Trading Insights: Data vs. model – what’s more important?

[Music]

Edwina Lowe: Hi, I'm Edwina Lowe, product specialist within the Data Assets & Alpha Group. And today I'm joined by Eloise Goulder, head of the Data Assets & Alpha Group. So, Eloise, thank you very much for joining me today.

Eloise Goulder: It's a pleasure. Great to be here.

Edwina Lowe: So as we are fast approaching our 100th episode since we launched this Trading Insights podcast three years ago, during which time we’ve interviewed about 20 external buyside guests. I thought today we would do things a bit differently. I would turn the tables and ask you two of the key questions that we’ve asked most of our guests. So the first question is: what is more important, the data or the model?

Eloise Goulder: Absolutely, Edwina. Well, this is one of my favorite questions. So, of course, I'm very happy to talk about this one.

Edwina Lowe: Excellent.

Eloise Goulder: So I'm going to start by giving you a very straight answer that I would lean on data being more important between data and model. And that's because the data is right at the start of the investing decision, the way I see it. And there is such a broad range of data that one can use to have a lens on what's going on in markets and what's going on for a given stock. And I think if you use the full breadth of that data and if you use it wisely, and by that I mean if you normalize it in a sensible, thoughtful way, which is incredibly time consuming, then it can be very, very powerful. And even with a very simplistic model, that can be a brilliant tool for investing decisions, but I've got a lot of nuance to add to that answer. Perhaps to give some context, when I ask this question, I'm asking with a view to what's more important for making predictions in markets. And that data could come from many different sources. It could come from macro fundamentals, GDP, PMIs, ISMs, Nowcasts, so a real-time measure of the state of the economy, which has become such an important tool within a macroeconomics framework. Micro fundamental data, sales, margins, return on capital, earnings, dividends, revisions to all of those things, which are in many cases more powerful than the absolute numbers themselves. And then, of course, there's the wide range of technicals data, price volume, correlations, market depth, implied volatility, RSIs, And then there's the flows and positioning category of data, an enormous category that we often discuss from our team, because obviously we look at hedge fund positioning, retail, ETFs, CTAs, long onlys, put call ratios, you could add to that valuation measures as input data sets. Sentiment, a massive area of data and one we're focusing on as a team, whether it's social media sentiment, news sentiment, survey sentiment, research sentiment and recommendations. And I guess a final category of data I'd add, which I think is becoming increasingly important, is external or other revenue proxies, rather than following the company and listening to the company in terms of their guidance for earnings, using other arguably more real-time proxies for the health of their revenues, like credit card data or perhaps even satellite imagery data. There's such a breadth of datasets that are available to investors and are really important and are, in an ideal world, somewhat orthogonal in that they give you a different lens on what's going on in markets and for given companies. But it's not just the data itself. But actually, if I had to really hone in on my answer, it would be the data processes and normalisation, because most of those datasets, if you use them in raw form, wouldn't give you much of an alpha. One small example, when we look at social media sentiment via the dataset that we make available to clients, we've noticed that there's a real seasonality and it's not a monthly seasonality, it's a weekly seasonality. We find that the posts that come on a Friday are very different to the posts that come on a Monday. I mean, one of the most trending topics we see on a Friday is Happy Friday.

Edwina Lowe: That's fascinating.

Eloise Goulder: So if you don't normalize, seasonally adjust on a weekly basis, that data actually is fairly flawed. So that would be my nuanced, but highest conviction answer on that question.

Edwina Lowe: So you've given examples of the breadth of data sets available for investment decisions… but why would you say they are all so important?

Eloise Goulder: So I'd argue that to have a really robust strategy, you need to know that there is data at the start, which is really giving you an edge, giving you an underlying signal. And in an ideal world, you also know why such a signal exists. Is it a result of a persistent bias or a persistent investor behavior that exists in the market that you are counteracting? Or is it the fact that you have data that not many other people have? So I think you really need to have that underlying signal. You need to have that source of alpha. You need to be aware of the market inefficiency that you're capturing. So if it's, say, sentiment data or earnings revisions or PMIs, then the alpha may simply be the fact that this information will take others more time to price it in. So it's the fact that you're there early. It might also be the fact that certain variables are autocorrelated or they trend. And we really see this with earnings revisions. Why on earth is it that companies tend to outperform up to a whole month after they've seen an earnings upgrade? It seems a bit strange because it's not as if you're having to act on that information that quickly. But actually, we find that the best predictor of a future earnings upgrade is a past earnings upgrade, probably because companies are often benefiting from a structural advantage in the case of earnings upgrades. And that structural advantage is likely to persist, whether it's an underlying structural demand for a product or a company's product is outpacing its competitors. These things do tend to trend. Another edge you might have with data is if it's proprietary. So is it a proprietary survey? Positioning data might be proprietary. That's an edge that others don't have. Another reason that data can have edge is because it took a lot of analysis even to create the data. And this is where the lines blur really between data and model. But I think our social sentiment data set is a great example because it's taken such an enormous engineering heavy lift to ingest all of these posts from social media and normalize all of that with a history so that we can run appropriate back tests. And getting hold of that data is difficult. In a sense, it's a form of computational alpha and not everyone has access to that. So essentially, there is a higher barrier to entry here.

Edwina Lowe: Well, it's clear that you really see data as an integral part of the investment process. But I know that you also think the model is important too. Could you touch a bit on that?

Eloise Goulder: Yeah, when it comes to the models, there are obviously so many different models one can use to assess the relationship between the data, all of the data sets we just talked about and markets. You can simply rank, you can simply bucket. And in a sense, if the data has an edge, as we just discussed, it's very early, it's very timely, it's very differentiated. You can often manage with a very simple model and arguably a simple model is better because at least it's more robust. Equally, using machine learning techniques, you can also look at this on a nonlinear basis. I also think there's an enormous source of alpha, in combining all of your different strategies. Let's say you have lots of different strategies based on lots of different data sets and lots of different models. Then the question is, how do you combine them? Hopefully they're uncorrelated strategies. How do you apply your portfolio optimization? We find that supervised machine learning strategies can be really powerful there, in determining how to weight different strategies. So I think all of that goes into the complexity of the model. It's obviously an endless topic that is worth a lot of discussion as well. Finally, I mentioned execution and transaction cost analysis. What's so key here is that when you're looking at an individual strategy, we always look at the profit per trade based on that strategy. And if the profit per trade is actually in the tens of basis points, let's say it's 20 basis points, then whether or not it becomes a profitable strategy entirely depends on what the transaction costs are and whether they eat away at it. And if the transaction costs are above those 20 bps, then of course it's no longer a profitable strategy. It might be a systematically unprofitable strategy. So transaction costs are completely critical to look at. And in that sense, there is a real blurring of the lines between the ultimately really the PM role and the trader role. In this case, if the PM is determining the data and the model, and the trader is determining the execution, well, they really all need to be one and the same because data plus model might only be profitable subject to appropriate execution and accurate analysis of the transaction costs. And in reality, you really need this feedback loop where you test models, you test strategies, you might A, B test different execution strategies and only some of them, based on certain executions, will make sense as profitable strategies. And also it's worth saying that monitoring the sector skews, the factor skews, the portfolio skews is so unbelievably critical. And this is where human in the loop becomes so important because there will always be imperfection. To quote James Benford on our last podcast, the data is just a reflection of the past and history doesn't always repeat itself. And the models, every model is wrong. Some are useful. So of course, we've got imperfection in the data, imperfection in the model. And so the end result might give you these skews, which the human might have additional nuance to throw in. If there's news flow overnight and it's not captured in the data and it's not captured in the models, then you might still have a need for human judgment.

Edwina Lowe: Thank you, Eloise. Well, clearly there's a lot of nuance to your answer.

Eloise Goulder: Totally. And that's why I love asking this question. I love hearing what all of our guests have to say on the topic.

Edwina Lowe: Well, actually, on that point, I've had a look back at the answers that we've had when you've asked that question, which is more important, the data or the model? And really for the benefit of our listeners, since we launched our series, we've hosted a wide variety of external guests from macroeconomists, to CIOs, to chief risk officers, from both the discretionary and systematic ends of the spectrum. Roughly 50% of guests have been squarely in the camp of data being the most important thing.

Eloise Goulder: Great, I'm not alone then.

Edwina Lowe: No, far from it. The minority, in fact, have been in favor of the model as the most important aspect. And then I would say roughly the remaining 40% have been a slightly more nuanced answer, primarily leaning on the importance of human judgment or the role of human in the loop.

Eloise Goulder: Yeah, it's interesting how many people have actually not answered the question (laughs) and have gone with something different altogether, which shows that data and model are not the only important variables in the investing decision. And human in the loop, I think it makes absolute sense. I mean, arguably, you need human judgment in how you normalize the data, definitely in which models to use and how you combine the models, definitely in the transaction and the execution, and definitely in monitoring portfolio skews to anything, to region, to sector, to factor, to thematics. I would agree that the importance of the human in understanding the pitfalls or the flaws in any systematic process is completely key, because in an ideal world, you can use that to iterate to improve your model and or to hedge out some of the exposures and risks. But Edwina, thank you for that. I love the data-driven approach you've taken to seeing what people have answered historically. And I'm pleased in a sense to know that I'm not alone in valuing the data so highly.

Edwina Lowe: No, definitely not. So Eloise, if we could turn to the other question that we usually ask our guests, what’s next?

Eloise Goulder: Yes, a very high-level question that we do indeed ask all of our guests. So LLMs, they will obviously continue to be so significant in the investing process, on the buy side and on the sell side. I do believe it's such a rich and exciting source of alpha. It's opening up so many opportunities that didn't previously exist. I mean, we see it, don't we, on our side in terms of the client demand for machine readable forms of content.

Edwina Lowe: Absolutely.

Eloise Goulder: So many of our institutional clients are now building up large textual databases on which to apply their LLMs. We heard from Chris Pullman at Balyasny on this in December. Chris Pullman gave a very compelling case for how you can use LLMs in the discretionary process. It speeds up the research process. So LLMs really allow you to create that many more ideas, which in theory is Sharpe accretive. If the edge is in being among the first or among the best, and you can't afford to be neither, then with LLMs, if you can be among the first, but it's imperfect, that could still be pretty good and pretty alpha generative. And of course, LLM models themselves should get much better over time. And so to the extent that you're trading perfection for speed right now, perhaps you can end with slightly better perfection and speed in the future. But of course, one of the great challenges with LLMs and systematic strategies is back testing. It's so hard to back test reliably, especially given so many of the data sets that feed LLMs have limited history. So coming back to your question about what's next, as a team, we're doing so much work on sentiment, sentiment based on the news, sentiment based on social media. We already make some of this analysis by stock available to our clients via our Fusion platform. I also think the relationship between positioning and sentiment is absolutely key because the two trend together, but they're not always coincident.

Edwina Lowe: It's a great point, Eloise. So I think this is probably a good point at which to wrap up our conversation. We've obviously cited several the guests that we've featured on our podcast series over the last three years. And if any of our institutional investors would be interested in receiving a hardcopy magazine in which we have summaries of all of those episodes, please do get in touch with us. So, Eloise, thank you very much for joining me today.

Eloise Goulder: It's been a pleasure. Thanks so much, Edwina.

Edwina Lowe: Thank you also to our listeners for tuning into this biweekly podcast series from our group. If you have any feedback or if you'd like to get in touch, please do go to our team's website at jpmorgan.com/market-data-intelligence, where you can reach out via the contact us 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, YouTube, and jpmorgan.com. 

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 JPMorgan Chase & Company. All rights reserved.

[End of episode]