Trading Insights: Exploring trend-following strategies with Lynx’s CEO

[Music]

Martin Kallström: I think we are in the most exciting times ever for quant managers. So I'm very optimistic about the future. It's partly because we see the full explosion, with new data sources coming at hand for quant managers. Large language models are extremely interesting. If handed correctly, I think this could be just as fantastic leverage to the research process of quant managers and how to make use of unstructured data.

Eloise Goulder: Hi, welcome to ‘Market Matters,’ our market series here on J.P. Morgan's Making Sense. I'm Eloise Goulder. And today I'm delighted to be joined by Martin Kallström, CEO at Lynx Asset Management, a systematic hedge fund based in Stockholm and founded back in 1999. So, Martin, thank you so much for joining us here in London today.

Martin Kallström: Thank you very much, Eloise, for having me here.

Eloise Goulder: And Martin, could you start by introducing yourself and your background?

Martin Kallström: Sure. My professional career started after university studies as an actuary with Watson Wyatt before taking on a role as head of investment consulting and actuarial practice for Aon Consulting in Sweden. And after having done that for a few years, I moved over to AP1, where I started the alternative investment department. And with the hedge fund portfolio, I had a particular focus on CTAs and global macro and got to learn the industry very well, invested with many managers, including Lynx. And in 2017, I sat down with Svante and we started to talk about Lynx and how it could develop in the future and ultimately, he came with an offer to me to join the firm in a leadership position. So, since 2018, I've been with Lynx; first as a deputy CEO and now, during the last year to become the CEO and taking over from Svante.

Eloise Goulder: And how would you describe the investment philosophy at Lynx?

Martin Kallström: So, first of all, we believe there's anomalies in markets that could be captured by systematized processes, rigorous scientific research with mathematical and statistical methods. Our flagship program is very much centered around trend following. We have, since inception, had that as a core of our strategy, but complemented trend following with other type of alpha sources to generate a program that is more robust over time and generate a higher expected Sharpe ratio.

Eloise Goulder: And the alpha potential in trend has been around for many decades. Why do you believe that it exists and that it's so durable?

Martin Kallström: So the strategy of trend following is really at its core, it's a loading and exposure to change. And change is natural. It comes through changing conditions for economies, for countries; they go in prosper and decline. We see behavioral changes, sentiment changes, technology changes. And with all of those major shifts those changes take time to disseminate in markets.

Eloise Goulder: So there are other quantitative investment firms following trend. We've heard from several of them on this podcast series. Martin, what would you argue differentiates Lynx?

Martin Kallström: So, I think being based in Sweden is probably the core of it. The country has a very entrepreneurial culture and constantly comes up actually as one of the countries in top of the list of countries that innovate well. The tech unicorns of Spotify and Klarna is something that comes up many times talking about innovation in Sweden. But we have other strong engineering traditions as well, companies like Atlas Copco, SKF and Volvo. So when it comes to innovation, we have a culture of that in Sweden. We have a very healthy life work balance. People are staying long term at Lynx, we tend to have a strong pipeline of researchers coming to us.

Eloise Goulder: And you have a comparatively big research team, but I understand you run a few core strategies. Could you talk about what those strategies represent?

Martin Kallström: Yeah, so our flagship program, the Lynx program, where we combine trend following with other complementary approaches that are seen as alpha overlays to trend following. This program has a dual mandate, for one, to generate attractive risk-adjusted return. But the second part of the mandate is to generate returns that are highly complementary to our client portfolios, especially when times are difficult. So that tends to be during long drawdown periods for equities.

Eloise Goulder: It's really interesting to hear you articulate that returns or risk adjusted returns alone is not the only objective, but you also have this protective, protecting the end investor's portfolio objective. And given that trend following has been around for such a long time, how do you go about constructing and ultimately improving upon your strategies?

Martin Kallström: We still have a couple of models running in production from the inception with a 25 year track record. But we have continuously developed new models, new approaches to trend following that we believe on a standalone basis, are superior to the classical ones. We really saw a big jump when moving from univariate trend following approaches to multivariate trend following approaches that took place 2008-9.  But other areas that are also very interesting is other type of alpha sources that could complement trend following. And those alpha sources tends to be focused on shorter term price movements, they tend to be utilizing machine learning techniques that are essentially capturing nonlinear, highly complex patterns in markets. And they tend to use new type of data with macro based ex ante type of hypothesis, which is highly complementary to trend following. So when you compare models it's a big challenge. And I would claim this is a big challenge for any investment strategies that are back tested. Because what you're left with is models, that you have a long term, out of sample track record from, realized track record, those should be compared to newer, back tested, solely in sample generated models. And what we know is that there's an increased risk of overfitting with new models. So how do you compare the two? The way we do it is that we are borrowing drivers of returns that we've seen work before. So even a new model could share and could benefit from something that we've seen work before and we want to shrink the part that is idiosyncratic and unexplained, but also highly sought after. So that unique piece of a new model is, of course, something that benefit the program very much, but it's uncertain that we'll get it.

Eloise Goulder: What would you argue are the key benefits and or drawbacks of utilizing machine learning?

Martin Kallström: So we consider ourselves somewhat of a pioneer in the space of machine learning. In our industry, we started doing research in 2009 and have traded alpha signals based on machine learning since 2011. Over the last years, we've seen tremendous development, both in the technique and the type of data that could be used by machine learning models. But our approach has been fairly consistent. We are very careful how to use these tools. The algorithms are so powerful that they will find any pattern in historical data. Most of those patterns will turn out to be just noise. So how you design your models, how you train them, how you control for overfitting, is absolutely critical. But overall, this has been an amazing success.

Eloise Goulder: And how do you judge which data makes most sense to feed into your machine learning models? Because we all know that history is incredibly valuable when it comes to machine learning. And price-based input variables have a lot of history and therefore can be quite powerful. But when you talk about newer sources of data, to the extent that some of them have less history, how do you determine whether or not a machine learning model would be appropriate to use?

Martin Kallström: So, when building features for machine learning models, you want to serve as information-rich data as possible and with a length of history. So of course, price-based data has a very direct relationship to what we try to predict. So they are highly information-rich and they come with great length. So what we have done is that we have used higher frequency of data. but also to look at new data that could be information-rich. And here the challenge is of course that the data is not always so direct related to what we try to predict. And often with limited length of history. So you need to be careful because there's increased risk of overfitting to noise, but at the same time, potential information here that could be utilized. So this might be the art in building machine learning models. It's about serving the right features with enough information in it.

Eloise Goulder: And I've heard others say that it's quite humbling to use machine learning models and to see their efficacy in predicting price performance and alpha opportunities, because of course it takes some of the human domain knowledge out of the picture would you argue the skill and the role of the human has changed with the use of machine learning models.

Martin Kallström: That's a good question. Yes and no, I would say to that because the humans are critically important to design how the model should be constructed, and also determine what type of data and how the features should be built into the model. So maybe a little bit less of domain knowledge of the financial markets, but it's still a critical component of designing a well-functioning machine learning model.

Eloise Goulder: So, you mentioned that you're constantly looking to enhance your strategies with new sources of data. How would you argue this process has evolved over time?

Martin Kallström: I think we are, overall, in a very transformative period for asset managers in general and in quant in specific. So, we're seeing an absolute explosion in data, as well as tools that we have at hand to analyze that data. And ultimately, this means richer, more nuanced information to make investment decision upon. At least for those that have the capacity to use that data. And here, I think it is a clear advantage to be a little bit larger. There's also very interesting developments when it comes to large language models and how to structure unstructured data that couldn't be utilized before in a quantitative process. But with the use of large language models, there's all of a sudden a new field that we could capture with data that is in text format, it could be audio or video data that could be structured and traded upon, which is very interesting.

Eloise Goulder: Are there any particular textual sources of data which you find to be most powerful or perhaps most orthogonal to your existing alphas?

Martin Kallström: I think at this stage, when talking about the futures programs that we're running, there's big challenges to see high information from textual data. So, we're making a lot of interesting research, but yet to be seen the strong evidence from it in production. When it comes to trading single name equities, there's a total different landscape. In a year's time, we're planning to launch a systematic equity program, long short, and that will have a lot of focus on large language models, making use of text data, for instance.

Eloise Goulder: Well, such an exciting area for development. And you referred to the increasing use of macro data and I find this space fascinating because, of course, there should be a relationship between macro datasets and groups of assets. And in fact, we heard from our colleagues at Macro Synergy about forecasting asset returns using macro variables just a few weeks ago on this podcast series. How rich an alpha source do you see macro variables to be?

Martin Kallström: Yeah, we definitely have models running on macro data today. And something that we are researching intensively. The challenge is to find higher frequency macro data, but it's definitely out there and something that we are constantly on the hunt for.

Eloise Goulder: So, we've spoken a bit about the model and you've talked about the enhancements in your models and the fact that you make great use of machine learning. We've also spoken about the enormous data availability. We always ask our guests, what's more important, the data or the model? Do you have a view on this one?

Martin Kallström: Well, can I say something else?

Eloise Goulder: Please do.

Martin Kallström: Okay, then I say humans, because humans are the ones that are developing models. They steer the parameters, the hypothesis behind models, and they select the data that models are built upon. But of course, both models and data is super important. But the humans behind are the ones that are making this happen.

Eloise Goulder: And therefore, the importance of culture, the importance of the research effort and the collaborative approach must be paramount if you're putting humans at the epicenter of this process.

Martin Kallström: Yeah, because even if this is a fully systematized approach, and we are trying our best to make an automated program, it's still the humans that are behind selecting models and data and making sure the program is running according to the mandate.

Eloise Goulder: It's so interesting. I mean, we've heard this answer from other CTAs as well so it's not necessarily a surprise to me to hear your answer. But when you take a step back, the fact that you are a truly data driven and model driven firm, it is fascinating to hear that humans are still so paramount in your process.

Martin Kallström: Yes.

Eloise Goulder: Before we close, Martin, could I ask you what's next? You've mentioned the enormous availability of data. You've mentioned use of LLMs, use of textual data. What excites you the most when you look towards the next few years?

Martin Kallström: I think we are in the most exciting times ever for quant managers. So I'm very optimistic about the future. It's partly because we see the full explosion with new data sources coming at hand for quant managers. Large language models are extremely interesting. It's not without risks, but if handed correctly, I think this could be just as fantastic leverage to the research process of quant managers and how to make use of unstructured data.  And then lastly, at Lynx, even though we focus on a few selected areas where we think we have a strong position, we are moving into systematic equities with a new program, which is highly interesting. We dedicated a team for a couple of years already and see very interesting opportunities in that area.

Eloise Goulder: Well, it's wonderful to hear your excitement behind this systematic investment space given the enormous increase in availability of both datasets and analytical techniques, including LLMs. We've really seen such a fast pace of improvements in both over the last few years. So thank you so much, Martin, for taking the time to walk us through your strategies and your business model. It's been a really wonderful conversation.

Martin Kallström: Thank you very much, Eloise.

Eloise Goulder: And thank you also to our listeners for tuning into this bi-weekly podcast series from our group. If you'd like to learn more about Lynx, then do follow the link in the show notes. Otherwise, if you have questions for our team, then do go to our website at jpmorgan.com/market-data-intelligence. And there you can always contact us via the form. And with that, we'll close. Thank you.

[Music]

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, and YouTube.

The views expressed in this podcast may not necessarily reflect the views of J.P. Morgan 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.

© 2025 J.PMorgan Chase & Company. All rights reserved.

[End of episode]