equity value strategy, a volatility carry strategy where one sells options and delta hedges them. One can combine these strategies together in a multi-strategy, multi-asset portfolio. And in our research you find Sharpe ratios of about 1.6 when one builds these portfolios. And these are then overlays investors can add in their allocation. The benefit of some of these QIS strategies are they’re usually accessed via swaps. So they’re not funded like funds are. So investors can invest in a hedge fund, they can add these QIS strategies on top as an overlay. Or some investors they use these QIS really as the core driver for their alpha in the portfolio, as they allow a non-correlated return profile. Because as we come back to this portfolio construction mentioned earlier, we can combine these together in a way that they don’t carry much factor exposure, that they’re neutral to equity markets.

They’re neutral to rates markets, they’re neutral to macroeconomic shocks. And so as we build this kind of core say shock neutral, pure alpha portfolio of QIS, many investors add those then as the alternatives diversifier to their portfolio and you know the ability to not having to finance it on one side, but also transparency and portfolio construction and strategy design is something that is valued by investors that we talk to. And you know these pure alpha diversifier portfolios I’ve just described, they can be comparable to a broad portfolio of hedge funds. They of course don’t provide say the selection alpha that a good hedge fund manager provides. That’s not something that a QIS strategy captures but a QIS strategy captures the core drivers, the systematic drivers of investment strategies quite well and delivers attractive long term returns on the back of that.

S. Aneeqa Aqeel (13:14)
Stephan, so how has the perception and application of quantitative strategies, especially now with AI in the mix, how would you say that’s evolved amongst institutional investors over the last few years?

Stephan (13:26)
I think the perception has actually evolved quite positively. So if you think about the history of these strategies, they really go back to some academic publications in the early nineties, like the Fama French factor paper from ’93, for example, which founded this perception that there are ways to systematically screen equities that deliver long term attractive returns in a market neutral or cash neutral format. So that’s been around for a long time as a concept and a lot of money has been raised in these assets. Trend following strategy is a prime example. They’re around for a long time. According to our analysis, about 300 billion probably in trend followers, probably a little bit more actually on that. So it’s already a core part of what investors do. The absolute numbers have increased I would say, steadily. But what has changed more recently is I think this understanding of quant through this AI lens. So if you think about the perception of experience that your average person has with technologies like AI, today everybody is using these tools. And so everybody starts to feel confident talking to a machine and getting an answer and also sees that this answer actually is reasonable. And this experience has in my view led to a little bit of an acceptance that machines provide very useful results and that acceptance has increased. I think many investors was always there, but essentially there’s some academic research that shows the machine has needed an edge versus human in the past to be considered on par. And I think this perception is changing now.

Where you know you needed to be really a very deep quant. You needed to have a statistical education to understand what do many of these strategies do. That’s you know until very recently that was really the entry point. Today I think many investors are more willing to say, well I don’t need to understand all of the math behind it because when I use my AI I also don’t go that deep, but I trust the answers. And like this I think this transfer has helped. On top of that, the quality of the performance has of course helped. Like this performance has been strong. And so strong performance is always something that helps to make the case. And so that I think is supporting the inflows as well. And the final point I would make is more and more every investor will need to be some form of a quant in order to get that kind of cutting edge in terms of information processing. Now quant doesn’t necessarily need to mean you need to know all the statistics, doesn’t necessarily mean you need to code a lot. It can all be done now by front end, which is like a Claude command line interface approach but more and more investors will rely on that and so there will be more and more systematic quantitative inputs in portfolio management to get to better results. And I think that’s even with very fundamental managers, I think they will layer in more and more of a first step that is a systematic screening process, a starting point to then layer in their fundamental insights which contain alpha and elevate the performance even further.

S. Aneeqa Aqeel (16:25)
Excellent. Thank you so much for your time, your insights. I really appreciate it. thank you for coming on the show.

Stephan (16:28)
Yeah, my pleasure. Thank you so much for having me.

 

IMPORTANT INFORMATION ABOUT THIS CONTENT

For important Morgan Stanley disclosures, please see the Morgan Stanley Research Disclosures website at www.morganstanley.com/eqr/disclosures/webapp/generalresearch

HFR is the global leader for hedge fund data and indices in the financial world. The HFR Podcast features interviews with allocators and subject experts. Contact us at hfrmedia@hfr.com

The opinions shared in this video are those of the individuals featured and do not reflect any interest, recommendation, or policy of HFR or its subsidiaries.

HFR has no obligation to provide revised assessments in the event of changed circumstances. HFR cannot guarantee the accuracy of the information provided. This material may contain “forward-looking statements” including projections, estimates, and information about possible or future results related to market or recent developments. No portion of this podcast may be published, reproduced, transmitted or rebroadcasted in any media in any form without the express written permission of HFR Media LLC. © 2026 HFR, Inc., all rights reserved. HFR®, HFRI®, HFRX®, HFR.COM™ and HEDGE FUND RESEARCH™ are the trademarks of Hedge Fund Research, Inc.