Sentiment ETF 'BUZZ' Takes Off in 2025

Bloomberg

Watch Sentiment ETF 'BUZZ' Takes Off in 2025 - Bloomberg on Dailymotion

Transcript

00:00When you come across an ETF like this, it's so fun to talk about the methodology because you know

00:05better than most that sentiment is fickle. It changes on a dime. So with that in mind,

00:10what does your methodology look like when it comes to staying on top of these trends? How

00:15frequently do you rebalance and things along those lines? Yeah, that's a great point. Sentiment

00:19can change quickly. And when you think about sentiment, much in the way you think about

00:22volatility, even it depends on the timeframe that you're trying to measure it. What the index does

00:27is it rebalances once a month. So it thinks about a little bit longer term sentiment. This

00:31isn't a high frequency, high turnover, capture the next tick or movement in the stock based on

00:36sentiment, but think more about a sentiment trend. And so what we're doing is looking across

00:41online platforms where people are talking about stocks from an investment perspective,

00:46measuring that conversation for sentiment, doing that over a month, and then aggregating that

00:51sentiment into the buzz holdings every monthly rebalance. So you're basically buying companies

00:56that are in the zeitgeist, that have a lot of positive buzz around it. Does it matter? Does

01:01your technology make the distinction between companies that are kind of on the way up versus

01:05those that like have peaked and perhaps are, you know, it's really natural language processing.

01:09So it's what people are saying about the stock from an investment perspective. It's not overlaying

01:14anything more than that relative to are people positive about the stock and its prospects. And so it's not

01:22trying to overlay any kind of thematic or any kind of directional signals or stock trading history.

01:28It's simply based on the collective conviction of the online community.

01:33When I look at the holdings here, obviously, these are buzzed about stocks. How do you

01:36account for just the volume? Like everybody's talking about Nvidia, it's the most traded.

01:41How do you get into some of the smaller names that might be positive? That way, you're not just

01:45stuck with like the cues.

01:47Yeah, sure. So there's some thresholds around the index construction. So we should talk about that.

01:50So the index looks at all stocks that have a market cap of 5 billion or higher.

01:55There's about 1,500 or so of those stocks in the market today.

01:58But it's not every one of those stocks. We need to understand and we really want to see

02:02that within those stocks, which ones are being talked about most frequently,

02:06most consistently online. You can imagine if the four of us are sitting around talking about

02:10a stock, but no one else is. The insight from that conversation isn't all that relevant.

02:14But if you can see that there are stocks that are consistently talked about by a large

02:19audience, then you want to know what that average or the overall aggregate sentiment is

02:23from that conversation. So we we set up a quarterly eligible universe. So from that

02:281,500 stocks, we'll distill it down to somewhere between two and 400 every quarter. And that's

02:34the conversation that we drill into each month. We rank that conversation from most bullish to least

02:40and the 75 make the index. And so when you rebalance every month, does it start out as equal

02:45weight between these names? Or how do you decide on how to actually weight the different stocks in

02:50the portfolio? A great question. So we weight it by sentiment. And that's important because we think

02:54of sentiment when we started this 10 years ago. So the index has been live now for a 10 year period.

02:59We really thought of sentiment as a potential factor. So if you think of explaining security returns,

03:04we all know that there's growth and value momentum and size and all of these other things. And

03:09everything else was alpha. And in that alpha bucket was sentiment. And our thesis back then in 2015

03:14was, if you can measure sentiment, you can actually strip out of that alpha bucket, a factor called

03:20sentiment, and it can be its own unique thing. And so when we weight it, we don't want to market cap

03:25weight this index to let certain stocks drive it like is in the case of the S&P 500, for example. So we cap

03:32each stock's weight at each monthly rebalance at a max of 3% to keep that diversification there.

03:38If a stock had a greater than 3% weight because it had a higher sentiment aggregate view, then we

03:43would just distribute the excess equally amongst the remaining constituents. You're tracking positive

03:47sentiment. Are you looking to attract negative sentiment? I mean, could you use this as a way to

03:51short certain companies as well? You could for sure. So the NLP measures sentiment based on what people

03:56are posting and talking about. That could be positive. It could be neutral. It could be negative.

03:59What buzz is, is a representation of aggregate large cap positive sentiment. Of course, we have

04:05the data set of negative sentiment. We actually use this data set and some options related strategies

04:10within our hedge fund business at Periscope Capital. So there's lots of different use cases.

04:15Buzz to us is the ability to give back to the community, the aggregate sentiment factor in its

04:21simplest beta form. Does the computer ever pick up like sarcasm? Like, oh, this is the greatest stock.

04:27Yeah. And do you ever get false positives? It was a huge challenge when we started 10 years ago,

04:31because we had to build our own NLP to do this, right? So we're Toronto based. We had a

04:36collaboration with U of T. We brought in, we trained our models. That was a really difficult exercise 10

04:42years ago. And then training them based on examples of sarcasm. And then all of a sudden, emojis come

04:47into play. You know, with emojis in finance are very different than conversational emojis. You have rocket ships,

04:53you have all these other things. So continually refining and training and updating the model in

04:57those early years of buzz. Today, we have LLMs and you can take any post, put it in an LLM

05:02and the accuracy score of its sentiment, sarcasm, emoji, anything will be very high. So it's really

05:08table stakes, the raw data today. It's just the longevity and the understanding of what it is

05:13you're looking at.

Category

Transcript

Be the first to comment

Recommended