This is Ken Froot, and this is a presentation on investor and market behaviours focusing in on behavioural, what we call behavioural benchmarking. It's contributed by the entire team of behavioural indicators at SSA. What I'd like to do first is basically just say a couple of words, if you would advance the slides. There are a number of things that we're doing in IBRT. We are advancing the ball in a number of ways. Please back one slide to the second black slide. Slide number two. Thank you. What we're doing there is we're advancing a number of things. One thing that we're advancing is called trade typology. I won't be speaking about trade typology today. Trade typology is our ability now to be able to look in and see at the point of a trade whether a portfolio is holding long or underweight or overweight, a security, to understand essentially whether or not we should call it a conviction trade, an unwinding trade. That's an exhaustive way of splitting up trades, or whether we should think of it basically in addition as a trade coming from an overweight portfolio or an underweight portfolio. This ability, basically for us to divide upflows based on the positions that are inherent at the funds that are making them gives us tremendous additional insight into what's happening in flows. I won't be speaking about this, but Alex Cheema-Fox will. I urge you to go and enjoy his presentation. Today, I'll be speaking about active positions and risk assessment, and so, on the slide that's titled Active Positions and Risk Assessment, if you would show that. If you would show that slide; Active Positions and Risk Assessment. Very good. The objective, of course, we all know that active positions and using active positions to assess risk is a very common thing that we do across the industry, and the procedure there is we often define what we might call a neutral position benchmark, a benchmark that reflects what our tendencies would be, if we didn't have any particular ideas at the time. Then basically, we measure our active positions by looking at the difference between our actual and benchmark positions, and the only math I'm going to show you here basically is just the computation of active risk based on those active positions. That's really the workhorse essentially of a lot of the way we assess and budget risk and the like, and you're all familiar with this list of the ways in which we can develop and refine strategy on that basis. We can characterise our biggest bets, we can classify risk in different ways, we can aggregate accordingly and of course, at State Street, what we do is we take active positions in order to understand holdings rather than just the holdings themselves, because after all, clients have very, very different benchmarks. If you would advance the slide to the next, what is old in this discussion, what we've already done for the benchmarks, is to create benchmarks based on many, many positions. Billions really of positions over time across tens of thousands of funds. Anywhere from a few to thousands of positions in an individual portfolio. We built a large hedonic model with over 100 factor attributes, attributes that point toward the market, that point toward the fund, that point toward the instrument or the asset. It's a little like a Zillow hedonic model, where Zillow of course comes up with a price of your house depending on its attributes, like the number of bedrooms and bathrooms and the like. We've been doing that for some time in order to establish analytically what the underlying benchmarks are fund-by-fund, so we have tens of thousands of not only fund data, but tens of thousands of benchmarks for those funds. Then we can aggregate those up, in a secure, of course, all this being done. We can aggregate that up into large aggregates that reveal nothing, of course, about the client, individual client portfolios, but tell us a tremendous amount about the bulk of professional managers and what they're thinking and doing. That's what's old. Now if you advance the slide again. What's new today. Today, what's new is that we're going to focus a little bit on the benchmark, and the benchmark turns out to be very, very important. In other words, in just computing active holdings in the past, it mattered a lot whether or not we were adding a bunch of Russian clients. If we added a bunch of Russian clients at State Street, what would happen, of course, is those Russian clients would likely be holding a lot of Russian securities. If that was simply reflected in holdings and all we looked at was holdings, we might see a big bump up in Russian holdings, but in fact, what happens is that those clients have a benchmark that is Russian-centric and as a result, by measure of their benchmark, it wouldn't necessarily be overweight Russian securities at all. We get a tremendous amount of variation in benchmarks because of the individual attributes that affect funds. Our revelation here really was this model environment that we have built, this large hedonic infrastructure that we can use creates an unmatched set of capabilities that will allow us to generate benchmarks from client-chosen reference groupings, so that we can interact one client at a time with a client in a secure environment. It'll allow that client to choose in a sense a reference group by managing the attributes that are involved. Are we going to look at Russian-centric portfolios? Are we going to look at tech-centric portfolios? Do we want portfolios with a lot of momentum? Portfolios that are large in size? Et cetera, and we can alter the reference group based on what is interesting most to the client, and then allow the client further insight into how our benchmark would apply in those circumstances. Really, what we're building are the capability to provide managers with proprietary totally anonymised customised benchmarks, populated by thousands of funds based on client-chosen attributes. It's a completely different client experience here because instead of pushing out to everyone, the same indicator. What we're doing here is we're offering a set of interactive capabilities, one client at a time in a secure environment, where the client can pass certain information to us or pass certain information to a neutral server. We can pass information to that neutral server and the neutral server could compute essentially the things of interest to the client. I'm going to take you through this. That's the point of the talk, to give you an idea of what we're doing here and hopefully you can ask some questions and we'll introduce you to the notion really of this behavioural benchmarking that we're quite excited about, because in this industry, investment management, we actually have an astonishingly narrow ability to really see what others of interest are doing or play with others of interest. People who are buying tech stocks recently, what are the things that they tend to be buying and what are the things they tend to be selling? That's a very large group of clients. We can help characterise that basically for a fund and show a fund basically where they stand relative to that group. There's a tremendous amount of capability there for one to understand the environment one is facing, what systematic things these different reference groups are doing, and how my fund compares in fact in terms of the systematic things I'm doing with that. If you advance it to the next slide, where I have some examples here. We can overweight a stock, how overweight am I of a given stock, say like, Apple compared to portfolios that are long, long Apple, compared to portfolios that have recently bought or been buying Apple, funds that are actually, maybe I'm not overweight Apple, but I want to get an idea of the things that people actually are buying in the tech space. I can learn about that. I can also ask questions around whether funds that are overweight energy, what does their energy portfolio look like? What are they overweight and underweight in the energy space? Given that they tend to energy-focused funds. I can get a best practice view of what active positions look like in the energy area. Lots of ways to essentially cut this and lots of variation in benchmarks across funds. If you advance the slides again to the black one that says, behavioural benchmarking against customised reference groups. I'm going to take you in now to a bit of an analysis of how we do this. It's very bird's eye view, of course, of the analytics, but I think it'll give you a good sense of how protective we are of individual client information, how careful we are to be confident that we don't reveal anything about any given client's portfolio in this activity when we show others what these reference group benchmarks look like. Here are the five basic steps in the next slide which says, deriving a fund's expected position. If you advance the slides to that. You can see basically, there are five steps here. The first is SSA's universe of anonymous information. We have billions of data points by position, by security across time, in many, many funds. Next, we basically have built ourselves a model. That model has something like 100 different factor attributes. Factor attributes that point back to the fund, that point back to the security, and the like. Using that model, a little like Zillow, we estimate factor attribute sensitivity. We estimate, if you like, a set of betas on a set of factors, in order to understand what the average tendency to hold a security would be across the entire universe, say, or across any given universe of funds. That's basically what we do within our servers. Then on the client side, our steps four and five, next, the client can choose attributes that define an anonymous reference group. Those attributes could be, well, I'd like to see funds that are heavy on momentum, I'd like to see funds that are long Tesla, I'd like to see funds that are large in size and relatively heavy in tech. Provided those designations have a sufficiently well-populated universe of funds that come out of our many tens of thousands of funds, so that no one is exposed, we're very careful about that. We've always been careful about that and this is much more, this works very well, a lot like when we add upflows across these tens of thousands of portfolios. We're always very careful not to publish anything that would have less than 50 or 100 different voting portfolios, in order to make sure that there's deep averaging going on. The last step is essentially that designer benchmarks can be computed. We don't even need to see the exact designer benchmarks. This could be done at a secure server away from us. Where basically, we combine our sensitivities with the attributes in our model and the ones that the client is interested in, in a reference group, in order to generate a client reference group that's designer-made. If you advance the slide again. I'm going to take you through it in a little more, a slightly more visual version of this with a little better sense of the computations. The first step is basically SSA's universe of anonymous information that you can see. Advance the slide and you can see basically number two is we identify all these attributes, roughly 100 attributes that are functions of the fund and the like. Advance the slide again and we get to number three, which is basically based on that model of attributes and the data we have from these billions of data points, attribute sensitivities. There, you see basically a list of betas, if you like, sensitivities for each one of these funds mocked up. The next slide shows, now the client contributions, so the client is going to provide attribute values or a range of attribute weights that matter most to the client to pick up and include in the reference group. Of course, we have the capability here not to do just what a Morningstar or a Lipper would do by, we either include or exclude a given fund. Here, basically we can use weightings, so a given fund might have a lot of the momentum characteristics you want, even if it doesn't hold a lot of the technology. Nevertheless, it should get some weight in the overall benchmarking reference point. Basically, then we have the last step where we take again, the client-specific attributes. Now we multiply them again by our sensitivities in order to come up with a client-generated designer benchmark. If you advance the slide again, you can see this big giant hopper. We have a factory in SSA, of course, that's doing proprietary analysis on all these billions of data points. We then throw into this hopper, essentially the sensitivities we build, we put them together with client attributes and the reference group the client is interested in, and out will pop, here for an individual security but for any number of securities you'd like, here pops out what the client would see, which is not all the things that go into this calculation. It's sort of like the vodka that comes out after you distil the potatoes. You can't get the potatoes back out, so you can't see any of the data points that went into this, but you can see that the reference group benchmark for Apple here is 1.7 per cent, 1.87 per cent. That's actually instructive because, believe it or not, if you advance the slides, what makes our reference benchmarking so different than normal is that we're actually out to capture what expected holdings are or average holdings or normal holdings, given the attributes. These differ quite a bit from the simplest ways of thinking about things, say market cap which is in the upper left of the slide you're looking at. Where you can see basically the data cloud is very tilted. That mid-size stocks of which there are reasonably numbered, tend to get a reasonable amount of overweights across funds and that comes at the expense of underweighting very large cap stocks, like Apple for example. Our benchmarking procedure is the one on the right where it eliminates these forms of bias that you would get if you were to use like, a cap-weighted benchmark or an equally-weighted benchmark or other things to analytically generate simple forms of portfolios. If you advance the slides again. Now we're going to do this and work through this activity basically for a specific benchmark for a specific portfolio. If so, you advance it to our case study slide. You can see what we're going to do here is have a case study of the ARKK fund, innovation fund, Cathie Wood's innovation fund that everyone is aware of surely. One reason that we chose this as an example is that Cathie has structured this as an ETF, so we all have access to positions and the like, so we can do this in a way that is not informationally invasive in any sense. Nevertheless, give you a sense, if you have a fund that come with proprietary positions and want proprietary information, how we would go about generating it. You can see here basically that we've listed the largest ARK positions and the smallest as of September 30th. No surprise, Tesla is a very big weight. It's almost 12 per cent of the portfolio. If you measure that against the SNP where Tesla has a two per cent-ish market cap, it's almost ten per cent overweight there. In terms of the smallest positions, there are a lot of zeros, of course, in SNP firms. The biggest underweight is going to of course be Apple which is the largest market cap at 6.05 per cent. ARKK is quite underweight, of course, Apple. A lot like the picture I just showed you, large funds or any fund really tend to be relatively underweight on average, these largest cap stocks, and the FANG stocks below all tend to be very underweight in ARKK. If you advance the slides again, you can see in a sense what we've already shown you, which is the ability to look at this versus the full universe reference benchmark. We're going to just put in all the funds that are available and use this in the widest sense against all professional managers regardless of their mandates. You can see here, what's interesting, if you look down to where Apple is at the bottom, you can see that 1.87 per cent is the full universe reference benchmark. That's our smaller number than the 6.05 market cap weighted fraction that would be appropriate for Apple. We have a big underweight there essentially for Cathie Wood from this point of view. On the other end of this, for the largest positions, we have Tesla there with the 12 per cent position. It's only two per cent in the SNP. The full universe has it about a third of that actually, about 70 basis points, so Cathie Wood is even longer against that reference benchmark. We can do much more than this. That's just the full universe. Essentially, what we can do is create a secure environment where we start with our universe and we allow the client to choose the importance to them of all kinds of different attributes. Maybe it's the fund weight in Tesla, maybe it's the beta, maybe it's the active share, maybe it's momentum. Any number of factors essentially that, maybe end up being long energy or maybe it's having purchased recently a lot of technology shares. Any number of factors basically that are important, we can allow the client to weigh in and sort of say, here's how I want to think about my reference group and here are the things that are most important to me. What we're going to do is we're going to take the original universe and we're going to re-weight it, so that coloured ball is essentially now the re-weighted version, and we can compute designer benchmarks from that. If you would advance the slides again. Unfortunately, we were behind there, so this is the data. These are the little weighting slides that I can apply factor-by-factor, again for many different factors, I can apply. I get a re-weighted universe and from that universe, I can actually create new benchmarks that are appropriate for the designer wishes of the client. Advance the slides please. You can see just one quick example for three different securities; Tesla, Teladoc Health, and Apple, for how this would appear for an SNP 500 as a benchmark, for a full universe, our full universe benchmark, for a growth-tilted benchmark of ours, and for a Tesla-focused benchmark of ours. You can see how these vary. For Tesla, of course, we get a lot of sensitivity. The overweights are massive against the full universe benchmark, as we already said, over ten per cent here. Once you zone in on Tesla-focused funds, the overweight is much smaller. If you go over to the right basically, you can see there's a lot of variation for Apple too. It specifically comes from the full universe, a lot like I referred to, that our full universe tends to be quite underweight, the very largest cap stocks. As a result, Cathie Wood's underweight in Apple doesn't appear as extreme, versus these other kinds of benchmarks which of course, growth-tilted, Tesla-focused, SNP 500, all would tend to generate bigger positions in Apple than the average fund in our universe. Advance again, and you can see we can array this across securities, all the securities in a benchmark environment for a given fund. We can show you here between 12 and one o'clock, we've arrayed this big overweights essentially, and you can see for different benchmarks or different benchmark approaches, what this dial looks like, and of course, because overweights have to be matched with underweights. We can see that because there is Mount Everest here between 12 and 1 o'clock, there's also a trench in the Pacific here around between 6 and 7 o'clock for the FANG stocks and the like. These things basically, we can look at the world this way in order to get a bird's eye view of what's happening across a fund or its underweights and overweights against a given reference group. Finally, if we go to, how does my portfolio compare against other real-money institutions? We're going to show another way that we can present this kind of analysis. If you page forward. Here, we're going to show you different fund distributions, so we can lay out essentially where a given fund, here, the ARKK fund again, Cathie Wood's fund, lies in these different spaces that we can come up. Volatility versus beta, book to market versus dividend yield. Two versions, two spaces where high-beta funds tend to be higher volatility or high dividend yield funds tend to be higher book to market. We also can show you lots of different versions. We just have five here, but all the way on the right, you can see active share versus concentration. This is a grouping that allows us to look at active risk versus concentration. On the next slide, we map quite well what's happened to Cathie Wood's fund over time. You can see as it's gotten bigger, it's become more concentrated on the X axis. It's moved to become a more concentrated version of what it used to do given its flexibility in the past. Let me stop there and just conclude, on the last slide. You can see basically client-chosen reference group benchmarks can reveal a lot about what's going on in one's portfolio and what one is doing differently from others, and that's the kind of insights that we want to allow clients to be able to gain.
Great. Well, thank you, Ken. I think this will certainly be a really valuable analysis for our clients. I'll open it up now for questions from the audience, and again, if you'd like to ask a question, please use the chat function on the side of your screen. Great. Hi, Ken. Happy we got the technology sorted out. We have time for a couple of questions here. One question, so ESG is certainly increasingly important to investors, so could you incorporate ESG factors or ESG scores into your reference group framework?
Absolutely. ESG is one of many factors, but certainly a factor that's at top of people's minds to say, let me score and understand the benchmarking that occurs basically for funds that are highly ESG sensitive, that weigh heavily into ESG stocks. That's something we can offer to people, so they can gain an insight specifically into that factor. Again, that's one very important factor, but still only one factor and one can find one wants to add all kinds of designer attributes. For example, many ESG stocks are small stocks, so if you're a very large fund, you might be interested in benchmarking also against larger funds that really need to hold larger cap ESG positions, and that may make a difference versus the average fund.
Great. In terms of a typical use case here for investors or for clients, do you think of this more as about a risk-management tool or more about coming up with alpha-generating ideas? How do you think about clients, how they would use this?
Very good. I think it's very much both. I think it's a way for clients to have a broad view periscope, so that they can come up above water and look around and see what others are doing, and what others are doing might say, they're moving back to benchmark, they're cutting risk, or it might be that they're specifically, they seem to be buying these stocks in the technology space and underweighting those. It could be any number of insights, insights that allow one to better manage alpha, but also better manage risk. Really, it's better fund management through awareness of the environment one lives in.
I guess as a follow on to the question, how and when do you think this type of analysis will be available for clients?
Well, we're generating already some crude versions for clients, for individual clients to have a look at, and so we're very interested, by the way, in just getting your reactions to this set of capabilities. We've been excessively careful to make sure we protect client confidentiality in all of this activity, and we're very eager to show people what we're able to do, and so we'd love to have your feedback. If you are interested in having a look at this, even at still this rudimentary level that we're rolling it out, we'd be very keen to talk with you.
Great. Well, thank you so much again. Again, really excited to engage with clients on this. Unfortunately, we didn't get to all the questions that are here and we'll have to move on, but we'll try our best to get back to some of the questions separately that we didn't get to here. Thank you so much, Ken. This sounds exciting.