My guest this episode is Ralph Smith, Head of Research at BlueCove.

BlueCove offers long-only and market-neutral mandates in corporate credit and interest rate markets, with an emphasis on utilizing a scientific approach to portfolio construction.

We spend the episode discussing how the unique nature of fixed income markets present both opportunities and risks.  For example, how the differing breadth and liquidity in corporate credit versus rates markets impacts the types of strategies that can be implemented.  Or, how the assumption about a bond’s availability or liquidity can materially impact a portfolio backtest.

As Head of Research, Ralph also has some strong thoughts on the research process itself.  He shares his views on structuring a research organization, performing research in changing market environments, and even the appropriate use of backtests.

Please enjoy my discussion with Ralph Smith.


Corey Hoffstein  00:00

Okay, then well, Ralph, you’re ready to get going. Yep, sure. All right. 321 Let’s go. Hello and welcome, everyone. I’m Corey Hoffstein. And this is flirting with models, the podcast that pulls back the curtain to discover the human factor behind the quantitative strategy.

Narrator  00:24

Corey Hoffstein Is the co founder and chief investment officer of new found research due to industry regulations, he will not discuss any of new found researches funds on this podcast all opinions expressed by podcast participants are solely their own opinion and do not reflect the opinion of newfound research. This podcast is for informational purposes only and should not be relied upon as a basis for investment decisions. Clients of newfound research may maintain positions and securities discussed in this podcast for more information is it think

Corey Hoffstein  00:55

If you enjoy this podcast, we’d greatly appreciate it. If you could leave us a rating or review on your favorite podcast platform and check out our sponsor this season. It’s well it’s me. People ask me all the time Cory, what do you actually do? Well, back in 2008, I co founded newfound research. We’re a quantitative investment and research firm dedicated to helping investors proactively navigate the risks of investing through more holistic diversification. Whether through the funds we manage the Exchange Traded products we power, or the total portfolio solutions we construct like the structural Alpha model portfolio series, we offer a variety of solutions to financial advisors and institutions. Check us out at www dot Tink And now on with the show. My guest this episode is Ralph Smith, Head of Research at Blue Cove. Blue Cove offers long only in market neutral mandates and corporate credit and interest rate markets. With an emphasis on utilizing a scientific approach to portfolio construction. We spend the episode discussing how the unique nature of fixed income markets present both opportunities and risks. For example, how the differing breadth and liquidity in corporate credit versus rates markets impacts the types of strategies that can be implemented, or how the assumption about a bonds availability or liquidity can materially impact a portfolio back test. As head of research, Ralph also has some strong thoughts on the research process itself. He shares his views on structuring a research organization, performing research and changing market environments, and even the appropriate use of backtests. Please enjoy my conversation with Roth Smith. Ralph Smith, welcome to the program excited to have you here. It’s not often I get to talk to someone who is doing quantitative investing in fixed income. So I’m really excited about this episode. So thank you for joining me.

Ralph Smith  02:58

Thank you as well, nice to be here.

Corey Hoffstein  03:00

I suspect many of my listeners may not have heard of you or blue Cove before, given you’re on the other side of the pond. So why don’t we maybe start with your background?

Ralph Smith  03:11

Sure. Yeah, I started back in the 1990s. So I’ve been around that long. And I left university having done a mathematics degree was not really a clear idea of what I wanted to do in finance, I had a clear idea that I wanted to spend my career in finance doing interesting mathematical things. I was fortunate enough to be taken up by a little boutique firm doing dynamic hedging in foreign exchange, which is a small corner of finance. And they actually do some cool things back then in 1990s, in the first days of Windows operating systems, and even before we had email, they were running 24 hour, five days a week dynamic hedging programs. So I was a researcher for that firm for a while. So that’s how I where I cut my teeth in finance. I would say after a while I wanted to kind of branch out to bigger things. And I joined a bigger investment managers. So I’ve joined the investment arm of State Street Global Advisors to be a portfolio manager indeed, in foreign exchange space. And I was running some relatively large foreign exchange portfolios, then, which was an interesting experience. Back in the times, this was around the time of the Asian crisis, or just after the Asian crisis with massive currency volatility. It was around the time I don’t know if you can remember, probably not. But the euro was introduced. So it was introduced in 1999, and sort of sank like a stone for the first year or so of its existence. And so there was lots of FX intervention by central banks, lots of volatility. And it was an interesting time to be a portfolio manager. I think really, on the back of that experience, I learned quite a lot about markets and when models work and when they don’t, but I also learned something about myself, which is I didn’t really function that well with all that kind of stress, and that I was actually really interested in building models not running them on a daily basis. So I switched then to building a research team back at SSgA. So I set up a quant team designed to look across the areas equities, fixed income and asset allocation etc. So that was my journey there. SSgA And then after that I joined BGI, which folks may know, before it became Blackrock was a pioneer of scientific active equity investing. I joined in 2007, when they were just going out about the process of setting up a systematic fixed income. So we joined them to build the first set of fixed income systematic models there. And obviously 2007 was just before the financial crisis. So another opportunity to learn about how models perform outside their back test period, a pretty interesting time, and one to really observe what happens when markets break down, I often think that you learn the most interesting things about your understanding of market function from stress periods, you know, so called correlation breakdowns, which I guess, to my mind are better thought of as new factors that come along in stressed market periods that cause assets to go in different directions. So we had a lot of experience back then. With that, after spending some time building weights models at BGI I then took a new role to the head of the research innovation team, which was a team charged with enhancing and producing the new wave or new versions of content models across the fixed income process. And then in 2018, I had the chance to join blue Cove, which was a newly formed scientific fixed income manager really formed with one purpose, which is to do scientific fixed income really well. And that’s where I’ve been for the last three years building out the investment process with my with my colleagues,

Corey Hoffstein  06:21

before I dive into sort of the nitty gritty of fixed income itself, and we’re definitely going to get there. I want to maybe take a step back and just talk about the research process itself. Because prior to Blue Cove, as you mentioned, you ran the research innovation team at BlackRock and you now sit as the head of research seat at Blue Cove. I’d love to get your thoughts on running a research team, you know, things like how do you prioritize projects? How do you spark innovation and avoid stagnation? Even the nitty gritty things like how you think about resource management?

Ralph Smith  06:57

Sure. I’ve talked about this topic for absolutely hours, because I’ve spent a lot of my career running research teams with a lot of fantastically smart people. But let me think what are the main points? I mean, I guess I first I would admit, it’s important to be humble personally, because I don’t have all the answers. But what what Little can I add? I would say this culture is absolutely key. So getting the right culture in the research team is a prerequisite to success, not giving the incentive to produce good back tests, because anyone really can produce a good back test going from bottom left to top right. But getting people incentivized to produce long term alpha, that’s key. How do we do that here? Well, everyone is an equity investor in the firm, and everyone and therefore has the long term interests of clients and the firm at heart. Getting the right degree of peer group critique in the team is key. No person works on a project by themselves, everyone always has to justify the design decisions they’re making and to defend their research. And we encourage people to be as dispassionate as possible not to get sucked into this trap. That often happens in academia, which is positive publication bias that the thing has to go ahead just because I’ve started working on it, we don’t want that we want people to be very critical of their own, and each other’s work, on prioritization, it’s super hard, but also the most fun thing that we have a chance to do. So every few weeks or months, we’ll get in a room, whether it be credit or rates, we’ll talk about everyone’s ideas, the entire research team is invited. And we’ll talk about the merits of those ideas and how much they could contribute to the investment process. Trying to be as scientific as possible, I think, is important. So trying to say, what’s the value of idea how much additivity? Will it bring to the models, if we put it in? What’s the time cost? That’s the only real way I think you can rank projects. And it’s obviously not easy to do. But you can sort of guide and get a sense of what projects additivity is likely to be. Let me give you an example. Let’s say we’re developing a signal in natural language space, using some textual information about the firm, we would kind of expect that to be correlated to some other signals in the sentiment category. And therefore, it’s going to be a correlated signal to those that you already have. Does that mean we wouldn’t work on it? No. But it doesn’t mean that it’ll have to compete with the other signals in our models in order to get a weighting in the final strategy. And in that way, you can kind of gauge the likely additivity of an idea in advance or get get a sense of it. Whereas something completely different a kind of insight that we don’t use already might be more likely to be additive in the end. In terms of other bits of prioritization. Well, once you’ve decided to work on some projects, you then and you’re starting to work on them, then you need to know which ones to continue with and which ones to give up. And that is a different challenge. But it’s even harder, I find I have a behavioral bias to wanting to keep working on stuff. And you know, I’m a curious individual, we hire curious individuals. So the tendency is to want to work on projects and carry on answering questions about them. But you have to prune the tree right? You have to not only cut off some questions which are interesting, but not important. You also have to say you have to be prepared to say when’s enough when enough is enough? For a given project, I find that personally super hard, but it’s important. And it’s thankfully the case that the researchers here at Blue Cova are very good at this better than I am. So what will often get in the room? And somebody will ask the question, should we continue to work on this? What are we going to get from this? And then it’s on those people who are sponsoring the project to justify what we’d expect to get from it. So that process of terminating or culling projects is super important. In terms of resource management, I mean, you know, that one is really hard to answer that there are always far more interesting ideas than you can possibly answer. So, you know, my job as head of research is to say, Are we delivering on the firm’s commercial objectives of building out strategies in the various areas of fixed income, made sure we’re focusing enough time on commercial tasks versus enough time on sort of bleeding edge true innovation?

Corey Hoffstein  10:49

So let’s set the table for the rest of this conversation, which is going to dive into the nitty gritty of scientific investing in fixed income. Can you describe at a high level for me some of the mandates you help oversee at Blue Cove?

Ralph Smith  11:03

Sure, yeah. Briefly, as I said, we’re a scientific active manager. We cover long short mandates, we cover long only mandates, we are solely doing fixed income in a systematic way. For us in terms of asset classes, that means liquid corporate credit, both developed markets and emerging markets, and then rates and going on in future that will mean more fixed income asset classes. And then in terms of how we bring that into the investment process, we have one investment process in terms of it being systematic from the start in terms of universe to data signals, risk modeling, etc. So all of our client portfolios are within the same investment process.

Corey Hoffstein  11:41

Maybe the best way to dive into this topic is from a footing that I’m more comfortable and familiar with, which is factor based equity investing. So let’s start with maybe a bit of a compare and contrast. How would you think about comparing quantitative equity factor investing versus a quantitative approach within fixed income? Well,

Ralph Smith  12:04

that’s a big question. I think maybe one place to start is just to think about the evolution of current equity investing, and then how fixed income investing has kind of compared to that. So at least my perspective on current equity investing goes back to the 1990s, the 2000s pioneers like granola and Karna, at BGI, as I mentioned, before, Sorenson, gn, Putnam panagora, etc. You know, Wells Fargo, that kind of origin of scientific equity, fixed income has been around a while, I would say 20 years plus from the BGI experience, and indeed, the market in terms of assets is as big as equities, the opportunity is as big, but the development has been much more gradual. And I think that’s because of greater complexity in terms of the modeling challenge, and perhaps even more importantly, a sort of lack of clean and available data. What do I mean by that, that complexity? I think it’s the thing that, you know, unlike equities with which there’s kind of one equity per firm, yes, there’s some corporate actions, but the data, or the structure is somewhat simple. I think on the fixed income side, obviously, there are multiple issuers, very many issuers, both private and public, each issuer can have one or more bonds at any given time. And those bonds are a finite life. So they get issued, they live for a while, and then they die. So if you think about that, that challenge of mapping and ordering the data, it’s a different order of magnitude to equity space. And then the bonds themselves obviously, differ. Fundamentally, you’ve got a set of cash flows, which are assumed fixed, although they can be risky. But you’ve got all sorts of different characteristics like coupon maturity, whether they’re callable or not, whether they’re senior or junior in the capital structure, etc. And you have to model that. And then there’s the question of liquidity and straight through processing. So my point is the complexity of the data set and the environment has caused fixed income, scientific or systematic investing to evolve slower. We think big, everything renascence is happening now, because data is more available. And there’s been a real, you know, uptick in the interest of scientific fixed income. And it’s just kind of getting going. Now. In terms of your question, though, how did the models actually differ, I would say, high level, folks that are familiar with equity modeling, we’ll find at least our approach to fixed income modeling pretty familiar. So at the high level three main elements, I would say a set of insights or signals is the first element. So we’re trying to forecast the returns of a security, then a risk model where we’re obviously trying to forecast the forward looking risk of any position or any portfolio. And then finally, a portfolio construction process or algorithm, where we’re trading off the expected returns from the insights, the risk on our portfolio, but we’re also taking into account transaction costs, liquidity, any other client constraints, so high level, it’s pretty similar to the equity models that you and the listeners will be familiar with. I think where it differs is all of those components, those three components and the detail Well, design choices within them are all different in fixed income than equities, I would say. And they’re different within different areas of fixed income. So as I said, here at Blue cave, we do corporate credit and rates, the design choices for insights for risk modeling, for liquidity, etc, are very different across those asset classes. So it’s a familiar pattern or a familiar structure, but with a very significant different implementation and devil in the detail.

Corey Hoffstein  15:25

Well, maybe we can dive into that a little, because I’m curious, given all those differences that drive a wedge between equity and fixed income that you mentioned, how does that knock into how the research process has to be different in response?

Ralph Smith  15:38

I think probably the biggest difference in the research process, what does the research process start with? For us? It doesn’t start with data. It doesn’t start with jumping into the back test. It starts with a process we call experiment design, where we say we firstly say for a given opportunity set, how do we want to trade so in corporate credit, we’re looking to take firm specific risk, we’re looking not to take market risk or sector risk or curve risk, generally, we’re looking to take firm specific risk. So what’s the trade? What’s the exposure that we want to explain? And then having said that, so then the question is what drives firms spreads, say, you’ve got to think fundamentally about what the drivers are, are, they’re sentiment factors that drive spreads, how to think about value, how to think about quality, defensive factors, etc. And although that the structure of that process is the same as in equities, the details of the mechanisms that drive corporate bonds versus equities will be different.

Corey Hoffstein  16:32

One of the things that you and blue Cove really stress is that you apply what you call a scientific approach to fixed income investing. And I’m curious, how do you see this as being meaningfully different than anyone else that says they take a quantitative approach?

Ralph Smith  16:48

I think the two are related, I think the best way to describe the scientific approach is to think about the mission or the vision behind the founding of blue Cove. And this is coming from our founders that set out this agenda back three or four years back when we founded the firm at heart, what is an investment management business? It’s managing knowledge, large amounts of financial information and its output is is investment decisions. And so their vision was to build a firm which is optimized to do that. And how do you do that best? Well, you use the scientific method, which is the being the best way of turning information into knowledge that humans have used for the last, say, 400 years. And it stood a great test of time in terms of producing the technology that we all use today. So here at Blue cave, I guess we think about the science concept of scientific at both the firm level and you know more in my domain, in terms of the research team. At the firm level, we consciously didn’t want to build a firm, which was a whole bunch of siloed investment processes in different areas, which is what a traditional investment management firm can look like. There was a conscious desire to build one investment process. So again, from kind of data sourcing and thinking about modeling of assets, through to signals through to risk modeling, and portfolio construction, and the feedback loops, reporting, execution, etc. And take that one investment process and use that everywhere for all products. And then you’ve got the decision of how you do those individual elements. And we want to use technology, obviously, as far as possible, where technology has the edge, which is in manipulation of modeling of information. And we want to use humans where they have the edge, which is in terms of interpretation, and assessment of the outcome. So we don’t operate a fully process without any hands on the wheel, we use human insight to look at the outcome of models, to gauge whether it makes sense. We also critically use humans to build the investment process. So the key thing I think, at the firm level is using technology and people in the right ways to optimally harness information and turn it into decisions. At the research level, as I was kind of saying before, it’s down to that subtle difference between just being systematic and being scientific. And I think I would probably express that as being hypothesis driven. So as I was saying before, when we start a research project, we don’t go to the data, we don’t start looking at back tests, we try and set out our rationale, economic, financial or otherwise, for why a signal should work. Why should this particular signal of drive spreads or driving expected returns? Having got that hypothesis, we then look to test that on data. And that is the scientific method. So that’s really what we mean by being scientific. Just one more observation. I guess this is a familiar theme that comes back through these conversations. We recognize that back testing is a powerful tool. It is even a kind of defining tool of systematic systematic investing, but it’s not a perfect tool. Hence, I would say an important element of being scientific is to recognize the limitations of your setup, the fact we can back test, but we only have limited data. We don’t have multiple samples. So being very appropriately cautious and evidence driven is part of it.

Corey Hoffstein  19:54

In the pre call we had in preparation for this episode. You mentioned that While there might be a similarity or even overlap in the factors that exist in both equity and fixed income, you think it’s really important to define and defend the market mechanisms for why they exist in fixed income. Can you explain what you mean by that? And maybe provide an example?

Ralph Smith  20:18

Sure. Yeah. I mean, I think I think fundamental to the research process to building a systematic strategy is to articulate the mechanisms driving returns. And as you said, there are some overlaps between fixed income equity. But I think it’s important to start with first principles, and just, you know, uncover and unpick those mechanisms as far as possible. So, as I said, we do two areas here corporate credit and rates, I think the closest relation to equities is corporate credit, where you know, and what drives the corporate bond price? Well, if we take out the rates element, which is what we do by default in all our modeling, then you look, you’re left with a credit spread. Now, you could say that credit spread is similar to the equity price in sense that good news for the firm in terms of earnings or cash flows will have a positive impact on the spread, it’ll narrow spreads, and it’ll make the equity price rally. So you could say, well, there are some strong similarities there. But looking in more detail, the corporate bond investor and the equity investor have a different claims on the firm’s assets, the corporate bond investor has much more limited upside than the equity investor. And they get compensated for that interesting upside by earning a spread. So although some of the earnings information is common, the reaction to it may not be there’ll be an asymmetry, good news for the equity investor or for the equity will carry on being good news. And there’s not really a limit to the upside there. Whereas in corporate bond terms, there’s a limit to the upside in terms of, well, once the earnings picture is really good, or the cash flow picture is really good, there’s a limit to how much the spread can narrow, because it’s also driven by other things like liquidity premium. And there are also things like call ability, which stop a bonds price rallying above or significantly above par. So again, it’s about articulating the mechanisms in detail that drive the assets price. And that goes for signals, insights. And it also goes for any other component in the investment process that I described. So risk models, I think valuation is a massive area. So when we’re thinking about valuation in equities, I guess we’re thinking about some relatively simple price ratio based models. On the fixed income side, it’s very different, you got to think about the risk of the firm, its leverage how far it is from default. And then any one of a number of structural models that help you gauge what what the fair spread is. So again, it comes back to this thesis of a common framework in terms of thinking about signal categories, perhaps, but always the differences, the detail has been different. So another area that I think is very important, and very interesting to think about in terms of systematic fixed income is the actions of particular market participants. What do they do in terms of infecting corporate bond prices? So there’s a paper that we wrote called reaching for safety. Back in my Blackrock days, I wrote it with a guy called John Kang who did most of the work, so I should give him a hat tip. And there’s a decent literature on this topic, actually, not just our paper. So this paper explores the impact of ratings constraint investors on corporate bond pricing. So if we take insurance companies, for example, which are some of the biggest holders of corporate bonds, they are typically constrained in terms of the exposure to lower rated names, they can take a face higher capital requirements safe for taking on low rated exposure, so tend to avoid low rated bonds, but they also need yield. So they want to own the higher yielding bonds within the decently rated groups. And now, Why would some bonds in a given ratings group be higher yielding, it could be random? Sure, but it’s more likely that there is some extra risk for that name that the market has priced, and that the consequence of that is a higher yield or higher spread. But by focusing only on rating, the investor is ignoring that and saying, hey, I can get some carrier, I can get some yield. So the consequence of that investor ignoring any other measures of risk than rating, treating the ratings as a whole proxy for risk is those those bonds will be attractive, their prices will get get bid up, and their spreads will be lower. And that means that these bonds, which should trade right, for a reason will be rich. And the consequence of that is if you actually realize that dynamic holds, if you avoid those bonds, which will have lower risk adjusted returns, you can get some alpha in corporate bond space. So I think that’s just that’s just a nice example of a market structural effects where investors with with ratings constraints that are motivated to get as much yield as possible or create a mispricing which can be exploited by investors who are realized the underlying mechanism. And I think fixed income is full of whether it’d be the right area or the credit area of situations where the actions are one or other types of market participants will create alpha opportunities.

Corey Hoffstein  24:51

So sort of in the same vein, though, not specifically talking about Alpha opportunities, just sort of the structure of the market when we look at the corporate credit A market we’re talking about a set of securities with potentially much greater breadth, but much less liquidity in comparison to say the rates market that might have a much more limited breadth and scope of the universe but much deeper and deeper liquidity. I’m curious how does that affect the type of signals you can pursue or the strategies you want to implement when we’re talking about a credit mandate versus a rates mandate?

Ralph Smith  25:27

What you say is absolutely right. And it does influence how we build models. As as you say, corporate credit as natural breath, we have literally 1000s of issuers, and more 1000s of bonds, but each of them is lower liquidity. So the key there is to harness breath as far as possible by building a very diversified portfolio, but also to be aware of the liquidity or potential illiquidity of those positions, can you trade into their positions? And what’s the consequence of building a big position in terms of in terms of illiquidity? So the focus there is very much on liquidity modeling, optimization, transaction cost modeling, and exploiting the natural breath rates is a different area, there is more natural liquidity in government bonds in interest rate swaps and futures, etc. But there’s much less natural breadth. Why is that if you’re looking in developed market space, say interest rate swaps, you’ve probably got maximum of about 15 Different issuers a few more if you go to em space, but then you start going down the liquidity spectrum. So there are fewer, there’s much less issue of breath. So what do we do given that we want breath, it’s the single most important thing in the whole of investing to have breaths and diversification. We create it by building trade exposures, which are uncorrelated, or at least not fully correlated to each other. So in rates, for example, we trade the level of the yield curve. But we also actually, we put much more risk into trading slope curve, we’ve recently built a volatility strategy. And then there are plenty of other things that you can do to create uncorrelated trades, breakeven inflation is one of them, basis, etc. So the breadth that we create there is by building models across different types of rates exposure, and that the result of that is a process where you can you can create diversification, I think it’s worth mentioning at this point, that is a key advantage of systematic investing, where you can build a back test is you can show what these correlations actually are. So we wouldn’t be able to create as much breadth in Brightspace, without the ability to build models, and to back test them. That said, probably even with all that the breath is naturally less than corporate credit. And hence, you would expect a lower quality of returns in terms of information ratio, or IC than in corporate credit space. But there’s less risk and less need to model illiquidity. So that’s, that’s probably where the relative focuses our liquidity and optimization and credits, and their breath question on the right side. One of the

Corey Hoffstein  27:52

other subtle nuances that came up when we were having our pre call was talking about market neutral strategies and fixed income, and just what it even means to be beta neutral, right? Something I don’t think a lot about in the equity space, because equity betas sort of well defined and agreed upon. But when I suddenly shift my focus to fixed income, whether it’s a credit mandate or a rates mandate, the definition of beta suddenly becomes a little bit fuzzier. It could be level, it could be key rates, it could be slope and curvature, or it could be different sectors of the credit market that are important. How do you think about defining beta for a market neutral strategy?

Ralph Smith  28:36

Oh, that’s a big question. Again, I think it’s going to end with my kind of theme, which I think it differs depending on the particular area of fixed income you’re looking at. On the corporate credit side, I think the first decision you have to take for market neutral strategy is whether you want to include the rates component of a corporate bond, as I said before, we think it’s more convenient not to say if you think the underlying instrument, it’s a corporate bond with a rates hedge. So then, having done that you’ve got a universe of similar corporate bonds on a rates hedge basis. The question you’ve got to answer then is what are the common factors within those corporate bonds? The way we do it is we define a market factor. We find sector factors, curve factors and other common factors. And we use something that your listeners might be familiar with a bearer style, cross sectional factor risk approach that we’ve built in house here to extract those factor returns. So once we’ve done that, we’ve got a decomposition of the returns of a bond into its spread component, its rates component, and then for its spread component, we can decompose that into its factors, market factor, sector, curve, etc. Having done all of that, we can then answer the betta question. And depending on how you define it, the beta really is is defined by the factor returns really so it’s either just the market beater or it’s all of those common factors and how they co vary with market portfolio. That’s very different from just doing a statistical beater against the global lag. It’s different but because we’re using a factor model, and we’re passing out the various return drivers. And it’s different because the global lag itself is an extremely rates heavy, high quality corporate bond plus a big chunk of treasuries benchmark. So probably isn’t isn’t that relevant for our high yield investing, or it’s not the best choice of theta for, for high yield strategy.

Corey Hoffstein  30:19

One of the considerations that you’ve brought up a couple of times now is the liquidity issue in fixed income. And I want to dive into that directly. Because again, it’s so different than the world of equities. If you and I are equity managers, and we both want to buy Coca Cola stock, for the most part, we’re going to acquire the exact same shares as each other. And we can acquire just about as much as we want, unless we’re some of the largest institutions in the world. The same can’t be said in the world of fixed income. It’s entirely possible, for example, that the bond your model wants you to buy is not even available in the marketplace. So how do you deal with these sort of liquidity considerations? Both from a research and back test perspective, as well as from a forward portfolio construction perspective?

Ralph Smith  31:10

It’s a great question. It’s something we spend a lot of time thinking about. And again, as I was saying before, probably the principal focus is in corporate credit. In summary, you’ve got to regard liquidity is uncertain. It’s therefore something you shouldn’t look to measure exactly. It’s not a question of looking on a screen and saying, Oh, well, liquidity is this, you’ve got to model it, you’ve got to treat it as a probabilistic process. So I guess what I have in mind is there is a certain probability of a bond trading tomorrow, and that it’s that latent probability you’re trying to forecast. And if you can forecast that, well, then there’s more chance that when your model generates a trade, you’re actually able to execute it. So how do we approach that? You got to think about a liquidity model for each particular area. So not just all corporate bonds as a whole, but it’s a European bonds different from us bonds, IG bonds differently, liquid to high yield, how about emerging markets or be specific, there are obvious drivers have liquidity at the bottom level. So newer bonds trade more, because they’ve just been issued, and there’s lots of activity, older bonds tend to get locked up with institutional investors, then there are patterns of liquidity across sectors with spread. And obviously, large firms have more liquid bonds than than smaller firms. I think probably more fundamentally than that having sort of modeled the characteristics at the bond level, you’ve got to decide if a bond doesn’t trade too much over the recent period. Is it because the bond is just illiquid? Or was it because there was no particular interest in in it, but it would have been available to trade if somebody had wanted to trade it. And that’s where dealer quotes can come in. So there’s another decision to make about how much to reflect model based on the past liquidity, versus how much to take into account to dealer quotes. And using a mixture of those things, I think we would believe is probably the ideal, then there are questions like how much liquidity varies through time, both on a structural basis, as in generally there’s been improving liquidity and fixing over time, and the mechanisms associated with that. So the presence of ETFs, which tend to increase liquidity, but also what happens during shock periods. So I think you have to have an answer to all of those questions before you build a reasonable lead accurate liquidity model. But then you also have to recognize that it won’t be fully accurate with time, the only way to really test these things properly is to use them out of sample. There are probably a couple more points, if we’re thinking about back tests, building strategies. The question is how aggressive to make the model. So we have this trade off, which we typically think about, you could build a model only trading the most liquid instruments according to your model, in which case, the chance of being able to trade that and get the trades on that the model wants to do would be maximized, and everyone would have an easy life, but there’s probably less alpha in that. Or you can build a more aggressive model, which goes down the liquidity spectrum, which means there’s, it’s going to be harder to follow the model in practice, that transfer coefficient won’t be 100%. By transfer coefficient, I mean, the correlation between actual positions and model positions, but there will be more alpha in it. And that’s the debate we have all the time. One more point on back testing, you’ve got to be aware of whether you’re being optimistic. So let’s say we do a back test with our liquidity model. And we assume that the model is giving us a good guide to what liquidity is there. But that could be wrong. So it’s important to have a robustness concept from the point of view of backtest. Or what if what if liquidity turns out to be lower than we’ve assumed, according to our model? What’s the impact on back testing? And you’ve got to be in a position I think, to answer those questions. And then I think finally, something that we spend quite a lot of time on here at Blue Cove is the feedback loop. So we have a process called Model Management and we look at lots of things in that. But one of them is questions around liquidity and transaction costs. So how are good are our models in terms of forecasting liquidity? on how many occasions did we look for liquidity and it wasn’t there? How well are our transaction cost models doing, etc. And then taking that analysis and feeding it back into the design is obviously key in terms of evolving those models to be as good as possible.

Corey Hoffstein  35:13

from a research perspective, fixed income strikes me as having both a big data and a small data problem. It’s sort of this big data problem, because the number of bond issues dwarfs, for example, the number of equity issues out there, but it’s also a small data problem, because you constantly have all these new issues coming to mark it all the time, that have no historical data. How do you think this presents unique challenges? Or perhaps even opportunities to quantitative investing? And how do you tackle them?

Ralph Smith  35:49

You’re right, it’s both it is challenges in terms of doing the work, but it’s opportunities because not everyone will be in a position to tackle those challenges. I think on the big data part is absolutely right. There are lots and lots of bonds out there with their finite life that I was talking about before and their characteristic details. And we spend, we spent at the start of the firm, a lot of time on getting the details and the mapping, right. So looking across the universe of corporate bonds, you got to make sure that they’re mapped to the right issuer, that’s the issue who’s going to provide the economic support, not just the subsidiary. So there’s some degree of complexity in that, you got to think about getting the characteristics right, like the domicile, the call ability, anything else that’s unusual about the bond, green bonds these days, etc. Having got all of those characteristics correct, you also got a map to the equity because there’s information from the equity market that can be very useful, and clearly with the presence of corporate actions, etc, that process isn’t always simple. So once you’ve done that work, you can kind of solve the Big Data Challenge. And then probably the next area in which the Big Data Challenge kind of appears, is in the in terms of optimization. So some folks take simple ways around this a lot of the academic papers to pick a representative bond for each issuer that might vary through time, but they always have a one to one mapping between issuer and bond, we chose not to do that, because we wanted to model the whole spectrum and go back going back to your point about Coca Cola stock before and whether there might be liquidity in one bond or the other. If you only model one bond, then you can’t really say what happens when liquidity dries up in that. So we decided to model all bonds in the universe. But the consequence of that is you have to decide whether your model is going to take across issuer exposure. So exposure between issuers, or whether it’s going to trade bonds within the issuer, it’s going going long, one bond short and other one for the same issue. And getting that balance right is another sort of big data complexity. And then on the new issues point, you raise kind of like the small data, yes, we have that issue, that when a new issue comes along, that it’s going to be released tomorrow, we need to know whether we’re going to buy it, whether we’re going to go short it and you have to make that decision without any data, we have to decide whether it’s this value, whether it’s going to be printed expensive when it’s issued. And maybe more importantly, we have to be able to model the risk of that bond. And I guess that that’s actually what makes the factor risk approach that I was talking about key, because we’re factor risk models, as you know, as opposed to statistical risk models. So you don’t actually have to have a price history. All you need is the bonds characteristics. In order to model it’s factor risk, at least. So that’s the way that we solve that challenge is to employ factor models rather than statistical models.

Corey Hoffstein  38:30

And know that from an actual portfolio construction perspective, a lot of your process boils down into a grand optimization. Can you walk us through a little bit how you think about designing the objective function for that optimization?

Ralph Smith  38:45

Sure, it’s really the three elements that I was talking about before, so it’s expected returns. So we’re obviously looking to maximize expected returns, but minus a risk penalty minus a penalty for transaction cost. And then in terms of the rest of the optimizations, constraints, and penalties, they will reflect individual asset bounds, liquidity, both in terms of position liquidity and trade liquidity, how quickly can we build the positions and how a pickup position should be at maximum? So those are the components of the objective function. Overall,

Corey Hoffstein  39:19

optimization is sort of notoriously famous for its instability, you know, arising from poorly defined objective functions or just input noise or numerically unstable variance covariance matrices. How do you think about addressing these issues, particularly given the vast dimensionality of the investment universe that you’re tackling? I think going

Ralph Smith  39:43

back to the risk modeling point factor models are key there. So getting some structure and managing the correlations in your in your risk model in your covariance matrix is key to avoiding some of the worst issues of Optos as you’ve described. The factor risk modeling process which effectively describes the columns factor structure across bonds via their common loadings on on readily interpretable. Indeed, designed factors is a great way to kind of supervise and shrink that covariance matrix, which then takes away a lot of the issue around high correlations, which then can cause those unstable solutions that you were talking about, then I think, it’s wrong to imagine or at least we believe it’s wrong to imagine you can only do one stage optimization. So we have a number of phases in optimization designed to effectively improve the result from the initial, the initial run, and to handle things like the trade off between within issuer risk and cross issuer risk, as we were talking about before, to handle things like minimum trades, maximum trades, etc. That’s kind of the detail of how we think about it, I think sort of more fundamentally, is in terms of the culture and the process. Our models are always designed as hopefully is kind of clear now in terms of how I’ve been talking to be transparent. So the signals have an economic rationale to them. The risk model is expressed in terms of, you know, readily observable, meaningful factors, not statistical factors. So by keeping the components pretty simple and transparent, we can always interrogate the final portfolio. And we can say why the optimizer likes these trades and doesn’t like those buttons, we can attribute that back to signals. And it won’t be surprising that the optimizer is not a linear sum of its inputs. But it is something that can really explain, you know what the moving parts are. So from day to day, you can track that back. And I think a lot of that helps to reduce the uncertainty that an optimization process might otherwise generate.

Corey Hoffstein  41:35

Given everything that we’ve spoken about already, how does the process change? And maybe what what are the impacts? When we move from talking about long short mandates to long only mandates,

Ralph Smith  41:48

I would say overall, the process is the same. It’s the same building blocks, as I said before, so insights signals, risk model portfolio construction, it’s just that some of the settings differ. Now, I guess, even then, the differences aren’t huge, because between the long shorts mandate, which is looking, which has a benchmark of zero and is looking to maximize expected return, and a long only mandate, which is relative to benchmark, those problems are actually pretty similar. If you’re if you’re managing versus a benchmark, you’re looking to build a set of active weights which sum to zero, which maximize expected return and minimize risk relative to the benchmark. So I would say long, short and benchmark relative long only are pretty similar. Yes, they differ in terms of the constraints. So obviously, the no short constraint in the benchmark does change things a little bit, I would say where it differs is if you think about a long only portfolio, which is absolute return to there, your objective is just let’s just generate the best Sharpe ratio possible, not relative to a benchmark. Typically, those mandates will take a much higher degree of factor risk than security selection risk, there are other mandates she would take. So by default, in a long, short mandate, we’ll take very little factor risk, and most of the alpha will be security specific, and it’ll be market neutral in not only in market exposure, but also sector curve, etc. In terms of an absolute return mandate, it’s a different challenge, it’s about mixing in those factor returns a return from markets, either tilt or timing, a bit of both return from Sector tilts from curve, as well as that security selection piece. So there, the design process is that the components are the same, but the design process is a little bit different. And the trade offs are interesting, I would say, clearly, the market neutral product is the more pure product from the point of view of being uncorrelated to other market factors, high art, etc. But the consequence of that approach is much more leverage per unit risk. So you can’t get as much scale, more tea cost as well, because you’re hedging out the market factor all the time. Whereas in the absolute return mandate, by getting the balance between factor risk and security selection risk, right, you can get a pretty nice combination lb with correlation to the market that is lower in terms of leverage, and in terms of transaction cost. And I should probably note that the nature of our security selection in incites tends to be a little bit defensive. So they they combined quite nicely with some market exposure in terms of evening out drawdowns. I know you have some pretty strong feelings about back testing, and both the right and wrong way of doing it as well as the appropriate and inappropriate use within the research process. So I’m going to just maybe set up a little bit of a soapbox here and give you the platform and let you share your thoughts. Oh, it’s a dangerous thing to do. But yeah, you’re right. I do have lots of thoughts on it, having seen it done sort of well and badly over my career. I think as I said before, probably my main point is back testing. The concept of the back test is a tool. Like any tool, it’s not a perfect thing. It should be used carefully to avoid data mining really That’s That’s it in a nutshell. So as I mentioned before, that experiment design process is key to talk about the idea before you go into back test. Why because we only have one data sample it would be different If we go to Mars or mercury and get a different history of corporate bonds, but we don’t have that sample period has typically been looked at before. And although you can use all sorts of controls, like in and out of sample analysis, those things will always be imperfect. So have some sense that that back testing is potentially a dangerous thing. We try and do ancillary testing where possible, so we try and prove out the mechanism of that idea, rather than just jump straight to the back test. So to give an example of that, if we think we’ve got a new signal, which can forecast say, future cash flows for a firm, and if it’s positive, therefore, cash flow should improve and spreads should narrow, don’t just take the signal and back, test it, do the ancillary analysis and say, Well does that that claimed signal actually Forecast Cash Flows. And if you’ve, if you can do that, you’ve proven the mechanism, not just the back test results, it’s far more likely you’re going to understand your idea better that way. Another area, probably that’s very big, it’s to think about the bias inherited from the back test period that you have available. So as I was saying before, the data quality and fixed income, especially in corporate credit is not great, we have limited samples. And you’ve got to be aware that your sample is just a sample, it doesn’t necessarily define all that can happen. And you’ve got to be conscious of regimes, both within your data. And in regimes that can occur outside of that period. No perfect fix to that, you’ve just got to be aware of it. What do we do here, we have this process, as I mentioned, that we call Model Management where we really looking to kind of supervise or monitor the models in terms of back tests and actual and try and capture or catch any issues that might have arisen from the issues around back testing. So we would look to answer two questions. Firstly, is the strategy performing as expected given the back test? And then secondly, is the market environment the way in different to the back test period? Or is it more similar to particular parts of the back test period. And between those two tests, we can kind of capture on the one hand, you know, any potential data mining a the strategy is underperforming, its expectations. And then secondly, we can monitor whether the back test period is different to the current one, or the weather the current period is different to the practice one, I would say, all strategies have out of model risk, because not every event is quantifiable. And some regimes are rare events. And this this is a problem faced by the way, not just by systematic investors, but by every investor because every investor draws on history and experience. And if the regime changes, every investor can be wrong footed by that, I would say systematic strategy is given the drawing back test quite a lot, I probably need to be especially conscious of this issue. So then, we’ll go on in our model management process to think about, well, if the market environment has changed, if it looks characteristically different, then how do we expect it to impact the model and what changes to the model do we think might be necessary? So to give an example, from corporate credit, we’ve lived it seems a long, long time ago now given me where equity markets are in recent volatility, but we’ve lived through a period of very low dispersion in spreads due to the post COVID intervention by central banks and etc. that low spread environment leads to low dispersion, low changes in spreads. And that for our corporate credit models means a significantly reduced opportunity set much more tea costs per vial signals don’t work as well, etc. So, in facing that period, we asked ourselves the question, which period in history is this similar to? And what can we learn from that, and the period leading up to the GFC, was somewhat similar 2006 and low spread period, everything risk seemed low, there was not too much opportunity. So I think by intelligent use of that back test period, you can kind of learn lessons that should tell you what to do, what are some things you can do now in order to improve your models? What did we end up doing? We shifted risk budgets around a little bit towards where there was more dispersion, and away from areas where there was less dispersion. And we also recognize that during a period of low spreads some sort of low natural opportunity, the issue of out of model risk, which for us can mean the drivers of corporate bonds, which are sort of news related corporate action related m&a, that kind of thing was in some areas more likely to occur. So we shifted risk a little bit via tightening constraints away from there as we thought were vulnerable. So it’s a learning process really knowing the deficiencies of your back test and thinking about regimes. And I would say, it’s probably as my final soapbox point, you can’t be agnostic to the macro environment, you have to even if you prefer to spend your time to building all sorts of lovely models, you have to think about the real economy, what that means in terms of spread drivers. So as I was saying before, intervention lowered, spreads lower dispersion, but we knew even while that was happening, there will be a time that the central banks and the Fed especially with take away their support for markets spread would widen. Firms wouldn’t be able to finance so easily financial conditions would be tighter, and that higher dispersion period would return. So you have to invest time in that forward looking macro view, which helps us on the credit side and indeed on the right side.

Corey Hoffstein  50:06

Well, you’ve teed me up perfectly, because that’s exactly where I wanted to go next, which was this idea of how is the scientific process that you employ challenged by an ever evolving macro economic landscape? I mean, you touched on many of the points that I had in mind, you’ve had over the last several decades, Central Bank’s become more and more involved in both rates and credit markets, very active players in the fixed income arena, through their quantitative easing, quantitative tightening programs, and then during periods, like the COVID crisis become very active players in the credit market, which we would presume would have structural impacts on people’s views on how tight or wide spreads should be. And now more recently, you suddenly have inflation as a very real risk to markets, which is a very important premium within fixed income, whether it’s baked into the term spread or other spreads. But the problem, I guess, from a scientific investing perspective, is you have very little data or even relevant precedent to explore. So how do you attack these problems?

Ralph Smith  51:18

I think that’s a great question. I think I covered some of it before, but I think it’s important to kind of go into a bit more detail there. I would want to reiterate that these problems face all investors are not unique to systematic investors. But we do have to think about not regard the back test period, is that by default, as a good forecast of what’s going to come? And how do you decide which regime you’re likely to live through, you have to have that macro view, as I was saying before. So I mean, if you think about the list of things you mentioned, I think quantitative easing the various experiments with low rate policy that we’ve seen over the last few years, that was all, always a learning experience, as we saw it, I think now we’ve seen pretty much dare I say, the full central bank toolbox. But what the current theme is, obviously, high inflation, stagflation and potentially looming recession. I can talk a bit about that how we’re thinking about that across the business. It’s a very unusual market environment and unusual indeed, in terms of the backtest periods we have available. So the the high inflation period, the stagflation airy period in rates means much higher than than usual volatility in the yield curve level, in the shape, much more violent moves of the curves, much flatter long end curves, especially in Europe, generally increasing dispersion in credit, as I saying, it means at the moment, having gone through a period of easy financial conditions and low spreads, it means much tighter financial conditions. And then the recessionary parts means poor fundamentals, and those do things for firms then mean wider spreads, because lack of financing and wider spreads because of lower quality of balance sheet, so more dispersion. So the current period provides kind of, you know, both sort of challenge a little bit on the right side, and lots of opportunity on the credit side. So it’s rare, that environment brings always positive or always negative, how do we handle it. And as I was saying, before, it’s about understanding the strategy drivers and the economic rationale, and not just being data driven. That’s the only way to kind of understand what a new regime might bring. And then a little bit more detail. Thinking about our rates area, which I haven’t spoken too much about the way we think about sort of strategies there is broadly into groups. So relative value kind of convergence type strategies on the slope and curve. And then sort of directional positive convexity strategies. Clearly, from what I’ve said, the impact of the current environment on those strategy types is going to be different, the former war will suffer, and the latter will benefit. So I guess that takes you to one immediate point, which is risk budgeting and diversification across different strategy types is the principle I think, in finance that I would always cleave to make sure that you don’t have all strategies to one type that you can diversify across across different types. There is a need to be creative in terms of the backtest period. So I think I said before, you can only get liquid swaps data, which is where we build most of our rate strategies, going back to say 2000, maybe a little bit more, it’s generally a low inflation period. But one can think creatively about what data is available in the US there’s rich yield curve data, maybe not in interest rate swaps, but in cash bonds, that goes back to the 60s even. And we’ve obviously seen plenty of high inflation periods in the US and other countries in the 70s 80s. And even in the in the early 90s. So, okay, we can’t get full cross market data. But we’ve definitely got plenty of instances of how yield curves evolve in an inflationary environment, which we can learn from in terms of modeling, but you have to be creative in your in your use of backtest. And thinking about how you generalize lessons that one market might teach you.

Corey Hoffstein  54:47

In the late 2000s 10s, fixed income was being promoted as sort of one of the last frontiers of quant and yet fixed income never really seem to get its smart beta moment. The way equity factor investing did. Why do you think this was the case?

Ralph Smith  55:05

I think it probably goes back to that point I mentioned about complexity, it just is a bigger challenge in terms of being a manager in the space to actually get set up, because of the difficulty in terms of getting the data and probably a little bit in terms of expertise, because some of the things that I talked about before in terms of building models in systematic fixed income, do require specialist skill sets. So you obviously need the current skill set in terms of building back test models, building risk models, etc. But you need fixed income relevant experience, you need to know how our balance sheet looks, you need to think about capital structure, you need to think about what drives the value of a firm, and structural modeling, say, for example, in credit. And finally, you need to have people that understand the analytics of CDs of cash, bonds, etc. So that that sort of work that goes on the background to kind of derive returns and exposures for all the bonds. So you need those kinds of things in place to get table stakes, I guess, to be a systematic fixed income manager. I think it is ripe for being exploited the space and ripe for being managed to systematically and there are definitely a growing number of entrants that we’ve seen, not just blue code for plenty of others in the last few years, you can see clients looking for that offering a little bit more, I think, okay, it isn’t a sufficient condition. But I think a necessary condition is that clients are willing to embrace managers beyond the traditional fixed income active offering. And indeed, you’ve seen a fast growth of passive investing in in fixed income, according to our figures, passive is growing probably about three times the rate the activists, so I think the preconditions are there of scientific active fixed income being attractive, we can also show that it’s traditional that it’s complementary to traditional investing in terms of in terms of low correlation. But those barriers to entry exist. And that’s why I think it’s been a somewhat slower process to build up, you kind of have to put enough energy into it to make sure that you can do all of those things I talked about and get to the end of it, it took us probably 18 months of r&d from founding the firm to have our first model, and therefore the number of entrants that are willing to do that is going to be limited. But in our view, it should be the dominant way to manage fixed income. I think most folks, I’m sure you would agree that most folks listening would agree that systematic investing in general, is the best way to invest. I don’t see that fixed income is any different, just because the complexities are there. That doesn’t mean they can’t be surmounted. And indeed, until fixed income, active investing is ubiquitous, those that complexity will probably mean that there are more inefficiencies. And then as I was talking about before, in terms of the tactical environment, it’s actually very promising the systematic strategies thrive off volatility in general, the fact that spread dispersion is increasing, and risk assets faltering, should probably mean, an increased focus on fixed income. And then where should people come? Well, the strategies that can extract alpha in those periods, and risk managed well, and again, systematic strategies are good at both of those. So I mean, bottom line, we would expect to see a significant interest from clients and from new providers in the space. And we very much welcome that last question of

Corey Hoffstein  58:11

the episode for you. And it’s the same question I’m asking everyone at the end of this season, which is to reflect back upon your career and answer what was the luckiest break you had?

Ralph Smith  58:23

Oh, gosh, I think, given how fascinating I find finance, I think the luckiest break, was even getting my first job there. Because I left university probably a little bit arrogantly having a math degree thinking, I know, this stuff can’t be that hard. It’s all math in the end, and knowing absolutely nothing about the financial industry, including the difference between the sell side and buy side. And I mentioned before the first firm I worked for, which was a kind of quasi scientific, I would say, boutique focusing on FX, they took this relatively nerdy, freshly minted mathematician, and they put me through some interviews and they were kind enough to offer me a job on the first day. So I think that was my, my break because I learned a lot from that experience, both in terms of kind of how to think about or basic entry to finance but also when models break down, when you know, markets jump in massive kind of ways in terms of fat tails, etc. So I think that’s probably the one I would cite.

Corey Hoffstein  59:23

Or Ralph, this has been fantastic. Thank you so much for joining me.

Ralph Smith  59:27

It’s been a real pleasure. Thank