Trading Insights: Retail vs. institutional investor divergence

[Music]

Eloise Goulder: Hi, I'm Eloise Goulder, head of the Data Assets and Alpha Group here at J.P. Morgan. And today I'm so pleased to be joined by my colleagues Edwina Lowe and Luca Rainero to discuss the very divergent trends between the retail and the institutional investor in U.S. equity markets this year and what this all means. So Edwina, Luca, thank you so much for sharing your thoughts on this topic today.

Edwina Lowe: It's great to be here, Eloise.

Luca Rainero: I'm glad to join the conversation today.

Eloise Goulder: Well, I think it's fair to say that the retail investor in particular has been a really significant driving force in equity markets this year and especially in U.S. equity markets. Ultimately, U.S. equities have made fresh all-time highs. S&P 500 is now up about 8% year-to-date. And all of our data suggests that the retail investor has been the greatest net buyer of U.S. equities over that period. We've also recently seen the resurgence in the meme phenomena with our popular retail basket up a full 30% year to date but this bullishness from the retail investor has not been matched with bullishness from the institutional investor. In fact, the trends have been pretty divergent. I think this was most clear in the sharp equity market selloff from late Feb through to early April and equally in the sharp equity rebound through April and May. So if we cast our minds back to that equity market selloff, which took place around the time of Liberation Day, the S&P 500, it sold off almost 20% over that period. And institutional investor flows over that period were cautious and yet the retail investor remained consistently bullish. Data from our research team suggests that March was the strongest month on record for retail buying. And April beat that record with an even higher magnitude of retail net buying. All to say, tracking the retail and the institutional investor has been very important this year. So Edwina, can we start with you? From your perspective, how are we going about tracking these divergences?

Edwina Lowe: Well, I think that was a really great summary, highlighting the divergences that we've seen in these different investor types, so when it comes to what we look at from a data perspective, it's important to make the distinction between the flows and the sentiment. John Schlegel, who heads up our positioning intelligence team, has talked extensively about the flows so today, I think the focus of our conversation should really be on the sentiment.

Eloise Goulder: And we can dissect flows and sentiment across both the retail and the institutional investors, can't we? And there are typically divergences between both. And arguably, the sentiment is just as important from a predictive perspective as the flows.

Edwina Lowe: Absolutely.

Eloise Goulder: And it's worth just caveating here that when we refer to the retail investor and institutional investors, of course, each of those groups is enormous and heterogeneous. And we're using these phrases as somewhat of a catch all. But in reality, the retail investor could span any age group, any wealth bracket, and any experience level. And I think that's really important given the increasing availability of information and toolkits to the retail investor. But equally, the institutional investor includes hedge funds, includes CTAs within hedge funds, includes mutual funds. The institutional group is also wide and very heterogeneous. So, Luca, can we dive into all of your work around sentiment?

Luca Rainero: I remember when we discussed last year our work on social media, we used it as a lens to get insight into the retail investor activity. Now, since then, we try to expand a little bit what we're focusing on and we try to bring into our spectrum all sources for institutional investor sentiment, such as reports from our industry. This is particularly relevant given the rise of new AI tools that allow us to analyze these rich textual sources in quite different ways. And as you mentioned the distinction between retail and institutional is critical because of the divergence in behavior from both groups. Let's start with retail sentiment. As you mentioned earlier, the meme stock phenomenon is far from over. Recently, we have seen stocks like Opendoor, Krispy Kreme, GoPro being at the centre of social media attention. And one thing that's very helpful about retail investors is they love to discuss their trading ideas on social media. And we've seen both these aspects, picking up again in the latest resurgence of the meme phenomenon. So what have we done on our side? Well, with the help of the team, we're trying to focus on intraday momentum, or as it was defined in an article by Costola, Iacopini and Santagiustina, the “Mementum” rather than momentum, just like momentum, but applied to meme stocks.

Eloise Goulder: And I wonder if that term will take off given the presence of meme stocks.

Luca Rainero: Jokes aside, what we really notice is a strong persistence in intraday trends when stocks enter meme territory, which in other words means that when a stock is talked about on social media, the price action that we see during the day is likely to persist throughout the period. This gives us quite an obvious indication of how social media can be used as a tool to detect different shifts in market regimes. Now, given the strong results that we found in the retail sentiment, we thought it would be good to extend our analysis to gain insights into different investor types and track divergences in the sentiment between institutional and retail investors.

Eloise Goulder: It's fascinating that there's this predictive power for a certain subset of stocks, let's call them the meme stocks, but equally only in a certain subset of time periods. And this idea that you can use social media data to those regimes, a meme regime, so to speak, where the social media data becomes particularly compelling. And I think we're all so familiar with those stocks and those time periods when a stock is in play the social media data becomes so important, so we spoke earlier about the enormous divergence in flows between the institutional and the retail investor. Luca, have you seen the same level of divergence in sentiment between both investor types?

Luca Rainero: Yes, so let's just take one example, the so-called Liberation Day and the related market volatility. Now, you mentioned earlier, indeed, there's a difference in what we saw in institutional versus retail flows. And if you look at sentiment, we did see actually something very, very similar. On the one side, institutional sentiment dipped much more significantly than retail sentiment in the sell-off, and even fell below the level that we've seen during the early months of 2020, at the start of the pandemic. In comparison, retail sentiment didn't fall as much, and actually picked up from the lows relatively earlier. And this difference in sentiment with retail sentiment being more positive than institutional sentiment on a relative basis can explain and to some extent predict why retail flows were much, much more positive coming out of the market lows in April, when compared with institutional flows.

Eloise Goulder: So the retail investor just never got quite so bearish.

Luca Rainero: That's exactly what happened. Now, it's not always that straightforward to make a comparison between the two. If we look at retail institutional investor sentiment at the moment, I would say it's probably in the 50 to 75 percentile compared to the last five years. Now, historically, such a level of sentiment usually is associated with a relatively benign scenario, with market drifting or trending a little bit higher over the next few weeks, although without a strong directional sense. However, there are other points in time, like in April, when the sentiment was a much stronger indication of what we should expect in the market.

Eloise Goulder: Well, tracking the various types of investor sentiment is clearly a really important toolkit but it's also worth noting that the aggregation of all investor types is also an important measure to be aware of. And our positioning intelligence team have their tactical positioning monitor. This pulls together all of the investor types that we track from institutional investors. And then Luca, you've been developing an equivalent for sentiment, the tactical sentiment monitor, which brings together the aggregation of sentiment from the retail investor via social media, for example, with sentiment from the institutional investor.

Luca Rainero: Yeah, that's actually quite a good comparison. Although in reality when we look instead at different sources of sentiment data, the picture is nuanced. And there are several complexities, both linked to the actual sources themselves and their availability, as much as the models that we use to analyze that. And the recent developments have opened up many opportunities, but we need to be cautious about how we go about it, given the complexities related to their applications.

Eloise Goulder: Luca, how much would you argue that the advent of LLMs is transforming the sentiment space?

Luca Rainero: I think it's revolutionary, but not always in a positive sense. And I think when we look at our sentiment analysis, there's two parts to be considered. First of all, is the scale of the work that we do and the amount of data we're able to ingest, thanks to improvement in our processes. And the other part is on the model side. we tend to prefer to use traditional NLP models, like Google's BERT or Facebook Roberta's, as in production they are more efficient to run and can actually be fine-tuned and controlled a little bit better for specific use cases. It's quite well known that there's a lack of clarity on the sources behind LLMs and how they are trained. And an excessive reliance on such tools might lead to different type of biases, like anchoring and self-confirmation, which we may be actually unaware of during the development if we don't pay close attention. So while I sound somewhat negative on LLMs, the truth is that they are actually very, very helpful in speeding up the model development cycle. Their big advantage is that they can produce decent results with almost minimal prompt engineering. And as such, they can be very helpful for speeding up the annotation, the data labeling, or actually serve as an early benchmark for other models to see what are the deficiencies of our models and how to go about improving them. But it's always the case that a human should review this data, and to add to that, cost and speed remains a key constraint for large scale LLM use, as well as its explainability, as we were just discussing. And it's hard to find the right balance between how quickly you adopt the technology versus how careful you are into it, especially in this rapidly evolving landscape where new models and new opportunities come up almost every day.

Eloise Goulder: So when it comes to these new models, these new opportunities, where are you planning to focus?

Luca Rainero: I think we could talk a lot about this space. on the one side, you have the coverage universe. can you bring in more data, more sentiment sources? Another area of the work is how do we combine the analytics that we get from our sentiment data with more traditional sources, like market price or flows? I think that major gains can be done if we are able to bring together these different sources and build a more coherent picture about what they are telling us.

Eloise Goulder: Well, Edwina, could we turn to you to close and could we turn back to the client lens? how do you see this space evolving from here?

Edwina Lowe: I'd start by saying that the client feedback loop really is critical for us as a team in terms of determining our pipeline and where we dedicate time and resource. I think it's fair to say that the increasing demand for data, in particular for textual data that can be fed into these models, isn't going anywhere. And in fact, it's continuing to grow. From our perspective, there is some debate about whether to prioritize breadth, and by that I mean expanding the underlying sources that we analyze, versus depth, i.e. getting more granular in terms of the work that we are already doing. So, one area that we could explore is applying more nuance we could look at tone in delivery as well as the underlying language that's being used. Another area could be distinguishing between mid-frequency versus higher frequency intraday signals. So, there's certainly a lot for us to consider. But Luca, do you have anything to add to that?

Luca Rainero: I think there's one thing worth mentioning is that we are actively involved in many conversations with clients on the use of agentic AI and how we can build LLMs as part of a longer, more complex process to achieve the results they want. So, that's definitely a theme that's going to expand, I believe, in the next few years.

Eloise Goulder: Yes. And equally, the topic of where LLMs are useful and where they're not. To the extent that you run an agentic process, you probably want some toolkits not to be AI, but based on fundamental statistical pre-programmed relationships and the connectors to be based on LLMs or AI to help you guide between those processes.

Luca Rainero: That's absolutely true. And the balance between power of the tool versus control and explainability is something that is very much in flux at the moment.

Eloise Goulder: Well, This is such a fascinating space with some really sharp divergences between investor types, between what the retail investor has been doing with such optimism and what the institutional investor has been doing with arguably more caution. I know we'll continue to monitor this space closely. So, Edwina, Luca, thank you so much for taking the time to sit down and discuss this all today.

Edwina Lowe: It's been a pleasure, Eloise. Thank you so much.

Luca Rainero: Thanks, Eloise, for the amazing opportunity.

Eloise Goulder: Thank you also to our listeners for dialing into this bi-weekly podcast from our group. it’s worth noting that many of the datasets that we have mentioned today including the retail social sentiment are available to our institutional clients. If you have questions, we'd love to hear from you. So, please do go to our website at jpmorgan.com/market-data-intelligence and 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, 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 JPMorgan Chase & Company. All rights reserved.

[End of episode]