Clayton Gillespie - A Fundamental View of Quant Equity (S7E5)

Hey everyone, Cory here. Thanks for tuning in to another episode of flirting with models. If you're enjoying the show, I'd greatly appreciate it if you take a moment to rate, review, and most importantly share with a friend. Word of mouth is how this podcast grows, and if you'd like to learn more about new found's platform of return stacked mutual funds, ETFs, and model portfolios, head over to returnstacks.com. Now, on with the show. You ready? I'm ready. All right, three, two, one. Let's jam.

Hello and welcome everyone, I'm Cory Hofstein, and this is flirting with models, the podcast that pulls back the curtain to discover the human factor behind the quantitative strategy. Cory Hofstein is the co-founder and chief investment officer of new found research. Due to industry regulations, he will not discuss any of new found research's funds on this podcast. All opinions expressed by podcast participants are solely their own opinion and do not reflect the opinion of new found research. This podcast is for informational purposes only and should not be relied upon as a basis for investment decisions. Clients of new found research may maintain positions in securities discussed in this podcast. For more information, visit thinknewfound.com.

My guest this episode is Clayton Gillespie, VP at Deutsche Bank, where he works in quant equity research for the QIS team. Clayton began his career at Credit Suisse, where he got his hands dirty and extracting fundamental information. This formative experience dramatically impacted how he views how fundamentals should be incorporated into quantitative equity strategies. Today at DB, he strives to improve quantitative equity strategies by anchoring them with a strong fundamental understanding.

We discuss how fundamental and statistical interpretations can be at odds, how a strong fundamental understanding can help with the identification of emergent risk factors during regime changes, and how best to incorporate fundamental insights while avoiding potential biases from the analysts that deliver them. Please enjoy my conversation with Clayton Gillespie.

Clayton, welcome to the show, excited to have you here. I think this is gonna be a great conversation that really provides the listeners with a different perspective on quant equity compared to many of the guests we've had in the past. So really excited, again, thank you so much for joining me.

Hello everyone, Corey. Thanks for having me on, my pleasure. Look, I want to start with your background because I think a lot of what informs your quant equity work comes from having somewhat of a differentiated background than the way most people get into quant equity. So why don't we start there, give the listeners a view of sort of where you started and how you got to where you are today.

Absolutely. So I guess I grew up on the other side of the fence. After University, I decided not to go straight into a quant role, actually went into a fundamental equity research role, and from there, I spent quite a few years really on the fundamental side, publishing on single stocks, talking to active managers and I worked for a team called Credit Suisse Holt, and the first thing they really get you to do is this thing called QC. So they have this big systematic equity valuation platform, where they try and model every stock really through one big model. And every year, as companies bring out new data, it's the juniors jobs right at the start of their career to sit down and crunch the spreadsheets, you've got to go through those annual reports, and you've got to update every adjustment that they're making. If they're capitalizing leases, you've got to work out what this company is spending on rent, what the company is borrowing in their lease debt numbers. You've got to think about how they're capitalizing R&D and so on. And that means that you have to go through hundreds and hundreds. I think I probably did a thousand annual reports, maybe more, and you're really getting to grips with what are these notes mean?

What are these companies doing? What is the accountant's way of thinking about how a company invests, how a company measures its return on capital, how a company pays its executives, and I really enjoyed that. I think I had a very good time there, and I really respected the fundamental skill set. But I also slightly lost interest in the kind of analysis that the fundamental world really thrives on, and I started to get more and more interested in things like programming and what used to be called, I guess, big data analytics. I remember thinking about that word a lot of the time but not really hearing it that much these days, and so I moved more and more into what was called then the quantum mental side, and I got involved in answering questions that required a bit of accounting expertise. But we're really trying to help active managers, sell side brokers think about things. Well, maybe more from a big data perspective. So, things like Holt had been capitalizing leases for decades, but how is it going to deal with IFRS 16 or changes to provisions for banks, or all of that kind of thing as those things came through.

And then I got asked to work on the team that was thinking about trying to improve their cost of capital framework. And that's what really introduced me to factor investing, right, because the standard CFA Damodaran type way of thinking about cost of capital really just relies on beta. But we know from factor investing, we know from the quantum world, that there's a lot out there that isn't just beta, and there's a lot more to be said in that world. And is there a way that we can try and think about? Bringing some of those insights into single stock investing. So, as I said, that's what really introduced me into factor investing, and then after a stint there, I actually jumped further into the quantum world, and I was contacted by Deutsche and they asked me to come and join their equity Qis team, so I'm now really more in the quantum world. That's where I am now, and I'm trying to bring my fundamental background into what they do day-to-day, into what we do day-to-day.

Well, my perception, based on our conversations and because of your background, is that you spend a lot more time than maybe most quants to trying to reconcile that viewpoint between quant and fundamental in things that don't always naturally line up, things that maybe we as quants don't even question, that might seem odd to someone taking a more fundamental view. And as an example, in our pre-call, one of the questions you asked was, well, how do you make Fama French make sense to people who pick one stock at a time, and there was a question I hadn't even ever considered, and I was hoping you could expand a little bit on what is the friction there from your perspective? Why is it so difficult to communicate, and then how would you actually overcome that conversation?

Sure, so I guess especially after the kind of COVID rallying in growth stocks, one thing that people asked a lot was, how can growth stocks have lower expected returns, aren't they riskier? And I get where they're coming from, really. They're thinking about risk in terms of beta, right? Generally, you're going to be expecting, especially a lot of the early-stage growth stocks, to be higher beta and to therefore be riskier, at least more volatile. But I think the best way to do it is really to try and take a perspective from a different asset class. One that I like to do in that context is really to think about bonds and yields, right? So if you've got a bond with a very low yield, you know that it's quite low risk, right? Generally, it's gonna be quite low risk.

And when you look at some of these Make a cap growth stocks, for example Yes on one hand you can point to their corporate debts and say Oh Amazon's raising capital at only few bits above treasuries. But also you can say well actually stop thinking about your P multiples and start thinking about your earnings yields flip that upside down. And say okay well Amazon's actually trading at really really tiny earnings yield and What kind of bonds does that make the stock of Amazon and you start to see how you can then think about? These kind of growth stocks trade on these really high multiples. It's actually being priced for very low expected returns.

I want to stick with that Amazon example because it's another one that you mentioned specifically in our pre-call is sometimes being one of those quant verse fundamental conundrums and this idea that you specifically brought up was as It changes from an early stage company into what has become this compounding wealth machine that increasing profitability actually increases the cost of capital and which was Candidly to me when you first mentioned it's something that didn't immediately click and so I was hoping you could explain how a factor versus Fundamental viewpoint again using Amazon as an example might cause conflict here.

Yeah, absolutely. I think this is a really nice example of why I like trying to blend fundamental and more Systematic approaches because it's a good example of where they're sort of multivariate dynamics and some of them are intuitive and some of them aren't. Very intuitive. I mean I was actually reading in Bloomberg Yesterday someone was recommending investing in small caps and then they said you should buy names with higher gross profitability to kind of reduce risk and You can see how that language is pervasive in thinking that higher profitability is is lower risk. And again, you can see where it comes from in sort of beta terms.

I actually heard Novy Marks talk on this on another podcast and he really emphasizes that it's important to put them together and what I mean when I say "them" as kind of value and Profitability at the same time because you don't get much out of comparing just two expected returns numbers, right? if I tell you that two stocks both have a p-multiple of 25 or Adding to the four if you want to flip the upside down. It doesn't tell me much about the company. You know, is this a growth company? With like reasonable expectations or is this a kind of quality company with a lot of stability priced in or it's quite hard to measure to understand the difference between the two and I think it makes more sense especially in the kind of whole frameworks again drawing my background of things like fade and life cycle and what I mean when I say Fade is the ability of a kind of quality company. I guess you can keep thinking about Amazon to be able to Continually compound and produce the same return on capital year after year after year rather than reversing to cost of capital reverting to average.

And in that sense when you're buying that stock one risk that you're taking is the chance that it might not be quality. It's a chance that fade might happen very quickly. You might lose that quality and that kind of insurance that when the market goes down that company's still going to be producing profit. Doesn't pay off and I think life cycle in the sense that as the company matures it goes from growth To quality to value and it's what cost of capital rises as its tech gets old and as competition comes in you know starts out very kind of venture capital world and then as it breaks through into Having product market dominance it's able to really put years and years of good profitability in and then over time it fades out as it becomes too big and bureaucratic and New tech comes in and I think expanding on that when I try and explain it to Non-equity quants.

I like to think about covered calls. So let's imagine that you're choosing between an early stage growth stock and a quality growth stock. And the difference is that maybe the early stage stock really still has the potential to go really, really big, that really high upside potential, take over the world if you want. Whereas actually the quality stock probably doesn't have that potential anymore. It's probably realized a bit below that, so you can kind of think of buying the quality stock as a bit like buying the growth stock and selling a really out-of-the-money call: you lose that potential of the kind of super upside. But in return you get a little bit more yield, which equates to a little higher expected return and a lower multiple.

When I asked you to summarize the overarching theme of your work, you said that your guiding question was quote, "How does this fit into my bottom-up valuation," end quote. I was hoping you could explain why this question is so important in guiding your research and maybe provide some examples of how it has actually come to inform the research.

Sure, so I try and use this question to really anchor my thinking, and I think that helps protect against data mining. But what do I mean by it, though? Well, let's keep going with quality. So let's say a fundamental analyst comes to me and he says, "I think this stock is going to go up because the markets underappreciating quality." What I'll say to him or her is, "Do you mean cash flow is going up at the same multiple, or that the multiple is wrong and the multiple needs to go up? In very simple terms, your price is your earnings times your earnings multiple. Is it that the earnings is too low or that the price-earnings multiple is too low?" And just really being able to force that question on my framework is quite helpful when I've got this nice data point that correlates very well with quality.

What about that data point is really capturing where the multiple is wrong about this company, or where the cash flow is underappreciated about this company? Maybe ESG is another good example. I mean, I'm not the first to point out that a lot of research was produced that said ESG stocks are both lower risk and higher returning at the same time. But the question was, are you saying that the market is underpricing the cash flow risk? Let's say the risk of a carbon tax, or are you saying that ESG funds are going to flow into this stock when they realize how green it is, right? That the multiple is underappreciated and there's going to be a sudden wave of new buyers.

I actually think that's a good example of where maybe fundamental and quants didn't talk to each other very well, and it slightly ended up a bit messy, right? I looked at the fundamental research and I thought, "Oh, these guys aren't thinking well about expected returns, right? This is some of this stuff doesn't follow very logically." But then I turn around and look at some of the stuff coming out the quant side and you think, "Oh, I'm not sure they've thought really about how this is impacting cash flows. Is this a framework of how much?"

Climate change is going to impact this company and hurt its ability to function and make profits or is this a framework of how much this company is going to be hit by changes that are going to stop that and I think there's got a lot better over time. But it's really a good example of where you need a blend of being able to talk to both sides and get them to collaborate productively and in a specialized way.

Well, I interpret a lot of your focus as a quant as getting a better understanding of some of these nitty-gritty fundamental details but one of the views I've always had is that quants have largely avoided having to do that because they're able to just diversify away granularity risk. They're able to look at a large cross-section and use diversification to avoid having to get into that minutiae, and my thought here is that if you spend more and more time getting into those little tiny important definitions, where do we end up drawing the line between quant and fundamental discretionary?

Sure. So I think the boring answer is a practical one: you're a quant when you can't intervene to change the rules and your discretionary when you can. But philosophically, I think it is a better answer. For me, you're a quant when you can't guess your signals anymore. So if you give me a stock, right? Let's say let's take Tesla that's been in the news today. I can think of Tesla and I can go okay, that's medium to high quality, especially if you add back R&D and maintenance capex as we do. It's gonna be a great stock. It's got pretty good momentum, especially if you're taking first 11 months, and it's gonna be quite high beta, right? For me, that's quantamental because I'm able to guess those signals. But if you try and ask me what's Tesla's cross-sectional reversion score? I have no idea and I don't know anyone who would have the brain to be able to guess that, especially in a kind of global or developed markets way.

And that's really where you're sort of starting to move between the two. In terms of your first question, can you not just sort of diversify away granularity risk? Well, I think the point is that you can always do better by trying to take those little bits into account, right? Obviously, if I'm the only person and I'm running money by myself, it's gonna be a bit lower priority. You've got to prioritize these things, but I think you can do better by using kind of yet collaborative specialization. And one thing I read in Peter Thiel's zero to one actually that's really stuck with me, he points out that a tiny computer can sum a million numbers very easily but you need like a supercomputer to recognize a cat in a photo, right? Whereas if you take it a tiny human, right to take a baby. It has no hope of summing five numbers at once, let alone five million, but it can recognize a cat in a photo, right? So there are quite obviously ways in which a computer is much better than a human and a human is much better than a baby and just trying to find the ways in which those work together. Well, when you're eking out that extra bit of alpha or that extra bit of craftsmanship as they say, that's really where I think the improvements can come.

As a follow-up to that question. How much of those nitty-gritty details are cross-sectional details, cross-sectional improvements versus idiosyncratic improvements, something truly unique about a single company that has to be adjusted? So I always say that one of the ways which I really see fundamental analysts helping quant analysts isn't kind of identifying the like false positives. So one story that I love to talk about is Adobe, where they moved probably a decade ago now. From a very kind of we sell this product for a certain number of dollars model to a subscription-based model and what you've seen is the stock just fly since then. It's been one of the best business model changes I'm sure it's gonna be talked about in business schools for decades to come. But it's retirement capital collapses, right?

Because in the short term you go from getting $90 a year from your key customers to getting nine because now they're on the cloud subscription model. And it's those kind of instances where I think the fundamental guy has to come in and say look, You're actually under appreciating the quality of this company because you're taking a very short-term view on the PV of its future cash flows. and that's a nice example where you have to either start forecasting longer into the future or you have to start trying to do like a model on top where you say right, if all these subscription models were converted into non-subscription models, What would that look like?

I think that's a good way of harmonizing the idea of fundamental versus statistical views. Goes well beyond just looking at individual companies. You gave me a really interesting example in our pre-call where you talked about how equity reversion strategies for example, are a case where there has been a significant structural change in. The best way to apply that strategy, particularly is seen through the lens of fundamental versus statistical.

I was hoping you could talk a little bit about that as an alternative example here, as to how that strategy has changed and how implementation maybe has become more or less fundamental versus statistical over time.

Yeah, absolutely. So, I love this story because I think it really gets at what I find interesting about finance. And I think the way that you can tell the story of reversion is that it starts out as someone just being a bit opportunistic as a discretionary fund manager. This theme is overplayed, I'm gonna take profits on this company or the markets ever corrected for this with this stock. Here's a good entry point, right? Then you start to do a bit of pattern recognition, right start to bring the quants in and they can identify those stocks more easily, maybe automate a bit of that research and you start to look at it on a kind of sector region relative basis, where you start to look at which stocks have out or underperformed their region sector counterparts and that works quite well through the noughties, right, that's actually quite profitable strategy in that sense.

But that stops working, right eventually that gets maybe arbitraged away. We can talk about crowding, whatever might be the reason and then you find that actually, this can be enhanced statistically using principal component analysis, right? So if you stop thinking about region and sector start thinking about principal component analysis, strip out enough PCAs and control for your signal well, then this becomes profitable again. And then you're starting to really get an exciting strategy. That's not correlated with other factors and it's kind of reasonably intuitive.

Now, I think the next step that people kind of expect is when PCA stops working we need to move on to the next statistical thing. But actually, I want to stop and say no, the next thing that really is coming out is how can we understand this better from a fundamental perspective. So my colleague John Paulo who built the strategy at DB published recently last year, some improvements to the strategy and one of the key ones is really making sure we take into account when a company's reporting earnings, right, and what's the move of that company on that day and not treating that as idiosyncratic, treating that as an earnings move.

And putting the market returns in instead and actually it's just a quite a nice example of where the guy has a postdoc at Yale, one of the smartest minds I know. But actually what's really making a difference to the strategy is understanding a bit more fundamentally, what's moving the market as a result of that stock on that day? What's coming back to my cash flow, my bottom-up valuation and then from there you start to go right, how do I understand the explainability of this factor?

Am I just shorting the latest trend or when I put this in a risk model it all falls out as idiosyncratic? So being able to understand the kind of companies that you're buying at each time. Requires a fundamental overlay and is actually quite important to feeling confident in the strategy. Well, another really interesting example of this that you brought up in one of our prior conversations was the emergence of different factors during COVID. And I've always had the bias that looking at latent statistical factors allowed you to potentially capture unidentified but emergent factor developments in the data without any bias.

And you brought up, I think again using COVID as a really fascinating example, that this is a case where fundamentals actually were really important to understanding what was actually emerging through that period. I was hoping you could share that example, explain from your view why a fundamental knowledge was so important to understanding the emergent factor structures during that period.

Yeah, absolutely. So I think that statistical shines when you have a lot of data and you have a lot of historical data; fundamental tends to shine when you don't really have a lot of historical data and there isn't a lot of pandemics in the historical data that we have to play with as quants. And so I think that when you've got a case of such an extreme, unexpected exposure, you really have to think about bringing in stuff that you can't capture easily statistically. And I think that there is some element that you're gonna be able to capture better statistically. Don't hit me wrong, I’m never gonna say that you don't want to use statistical measures, I always want to be saying use both, but I think it's quite important to recognize that things like GICs, I always like to joke that GICs is just one person's opinion. GICs can't really capture what's going on in COVID, right? You can't say that tech performed because Uber crashed and Zoom rallied. So it's very difficult to say, oh, well if I just look along the kind of axes that I'm used to looking at, I can see what's going on here and I can position. Well, no, you need to actually fundamentally sit down and get to grips with what's the exposure of this company to the stay-at-home trade, to the vaccine, to the pandemic.

I think another good example maybe is some of the Middle East trade routes disruptions that we're talking about at the moment, you know, that's a that's quite a good one I think where I'm sitting on a desk where there are a lot of fundamental analysts around me. I'm hearing 24-7 them get them on the phone to IR going how are you changing your trade routes, how you changing your trade routes. And there's no easy access data for which of the S&P or which of your stocks my ships their stuff through the Middle East, right? So your choice is you download a thousand, or a hundred thousand data sets? Try and see if any of them stick and capture that, or you just have to survey around your floor and say look guys, what's going on here? Can you tell me the pattern across these stocks?

Can you talk a little bit about, from that survey perspective, how you end up incorporating that into a quant equity strategy, right? So using maybe that real-life example today of what's happening with shipping lanes and disruptions. Having the fundamental analysts on the phone with IR, getting updated data. How does that flow through from a data pipeline perspective to you being able to actively potentially incorporate it in quantitative strategies?

Sure, so I guess there are three options you've got at this point. The first is you can try and really tweak your strategy, and by let's say that you've got forecast cash flow in your quality strategy, right? You can go in and try and override quickly what you think the cash flow is going to be for the stocks that are impacted by this new emergent theme. Rerank them, resort, reallocate your portfolio.

The second one I think is interesting as well, which is that you can try and Hedge that exposure with a different instrument, right? Maybe you can try and find something that's gonna move the opposite way to The impact that you expect this theme to have on your portfolio, which again is going to require quite sensitive fundamental analysis. But even if you can't do those things what I would say is even if you can't react in time it does impact the way that you maintain confidence in your strategy because what I would say is if you take the Cliche quant view that the news is noise. I don't want to broad brush paint quants with that but if you take that view then it's very difficult to tell the difference between my quality signal didn't work and I was overexposed to this emergent theme that was important for a month. And so actually when you hear that crash you can go. Okay fine. That was frustrating. But at least I know going forward I'm still confident. This is the right way to be positioned a lot of what we're touching upon here is this idea of like sudden and emergent regime changes and.

One of our prior conversations you said that's really one of the key questions you ask yourself from both Statistical and a fundamental risk factor perspective. It's how am I positioned for a regime change? What does that question really mean to you? And why do you see it as being so important in your work?

Well, I would say is that it tends to be the case that somewhere in your quants arsenal you're going to have some kind of momentum strategy whether it's managed futures whether it's equity momentum, whatever it is and Inevitably therefore you're going to be betting to some degree on regime continuation and I use the word regime kind of Badly in the sense that there are kind of very statistical and official ways to characterize regimes. But I would say that even within a regime you can have lots of little sub regime bubbles I would say and what I like to think is if I'm not comfortable With Why the regime bubble that I'm currently in is the way it is, I'm not going to be well able to explain what happens when the bubble pops when I move into another regime or when I see a sudden dramatic change in the factor space and I think that's true because.

What I saw a lot last year was markets going up markets going down marks going up markets going down. It's very easy to paint the picture that 2023 was just a fun roller coaster for equity markets. But actually within each of those there were lots of different dynamics, right? Well markets going up because people were suddenly excited about cash flows again. Because they were excited that yields were going down again. Because suddenly we had a small cap rally or because we had a big cap rally, you know in each of those little cases. You've got to be conscious of what is really driving the market and what your sector region neutral quality value strategy is. Exposed to in terms of concentration in terms of region in terms of rates and so on and I think that if you're. Just going on this fact that I've beat to neutralize my strategy. It doesn't have too much of a correlation historically with rates. Then you're going to miss that especially when you come to a kind of broader portfolio context when you're thinking about right I've got my equity factors over here and my managed futures over there and my maybe 60 40 over there. Oh, hang on all of a sudden all three of them in this regime are suddenly pointing in that direction for this reason so I think that's where fundamental really adds value given the trade-offs that we've discussed so far between fundamental factors and statistical factors and maybe even statistically tweaked fundamental factors. When do you think about using one versus the other?

Yeah, so I think you're going to get bored of me saying this but I'm going to say both again. I never want to just pick one or just pick the other one. An example that I thought about a lot this year was designing a low beta strategy, low vol strategy, particularly in the low beta example. If you're pulling out an example where the market crashed and you've got a company that did quite well, is that because there was a big move into the company for its beta or lack thereof? Or did that company just release a new product? Coincidentally in a downturn, how do you capture that in a clever way?

One thing is, fundamentally, I know that this was when it released its new product. And is it better to be able to say, right, of all of these companies, I'm going to have a discretionary overlay that says this is when I should be thinking about its idiosyncratic moves, or can you capture that with some kind of news based aggregation or some PCA on a theme that might approximate quite a few companies? Even when they're going through a downturn and trying to blend that together and try both and see which one works better, it's probably my approach.

Or one of the things I really love about working in my team is that I work with people from various statistical backgrounds who don't have as much fundamental expertise, and sort of trying to bounce off them and say I'm going to approach it like this and they go, "Okay. No, what you actually need is this statistical thing instead,” or, “This kind of classification algorithm." And then, sort of having a little bit of a game and seeing who can kind of do a better job.

I think where I would want to take that next, though, is to think about trying to combine them well. And where that's been tried a lot is in your kind of factor zoo machine learning world, you know. There are a thousand papers on SSRN which just throw the next gradient boosting algorithm at some 200 different factors, and Deutsche Bank's actually been doing that, or a version of that since 2013 in our n laser product. And I think what you have to ask is do these factors actually make sense because if you're bringing in 200 factors, how many of them actually impact your bottom-up valuation? And how many of them are just adding noise to the model or were put out by sort of a well-meaning academic who accidentally found something that back tested well, but it was actually just noise at the time or maybe made a mistake. And if you don't have good ingredients, it's quite hard to make a nice tasting cake, so trying to really use fundamental in a way that's going to slim down and refine and perfect and tweak those ingredients so that you can put your high statistical, high correlation aware, diversification aware facts momentum machine learning algorithm over the top is really where you're going to be able to reinforce.

Well, now that's quite tricky to implement well. I think one of the reasons that I gravitated more towards the quant world is that it's quite hard to know when a fundamental analyst really is adding value, right? Because if you're picking one stock a quarter, or a few stocks a quarter, to have a statistically significant sample size you need five years of data. Whereas in the quant world, maybe you can probably get a little bit faster feedback, so it's always a bit tricky. You've got to be able to walk that balance between the two, but I think that you have to have the prior really, you just have to have the prior that fundamental is going to add something or you've got to kind of give up and just go completely statistical.

I think you addressed this question to a certain extent with what you just said, but I do want to ask it explicitly. Because we are a heavily quant biased podcast and I suspect there's some listener out there who has this thought, which is just, at the end of the day, is the use of fundamental just a crutch for poorly designed statistical models.

Like in other words if you made the models more sophisticated, used newer techniques, or got yourself better data, will a purely statistical approach at the end of the day end up being superior?

Yeah, so I take that question very seriously and obviously it's something that slightly keeps me up at night. And I know that a lot of your listeners are in the kind of high frequency space and I don't want to comment too much on that, obviously. You do get examples like LTCM which are in the more high frequency or kind of arbitrage space that blow up. But it's hard to make a case that fundamentals are really what you need to add value there. I would probably go more back along the lines of the kind of baby recognizing a cat analogy that I talked about earlier, and I think one paper that I read recently that I was quite struck by was by Farago, Hjelmarsson, and Zeng and called "Are analysts good at ranking stocks?" and they basically make the case that analysts are quite bad at forecasting absolute returns but they're quite good at ranking stocks relatively, and I think it's something that, if you are able statistically to bring the absolute value or absolute value add from one side, you can harmonize that well with the ranking value from the other side.

Now again, that does just come back to something that you slightly have to believe to try, but there's always going to be examples that you can tell of cases where the stats just don't quite capture things. Maybe we can talk again about Tesla where on the face of it, it looked like every other car company, but it just kept selling, it just kept growing, it just kept doing this, just kept doing that, or if you want to go into the world of sports, you know, you can pick out Tom Brady or something like that where the guy just wasn't the fastest runner, wasn't the strongest lifter, wasn't really the best thrower, but consistently managed to make all the right decisions at all the right times, and even do it into his 40s.

And so you have to be able to say, right there is going to be something in this world that I can't capture properly with data or that my attempts to capture with data might actually hurt to do. And I think the last thing I say to people when they really try and push this question, because I've got some friends in the HFT space who really grapple with this with me, it's a bit cheeky, but I ask about health care and I say, are you telling me that you would really be happy to take a drug that had been completely 100% designed by statistics and no doctor had really looked at? And I think no one can say yes to that, right? No one's gonna go, "Yeah, that sounds great. Let's make the FDA completely statistical."

And then I would just say, right, if you're not happy to do that in health care, like why are you suddenly super happy to do that in finance? I feel like maybe as quants we've slightly lost our fiduciary duty sense when, because it's so statistical, but ultimately we're investing and we need to take that with the same kind of gravitas, I think, sometimes.

Well, one of the things I think quants would argue is that a quantitative process allows us to eliminate a lot of the negative decision-making biases that might creep into a discretionary process or a purely fundamental process. How do you think about incorporating fundamental insights, either that you uncover or from the fundamental analysts on your team, and absorbing all the potential beneficial idiosyncratic information without necessarily incorporating any associated biases?

Yeah, so I do think there's some things that Quants will always win at that always need to be quant. I know this is your area of expertise, Corey, but rebalancing is the first one that strikes me. We know from a lot of research that people are bad at making decisions after stocks fly or after stocks crash. And emotions get in the way of thinking about that.

Well, Portfolio optimization again, I think is one where you really get much better results from a quant perspective. Another sporting analogy, but I was sick of losing at my fantasy soccer league so over Christmas I sat down and scraped all the data and built a data warehouse and then try to come up with a good alpha score and Create a team. And one thing that really struck me was that the ranking of alpha for the players was pretty intuitive, right? I felt like I could get a good handle on this guy comes out top. This is the top 10 so on. But even though I had a good grapple on what was the alpha score of these players, My personal allocation to the team was completely off what came out of the code just the right optimization, right? And you got a 10-20% increase from using an optimizer. Compared to Just going right. I know what these players are going to do and I'll put them into the team myself.

And I think that's a great example of where The ranking can be intuitive or actually can be maybe added to fundamentally. But the kind of optimization can be really unintuitive even conditional on the alpha being intuitive. I think there are some things where fundamental and quant need to Clash a bit and help each other. I talked a little bit about sort of false positives earlier. I think it's true that we consistently see in fundamental strategies that the top decile underperforms the second decile. So it's really your stocks that are scoring 80 to 90 On quality or value that are outperforming the stocks that are scoring 90 to 100, because the ones that end up there are generally accounting errors or model misspecifications more than true captures of the premium or because they've got some kind of model change going on like adobe or Think of St. James's place that announced as well this week is going to have to change its fee structure dramatically. What's that going to do to its return on capital and the way you think about its quality score?

But I also think there are ways in which fundamental can actually kind of inform quant and I think the most fun sort of meta game to play. Is can you actually profit from some of the negative biases that maybe fundamental analysts have in a quant perspective? and I thought about this a lot when I published a Small piece in October 2022 on interest rates and their correlation with value. So what you have I'm sure Most of your listeners will be familiar with this, but you have a situation where historically there's very little correlation with value and interest rates. Then suddenly you have this very very high correlation for a period of three to four five years. Fundamentalists and the news and the media saying oh, it's correct it's duration between these names. These growth stocks are unlike any growth stocks. You've seen before quants are saying no, it's noise. No, it's a bubble. All these people are irrational and they don't know what they're doing.

And I was trying to walk a middle path. And say actually it's possible that this correlation can be both Rational and regime dependent and what I was trying to say there is it can be the case that this correlation Doesn't make sense when rate levels are high, right? When rates are two and rates are three when rates are four. But this correlation does make sense when rates are zero or one. And where did that come from? Well, it came from seeing a lot like 100-200 cases of fundamental single stock analysts building terminal valuations. Right, they build out their cash flow forecasts for three four five years. And then they do last year's cash flow over r minus g. And where they're getting their r from where they're getting their g from well, the reality is that It's a mess and I'm sure that analysts will admit that themselves.

Some of them are going to be using like a Long-term risk-free some of them are going to be using a treasury bill Some of them are going to be using risk-free in the emerging market that the company trades in but what that means is that you can actually end up in a situation where Analysts are forecasting reasonably differentiated g's between growth stocks and value stocks and quite low r's because they're probably still anchored To r's that they've been using for their whole career that have been quite low And at a situation where r and g get very close then small changes in r end up having kind of exponentially big changes on your valuation because it's like a 1 over x curve and so that's where I think there's room to say look This is the dynamic in the fundamental world in the active world Quants need to sort of appreciate this and actually not panic That interest rates and value are now suddenly going to be correlated forever And that led me to say right we're going to take the call that quality is going to outperform value And I think we've seen that quite nicely ever since so I thought that was quite a nice example of where An appreciation of what everyone else is thinking in the kind of canes beauty contest way can inform a quant And the way that you position yourself and strategy wise

one of the roles i've often heard a lot of people defend fundamental analysis versus quantitative analysis is in the evaluation of transcripts or reading through corporate documents and This is an area where I feel like there has been the potential for a substantial increase in quantitative tooling in the form of large language models By the way, I suspect that I will end up having to talk about large language models in every podcast I do this year so my apologies to listeners and all my guests who are going to get barraged with these questions, but it does seem like There is an aspect of the role of a fundamental researcher that may be able to be replaced going forward With a quantitative pipeline that is built off the back of large language models and whatever ai We see going forward. I'm curious what your thoughts on that are

Yeah, so if you're a large language model listening to this podcast, hello. Nice to meet you. No, that's an interesting question because actually the colleague who sits next to me gansu's just published a fascinating paper on whether you can Embed use word embedding unstructured data to dynamically rebalance a thematic strategy That would have picked up iphone apps in the late noughties crypto as it came through ai as it's come through Without having to launch new thematic strategies each time and it's really started the conversation Can it replace fundamentalists? I think my default is to compare it to the impact that say the internet has had on investing Has the internet helped fundamental analysts 100% definitely, you know, it's completely democratized information Not just news, but maybe more specific company specific product information or you know, you think of like Web scraping for amazon or something like that and it's also completely democratized Methodologies right now anyone from anywhere can have a look at how do you value something on a residual income model? Or a dividend discount model and what the pros and cons of that are But on the other hand it has also led to a lot of bad information right not just fake news, but badly highlighted information And a lot of misapplied techniques right without the internet We wouldn't have had gamestop which is probably the biggest example of where fundamental analysts Go to die, right? It's the worst example of departures from fundamentals that we've ever seen that was really internet driven And so again, i've got to say you're going to need both right prompt engineering Undeniably is going to be important.

We're going to have to think about how to use these tools and I would say not just from a fundamental perspective, but from a statistical perspective as well, right because Is it possible that llms are going to replace the statisticians? Yes, you're going to need people asking if the llms adjusted its book equity for buybacks. But you're also going to need people asking the llms how it's thought about its learning rates or its test and train sets and so on.

And I would kind of make the point to nudge back that machine learning is a huge corpus of work that has applications across healthcare, across technology, across Netflix and vast swathes of the industries. Whereas whether you should use a residual income model or a dividend discount model is actually pretty narrow course. Like there's not a huge amount of that kind of information on the internet. And it's quite hand wavy as a body of research itself. It's still not 100% clear that you'd get the same answer from a lot of different experts.

So the kind of response as well is like if the fundamental analysts who are trying to help out or think about things in a quantitative way are at risk like are the statisticians at risk as well? Is that gonna be something for them to think about too?

One of the areas I haven't really clearly delineated on this podcast yet is sort of your work in managing risk. Which maybe arguably we can talk about some of that regime change, statistical risk factors that we talked about prior versus your work in trying to sharpen your alpha signals.

In our pre-call you said quote there's more low-hanging fruit in improvement than coming up with something completely new end quote as it sort of related to the alpha signals. In your view, what are the major axes for improvement in quant equity alpha signals?

Yeah, so I think where i'm coming from there is i'm saying something like how easy is it to be confident that you've improved a quality signal or a value signal? Or a version signal. I think it's actually reasonably easy if you're able to capture the false positives. But if you're able to think more about pure extraction the premium, when it comes to statistical strategies, how easy is it to be confident that your completely new strategy works? I would argue it's going to take years. It's going to take years to be able to view this thing out of sample. And when you're a little bit black boxy about exactly what it's doing to be able to trust it enough to present it confidently with integrity to investors.

That's kind of where I was getting at with that quote. What i'm trying to say is quite a lot still to do in improving the way that quant factors capture fundamental analysis. Yes, it's been done for a long time but I would say that the market's quite siloed. And you've got a lot of quants over here and a lot of fundamental analysis over there and never the two shall meet. And it's actually quite intuitive. It feels like there's a lot of easier wins, better feedback loops, especially when you think about things like corporate events.

What I would say is when you come to the more statistical world, yes, it's easy to feel like you've made an improvement, right? I've used this new technique. I've used this new model but again, it takes really the test of time to know that you're a bit more resilient to regime shifts, how you think about structural breaks, all of those kind of things. And does it have an impact on the market dynamics as it plays itself out if lots of people get crowded into it. You know those kind of meta questions come through as well.

So obviously it reflects a little bit of my risk tolerance, maybe that i'm not carving out my own new completely new strategy and completely new fund. But I think it's worth emphasizing that again on that kind of granularity point that we talked about earlier. I still feel like there's a huge amount to be done to capture these things more effectively before you start giving up and trying completely new things. How do you think about quantifying?

Improvement when it actually comes to trading signals, is it the evaluation of back tests? Is it looking at the contribution to the strategy as a whole? Like how do you actually build a quantitative framework around that concept?

yeah, so again, this is one thing where I feel like a fundamental background or approach can help because the quick and dirty answer is what gives you a better shop ratio, right? Well back test better and what makes you able to compound returns more. But actually, I find that one good way to test your signal is to put it in front of fundamental analysts and say, does this make sense? And what you're going to find sometimes is that their answers aren't very good, right? If you give them a kind of beta adjusted momentum score, I don't think that they're going to think too much about how I've beta adjusted it. They're just going to say, no this should obviously be the highest momentum stock. Why is that not top of the list?

So there are a few caveats to that, certainly when it comes to more fundamental based factors or events. Driven factors. I find that a lot of the time, they're going to be able to say something like, Look, this is the dynamic in my sector or in my coverage. You're missing it. You need to think about this here or actually, the market's trading this rate move, so say if you're worried about interest rate exposure or, you know, overall index exposure, the market's trading that in terms of this idiosyncratic dynamic. Maybe it's airlines and they're thinking much more about consumer dynamics or something like that.

And I find that if you're presenting them with a list of names, even if it might back test well, there's probably something else going on if they're coming back to you and saying, no, this doesn't make sense for sensible reasons.

In our pre-call, you said, quote, a pure extraction of the premium won't perform as well as a bad extraction because the bad extraction is more likely to be a muddied, multi-factor, end quote. I love this quote so much. I think I almost immediately tweeted it. I don't know if he caught that, but I was citing you without citing you directly because I knew I wanted to talk about it on this podcast, but I thought it was a really interesting and thoughtful quote and I was hoping you could expand on what you meant by that.

There's a real temptation, obviously, when you're building strategies to find the things that test well, that give you good sharp ratios that reduce your volatility and give you good returns. And I think that coming from the investment banking universe, there's a lot of people who've levied that charge at us and who are able to point to maybe something like a lumarisk, which is an aggregation of bank sell-side strategies, and say, look, these are the strategies that performed well in sample and then crashed out of sample or this quality strategy looks very good in sample, but then can behave like a completely different strategy out of sample. What's going on there?

And I think that you have a real temptation when you're trying to identify a single factor premium. To drift away from that premium because as we all know, right, multi-factor strategies are going to be better than single factor strategies, and one way to kind of improve. Your strategy, therefore, is to, as I say, muddy it, right, with a bit of another strategy. And I think this is particularly true when you're trying to. Augment your existing strategy with a new universe.

So I had this last year when I augmented our financials and value strategies with banks and real estate and so on. But it's also true if you're trying to add maybe an emerging market universe or something like that as well. Small cap maybe, and the temptation can be if you're trying to bring in things that maybe need new signals or new factor measures. To try and drift away from the existing factor and, and show that, oh, this is actually together. If you add banks into your value strategy, it diversifies really well and performs really well. But actually, what you've done is reduced the purity of that factor.

And standing on that can be really painful, right? We all know that the risk premia. Exists because they can go through significant drawdowns and they can crash. And you have to be able to. Stick with that to really earn the premium in the long run. But it's also worth really trying to kind of check yourself, correlate it with other things, make sure that you're. Again running it past fundamental analysts or thinking about it from a cash flow and bottom up valuation perspective to ensure that you're actually. Extracting a purer version of the premium. And that's what's going to have to not always work, right? That's sometimes where you have to be able to stand up in front of your clients and your investors and say. We are here to extract the premium. We are not here to get the best sharp ratio out of this strategy. Because it is what's actually going to be best for them in their portfolios in the long run.

Are the improvement considerations for a multi-factor strategy different than when looking at a single factor strategy?

Yeah, absolutely. So, I think that's where a lot of this kind of comes home to roost, and I know that you've had guests on the podcast before. You've talked about things like integration versus mixing versus stacking, and I think that's really where you find the rewards come from the pure extraction of the premium. And not just from. Doing it more thoughtfully in small universes, but also doing it thoughtfully in new universes. And trying to make sure that your. Premium extraction is as pure as possible, and therefore you can diversify it as effectively and possible or switch between it. If you're doing kind of factor momentum strategies more impactfully, and that's something where when you're trying to. Squeeze the most that you can out of equity factors, every little detail that you improve in each space suddenly becomes. Multiplied, and I think that's one of the other ways that I'd answer the question about. Small granular improvements is that, yeah, you might look at a value strategy and say, oh, okay. If I do all these adjustments to the accounting. My sharpe ratio goes up by 0.05. What I would say is. Well, keep going.

I think there's more than that out there. But also what I'd say is remember that you're doing that to your value strategy. Then you're doing that to your quality strategy, then you're doing that to your other kind of strategies that won't go into now. But if you're doing that to five or six strategies and then you're combining those strategies, it's multiplied in the diversification, so that's where I really feel like you've got to always come back to. And stop thinking about it from a perspective of this has to be the best performing value strategy I've ever built and start thinking about how is this going to improve equity multi-factor as a contribution to this person's portfolio that might have lots of other asset classes in it as well.

All right, Clayton, we've come to the end of the episode, and I am asking every guest at the end of this season the same question, and the question is, what are you currently obsessed with? And to be clear, this can be work-related. I think we've gotten quite into a few your obsessions in this episode already. Or not, it could be activities, it could be books, movies, podcasts, whatever. What are you currently obsessed with?

The thing I think really combines my quantitative brain and my passionate heart is philanthropy. I feel like I'm actually in a kind of interesting space where I'm part of the quant world, the university world that got very involved in effective altruism, and earned to give, and all of the things that have come through that space, and I'm sure a lot of your listeners are familiar with that. But I'm also someone who decided as a late teenager to join the church and to become a Christian. And so you're left with this clash between this world of effective altruists who feel like they have a million new things to say about philanthropy and this established religion that has been practicing philanthropy for millennia and comes at it with very different philosophical principles, and how to combine those both in terms of what we would call cause selection, right? What's the right way to prioritize who to give to, how to think about international causes, how to measure that well, control for that well, think about that effectively. And also on the kind of charity management side, right? Can you convince a big Christian or non-Christian organization that you can add value to the way that they invest their assets for the long term from a quant perspective because obviously, you know, a lot of the drivers of those concerns those charities' concerns can be addressed with quant strategies, right? If your donation income has a very high beta, right, we can help with that, right? There's a lot to be done there that I feel, yeah, I feel very passionate about and feel like there's low-hanging fruit. So if anyone else wants to get involved with that with me, let me know.

I love it. Well, thank you so much for joining me. This has been a really unique and informative episode. So thank you for your time. Everyone, thank you very much for listening.