Podcast: The Importance of Exceptional Data in Systematic and Discretionary Strategies

 

Darrel Yawitch, Chief Risk Officer, joins Eloise Goulder, Head of the Data Assets & Alpha Group at J.P.Morgan on the J.P.Morgan podcast series Market Matters. Darrel discusses how data is critical to both discretionary and systematic investors, the importance of centralising data ingestion and analytics processes and the key risks he is monitoring amid current market volatility.

Recording date: April 2023

 

Episode Transcript

Note: This transcription was generated using a combination of speech recognition software and human transcribers and may contain errors. As a part of this process, this transcript has also been edited for clarity.

Eloise Goulder:

Hi, I'm Eloise Goulder, head of the data assets and alpha group here at J.P. Morgan. Today I'm delighted to be joined by Darrel Yawitch, who is chief risk officer at Man Group to discuss the role data plays in risk management at what is a very large global active investment manager. So Darrel, thank you so much for joining us on this podcast today.

Darrel Yawitch:

My pleasure. Good to be here.

Eloise Goulder:

Could you start by introducing yourself and your role?

Darrel Yawitch:

I am the chief risk officer, the CRO of Man Group. We are a global technology-empowered active investment management firm which focuses on the delivery of both alpha and portfolio solutions for our clients. We're headquartered in London, but we also have offices in New York, Boston, and across the globe. We manage about 143 billion worth of assets, and have five different investment engines that span a difference both of traditional discretionary type investing, as well as systematic investing. So I think that's going to make for an interesting discussion later about what role data plays between the two. As the head of risk, risk within Man plays the traditional second line of defense risk role that you'll have in other firms, but we strive to do more than that. We aim to add value to the investment process and we aim to influence the investment outcome, and we find that obviously analytical processes and data are obviously very key to fulfilling that role.

Eloise Goulder:

That's a really interesting point that you must have a really unique perspective looking at both the traditional discretionary, as you put it, but also the systematic businesses. So how would you say their respective processes with data ingestion differ then?

Darrel Yawitch:

Well, it's quite interesting, because I think people have maybe the mistaken impression that discretionary businesses are not as reliant on data as systematic businesses are. The difference comes in what they do with the data. A systematic business will ingest the data, generate a signal and trade automatically off that signal, whereas I think certainly in our business, the discretionary side is using data to the same extent, but then you pause and the portfolio manager makes the decision about how to use that data, rather than just following it systematically.

Eloise Goulder:

I love that perspective, the fact that discretionary investors are arguably equally data-driven versus systematic investors. They just use data in that different way, as you say, which I think is counter to popular belief perhaps.

Darrel Yawitch:

Absolutely. In our firm, we very much think about our firm structure as talent and technology working together. We try to have the right ratio somewhere around two to one between risk management and technology and data, but both are so critical to the way that we do our business. Good technology is really critical to being able to run the processes and the analysis that we need to be able to do our jobs well.

Eloise Goulder:

So you just mentioned that tech and data are key in both the alpha generation and also the risk management side, but aren't alpha generation and risk management just two sides of the same coin? For example, you can't really identify alphas without understanding how to minimize your risks. Is that fair?

Darrel Yawitch:

Yes, I think that is fair, but the line between where risk management starts and alpha generation ends can be blurred sometimes, and that's part of the challenge that we have. If you take, for example, crowding, crowding is a really good example of this. From a risk management point of view, we need to know how crowded a position is, but crowding itself is also a source of alpha. So we need systems, technology and data that enable us to do both parts of that process, but to look at it differently. We've seen such a big explosion of big data and alt data over the last few years, and we need to use those same sources of data to quantify risk and to generate alpha.

Eloise Goulder:

That's fascinating. So you have your function, the risk function, but also the PMs all looking at the same data, but through a different lens. You because you're managing the risks, and the PMs because they're harnessing the alphas. So how do you navigate the debate then between your function and the PMs when it comes to identifying new risks to markets and to stocks? I can imagine that if done in the right way, it's really powerful to have these two groups coming at it from different directions.

Darrel Yawitch:

It is powerful. It's a difficult balance to try to achieve exactly. It's worth bearing in mind the function of the portfolio managers and the researchers is, they're searching out for new alphas. So they're taking the datasets that they're looking through, they're analyzing to try and find a way of making a return from that. We're then taking that same dataset and saying, given that you're going to use it in this way, what risks does that introduce, and have you properly and fully understood those? Because depending on how you use the data, you may have a different type of risk. So that keeps the job interesting, but we're also coming at it orthogonally. Our purpose, our reason for using the data is very different to them. We assume they're good at finding alpha, and our role then is to ensure that the risks that they're taking in doing so are well-captured and transparently understood.

Eloise Goulder:

That's really interesting. And you mentioned earlier the explosion in availability of data, so can I ask, what are some recent examples of data that you've been ingesting and using which you didn't use say five years ago?

Darrel Yawitch:

That's a really interesting question, because part of what makes the role quite interesting is the markets are constantly evolving. So two topical examples. The one is natural language processing or NLP. We recently had Slavi Marinov, the head of our machine learning team here, do a demonstration at our shareholder day using machine learning and natural language processing to process earnings transcripts. And in the space of about eight minutes, he processed 814,000 earning call transcripts, which equates to about 10,000 times the size of war and peace. Now, that's just the speed of... It would take I think humans 48 years to process that amount of data. It's just not something you could have done five years ago or 10 years ago. And that's really harnessing the power of data, but also the power of machines.

An equally topical example which attracted headlines recently was in 2021 the meme stocks and retail stocks where what we needed to do there was to quickly access data about which stocks were on the radar of the retail traders. So it necessitated us gaining access to that information from websites like Reddit, being able to collect that data to know which stocks were under fire, so to speak.

Eloise Goulder:

I love both of those examples. Thank you so much for those. And as you say, they really show the power of both data and machines and using the two together. And there must be a real competitive force for you in being on top of these things, because if you're not and your competitors are, then isn't there this risk that they see a lens on portfolios that you don't, and that they can be aware of risks that you aren't? That competitive force must be quite a driver for you.

Darrel Yawitch:

Absolutely. That's why we describe ourselves as a firm of talent and technology. We sort of joke, I think there's roughly 48 languages spoken within Man Group, and Python, the programming language, is the second most commonly spoken one. We invest a lot of time making sure people are proficient in Python. Python today is very much like the Excel of 20 years ago in the way that... And I know in firms like J.P. Morgan and others, it's the first thing people will get trained on. We've been using Python, training people on Python for many, many years now. It's worth pointing out something else that we're super excited about now and really proud of, which is our ArcticDB database, which is our first venture into the commercialization of software that we've developed in-house that we think has applicability for us and has applicability for other investment houses, banks, financial houses.

It's worth talking a little bit what ArcticDB does. It's a high performance Python-native database. It's a Python data frame database, and it's built in response to the ever increasing amount of data that you have. So one way I would describe it, I think the tech would have a better description, is Excel on steroids, about storing things. It is about storing things that have millions of rows of data, and thousands of columns of data quickly and easily in a way that you can access and process that information. So if you are a firm that needs to process tick level data, you've got high volumes of historical data, you're looking for signals, you need to process that data, we've developed a solution that works really well for us, and we think it will work really well for other people. And as I said, I'm super excited.

We've now signed an agreement actually with our first client, and our first client is none other than Bloomberg. And Bloomberg, we have an agreement with them where they will integrate ArcticDB into their BQuant platform, which uses Bloomberg's analytics. So for us, that's good validation. You always wonder, are we really good at this, or are we just fooling ourselves? But the fact that Bloomberg are our clients is very good validation. So we think that there's lots of potential for that. Lots of people in the financial services who need to process high volumes of data will be using something like Arctic db.

Eloise Goulder:

That's fascinating. So for any of our listeners using BQuant, really interesting to know that there will be a Man Group engine behind some of that. And doesn't it seem like the industry is still finding its feet in terms of where the best centralized data and systems architecture should live? Is it on the buy side where it's often developed and is critical to the processes, as you've just described, or is it in the sell side where we see ourselves as centralized providers of all sorts of services, including data and analytics, which our team often speaks about, or is it the data providers, whether it's the larger ones like Bloomberg, as you just said, or the smaller fintechs, or does it end up living in all of these places? But this financial data architecture, amid this explosion of data that we've just spoken about, it requires so much investment, doesn't it? And some benefits to scale must exist

Darrel Yawitch:

Definitely requires investments, and that's something we've been doing over a number of years. I think everyone who wants to be competitive needs to be investing in this. And I think the two points which you touched on, whether it's buy side, sell side, vendors or not, centralization and scale here really is the key. We focused on centralizing data ingestion and the onboarding of data, and there's two aspects to that. It doesn't make sense for different portfolio managers to speak to the same data vendor, and then onboard the same data two or three times. So we have a data science team, they are the experts now at onboarding new datasets, liaising with the different vendors, negotiating the best commercial terms, and then making that available to whoever needs to use that data within the business. And that includes the PMs and includes risks. So that's an example where centralization at that point is very empowering.

But then you also have the reality, even within a firm like ours, that every firm has the challenge of making the data available to the different teams that want to understand it. How do you join up data that's got different identifiers between a Bloomberg identifying ICE and a CUSIP and very different things like that? So you need either code, data, a database, something that joins together that data that makes it much easier to use. And the hurdle, there's no barrier to entry to using that in different parts of the business. So that's really that centralization and scale. You need scale to be able to have a team in the first place that spends its time doing this exercise to make it easier for other people.

The other thing which we also are focused on is how you do use that data at the end of the day, how much of it you can automate, how much of the generation of the data you can automate, and how much of the processing of the data you can automate. So for example, data watchdogs are an example of one of these things where a watchdog is something that's going to flag that there's something wrong in the data that you received, and that's a real world problem as opposed to a theoretical problem. In theory, the data's all clean. In practice, lots of people spend lots of time checking that the data is clean, something that you can to a high degree automate.

Eloise Goulder:

That's a really interesting perspective. And these data watchdogs, are these people, and how are they structured within your organization?

Darrel Yawitch:

That's a bit like asking when I get on an airplane, is the pilot flying the airplane? The answer is yes, but there's an autopilot that's active as well. And in fact... Because many people who do the flight across the pond, London to New York. It's an interesting question. From what I understand, the pilot's only flying the plane for about 20 minutes, and that's deliberately takeoff, because we don't let planes take off by themselves, although they can land by themselves. We live in a world of automation. Your choice is about what you buy at Amazon and what's recommended next. You looked at this, therefore you might like that. The YouTube algorithm that says, you watched this, therefore you're going to watch that next. Some of this originally was human stuff. Much of that stuff these days is automated. Machines are really good at doing a lot of things, and the things that they're good at doing, we should let them do. There's always a role for humans, so it's a combination of both. We let the machines do what they're good at, let humans do what they're good at.

Eloise Goulder:

So when we think about the data watchdogs that you've just described, how do we incentivize the people working behind these processes to really strive to keep the data clean?

Darrel Yawitch:

Fortunately for us, the vast majority of our data we ingest is for systematic purposes, and that means the quality of our systematic signal, the returns of our alpha, the returns for our clients are all dependent on the cleanliness of the data. So it's really important to us that that data is clean, so we spend the time doing it. In other instances, another thing that's really helpful is to have the same data used in many parts of the business, and not to use different copies of that data. Because if it goes wrong in one place, you have say 10 groups looking at the same data, you're much more likely to pick up the error, whereas if you have different groups using different versions of the same data, it goes wrong in one place, and you're only reliant on one person to pick up the error.

Eloise Goulder:

So in addition to the explosion in data that we've seen, there's also been something of an explosion in volumes, hasn't there? At least in US equity markets. This was actually something we discussed on the last recording two weeks ago. So Darrel, how have you handled this, and perhaps more permanently, as volumes have become more electronic, how have you navigated this?

Darrel Yawitch:

So as a firm, we're constantly looking for new sources of alpha and new sources of data. So we are geared up that way, and we have been for a number of years to onboard new datasets. So the first thing is to recognize markets are at different stages of maturity, you could call them. There has been a vast explosion of data in the US equity markets, but other markets, even in the developed economies, still remain high-touch. Credit, fixed income, still high-touch. It's moving to electronic, and as it moves to electronic or cleared forms, for example, swaps, much more data becomes available which you then have access to and should use, especially for risk management, and arguably for alpha as well. So having your system being able to be dynamic enough and your processes being able to be dynamic enough to constantly looking for those new data sources, onboard them, and then use them appropriately.

Eloise Goulder:

So let's come to the present now. Markets have been really choppy to say the least over the last few weeks or through March as a whole, and Darrel, we're really grateful that you've been able to take the time out to speak to us for this reason, but what are your observations on current market dynamics, and with your chief risk officer hat on, which risks are you really monitoring right now?

Darrel Yawitch:

Well, there certainly is a lot going on now. The financial sector, the banking sector of course is very topical and something that we're of course monitoring very closely. Inflation is another issue. It's not new, but it may be new to many people in the room, so to speak. You take a look around the room and you say, show of hands, who was here in the 2007, 2008 GFC? And not many hands go up anymore. And how many people in the room have experienced inflation like we're experiencing now? Well, I grew up in South Africa, and I know what it's like to have 15% inflation, but many people in the developed markets have not seen this happen in a long, long time. So we are in a different market regime. As I say, there's nothing new under the sun, but we have seen this before, but not recently.

So it's important to recognize the skillsets in the room, the experience of the people in the room, and to be vigilant about seeing the market as it actually is today. And we go through cycles. Sometimes the market is more volatile than others, and different parts of the market are volatile at different times. So most recently we've seen a period of very high volatility in the fixed income markets, yet the equity markets have been somewhat calmer, and that's not something we've seen in a while. The other thing which our CEO referred to recently also is the connected nature of the markets now where we live in a world with great interactivity, and the role that social media plays in the transmission of information, how quickly certain things play out. So that's also I think different to say 2007, 2008, and different to certainly 1980.

Eloise Goulder:

It's interesting you mentioned social media there, because I often talk about the democratization of content, i.e the idea that anyone, including the retail investor, has access to reams of information online, including via social media, and of course they also have the ability to transact on top of it. Do you agree with that?

Darrel Yawitch:

Yes. The world has changed the way that retail investors and other people have access to information, whether it be via data that's available online, on their phones, and then to actually execute from that. But I like the word democratization of information. I very much use that internally in the design of our systems. We've moved beyond a stage when a chief risk officer would ask somebody for what is the number. The number should be available to everyone in the team. It should be available as management information to whoever needs to see that in the firm. And the conversation should really be, why is the number that number, not what is the number. And we have the ability, using the data and the systems that we have, to make the numbers available, and that's the democratization of information.

Eloise Goulder:

And surely it makes the job that much more interesting as well. Rather than seeking the number, you're talking about why it should be that, and whether you want that number to be what it is.

Darrel Yawitch:

Well, not only does it make the job more interesting, it makes the job possible. We manage a few hundred funds to be able to see all the information. If I had to wait for the answer as to what the number was, you simply couldn't do the role. You need to know now in a very clearly presented data way exactly what your answers are. One of the tests I have about how well-designed our reporting is, if I went on holiday and I switched off completely and came back, how long would it take to get fully up to speed without asking anybody, just using the reports that we have? And that's always a good test.

Eloise Goulder:

Absolutely. So coming back to the present then, what data are you watching at the moment to really have a handle on the risks in the marketplace as it stands? And presumably it's the unknown unknowns that are the most worrisome to you, so how do you begin to think about those?

Darrel Yawitch:

I think that's what makes the role quite challenging. At any point in time, the importance of a given set of data varies. And as I mentioned earlier, for example, fixed income volatility is higher now, so obviously we're watching fixed income data a little bit more closely, but it's the ability to process that data and to react quick enough. At best, you can react quick enough. If you're really skillful and thoughtful and maybe a little bit lucky, you'll be able to use that data in advance to avoid an issue. And that's really the ideal space where you want to be, where you have the ability to avoid things that haven't yet happened.

But it's constantly questioning, is my data relevant, am I looking at the right set of data, and what am I missing? And there's no easy way to do that, and I think that's where it becomes part art and part science. You can't just rely on the data, you really have to apply your mind to the problem to say, how have I processed the data, and am I being limited in the way that I am, do I need to look at it using a different technique, think about the problem a little bit differently?

Eloise Goulder:

So finally then, if we can look to the future, from our side, we've seen this enormous growth in the demand for data and data-related infrastructure from our clients, and your organization has clearly done so much to embrace this and to embed this throughout your processes, but when you look to the future, what do you think this all means for the industry at large?

Darrel Yawitch:

So as we discussed, our firm very much is about talent and technology, and a lot of what we do is trend following. So recognizing that trend that you pointed out, the increased use of data, the increased importance of data is where the trend is. As somebody with a physics background, the way that I think of this is, data is the fuel cell. It's the fuel cell that will power your future returns, it will power your business. You have to be able to harness that power that's sitting in all that data. So that's what we do. And I think if your firm doesn't put itself in a position to be able to harness that, then those firms will be left behind ultimately, because there's very powerful things you can do with the data we now have access to.

Eloise Goulder:

Darrel, it's been an absolutely fascinating discussion. Thank you so much for taking the time to speak today.

Darrel Yawitch:

It's been a real pleasure. Thank you.