Interview with Caio Natividade
RR
All right, thank you for joining us today, and I would like to start off by introducing myself. My name is Ying. I’m currently a finance student, and I would like to ask you if you could just please tell us about your current role at Deutsche Bank. What do you focus on day to day, and what are the problems that you currently solve?
Caio
Sure, nice to e-meet you, Ying and Angad. So my current role is as a global head of quantitative research here at Deutsche Bank, we work very closely with the bank’s investment solutions business. In other words, DB is along with other businesses as well, and most, most of the research that we do has to do with looking at the world from a quantitative angle, right? In fact, all of the research that we do is linked to looking at the world from a quantitative angle. I would categorize it as solutions-driven right? And in that sense, it opens up into basically two fields. The first field is about replicating return streams, and the second field is about introducing portfolio solutions. Now, on the first field, replicating return streams. What do we mean by that? Well, there are certain return streams that you often only have access to as an investor if you invest in an active manager. The idea of QIS and with us in DB is to provide a set of returns and an investable solution, basically, that is similar to what you would get from an active manager, right? So that follows that particular investment style. A simple example of that is trend following. For a long time, you would have only been able to access the return stream of trend following, using an active manager, with their fee structure and their entrance and exit structure and so on, and arguably, arguably lack of transparency as well. But as time progressed, banks have started being able to provide those return streams as very similar return streams as directly investable solutions, with the advantage of them being more transparent, having a different and more favorable cost structure, and also higher transparency, higher liquidity, and all of these things. So that’s the first type, replicating return streams. The second type is producing targeted portfolio solutions for our clients. And a lot of the work that we do in research is to introduce those solutions. They often come in the form of portfolio construction research. And the types of solutions that we look to achieve usually have to do with either A or B. A is a solution that is more defensive. So in other words, producing portfolios that will perform well in a given market regime, and the market regime that tends to interest our client base the most is the bearish market regime. In other words, construct portfolios that do well when markets are in a bearish mode, so that they can hedge that environment. And second, they are producing portfolios that are heavily diversified. In other words, this gives you returns that are diversified relative to those of the traditional portfolios, the traditional market premiums. Okay, so, so under the stream of producing portfolio solutions, there are these two basically key use cases that we focus on. Right one, producing defensive type solutions, and second, producing diversified type solutions across all types of uh scenarios.
RR
Thank you so much. That’s a very detailed answer, and I just want to pick your brain a little bit. Looking back, what drew you into quantitative finance? Was there anything specific, or did it happen more gradually?
Caio
Sure. So I’ve been with Deutsche Bank for 23 years, and out of these 23 years, after doing the graduate rotations, my career started as an emerging markets strategist. I worked in that for five years. After that, I did options research for just about five years. And then after that, I became a more formally quantitative researcher, first in the cross-asset world, and then after that, in a more managerial role, encompassing all asset classes. In this career progression, the main [thing] I was largely driven by was what area of work would I learn the most? Right? What choices could I make, from a career perspective, that would lead me to learn the most? And it was always, I always also had a passion for quantitative topics, and so for quantitative topics, I was always more interested in the formulaic aspect of financial markets, interested in looking at things as formulas or in a more systematic way. Let’s say I would. I wouldn’t just call it an interest in quantitative topics. Actually, I would say that it’s also the fact that I enjoyed looking at the world using a structured thought process, right? The quants are well known for that we like to create a structure, creating a workflow, and then following things using that, that workflow, using that structure, that process, right, step one, step two, step three, and so on. And I found that the more I went into this world of following, [the more it] taught a structured thought process. The more quantitative things became, and the more I entered into the into the control, the more I was being geared towards the quant world, right away from in the very first beginning, when I was just a just looking at things as a standard macro economic strategist, all the way into looking at things as a full prompt later on, a couple of decades later in my career.
RR
Regarding your past answer, I have a question about your options research. Can you explain some more? Was it a non-quant and how option research goes in a non quantitative role, because it’s very quantitative.
Caio
So there are two sure options for research there. You can break it down into, I would say, two main elements, right? One of them is the pricing element of it, and the second is the signal research element of it, right? In some ways it’s another label that I’ve seen people use to say that you can either be a Q quant or a P quant. Okay, so the Q quant, in other words, the risk neutral quant, is more interested in pricing discussions, right pricing matters, whereas the P quant, in other words, the empirical quant, is more interested in signal research and alpha capture and so on sell-side, options research tends to focus on the P quant side of things. On the signal research side of things. So it’s not so much about pricing, and the pricing element of it is the one that is much more mathematical. This is where you’re looking at all the various tools and techniques that can be deployed to come up with the right price and the right hedging mechanisms and and so on. The signal research in options markets is quantitative. Yes, It is quantitative, though, I would say that much of it at the time that I was doing it was more discretionary than systematic. In the quant part you identify a number of quantitative scores that tell you about the attractiveness or unattractiveness of a particular trade. And then once you put those scores together, you then utilize personal judgment, a specialist judgment, to identify whether or not it is a good trade to put on or not, or a good trade to write a research paper on or not. And that is, but that’s not fully systematic. When it comes to fully, fully systematic research, fully quantitative research, then you’re not really making those decisions using discretion. You let the algorithm make those decisions for you.
RR
And jumping off of that, I just want to ask, over your 20+ years at Deutsche Bank, how has the quantitative investment solutions approach evolved?
Caio
Sure. So QIS, I would say, had two periods, two to two and a half periods. The first period is the period before the global financial crisis. Okay? (The global financial crisis of 2008), and over that period, prior to that period, the algorithmic strategies, algorithmic strategies that banks would formulate, tended to be all-weather, so in other words, more like pure alpha-type strategies. What do I mean by that? These would be strategies that were supposed to perform well under any type of market regime? So the strategies that were built over that period. They had that focus right after the global financial crisis, and I would say, for most of the last decade, the types of solutions that the banking industry, the QIS industry, has introduced had to do with alternative beta return streams. In other words, these are strategies that are supposed to perform well over the long run, but they provide a beta. In other words, they provide a strong exposure to a particular market driver, okay, an alternative market driver. And examples of that will include, for instance, strategies that capture the Fama-French factors, right, which are alternative drivers, alternative betas that drive stocks, or the carry trade in many of the asset classes, right? There is also an alternative driver of those asset classes and so on. So the industry would provide investable solutions that give you access to those types of return streams. These have given lower projected Sharpe ratios. These give lower projected Sharpe ratios, because ultimately they’re still betas, right? They’re alternative betas, but there’s still a beta component to it; their future performances tended to be more in line with history-based expectations than what was the case before the GFC, and the more pure alpha-type return streams that the industry would offer. So that was the the the key change after the global financial crisis later, and this is why, earlier I was saying two to two and a half types of periods, is that the industry became focused a lot more on specific portfolio solutions, in other words, hedging solutions or fully diversified portfolios and and things that have more to do with portable alpha type considerations. And these became even more popular after COVID.
RR
I have a question, like you mentioned. This change happened after the financial crash, but the betas are more dependent on the market, like the beta portfolio will be directly influenced by the market. So won’t another crash cause more losses? Like Alpha ones will still survive in the market crash, but the beta ones will be directly affected. So why would a market crash motivate banks to switch from alpha portfolios?
Caio
Yeah, so the key distinction here is in the definition of the word beta, okay? I was talking about alternative betas as opposed to traditional betas. Okay, a traditional beta is something like the equity risk premium. A traditional driver is something like the equity risk premium. You just long the market, the equity market, right? Or the credit risk premium, or the bond risk premium, and so on. These are, yes, directly influenced by the various pro-cyclical and counter-cyclical gyrations of the economy and their effect on financial markets right now. Alternative betas, on the other hand, are different. They are things that drive assets on top of the traditional betas. Okay? So to give you an example the concept of value, right? I was talking about the Fama-French factors earlier, for example. Value is a concept of the performance of the value factor. And strategies that are linked to the value factor, they are not, as they cannot be as easily conditioned according to equity risk premia as equity risk premium itself. Okay. So in other words, you may have an environment in which markets are falling and the value strategy could be doing well or it could be doing badly, okay, so in other words, it follows its own performance cycle. Okay?
RR
Thank you. That was okay. You guys are very shy.
RR
In your most recent quant post episode 50, you did mention how portfolio 365 was navigating the latest Middle East volatility, and you had mentioned both pro-cyclical and defensive sub-portfolios performing well, which is rare. Could you please elaborate on that?
Caio
Sure, so I suppose, in a way, this almost relates to, this has some linkage to the previous question, right? The key here is the directionality of markets to risk appetite. So in an environment where global risk appetite is driving everything, it’s driving the majority of returns of over assets, then you tend to see defensive strategies performing in a way that is anti-correlated with the pro-cyclical strategies. So you have either A or you have B, okay, one is going to go up and the other is going to go down, or one is going to go down and the other is going to go up. Okay, so if markets are crashing severely, then defensive strategies do well, and procyclical strategies are going to do badly, boom. Okay. Likewise, if markets are in a bigger risk-on mode, then you have the opposite performance. Now, that said, there are other periods, especially periods of lower volatility, periods when markets are calmer, when it’s not just market risk appetite that drives everything right, it still has an influence, but it’s not as high in influence as in the previous case. When you are in that kind of regime, then other market drivers kick in, and as other market drivers kick in, they will influence a number of the strategies that we have that would otherwise be defensive, and a number of the strategies that we have that would normally be offensive, right? Pro cyclical, and they can influence them in such a way that they both perform well sometimes, and unfortunately, it can also influence them in a way that both of them perform badly. In the case of episode 50, we were talking about the former case, right? So it was one in which both types of strategies were being used.
RR
Well, okay, I will start a bit more informal session, like you are a quant head at Deutsche Bank, so you are hiring a lot of freshers and like right from college quants. I was wondering what we should do more of, or what we are lacking in, because of the job market right now, people are not getting jobs. So what are we doing wrong? What should we be doing better?
Caio
Yeah. So I think the first thing for those looking at a quant account role, right? The first thing to think of is that you need, you need your quant toolkit, right? So it’s not like you’re going to enter a quant role, and then you’re going to learn how to code. Okay? No, you’re expected to have that knowledge as part of your personal toolkit before you join. You’re also expected to be able to read a quant paper, which will be full of formulas, right before you join, as opposed to learning about mathematics and statistics and things of that sort, only after you join, right? This is the first thing, okay, so you’re expected to have your own, your toolbox, before you start. And a lot of the interviews, they test you on that. But then there’s the bit that is more specific to research, right? Quantitative research, and I would say that this applies both on the sell side and on the buy side too. Researchers are curious by nature. Okay, so a good researcher is somebody who’s always very curious, who always wants to dig deeper into a particular topic and understand what is it that is driving a phenomenon, right? What is it that is driving things? And this is particularly important, because the more that you understand what drives things, what drives the return of a strategy, or what drives the fact, or what drives a particular scenario, or what have you, the more that you can take a view on what’s going to happen in the future when it comes to the performance of that strategy or the sustainability of that scenario and things like that. Okay, so I’d say that it’s almost like a researcher, a quant researcher, is, in a way, like a small child who always asks why all the time, so why this? And then you answer that question, and then they say, why, and then you have to answer that. And then they say, why again, and then you have to answer, and then so on, so forth, right until at the very end of it, you get to the, you know, to the core meaning of things. So then the next question you could argue is, how do I then, how does one identify that in an interview context, right in the and so on. How do we test that? Well, there are ways that we do it in the context of assessment centers and in the projects that we assign to candidates and so on. But one thing that we can tell you exactly is that there are things on the CV that you can see which suggest that that person is a curious, very interested person, right? For example, what do you do in your spare time? So how much of your spare time is spent looking at things that have to do with the quant world, just because you’re interested in it, right? Just because you have to have a core passion to it, and and, and things like that, basically.
RR
And follow up, like for the finance part, like, how much of finance should account know, or are there just finance rules in quantities, like pure finance rules for the financial motivation and intuition.
Caio
Sorry, I did not understand your question.
RR
Like for a quant, how much financial intuition or motivation is necessary? Are there like financial analysts working with quant teams to cover up that?
Caio
It’s almost in a way, the more serious the quant the role is about. Let me rephrase, prior knowledge in finance is not necessarily needed especially for entry-level roles. What is more important is to understand. What is more important for the candidate is to have a structured thought process, have the quantity toolkit, toolbox, and see specifically for roles in research to demonstrate that they have a passion for curiosity. Okay, so if you have these three things, those are the key requirements. Obviously, one should show some interest in finance. If you’ve exhibited zero interest in finance, then the other three things that I was describing they’re not going to be enough. And, you can show that by again describing what you do in your free time. What are your extra-curricular activities and things of that sort. But going back to the first point, which is these three things that I’ve described, not requiring a finance background. I say that because you find that many of the candidates, many of the successful candidates that we have, have a background in STEM. So, science, technology, engineering, mathematics, that kind of background. That gives you the right toolkit.
RR
Yeah, that was a very insightful answer. And one more next question will be that in the future, we are seeing right now that AI is taking away lots of jobs, and AI is in software engineering. AI has replaced software engineers. Do you think AI will replace quants in the future as well, or people will use AI to trade or build strategies, or, or like, how do you think AI will come into the picture?
Caio
Okay, so far, that question is a bit different from the question that I had from you in writing. The question I had from you in writing was, when you think about the next decade, which one thing that you expect will shake up systematic investing, the most shift to market dynamics, new regulations, or advances in powerful, low cost AI, does that? How about this? I’ll address the second question, okay, which is the one that I wrote my notes for. And then in that I’ll I can address bits of the first question that you asked. And I say that because the second question, I think, is more interesting. The first question is pretty binary in the sense that, no, I don’t think AI is going to replace the quants in industry. But so let me address the second question first. So when it comes to the next decade, right? I mean, what are the key shifts in market dynamics? And so the key shifts in systematic investing, of the things that you highlighted, such as shifts in market dynamics, new regulations, and advances in artificial intelligence, I think all of them are going to impact the way that systematic investing operates. The problem, though, is that it’s hard to predict exactly which one is going to be most, most influential and how right. What I can say, though, is that especially when it comes to the topic of artificial intelligence, what we see today is that one should be conscious of two revolutions that are taking place at the moment. The first revolution has to do with big data, and the second revolution has to do with AI itself. Both of these, when it comes to quantitative investing, are fueled by the democratization of quantitative finance. What do I mean by that? Well, I mean that it’s cheaper to do quant investing. Computational costs are lower, access to data is easier, and so on and so forth. I mean, there are even platforms today, such as QuantConnect, where you can do your R and D, and you can plug and play and invest. Like a quant So, so let me talk a little bit more about the two revolutions themselves, the two aspects of this revolution, one of them is big data, and the other being, AI with big data, what I mean is that today, there are a lot more data sources from which one can Search for alpha than what we had before. Before the world was, in a way, more concentrated in what we call today the traditional data sources. The traditional data, which had mostly to do with either market data, which is price, volume, and derivatives of those, or fundamental data, which are things that you capture, either from the balance of payments or a statistics office, or the accountancy data, right? When it comes to companies today, and in fact, over the course of the better half of the last 15 years, we’ve had, more and more alternative sources of data coming up. And these alternative sources of data, it has become increasingly easier to distill intelligence from these alternative data sources. We start from something like, you know, satellite images. Satellite images which have become increasingly popular, but the problem is that these are just images, right? They’re not about execution. They’re not structured. It has become increasingly easy to translate those images into information that can be processed by a quant program, in other words, to translate from images to numbers, right, and from numbers into time series. And once you have those time series, then you can translate that into alpha. If there is an alpha in them, right, you can, you can process that data to identify the predictive power that it has, or the explanatory power that it has. So it’s not just more data, more types of data becoming available, but also more processing tools that allow us to process these, these various types of data, which brings me into AI. AI, I suppose, in many ways, has been preceded by the term called machine learning. Machine learning tools have been particularly useful throughout the finance world, over many decades, and more recently, in being able to translate unstructured data sets into structured data sets. But as we talk about the latest version of generative AI and predictive AI, and so on. What this is, what this has been doing for the quantum industry, is to allow us to provide more efficient Alpha hunting. And what do I mean by more efficient Alpha hunting? What I mean is, instead of humans building alphas, increasingly, we are bridging into a world where the machines build their own alphas without the humans. And the humans are there to supervise the system. This is the concept of the Alpha factory, right where you have basically a workflow process, almost like an assembly line process for the generation and implementation of an alpha, but before that process was done by humans. One can increasingly rely on machines to do it for them, almost in a way that that we can almost allow it, as in a way to how we see it being done in actual assembly lines where cars are built, right before you had the humans with the, you know, tying the screws and and putting the actual windows in the car and so on. And now, increasingly, we just have machines doing that.
Caio
So this is the second revolution.
RR
Also, like I have seen, notice this, some firms are coming up with their own LLMs that they are promoting, that they will find the alphas for us, like you mentioned, the alpha factories. So are the alphas being traded in the market, and are they profitable?
Caio
Well, yes. So certainly, you have the small language models and you have the large language models. That is one discussion in itself, with most of the small language models tending to be more domain-specific and therefore possibly better for certain specific domains, including finance, and then you have large language models that tend to be more generic. The question, though, when, especially when it comes to large language models, but then again, to some extent, with SLMs as well, is, how do you fetch the right information from them? And then once that information is fetched, how do you process them efficiently to design an alpha? And this is where RAGs and Agents [come in]. These two new concepts that are becoming increasingly popular come in. So you use one in order to more efficiently extract the type of information that you need, even if you’re dealing with a language model that wasn’t specifically designed for your use case. And then the second one, which tells you what to do with that information. Once you’ve got it right.
RR
Okay, thank you. It was, yeah, it was really nice having a conversation.
RR
Cool answers were very in-depth.
Caio
So thanks, guys. I can see that you have a transcript, so feel free to send me so please send me the transcript. Later, I would ask you if you can make our life easier, if you can, if there’s any obvious pre-processing that needs to be done, that you do it on your end, and then you send it over to me once it’s done. So, for example, if I say, you know, if the recorder misinterpreted a term and it’s obvious what the real term is. Then you can override, in writing, what the record has used, and then send it over to me. And then what I’ll do after that is I may further process it in order to make it easier for the reader, right? Who’s going to read the whole article, and then I can pass it over to my compliance for approval and so on. This should be a pretty quick process. Is that okay for you?
RR
Yes, that sounds good. I’ll send it as soon as I can. Thank you so much.
Caio
Okay, guys, it’s been a pleasure. When are we going to see each other in person? Hopefully, in the next conference.
RR
Cornell one?
Caio
No, unfortunately, that one I cannot make. That’s unfortunately not that one, but, but hopefully the next one after that.
RR
Okay. Ying, nice to meet you. Angad, same, and I look forward to the text from you.
Caio
Guys, have a good weekend, and we will talk again soon. Thank you. Bye.
Interviewers: Angad Bhatti, Ying Xuan Li
