Citigroup Inc. (C) Earnings Call Transcript & Summary

Hello. Good morning, good afternoon and good evening to everyone joining us today for our webinar. Very excited to kick this off. We will be discussing our favorite topic when it comes to derivatives, unveiling quantitative and systematic strategies. My name is Mila Kuznetsov, and I'll be moderating the session today. I lead business development for our derivatives data and valuation services within Market Intelligence, and we serve clients from anywhere from banks, funds, insurance companies, pensions, you name it for any of their derivative strategies, and I'm very excited to talk to you all today. Before we kick this off, I did have a few housekeeping things that I wanted to touch on. Very first: I just want to thank everyone for their time. If you do have a question, we'd love to keep this as interactive as possible. Of course, we know we're not in a big conference center. So please feel free to use that Q&A widget. It should be over on your right-hand corner answer and we'll answer any of the questions that you might have either during the session or after we talk to you with any answers that we might have, but please ask questions as we're speaking, and we'll try to answer all of your questions during the webinar. In terms of additional resources. We do provide our thought leadership and any other types of fact sheets, documents in our Related Content widget. There will be valuable insights there for you and other resources. If you're joining us for the replay, then here, you'll also find that in that Related Content widget button. And finally, at the end of this webinar, there will be a short survey. We'd love to hear from you in terms of how helpful you found this a valuable use of your time, of course. And 2 final things I'll mention is that this webinar does provide close captioning in English. So please click on the icon in the Media Player to activate that so that you could see the close captioning in English, if that's what you prefer. And in terms of Market Intelligence, I did want to mention that Market Intelligence at S&P is separate from the Ratings division and the content that we're providing here today is for informational purposes only and does not constitute investment advice. And with that being said, I'll introduce the speakers that we have for you today. We have lots of diverse backgrounds and perspectives across quantitative investment strategies. First, Riccardo Borghi, who's the Vice President of Equity Trading, IMI Corporate & Investment Banking division at Intesa Sanpaolo. Then we have Lini Gao, who is Head of Commodity Index Distribution for UK and MENA at Citi; and Ankit Gheedia, who's the head of QIS Research at BBVA. I did want to mention again that the opinions expressed in this webinar are solely those of the speakers and do not represent in any way those of the present or past employers for any of the speakers that you're hearing from today. Before we head into the questions, we did want to hear from the audience as well, and we have one poll question that we like to kick things off with. We'd love to hear from you. What are the main systematic and quantitative strategies that you are deploying in the audience? And while you're answering that question, we'll kick it off to our very first question. And perhaps I'll start with, Lini, with you. Given the current landscape and the volatility that we've seen in April and continuing to see today, of course, quantitative investment strategies have seen a lot of exciting shifts. But what are the major trends that you're really looking at today?

Yes, absolutely. Before we get into the trends that we're seeing, one thing that might be helpful to just break down QIS and particularly the QIS offering that banks offer just so the audience can better understand. Really, bank QIS can be distilled into 3 main value propositions: One, the research and development in "bank IP" that goes into creating an effective rules-based, transparent strategy which achieves a certain investment outcome, whether it's pure alpha generation or positive convexity with minimal negative carry to give you a couple of examples; two, the technological and operational scale as well as the increased access to liquidity in certain markets that banks benefit from in order to execute such a strategy on behalf of clients; and three, the ability for banks to help deliver these indices in an investment format that might be better suited for the different legal and regulatory frameworks that our clients operate under globally. So in the Europe region, specifically, for example, we work with [indiscernible] clients on commodity solutions because of the stricter guidelines imposed upon UCITS funds with respect to commodities. So there are range of areas that a client might lean on a bank's QIS desk for given those 3 main value propositions. One thing that we're very excited about is the growing adoption of QIS by more institutional client types as they start to better understand the benefits and use cases of the product and see it as complementary to their existing portfolios. So more clients are starting to see the benefits of QIS for what they are instead of writing off the product as a "black box", which has historically been this stigma around the product. So in terms of trends, given the recent market environment, we've unsurprisingly seen more defensive-type strategies become fashionable, whether it's a strategy which has historically seen more positive performance in sudden growth shock scenarios, such as some of our macro defensive strategies or perhaps in more sustained periods of stagflation, if that's the concern, such as commodity curve carry or momentum. So broadly speaking, as the interest grows and invested AUM in QIS expands across our client base, on the sell side, we are also very focused on expanding our offering across new strategy types and new markets in response to the themes and new demand that we're seeing.

Yes. Also talking about trend, I think like year 2025, we have obviously been in a very challenging and interesting environment to say the least, in an unprecedented time. We saw changing German fiscal policy, trade dynamics and transatlantic relationship. If you take a step back, essentially, what we have -- what all of this result in is, in a way, I would like to characterize as a short termism where day-to-day policy headline is driving a lot of market shift and market volatility. So you see that in terms of many of the risk derivative parameters. Obviously, we know that VIX has moved to the highest level that we have observed without a recession. And given that we have lack of visibility, the trend strategies have obviously underperformed, some of the operating strategies hitting a 10% correction since the start of the year. As Lini mentioned, obviously, defensive strategies have worked well. But if you're looking at your portfolio now and you didn't have defensive strategies in place at the start of the year, and entering right now new vol position, new long vol position becomes really challenging, especially if you believe that vol may reset lower going forward. But also when we talk about defensive strategies, obviously, in the first few weeks, first few days of April, we saw a sharp price in vol. Things also came back very quickly. Like S&P for the month of April ended flat, down less than 1%. So you really needed to monetize your hedges at the right time, which obviously is very challenging in the current market environment. And obviously, when we talk about portfolio construction and just remain in general, like traditional asset class correlations have broken down. It's very rare to see equities and bond selling off and dollar selling off at the same time that U.S. almost acting like an emerging market, which is not something that we have been used to see in most of our careers. Gold, obviously, has been a place to hide. So if you have an asset allocator, a strategy with gold in your portfolio, then obviously, that has done well. But ultimately, like we are in a very challenging environment. But every crisis, in our view, creates opportunity. And this time, that's no different. So we published our latest Risk Premia Outlook last week where our view is like if we're taking kind of a bold call non-consensus view, like we do believe that U.S. will avoid a recession. We do believe that things and volatility will come down going forward. And if you take those assumptions into account then maybe initiating new tariff strategies could be compelling. Because if you think about tariff strategy -- like right now, the rate differential between some of the countries are still quite wide. Some economies have started in their rate cutting cycle much earlier, while some countries are still hiking rates. This is creates opportunity in our view. I think that, that could be one place where investors could find good returns in the next few months.

Yes, I agree with both Lini and Ankit actually. Now that I moved on the buy side, I see a lot of use of equity vol strategies, both to generate carry like short range swap and short put options, but also to create a tail hedge for long-only portfolios with a long gamma profile, but trying to reduce basically the cost of carry so the cost of the insurance that you're paying. I think looking at more of a medium-term trend, I think that the demand -- the additional demand for these strategies and especially for the defensive one, started out in 2022. As Ankit was mentioning, it was a period when transforming strategies were not working well because of the breakdown in the correlation across assets. And so that's, I think, when clients have to look at other type of defensive strategies. I think the offering by QIS is very large. Now as I mentioned being on the buy side now, I see the whole spectrum is very large and diverse. And we have seen some large market shocks, both last year in August 5 and around the liberation day this year. So these shocks are really putting the strategies to the test. I think another recent development of the last couple of years, so in QIS that I've seen and I think it's interesting is the move to intraday strategies, especially on single stocks. For example, you can augment like intraday momentum strategy with single stocks. You can -- instead of buying the index essentially, you create a portfolio with an additional signal that it's boosting a bit the rationale behind intraday momentum. This is something where clients might be happy to externalize all the operational risk and also the execution behind intraday strategy, which are more like heavy from an operational point of view.

Thank you so much, everyone, for those insights. I definitely agree that we've also seen a rise in intraday volatility demand and also, of course, the rise of 0DTE options on our side. I thought maybe it would be good to take a look into what our audience is saying in terms of what systematic strategies and quantitative strategies they're deploying. It looks like the winner is trend following followed by momentum carry, of course, value. And then I would say the lagger seems to be a risk -- risk premia dispersion correlation, and we'll find out what the other is. That's very interesting for me. I guess to take a deeper look into some of the defensive strategies that you mentioned and, of course, the more traditional ones as well, would love to hear what are the specific instruments that you're looking at when you're thinking about your derivative strategies. And what are the specific values that are really governing your approach there and what you're exploring there when you're looking at those instruments. Perhaps, Ankit, we'll start with you.

So like carry strategies, that we are currently -- we think -- like even though it's the second best in the poll, but we think it makes a bit more compelling argument right now. And by nature, carry strategies are like picking up pennies in front of a steamroller, right? The key challenge is always to get -- avoid getting hit by a steamroller. While hindsight is always clear, having safeguards in place that has a bit more forward-looking bias, it's essential in delivering long-term returns while avoiding large drawdowns. And this is where use of some derivative not -- while also as a derivative inbuilt in the index is useful, but also as a timing signal to get in and out, becomes interesting. So for example, history suggests that the best time to be short volatility has not been vol is high, but it's when -- short volatility is all like best time to go long carry strategies, is when vol is high, at least from a return stand point of view. But the question is when the vol is high, are we entering into a recession? Like, for example, is the vol spike more like August last year, which was short-lived and maybe arguably April might prove to be the same, but we still don't know that. Or it's something more structural like COVID or 2008. So some of our strategies used a risk-based filter -- VIX, sorry, VIX-based filter to step out of the market when we see a 2-sigma move in volatility. We -- our goal here is to avoid short-term nights and focus only on sustained elevated volatility regime. And the idea behind this using VIX as a indicator is, obviously, this is one barometer that everybody looks at as a barometer of health of the economy, of health of market stress and market sentiment. And when you see a 2-sigma move in something like VIX, which is much more better followed than individual asset class volatility may be in rates or FX, it tends to gauge a good estimate of where the sentiment lies. And when you tend to see a big spike in these kind of indicators, you need to take a step back and start to assess where we are. So right now, this -- like in April, this indicator went obviously -- we had a big move in VIX and was suggesting to go a bit more risk off. But the goal is when -- if you avoid a recession and if you -- when -- the question is, is the vol spike an opportunity to enter new carry strategies? Or is it something that you need to be worried about of a more deeper recession, dip on prices coming upfront, well like something that we have -- that we tend to see only once a decade. And if it comes back, if the vol comes back, normalizes, then traditionally, that's the best time to add risk in your portfolio. And we use some of these indicators to help us gauge that sentiment. Just for example, our indicator saved around -- saved investors roughly 10%, both in 2008, in 2020 in terms of drawdown, which I think is quite valuable. So either you go in and out of QIS strategies, which is very difficult because hopefully -- these strategies in nature are built for long-term allocation. And you don't want to be trading it in the same way that you trade S&P futures. So having some of these safeguards in place, I think is quite powerful.

I would agree on the point about carry strategies. One of the longest-standing and most consistent premia in commodity markets over the past few decades has been curve carry, i.e., selling calendar spreads on commodity curves to capture the role yield differential. And the simple rationale for the existence of this premia is that commodity curves tend to spend most of their time in a concave contango shape due to carry in commodities, i.e., storage and financing costs and commercial and commercial hedging from producers and consumers. This is a bit of an oversimplification and it isn't always the case as the strength of the premia can vary through time and across markets depending on external macro and micro factors, seasonality, market participant composition, et cetera. But structurally, we think it is still monetizable in our markets, looking at the historical backwardation levels of each commodity curve relative to its own history. In terms of implementation, curve carry strategies can range from more static versions. So an example for curve carry is always selling the front month versus the 6-month point on a commodity curve in equal notional on each leg as the most simplistic implementation. Or they can be more dynamic, i.e., using signals to determine the trading parameters, such as changing promotional exposure, the legs or the future contracts that are selected to adjust for seasonality and also be more reactive to changing market conditions, as Ankit was alluding to as well. So from our perspective, there is no one size fits all. For certain factors or for more passive investors, it may be justified to have a more dynamic version of a strategy. For other strategies where there may be too many unnecessary adjustments, that might end up being dilutive to long-term performance. Or for clients who prefer to have a more hands-on approach to controlling their own exposure, simpler or more naively constructed strategies potentially might make more sense.

Yes. So on this type topic, I always think about another trend that I've been seeing lately, which is the application of machine learning strategies to investing. So I come from academia, where I was working on single stock liquidity analysis and then I move to quant research always working on like alpha signals. And now being a portfolio manager, I still am bit biased towards like single stock market neutral strategies. It is something that I think it's adding a lot of diversification to portfolios, even for like long-only and trend followers. There are various types of strategies that I tested in the past and some of them also use macro variables to try to calibrate models, which is random forest or neural networks to try to understand what kind of environment we are and trying to blend the predictive power of stock-specific characteristics with the one of macro variables. Obviously, these strategies are difficult to calibrate. They attract a lot of attention because you have many parameters, hyperparameters, you can play around with. So the results are going to look amazing in sample. But then when you actually start to trade them, the story is a little bit different. I wrote a paper on this in 2019, which was still fairly early for this type of research, where we try to build a multifactor signal exactly, as I mentioned, using single-stock characteristics around like 200, which are your traditional momentum, value quality of different types, historical and estimated and also micro variables. So you can create a leaner or linear combination between the variables, and you're already trying to see if a machine would be able to trade essentially. At the time and even right now, as I mentioned, there are many known published like working papers, and they show amazing results. But even though I worked a lot on this, I'm more on the skeptic side. So I think that once you factor in all the constraints that you need to use for us to create realistic portfolios, like turnover constraints, max weight or capacity assumptions, at the end, you find returns on distributions that are not that different from a well-designed linear model. So still like evolutions of like the simple Fama-French, but like well-designed linear models out of sample that I could use as benchmark were working in a similar way. So the question is, is it that not worth it, this machine learning? I don't think so because what was really impressive was the little overlapping between the portfolios. So even if you are reaching very similar results, very similar information ratios, if you have a portfolio that is completely different -- and we're talking about correlations as low as 20%, 30% and overlaps processionally of less than 40%, then you have another strategy that you can add capacity to. So there is still a lot of value in using this. But I've seen some nice offering also by QIS, and I think it's another trend that it's worth considering because it's really computational intense, data intense. And so probably externalizing this is not a bad idea either.

Riccardo, you mentioned model constraints and portfolio constraints. And of course, in today's environment, I feel like -- and even anyway, right, there's always a trade-off between, of course, return and risk. So when you're thinking about these strategies and when you're really thinking about today's markets, what kind of strategies are you thinking of that really help investors mitigate portfolio risk? And how are some of the ones that you've mentioned able to do that?

Yes. Single stock strategy in general are useful addition to long-only exposure and CTA-type exposure. Trend following is often seen as a quantitative strategy, and of course, it can be very complicated indeed. However, it's a diversification benefit. Sometimes defensiveness is based on the little or negative correlation among the components, as we mentioned earlier. This is risky because negative correlations change a lot and can be misleading. They can go to 1, exactly 1 -- we need the 1 then 2. Whereas instead by having a full cross section of stocks with like maybe 3,000 or more assets, you can construct ex-ante market neutral portfolios. Even without optimizers, they can easily build -- can be delta neutral, which -- except for low beta, is a good proxy for beta neutrality too. In terms of diversification, probably they are useful for 2 types of diversification. One might be thematic diversification. For example, in defensive baskets, you can add them to other defensive strategies that are option-based cross assets. You can add to quality and low beta . We've seen low beta lately has been impressive, right. Year-to-date, it has done amazingly well compared to many of the strategies coming off from like a market that was like highly, highly concentrated, where low beta stocks were disregarded by portfolio managers because of tracking narrow constraints and also because of the defensive nature of low beta. Quality is another very good one. I think that gives you convexity, but has quite a low carry cost compared to others. And finally, connected to what I mentioned earlier, because of the low overlapping between machine learning portfolios and more traditional portfolios, single stock strategies that are coming from nonlinear models can also create some alpha diversification, not only diversification from a risk perspective but also from alpha.

Yes. I would say that, generally speaking, the majority of commodity factors have orthogonal risk profiles to most of our clients' existing books given the idiosyncratic nature of the commodity markets. And so that explains some of the intuitive diversification benefits. Within many investor portfolios as well, carry-oriented QIS can also be used as an easy systematic alpha engine to then fund negative carrying or potentially worse carrying hedges elsewhere in the book. So taking a step back from a top-down portfolio perspective, the most traditional placement of QIS was within the alternatives or absolute return bucket of a multi-asset portfolio. And going back to the emergence initially of QIS in the early 2010s, these were originally constructed to provide comparable exposure to long-short hedge fund strategies at a lower cost with greater liquidity and more transparency. But through the years, banks have invested heavily in the underlying technological platform on which these strategies are built, 1 of the 3 value propositions that I mentioned earlier. And some of these strategies, like Riccardo mentioned earlier, can be -- involve multiple intraday fixing windows or particularly within QIS on the vol side of the offering, managing thousands of line items in a single index. So today, the platform has really been extended to a growing number of newer use cases for more client types, including, but not limited to, beta replacement, benchmark replication, bond replacement or option replacement, just to name a few examples we've seen in addition to the classic alpha generation that it was originally purposed for. So what this means is that the popularization of QIS really challenges clients and participants on their traditional ways of implementation for both systematic and tactical trading in favor of increased operational and cost efficiencies. And by extension, this has really opened the door for QIS to be considered in many more areas of an investor portfolio than just being the classic use case that they were originally intended for.

Yes. And I would also like to add, obviously, like we have seen dispersion between equity regions and sectors. And as Riccardo pointed out, between individual factors with growth in low beta, obviously, having very diverse fate this year. But also within the risk premia space, if you look at average pairwise correlation between different risk premia strategies that has been according to our calculation, it has made new lows this year, which is quite interesting in an environment of heightened volatility. You're starting to see different kind of risk premia. Obviously, in theory, they're supposed to be orthogonal source of return, but there's always some sort of inherent correlation, which has now fallen to new lows. And obviously, because the regime that we are in is not something that we have been in, in the past few years, which is driving different market reaction to the regime where these strategies have been optimized for, for example. And within each risk premia bucket, obviously, like when we talk about vol risk premia, the way you implement it, obviously, will have a big impact on your return and your outcome, for example, if you are selling puts, selling calls, delta hedged, not delta hedged, using 0 day option or longer dated expiry. And what we are starting to see, obviously, it's hard to say that one strategy is the best because obviously, some strategies work in some period of time. So having a portfolio which combines different ways to capture vol risk premia is probably a bit more prudent in the current market regime. And this -- I expect this to stay. So when we talk about this one source of orthogonal way of looking at different risk premias, but also within risk premia buckets that we implement is also driving diversification. So having a portfolio approach is obviously very useful in the current market regime.

100% agree with you there. I think that diversification in many different -- comes in many different shapes and forms. And maybe this will be a great way to transition to a question that we're getting from the audience. Diversification can also come in the form of now introducing alternatives into QIS. So the question here is, what are the main challenges and differences you see when applying similar strategies to, let's say, crypto coins like Bitcoin or Ether. Maybe Riccardo, you want to take that one.

Yes, sure. So we thought about this quite a lot. Obviously, it's a new asset class, right? So you would want to jump on to it, especially on trend following. So the couple of issues that we found was the time series data. Through how you start to have a bit of time series data on this. The problem is that the market has changed completely, right? I'm not an expert. I wouldn't sell myself as an expert in cryptos, but we have like a desk for guys that are pretty expert and they tell me that the structure of the market is pretty different. So that means that if I'm using pricing data, I might have a price, right? But what is the value of this thing I'm trying to predict? Is the price correct? It's a bit different compared to market on close and official exchange prices. And as of lately, yes, the market has probably become more efficient, then we might use data. But yes, for example, to calibrate models are not even too complicated even simple ones. You need like a good 10 years or so of data. Connected to this is not only if the price I see is actually a good benchmark or a good proxy for the value of the asset, but it's also if it's point in time. Is it actually a real number or was it restated? Point time data is expensive on equities. And so you'd probably want to buy some good, level data for cryptos as well. And the other thing is liquidity. Liquidity that you see as transaction cost, which is probably the clearest one, but the other one is price impact, right? What is your price impact? There are many models on price impact, decrement models. I mean on equities on single stocks, there's been like a lot of research on this. For crypto, I wouldn't really know exactly how to estimate what would be the pricing part and the effective transaction cost that, that would take. These are the main potential issues.

So I would also add, well, first, some quick background. I spent a couple of years leading the build-out of structured derivative crypto products at a firm called Galaxy Digital, working with institutional clients to get efficient access to those products. And my personal views on the issue when it comes to implementing systematic strategies on crypto, first, really agree with what Riccardo said. If you think about Bitcoin really being born in 2009 and option liquidity only becoming meaningful in the late 2010s, you neither have a long history of data to work with for back testing those strategies nor do you have a mature market structure to implement high conviction systematic strategies, especially in large size. The second thing I would say is that the liquidity for futures and options in crypto is very fragmented. The majority of volumes still go through crypto-specific exchanges, even though more traditional exchanges like CME, for example, have available moving products. So you can have very large discrepancies at times with valuation if you only rely on traditional exchanges, which is usually a case for more institutional traditional participants. With that being said, what is really interesting is that in the few months, the last few months that IBIT options have started trading -- IBIT is BlackRock's spot ETF, which was launched at the beginning of 2024. The incremental growth in overall Bitcoin option volumes has been driven by those IBIT options now on Cboe. And those volumes, although still being second to Deribit, which is the crypto specific exchange where most option liquidity is, has surpassed CME volumes quite surprisingly, which are options on the BTC futures, which have been listed since 2019. So in a world where option liquidity in crypto is increasingly moving toward Cboe, a traditional exchange, and continues to grow at a similar rate, I think it's very natural to expect that QIS could realistically be extended to include crypto over time. I don't want to turn this into a crypto podcast. We've already spent quite a few minutes on this topic. But happy to discuss if anyone wants to reach out after the call. The last thing I'll say is that if you look at the 1 month vol risk premium, so the difference between the implied to realized vol differential, in Bitcoin, it's currently around 9%. And it might surprise a lot of people on this call, but that is actually lower than the current vol premium in some normal commodities, like Chicago wheat and corn. I just had to say that in case my boss is on this call listening. Normal commodity vol risk premia.

Have to mention that one, of course. I do think that the crypto question brings about a very interesting point when it comes to thinking, that's why I asked, which is that you need the large history for back testing, that you need to be able to stress test any type of strategy that you're going to be looking on that needs not only like data but also the ability to execute after the ability to, of course, hedge. So this makes me interested in terms of what is your process when you're really thinking about any strategy that you're putting on when you're thinking about back testing, when you're thinking about stress testing. Ankit, I would love to hear from you.

Yes. So putting my research hat on, I strongly believe in using approaches that make intuitive sense and economic sense rather than just rely on back test. Because obviously, like flawless back test rarely hold up and we're looking at out-of-sample performance. I also believe in looking at where we are in the economy, what financial conditions are in deciding as a very important input in asset allocation and risk premia allocation. So in my current role, I place a significant weight on what these indicators are telling me to form market views. For instance, currently, our growth indicator is neutral while our financial condition index suggests that the environment is slightly tight. So combining the 2, so if you don't have recession, then -- and if the market is a bit too stressed, then that's a good buying opportunity. For example, if S&P test 4,800 again. So rather than looking at 1 single indicator, I think having a composite view gives you a much better toolkit on deciding your market views and where we are in the cycle. And second point I would like to highlight is like it's very hard to do market timing. It's much better or much easier and where I have -- think I believe I can add much more value is to take medium- to long-term view, which -- in terms of thinking more in terms of portfolio construction. When the vol -- S&P's vol realizing at 100, which was at the start of the -- of last month, obviously, because market timing is more a matter of luck rather than skill. But there are meaningful signals that we can use to help us guide in such kind of regime, particularly when we think about this premia QIS allocation, which tends to be much more stickier. You need to take a view of next 1 to 2 years. What do you think where we would be in that period of time rather than just 1 week or so. Just to give you an example, for example, FX carry strategy tend to produce the weakest forward return when the market volatility is low, as I said before. Because obviously, there's less lower risk premia to harvest, but then there are some indicators like rate differential. Because, as I said before, some economies are much further ahead in the rate cutting cycle, while some are still lagging rate. These kind of signals help us decide what makes a current risk -- a particular risk premia appealing. So what we have done is looked at all different risk premia, looked through different set of indicators and tried to identify the most relevant indicators, lead indicators for each strategy. So for example, carry, obviously, vol and interest rate, but for trend would be like how the vol -- how the asset class has been trending recently. And more defensive indexes, how cheap or expensive is to own volatility right now. So we try to use these forward-looking indicator in trying to identify the regime for each of our QIS strategies, which helps us differentiate between signal and noise. And I imagine this is a similar process that our clients are undertaking in deciding what they want to do, which makes it always creates a good opportunity to have interesting discussions.

Sure, I feel like we would all love to have that crystal ball in terms of market timing, but that's not quite possible yet. We'll see. Riccardo, what are your thoughts when it comes to back testing and stress testing given the quantitative and machine learning that you've been doing of course as well. I'd be interested to hear.

Yes. Obviously, we tried a lot. The problem of market timing, I agree with Ankit. It's very difficult, I would say, almost impossible because you don't have enough regimes to test it, right? So you might find something that looks amazing, but just because it's in sample, just because the indicator has a good explainability, that doesn't mean it has predictability or forecasting power. So yes, we tried in different ways. There are papers that tried to use like 100 years of data and they tried to use yes, leading indicators. But also there, the results are weak. On the factors, so on the equity styles, probably the only signal that seems to work is the momentum on the factor themselves, so the -- what's called factor momentum, which is something else that is sparking a bit of interest lately. That's because if you build market-neutral strategies that are actually neutral, not only to the market, but have a constrained exposure to other single stock strategies, then you might say that they are uncorrelated. And so you can capture their own trend, and essentially, you can augment your trend-following strategy. This is something that most of the attendees might be interested in. You could augment your trend-following strategies with new assets that are essentially long short factors. I've tried it and it looks -- it was good if they are well-built. And for well-built, I mean a very complicated process that starts with point-in-time data, especially on the analyst estimate. These are a bit obvious things, also corporate actions adjustments. But they're still expensive. They sound obvious, but they are expensive. Like there's no way to buy this kind of data in a cheap way nor is it to have it like very fast. There is still like some way of externalizing part of it, but still keeping the code to yourself. But the other most important thing is the risk model. If you want to create market-neutral strategies or even strategies that are neutral to certain risk factors, you need a risk model. And that's another big questions we need to ask yourself, more like conceptual decision you need to take. Decide if you want, like regional models that are calibrated for your region or global models that are calibrated for the whole universe. And they might be less granular, but better at generalizing. Then the thing I like about single-stock strategies and those that are alternative, even come from alternative data is that you can test them even without a back test. That's the beauty of it. If you build the signal, you can create a heat map where your value behind the heat map is the information coefficient, so the rank correlation between your signal and the forward returns. So you can lag the signal at different lag steps. You can predict returns at different future horizons. And then you're going to basically see very quickly if the signal works. And if it does work, when is the case. And because there's a formula that the information ratio is equal to the information coefficient times the square root of N, where n is your number of assets, you can basically just calculate the information coefficient before doing everything that a back test decides or the risks -- the back test know more or less if your signal is adding something or not. Yes, I think I'll stop here because -- on the back test. There are many more technical details that would probably bore everyone.

I can offer a little bit of a different perspective from the sell side. So within a bank QIS desk, our job is to create a platform of transparent, investable strategies. And within commodities, many of these factors, for example, curve carry, which we talked about earlier, are rooted in academia and well understood and studied and have been around for decades. In terms of the data sets that we use, we use historical data on exchange pricing, volumes, open interest, et cetera, for the universe of instruments referenced in our strategies. This is all, for the most part, very transparent and public data. And then when it comes to back testing, you typically back test these strategies for as far back as we have reliable data for. For most of the benchmark commodity features, this usually goes back a couple of decades. For options, it's a shorter time history and varies by market. It varies by when the product was listed by the exchange and when meaningful liquidity really became existent in those products. And so this isn't always a straightforward exercise, especially as you look to expand the offering and add newer products into the universe which can either be non-benchmark commodity underliers, such as lithium, cobalt, iron ore, et cetera; new points on the curve; newly listed option contracts, for example, weekly and daily options. So this can often require onboarding new data sources and providers, making sure that we do our diligence on those providers before we incorporate that into the platform. In terms of stress testing, we'll look at historical drawdowns, time for recovery. And we'll also simulate theoretical scenarios for performance by applying a set of shocks to futures and vol where applicable. We do within our framework capacity analysis on each new strategy that we launch to be cognizant of market impact and preserving the premia to protect our invested clients. So the obvious disclaimer here is that, as my colleagues alluded to earlier as well, historical analysis cannot be entirely predictive of future performance, especially as we seem to have a new "unprecedented" market event every year. But equally, not every strategy is meant to go up in a straight line all the time. And so our job on the sell side is to do the diligence and try to construct strategies where the inherent risk factors are as identifiable as possible such that clients can then decide which risks they want to add to their portfolios and in which sizes.

Thank you all so much. I think this is a great way to wrap up the Q&A part for the panel. I'd love to open this up to the audience as well. I know we had a question while we were presenting, but would love to hear from you. [Operator Instructions] One that I would like to ask as well right now while we're waiting for questions is, how are you really thinking about when you're thinking about your derivative strategies in QIS, is there a thought about where you would be thinking about execution? So for example, Lini mentioned, of course, all the data that's available on exchanges and executing on exchange. How do you think about OTC as an execution and really like a different way to structure strategies in terms of OTC derivatives?

Yes, I think it's a great question. I would say that one of the reasons that clients like to use bank QIS is for the transparency that it hides. And so historically, the exchange listed instruments were [indiscernible] increasingly so. We get questions sometimes and we are also exploring whether it makes sense to include other data sources into some of our strategies. And this isn't specific to commodities, but it always becomes a question of what our clients can get comfortable with and whether the returns are justified for potentially moving away from the most listed, most transparency-based sources and potentially giving up the track record and academic backing behind some of the strategies. So it differs by client type, and it's not for everyone. But I think increasingly incorporating potentially OTC execution, incorporating alternative data sources really challenges the old-school, glass box QIS approach. And so like we talked about, a key value proposition of QIS is really the tech platform. And so also when new technologies should be incorporated into the product as they become demonstrated and available, obviously, within a compliant and governed framework, which is part of what the banks are here to do.

Riccardo, Ankit, any thoughts on OTC versus exchange?

Not particularly from my side.

Yes. Maybe the only thing I'd like to add is something I was meant to say earlier on the trends. Something else that I think is useful for clients that I've seen is the packaging of QIS strategies in different wrappers. I mean the traditional way, it's a TRS and it's a swap. But I've seen lately even complicated strategies being packaged in funds, in USIP funds then some packaged into ETFs. I think that it is a trend that started. And from a client point of view, this is definitely something interesting. Some clients -- for some clients, it might be complicated to do a swap. Maybe they cannot do a note. So they have funds, well, ETF is obviously is the best, but there's some funds you cannot always be hedged easily. So yes, that's another interesting trend.

For sure, we see that one as well when it comes to derivatives usage within ETFs and anything from the likes of, let's call it, flex options, of course, 0DTE options. We've seen even like equity-linked notes, of course. So very interesting to see. We do have another question coming from the audience and perhaps this will be our last one. Lini, over to you. How do you do your commodity factor capacity analysis? And how do you think about factor crowding? And maybe, if others can take the factor crowding question as well, it would be an interesting one.

Yes, absolutely. I think the question around factor crowding in QIS always comes up. And I would say, although it is difficult to quantify exactly how much AUM is deployed in each strategy, whether that comes through bank QIS or whether that comes through participants directly trading themselves, we do have different metrics that we track in terms of robustness of each premia that we offer in our space. And so many of the well-recognized premia that exist in commodities, like I mentioned earlier, arise due to the structural corporate hedging, such as implied vol premia and curve carry to name a couple of examples. So one of the things that we look at is futures and options open interest from commercial participants via CFTC-reported data, which is, again, public information. And while it's not a perfect metric, it should give you a sense of how prevalent this commercial participation is relative to the noncommercial community. In addition, for curve carry, as an example, we monitor current backwardation rank of each commodity curve relative to its own history. And so what we really like about curve carry as a strategy right now is the fact that it is a constructive setup for many reasons. A, fundamentally, given the generally bearish outlook for time spreads across commodities where supply risks right now feel low, especially in Ags as we move through the crop season. But also looking at the backwardation rank from a 5-year lookback, is screening quite well for Ags and some metals in particular. And then lastly, positioning right now is a lot cleaner than it has been historically after the large-scale derisking event that we've seen, which can not only support the robustness of the premia, but can also help reduce negative slippage for such future base strategies around the contract rolls as those rolls become less congested.

Yes, I agree with Lini. It's difficult to estimate exactly AUM across strategies and factors. On the single stocks, there is a way. I mean some banks provide aggregate positioning. It's a bit rare. It has some limits, but yes, there are ways to get some positioning data that's aggregated and it's stripped out, obviously, of the identifiers and run through a risk model. So it's kind of hidden a bit, but you can get like a cumulative positioning on idiosyncratic part of the portfolio, which is useful to estimate crowding. And some banks even offer crowding factors, something like that. This is obviously, again, a partial representation. And if you get positioning data, it's going to have a big lag so difficult to trust it.

Then also the issue with -- when you think about crowding in general and positioning in general, it's obviously your most crowded positions are also your most profitable ones actually in the very recent history. That's the reason why they are crowded. So while we are all worried about crowding, you don't really want to be selling your winners. Like it's challenging. And if you get the timing right to when to get out, I think obviously, that's like difference between a success and a failure. Like if you -- like end of last year, you came out of AI, then obviously, you timed it perfectly. But given how the momentum was despite the teaming crowd over the past 2 years, when you came out of that scheme, the size to your portfolio return quite massively. I think crowding is very interesting to keep an eye on. But I think to take portfolio decision based on positioning and crowding is very challenging, similar to market timing, I would say.

For sure. Well, thank you so much, everyone, for all of the insights today. I'd like to personally thank Ankit, Riccardo and Lini for all of their insights. Definitely, it would be a great way to leave this webinar to say don't sell your winners when it comes to crowding. I think, Ankit, that's a great way to wrap things up. I also wanted to thank the audience for your time today. If you want to listen to the replay, of course, that will be available later on tomorrow. We'll also be following up. And in case you're interested in learning about systematic and quantitative strategies or anything in the derivatives data and valuation space, please feel free to reach out to us and to the guest speakers on this webinar. Thank you, everyone, again, for your time, and I hope you have a great day. That'll wrap it up.

Thanks, Mila.

Thank you.

Thank you, Mila. Thank you.

Thank you.