My guest is Rob Croce, Senior Portfolio Manager at Newton Investment Management Group.

This episode is all about what Rob considers to be the two super factors: trend and carry. More importantly, how Rob uses them to inform how risk is taken within asset classes, across asset classes, and over time.

Rob is not afraid to get in the weeds, either. For example, on the trend side we discuss details such as how to combine trend signals of different speeds, how to balance the probability of a trend signal being noise versus its likelihood of continuing, and how trend signals can be improved using clustering ideas.

From high level thoughts about diversification to low level details about measuring bond carry correctly, there’s a lot to unpack in this episode.

Please enjoy my discussion with Rob Croce.


Corey Hoffstein  00:00

321 Let’s dance. 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  00:19

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  00:50

My guest is Rob Krawczyk, Senior Portfolio Manager at Newton Investment Management Group. This episode is all about what Rob considers to be the two super factors trend and carry. More importantly, how Rob uses them to inform how risk is taken within asset classes across asset classes. And over time, Rob is not afraid to get into the weeds either. For example, on the trend side, we discuss details such as how to combine trend signals of different speeds, how to balance the probability of a trend signal being noise versus its likelihood of continuing and how trend signals can be improved using clustering concepts. From high level thoughts about diversification to low level details about measuring bond carry correctly. There’s a lot to unpack in this episode. Please enjoy my discussion with Rob crow check. Rob, welcome to the show. excited to have you here. It’s been I think, a long time coming. We’ve been chatting on and off for a couple years. I’m sorry, I didn’t have you on earlier. Great to be here, Cory. You know better late than never right? I’m making up for my mistakes in past seasons. excited to have you here. I’m not going to presume all of our listeners are necessarily familiar with your work. So I’ll start where I normally start, which is handing it over to you. So you can fill in your background for our guests talk about how you got into the industry and take us up to present day.

Rob Croce  02:21

Okay. Well, my name is Rob crochet. I work in Newton investment management where I focus on liquid alternatives. I work on risk parity strategies, as well as alternative risk premia and manage future strategies. How did I get here, I had the great opportunity to work in asset management almost right out of graduate school. I interned at Texas Teachers for a summer before going on the job market. I have an econ PhD. And I was hired by someone who was connected to Texas Teachers at an asset manager in Houston called selling partners. And selling partners had a legacy as a fund to funds. But they wanted to move towards being a direct asset manager, they were coming, inquisitive, they bought an MLP manager and they wanted to directly start managing systematically. And so we launched his parity strategies, managed futures, alternative risk premia auxilia, was selling for seven years, learned a ton. And some cases learning by doing and other cases, learning from other experienced people around me, and then joined a different Bank of New York subsidiary called Mellon. And some things were moved around at the head of the ship, but same seat I’ve been in for the last five years.

Corey Hoffstein  03:26

Well, we have a lot to discuss, in regards to what you’re doing today, I think we’re really going to get in the weeds quite a bit and how you think about building portfolios taking forward some of those lessons learned at salient and in your presidency? Just start me, well, let’s set the table for the entire conversation. taking a big step back 30,000 foot view, can you describe the mandate that you manage today? For example, things as high level as what’s the portfolio objective? What assets Are you trading? What sort of decisions are actually driving the process?

Rob Croce  04:00

Got it. So the single largest mandate is a risk parity beta mandate. And what we do there is we just implement an index that looks a lot like the middle of the road risk parity strategy, targets a constant level of risk, and it’s diversified across high and low inflation and high low growth types of exposures. So what’s in there, basically, futures, futures and tips. So that’s a it’s a beta mandate, because risk parity is really, you know, it’s a bunch of betas. What does it offer it offers balanced beta and futures allow you to implement balanced beta. If you were to be locked into cash exposures, you could get 70 or 80% notional exposure to bonds, but that, you know, you wouldn’t have a very compelling risk adjusted sort of absolute level of return, in that case. So you really need futures to build a balanced portfolio and then take it up to a level of risk that’s interesting to an institutional investor. So that’s what was fairy dust, it builds a balanced portfolio and then assizes it at a level of riskiness, and a level of expected return that can be relevant to an institutional event. Strike and compete with something like equities. Because in the absence of a view, you should really be balanced. Right. So risk parity, it’s really the idea that, in the absence of a view, you should be balanced. And so we think that risk parity is really an optimal beta portfolio. We also manage an active risk parity strategy where we’re using an active overlay to add additional resolution to the correlation estimate. So risk parity is literally the optimal portfolio under some pretty strong assumptions about correlation. And if those assumptions aren’t right, then actually it doesn’t take that big a step away from risk parity, but it’s probably not quite optimal anymore. You know, the assumption under there is that stocks and bonds are diversifying each other, and the commodities are diversifying stocks and bonds. And that’s not always true. There are some environments in which one or two of those asset classes move in lockstep with each other. And so we have a an active portfolio that uses nonlinear measures of correlation to take account of that, once you’ve got beta, right? We think that there’s a lot you can do in alternative beta naught alpha. Alpha is like, insight, have some unique insight, limited capacity, you have some identifiable edge relative to everybody else. But there’s stuff you can do in between beta and alpha factors for lack of a better term, even though they have some negative connotations. So we’ve done a ton of factor space. That’s sort of what seems most interesting to investors today, just given the environment, given how poorly stocks and bonds performed last year, there’s a lot of discussion of factors is trying to factor Well, it’s certainly not a classical equity factor. But boy, it’s, it’s one of the ones that investors are talking about today. So I have a feeling we’ll dig in on that.

Corey Hoffstein  06:42

Well, you read exactly where I was going. I know on the factor side, you tend to focus a lot on trend and carry. And I think if you ask most quants, sort of what are the two dominant style premia, they probably say, value and momentum, maybe more generic mean reversion and momentum. But your focus is more on carry and time series momentum? Why focus on those two factors in particular, especially versus say, a more diversified approach?

Rob Croce  07:14

It’s a great question. So I’m a multi asset investor mostly. And so value momentum, low volatility quality, those are equity factors, when we take that same approach or the same sort of factor based toolkit to multi asset space, and we look for explanatory variables. Carry and trend are really the cat’s meow. Now we’ve looked at the nuances what’s true and what’s momentum. For our purposes, today, I’m going to describe momentum as a cross sectional factor. It’s using recent performance to decide whether it’d be long or short, within the cross section of a single asset class. So that’ll be a momentum for me. And that’s different from trend, because trend is just going directionally with whatever’s worked, whatever has been up. So the middle of the point at which you go from long to short, is not the middle of the return distribution. It’s instead whether it’s been up or down. And so Trent has a drifting bias, long or short and asset class in its sort of conventional implementation, and we do it a little bit differently. But if you look at the cross section, and the time series explanatory power of factors in multi asset, space, trend supersedes momentum. So if I run a competition or horse race, where I have trend on the left hand side and momentum on the right hand side, I’ll get positive alphas that are pretty meaningful. When trends on the left hand side, I put tread on the right hand side and have momentum on the left hand side, I’ll get alphas that are negative and pretty meaningful. So we think that trend and the asset classes where we operate, which are futures, and stocks and bonds and commodities, credit spread indices, those are not technically futures, but they work a lot like them in FX rates. Those are all places that we play and trend is meaningfully explanatory. The other question that I think is important is is trend really a factor this trying to identify opportunities within sectors? Like is it telling me I want to be long the SP in short, Russell, is that where it makes its money? Or is trying really a beta timing factor? And if it’s really the latter, than there are some nuances in terms of implementation that I think are important when if you look at one trend has big years, it’s not making hay from spreads between related markets. It’s really making hay from being directly right across a broad swath of markets. And so that has some implications for how we should implement it. Carrie, is fundamentally not that different from a valuation signal, right? It’s a yield an earnings yield in this case, but like an actual how much I expect to get paid if the world doesn’t change type of yield. And we use carry, divided by recent measures of volatility or forecasts of volatility, to be sort of a Sharpe ratio estimate. And so what both trend and carry can do As they can tell me not just within a cross section of an asset class where I want to be allocated, but it turns out that they’re also predictive cross asset. So if the spread between high and low yielding currencies is big, relative to the spread between high and low yielding bond futures, for example, then I want to do more FX spreads. And it turns out to be the case that that cross asset carry metric is also predictive. In trend space, if one asset class has a bigger trend signal than another asset class, I typically would want to have more exposure to the asset class, this trend and more with some caveats, right? Like if it’s overextended, blah, blah, blah. But basically, trend is also predictive cross asset. And then in the third dimension, cross time, trends predictive if markets are trending heavily, maybe that’s a good time to do more trend if markets are not trending. Hard to explain to investors, why you have a big allocation to trend. So what we think about is this set of super factors carry and Trent, they’re super factors, because they don’t just explain the cross section of a single asset class, but they explain cross asset and cross time. And so they allow us to build a more nuanced portfolio. Frankly, they allow factors. So if you’re focused on these factors, they’re very complementary, that semi negatively correlated with each other, which is phenomenal, like value momentum in equity space. And they’re informative in these three dimensions. And so you can build a more nuanced portfolio, you’re not just sort of saying I want to allocate 5% of my risks, each of these 20 factors slash asset class combinations. And so you have a more intelligent portfolio.

Corey Hoffstein  11:33

I want to dig into the trend side a bit, and you said something in that answer, quickly, and it was a nuanced point, but I think it’s important to dig out a little bit. You mentioned that trend subsumes momentum. And I was wondering, you could maybe pull on that thread a little bit more. Why is that important? And whether you have any theories as to why you find that trend tends to subsume momentum,

Rob Croce  11:59

and the cross asset fence. So if I’m in future space, and I’m looking at a bunch of equity index futures, we don’t find a lot of predictive power and being long a set that has done better recently, in short, set this done worse recently, once we’re adjusting for riskiness, right, Russell 2000, is gonna be a little riskier than s&p. And so when we operate, and we build portfolios, we’re usually thinking about stuff and risk adjusted space. And once your risk adjusting the predictive power, the relative performance seems to go away. Why would that be? Well, it’s that predictive power. So let’s say Russell 2000 has a higher beta to equities in general than s&p, when beta is working, if you’re not risk adjusting, the Russell 2000 will be doing better. And when beta is not working, it’ll be doing worse. That’s the idea. Once you risk adjust, there’s really no, we don’t find any persistent predictive power in relative performance within related assets. So, you know, I think a lot of the equity research as well, it’s probably not risk adjusted, so that it is a beta timing phenomenon that you’re seeing inequity momentum stocks have done well recently, or on a relative basis, or the ones that have been had the right over under exposure to beta. In some cases, where to the right industry, or whatever it is, well, in multi asset space, momentum is good might work pretty well, if you’re just capturing relative beta exposures. But once you’re adjusting for that, and putting everything on the satellite level playing field that predictive power of the sword goes away. And all you’re left with is the time series.

Corey Hoffstein  13:28

Trend followers tend to think a lot about the speed of their signal. And that speed is really important, because it often informs the horizon over which the strategy is going to exhibit some form of convexity. So let’s talk about the speed element for a minute. What sort of speeds are you looking at and why?

Rob Croce  13:48

So if you’re just torturing the data, and you’re looking at stocks, I’m gonna keep using stocks as it’s a good starting place. But if you’re just torturing the data, and looking at equity index futures, that data is gonna tell you, you want a longer term signal. And the reason is because stocks go up over time, on average, and so the longer term your signal is closer to just long bias your signal is going to be, but that’s not the kind of return stream that a trend follower really wants to exhibit, because it’s a lousy complement. For the rest of what is showing asset. portfolio looks like you want to generate negative correlation to risk assets when risk assets are doing poorly. And it’s hard to do that when you have a very long term signal. Other problems with long term signals is they tend to hit the wall at 100 miles an hour meaning that you know you have this exposure on and it stays on until long after the reversal has occurred. And a lot of risk assets of stocks and 10 currencies actually seem to rip in your face once you get short, especially if you have a longer term signal. So it’s important to have something different from just a long term signal like a one year which was the cat’s meow. seems to us that just in terms of Sharpe ratio, but lousy properties at turning points. It seems to us that actually what works is gotten little bit faster in the last 10 years since we started doing this, of course, so we’re exponentially decaying our signals. And we’re using signals with half lives from 10 days, to about half a year. And so we’ve got a bunch of signals. And we do something non trivial in terms of how we aggregate them. But basically, we figure out how significant we think each measurement is relative to its own time period over which is measured. And then we aggregate those significance measures. And then we get a single metric for each asset class that rolls up all the look backs.

Corey Hoffstein  15:31

Well, maybe I can ask you to go into that a little bit more deeply if you’re willing, because I think that’s one of the non trivial components of being a trend follower, when you talk about using multiple trend speeds and trying to combine signals, the idea of the signals potentially being diluted to one another simply because of noise factors, or how do you combine them when their magnitudes might be materially different depending on how you normalize them. So I’d love for you to maybe discuss that aspect of of the mechanics of combining signals because it’s a non trivial component of building a trend following platform. You know,

Rob Croce  16:06

when we first started out doing this, what was common at the time and in the academic and practitioner literature was simply having the direction of recent performance inform the direction you want to take your exposure, and then having that exposure because of volatility targeting. And it turns out, it’s not so easy to beat that. But it’s hard to explain to clients, why you have you know, a full blown exposure to a market that is just barely past some midpoint for whether it’s trending or not having a binary trend signal is hard to explain to investors. And frankly, even though it works on average quite well, it’s not particularly satisfying. And so the first step away from doing that is to have 20 or 30 binary signals, and then take the average across them. And that’ll that’ll give you something that looks a little more nuanced. But that’s not what we do, what we really wanted was to find a way to have the magnitude of the signal be informative. And in order to do that, we have a number of signals, we have 20 Something signals in our standard implementation. And we need to find a way to measure the significance of each of them. And since we’re using an exponentially decaying moving average, it’s not immediately clear what the denominator for return over risks should be if I want to calculate a z score. And so you have to figure out what is the expectation, the expected standard deviation or the expected second moment of this exponential decay moving average. And so what you can do, you can do it similar, we haven’t figured out if there’s an analytical solution to this yet, but we can simulate it. And so you know, we do everything in constant ball terms. And so it’s straightforward to figure out what the expectation for a simulated 10 Ball return stream is, you’ve got that simulation, you do it a bunch of times, and you throw it through the exponential decay moving average. Any window you want, you get out standard deviations, boom, that’s your denominator. That’s the idea. And so you can take those denominators for all of your different signals, and you can take whatever your observation of the signal is, and you can divide it through by that expected standard deviation. Now you’ve got a z score, and you can aggregate across a bunch of z scores for whatever market you’re interested

Corey Hoffstein  18:09

in. You posed a really interesting question to me in our pre call when we were preparing for this podcast, and you said, as a trend signal gets bigger, should we have more or less conviction? How do you think about the balance of confidence as a signal gets larger? You know, it’s not less likely to be noise. But perhaps you have less confidence as a signal has room to run? How do you how do you find that balance? How do you how do you deal with it within your system.

Rob Croce  18:39

So the easiest way to do is to just use constant volume, which is what we did 12 years ago when we started doing this, but again, that leaves you open to having a big exposure when a signal is just barely past zero in terms of direction. And that’s not satisfying, it’s hard to explain to clients seems to work, okay, but I think we can do just a little bit better. So think about sort of the distribution of possible signals as a bell curve, like it’s not the normal distribution bell curve, but imagine that it’s shaped like a bell curve. And you start from a signal of zero this in the middle. And as you take steps away from zero in terms of signal, you become more confident the signal is not noise. But there’s this distribution of possible signals. And as you get further from the middle, you become less confident that it can continue out into the tail, because there’s just less room in that tail, right? There’s the area under the curve gets smaller and smaller as your signal gets bigger and bigger. And so ideally, what you would do is you would model those two characteristics separately. As your signal gets bigger, you have more competence, that is not noise, you want to invest more in it. But as your signal gets bigger, you have less confidence that could continue. And so you sort of pull away from investing in it. We model those two dynamics separately, and then we aggregate them and that’s what gives us our aggregate confidence in a signal.

Corey Hoffstein  19:58

Can you Explain how you think about applying trend signals to manage positions sort of not only within an asset class, but also across asset classes. And over time, you mentioned these three dimensions in which you’re applying both trend and carry, I’d love to dive into that a little bit how you think about using that same signal across these three dimensions.

Rob Croce  20:21

So trend, fortunately, if you’re using it in the way that I described, where you have more competence or less confidence, your signal gets bigger or smaller, and you’re accounting for the fact that there’s less room in the tail, you’re going to have a dynamic exposure to trend in an asset class. And you’re going to, you can combine some asset classes. And because we’re purposefully trying to have dynamic risk exposure to each asset class, what ends up being the case is you’re also going to have dynamic risk exposure to the combination of asset classes. And we think that that’s desirable, because we do want our strategies to have this dynamic risk, where if something’s really firing, if markets are really diverging, we may have long exposure to one asset classes doing well. And short exposure to another negatively correlated asset classes doing very poorly, our combined portfolio is going to start to operate at a higher level of riskiness. And exactly at the time when you would want that to be the case, right? So we do want the strategy to have a dynamic, I’m not gonna say the word convexity with regard to the strategy because I think that that’s an overused and poorly understood word in many cases. But we do want it to have a dynamic response to changing risk exposures. And so if markets are trending, and we’re long song and short others, and those shorts become pretty correlated with our lungs, then we can start to have a an exposure, that starts to be very, where we can capture trend the right way in both directions. And that’s kind of what exactly what you want, you would like a time series that is offering an additional sort of payoff, and exactly when the stuff that you own in your core portfolio is plummeting, or, you know, really under pressure. So we’ve got this dynamic exposure to trend. On the carry side, what we’re doing is also dynamic because we are dynamic in terms of where we allocate in the asset classes, right, so we’ve got these carry signals, and we’re basically we normalized Carrie relative to each asset class. So we’re always cross sectionally, market neutral within an asset class, but the amount of risk that we deploy in in one asset class totally different. We deployed another asset class because of the difference in opportunity. And so we’re dynamic where we’re using absolute magnitude. And so we do on the carry side, but we allowed trend to sort of take more or less risk through time, but have a constant long term average expected risk of about seven. So if we want 10 of all we’ll do seven vol trends seven well carry, putting together you get something actually price out of 10. To that we add vanilla equity factors, commodity term structure, stuff like that. And we still end up with something in a 10 ball range.

Corey Hoffstein  22:45

Measuring trend is, I would say a fairly consistent exercise across different asset classes. Obviously, there’s a tremendous amount of nuance and combining trend signals. But generally speaking, as long as the asset class has a price series, we can calculate a trend signal. The same isn’t necessarily true for carry, right, the way we might think about carry in commodities could be very different than the way we think about carry Within equities, or currencies. Can you talk a little bit about how you think about carry in each of those asset classes, and maybe some of the special adjustments you might have to make

Rob Croce  23:17

carry can be pretty consistent in terms of how we measure it in other asset classes as well, unfortunately, so. So a nuance in bonds is that you can take the term structure and sort of figure out if today’s tenure rate is 5%. And today’s nine year rate is 4%. There’s an appreciation implied. So if the shape of the term structure doesn’t change your 10 year bond next year, when it’s a nine year bond is going to have a 4% yield and a higher price, simply because you’re discounting by 4%. Now instead of five, and so that price appreciation becomes part of your carry measure, we call that the drop. So you get yield plus drop minus financing cost. So you know, your local LIBOR. So for right, and so you can do that for each of a bunch of bond futures. And you can do it for different bond futures across the term structure. So you can trade us two year bond futures versus us 10 year bond futures. Now, I wouldn’t just trade the carry versus the carry I would trade carry over vol vs. carryover vol. And you can do this a little bit better than just using today’s term structure. It turns out you can use the forward rate. So why would I use today’s nine year rate when the one year forward nine year rate accounts for expected monetary policy and today’s nine year rate doesn’t? Not in the same way anyway. So it turns out that expecting your bonds to drop to the forward curves implied rate one year from now makes more sense and accounts for the expectations of monetary policy. Now those expectations will still be wrong at turning points. But at least they’re accounted for and you’re not totally ignoring the fact that the market may be completely aware that the Fed is going to raise rates a bunch and this was true last year. The market knew that the Fed was going to raise rates a bunch and the forward curve priced it and today’s yield curve Didn’t price it. And if you’re using today’s yield curve to inform your carry measure, you were getting the wrong metric, you thought that the front of the curve was very juicy. And in fact, it was just pricing the fact that yields were going to rise in the near term. And so we think it’s important to use a forward rate to one of our favorite lessons learned for the last year and a half, when we talked

Corey Hoffstein  25:20

about those three axes of within an asset class across asset class, and over time, you use trend following to help manage risk over time, but not carry, why not use carry on that third axis. So

Rob Croce  25:33

the reason is because of beta, it turns out that carry is informative if you want to time beta. And so if I wanted to take more beta based on my carry manager, or less beta based on my character, it works really well for that. But when we neutralize each asset class, relative to its own average, carry over vol measure, we no longer have beta in the cross section of each of these asset classes returned, we’ve taken the beta out, and we’re building market neutral sub portfolios. And once we do that, it turns out that the care is no longer predictive through time. And so, you know, we’ve looked at this a bunch of different ways. And if we are using Kerris and Metro to decide whether we want more or less beta and asset class, sure, it works quite well for that. But once we take the beta out, and we start operating market neutral is no longer predictive.

Corey Hoffstein  26:19

Do you find that the efficacy of these factors is equal when you evaluate them within an asset class across asset class and over time? My intuition here would be you have a decreasing breadth of diversification that gets applied, right? It’s when you’re within asset classes, you can have a large number of bets across assets, you maybe get four or five major bets and then over time, it’s sort of one bet, right? Do you find, for example, that that trend works better to manage risk over time, whereas actually carry works better for tilting within asset classes versus across? Or do you find that there’s actually equal efficacy applied across all three axes of risk that you’re taking, the first

Rob Croce  27:00

year carry is more informative in the cross section, if we’re building a multi asset carry portfolio is going to have a sharp North one, just operating in simulation space, it’s going to be well north of one. And trend is gonna be more like point eight. So for sure carry is is the cat’s meow in terms of building those cross sectional portfolios? When you put trend and carry together though the, the impact is pretty profound, right, risk goes down quite a bit, return doesn’t really fall. And then your question of how much value added is there in the timing, the dynamic component doing more carrying this asset class versus that more turn in this asset class versus that at any point in time, you’re right that those are sort of icing on the cake, and rather than the main drivers of performance. That said, there are times when it’s very meaningful, and where intuition cannot support having meaningful exposure to carry in an asset class, and where if you’re just operating systematically and allocating constant risk to an asset class slash factor cell in the matrix, you’re gonna have this meaningful risk exposure, something that doesn’t make any sense. So we still think it makes it’s better and makes more sense to go with the carry is when you’re doing carry. So we really, we think about putting our carrier before we we’ve got a one shot portfolio construction, but within each asset class, we’re normalizing relative to the middle of carryover risk, and so that that carry portfolio is market neutral in each asset class, but the preponderance of the risk exposure goes where the carriers

Corey Hoffstein  28:29

How do you think about combining trend and carry signals? You mentioned it briefly. But I really want to draw this out. Do you think about using a sleeve based approach is this an integrated approach of marrying the signals together something more nonlinear a conditional based approach, why the approach you take versus the variety of others that you might consider? So

Rob Croce  28:53

I, at one time stood on a made a presentation at a risk premia conference saying that integrated was the only type of portfolio construction that made any sense. And I sit here today saying that we use a mixed portfolio construction, not just in multi asset space, but also in our equity factors. We started with an integrated approach and a move to a mix approach. You know, in theory integrated makes a lot of sense and it’s hard to argue with but once the realities of portfolio construction start to be accounted for it turns out mix works really well. You get more breadth more positions, more individual line items, say when we built inequity factor space, we were doing some integrated portfolio construction what we found is that even though we applied equal weight to a bunch of different factors, our portfolio at the end of the day the integrate portfolio had the characteristics overwhelmingly of one factor low vol. And so it turns out that when you’re integrating a bunch of signals and one of them is low vol what we found in the states where we were operating is that the resulting portfolio had a real low vol bias it did not feel as if it was capturing all the factors if you ran a regression of you know the for Paul French factors, but You know, sort of implemented in the same space that we were implementing with the same rules that we were implementing, what you found was an overwhelming preponderance of the risk exposure was going to lowball. And that is not diversified. And so we found that that mixed portfolio construction was an easy way to solve that. In multi asset space, it’s not clear to me how you aggregate, I have not figured out how to turn trend signals into expected returns, it’s not as straightforward carry is literally monotonically related to expected return. And in the space where we play, we can use a bunch of different approaches. And what we find is that, you know, there’s a linear relationship between the carry measuring the subsequent expected returns on average. And so it makes a lot of sense to use carry directly as a return forecaster trend for us. I mean, I can take my Trend portfolio that I’ve created with some other heuristics and use a covariance matrix to back into an expected return, I could easily do that, and then combine it with carry and then build an aggregated portfolio. We’ve certainly looked at that historically, and have not found it to be better than what we’re doing. I think what you’ll find usually in trend space is that there are other heuristics being used to turn signals into portfolios. Now I can start beating on what I think a good way to build trend portfolios is, if you want, like I think that there’s some, some room for improvement in this space, relative to what’s been done historically.

Corey Hoffstein  31:19

And let’s not leave that cliffhanger. Give me the juice, where’s the room to be improved? Do you think?

Rob Croce  31:25

So it just strikes me that so many of the trends strategies that I’ve seen historically, they really are built in such a way that it seems like they’re being built trying to capture cross sectional differences in return, as well as directional differences in return. And when we look at when trend works, historically, it’s really, you know, it’s worked really well when beta was moving. And when it captured that, and it didn’t really seemed to do much, it didn’t seem to, you know, generate a ton of alpha, when markets were kind of not trending. And so if we think that trend is really a beta timing strategy, then we should really try to figure out how to make it the best beta timing strategy that we can. And so when we dig in, and we look at how to do that, well, it turns out that we’re trading a bunch of related markets, right. So there’s, you know, 90 or 100 futures contracts in our opportunity set, but big blocks of them have correlations, NorthPoint seven to each other. So going back to our equity futures as an example, if I know that the Russell 2000 is point eight, or point nine correlated s&p 500. And I know that the Russell has been up recently, I also know something about SMP, even if I don’t look at his recent returns. And so it’s, I’m ignoring information that could be valuable. If I’m just looking at the s&p s recent returns to inform my current positioning in the s&p, it might make sense to use a bunch of related markets, that kind of D noise a series. And so what we can do is we can create, I don’t know, think of it as PCA factors, but not even you know, that’s making it more complicated. It has to be, if you’ve got a bunch of related markets, you can use them to filter each other’s recent performance and create a less noisy signal for trend. And that less noisy signal will be more informative in terms of which direction you allocate to a market, if we look at, I can actually use a basket of markets to signal the future performance of an individual market. And that seems to work about 60% better than using the historical performance of just that market. And so the point, despite 18 months ago, when the footsie was up over the last 12 months, and virtually every other equity market that we look at was down, is that an opportunity to get long footsie and short the others, I would argue that it’s not, I would argue that probably the footsy and the rest of the basket are going to converge on each other. And that you should be short all of them. And so that’s actually what we see in the data. We see in the data that especially in faster signals. In the cross section, the markets converge on each other. So if I’ve got 20 equity markets, and the ones that have done well on a recent basis, are going to converge on the ones that have done poorly on a recent basis and vice versa. So in the cross section markets mean revert, that shouldn’t be surprising. It’s the single dominant statistical characteristic of markets, related markets, and then in a time series, they trend. So when you’re a trend follower, you really want to eliminate this characteristic that mean reversion is working against you. So you may want to find a way to build a portfolio that’s betting on the time series of recent returns in related markets without making any bets on relative performance. And that should be have the ability to trend faster, or trade at a faster trend signal because the real headwind to trading fast trend signals is mean reversion. It turns out that mean reversion seems to operate largely in the cross section or in part, I should say, in the cross section of an assets returns asset classes returns. And so if we can eliminate that component, we should have less of that headwind for our trend strategy, and we should be able to trade a faster moving trend signal without giving away all of the potential of it. Now, if you go too fast You’re a mean reversion is going to be there. And then the question is, do I want to trade signals that I know are not enhancing Sharpe ratio? Because they reduce drawdown. So there’s other characteristics to return streams other than just Sharpe ratios. If I can add a signal to a strategy that’s going to reduce his propensity to drawdown, two standard deviations relative to its risk, and take that down to one, for example, without affecting the risk adjusted return too much, that may be a good trade for, for investors.

Corey Hoffstein  35:28

You mentioned faster trend signals, one of the ways in which we can talk about the speed of carry, which isn’t often discussed, but it’s the horizon over which that expected return is being calculated, right, you could have a one year carry metric, you could have a six month carry metric. How do you think about measuring the speed of carry within the systems you operate?

Rob Croce  35:49

We’re a little bit limited by the data, right? So if I had a really good metric for nine year and 11 month yield, and I could observe that in a frequency I wanted, and to compare that with tenure yield, to calculate the drop, that’s the one I would use. But what we find is that we’re more limited by the observable points in the yield curve. And some countries were quite limited by the observable points in the yield curve. And so we take the closest observable point on the yield curve that we can use. So let’s say we were doing this in sofr futures, we would use the consecutive software futures to figure out what we’d expect the drop to be. So you know, one quarter apart, but that’s not really possible. If we want to do that across 18. sovereign bond futures like South Korean Good luck getting South Korean nine year 11 month expected yield. Daily.

Corey Hoffstein  36:40

from an outsider’s perspective, what strikes me about my understanding of the way the portfolio is built is there’s significant opportunity for interaction effects to emerge. And there’s two areas I think I’d like to touch on. The first is this idea of both how you’re operating within assets across assets. And over time, and I’m considering a scenario maybe where we’ll use trend following as an example, you have strong trends within an asset class that promotes potentially upper level to say over exposed to that asset class. And because then also, there’s a preponderance of trends, you might overweight trend following as a style within the portfolio as a whole, for going potential diversification and leaning into the strength of that signal. But it’s compounding upon itself. It’s conditional upon itself. Curious if you can talk a little bit about some of those interaction effects the pros and cons of leaning into them at the risk of potentially sacrificing diversification because of signal strength.

Rob Croce  37:44

Yeah, so I started as a risk parity guy, and I still view myself largely as a risk parity investor as a starting point, there is not a from my perspective, there is no more defensible starting point than risk parity. So as we start to have information really taking, we started to take small steps away from diversification, maximum diversification, but not huge. I think it makes a lot of sense to favor diversification over alpha opportunity, because it’s just so much more reliable, there’s a million things you could say to counter that. But on average, we really believe that diversification potential is less bleeding than alpha signals. So that’s the starting point. And you know, the amount by which our trend signal can sort of run rampant over the overall portfolio is limited strictly by the portfolio construction. So it’s limited by the amount by which the assets in the portfolio can be correlated, and by the amount by which we allow the signals to manifest. Given that we cap there, we’re basically once a signal is way out in the tail of the expectation for what it can be, we’re not betting on it really very much at all. We don’t bet on continuation in that case, because there’s no more room out in the tail, the probability that it continues to increase in that direction is statistically quite limited. So that’s the first thing. So we do want to have more exposure to trends that are less likely to have been noise. But within limits, and actually the headwind gets stronger than the tailwind at some point, and starts blowing us backwards in terms of how much exposure we’re taking. The benefit of that is that we don’t get crushed at the reversals that inevitably occur. Once a trend has, once a, you know, exceptionally sharp market move is done. We don’t know when they’re going to end, but we know that they’re going to end and given that. And given that the reversal can be exceptionally powerful and sort of sharp, it’s best to not play and be fully extended. So when we look at and I’ve looked at this recently, so this is why I kind of have the numbers at my fingertips. But when we look at how much positive skew, let’s say to describe statistically we can get from a return series where we do trend and we allow that the signals to just go where we don’t have that headwind applied to signals that are way out there. We can get an exceptional level of positive skew skew of 100 on don’t even know what terms that said, but very big positive skew numbers. And then when we apply the headwind that comes from being out, you know, gosh, we’re really out in the tail. And what we think the signal can possibly be, when we apply the headwind that accounts for that, that skewness is still positive, but not by a lot. And that skewness, it’s still more positive than if we took like the sock Gen trend index or something like that, and looked at the skewness of that, what I’m describing is still more skewed than that, because we’re still getting out, we’re still making taking bigger positions and trends that are really moving. But it’s also not that much more positively skewed than just having cost and risk. And all of the markets where you’re long, and constant risk all the markets where you’re short. So the trade off is that we give up big positive skewness and kurtosis. In trends that are working. But the benefit is we don’t get our eyeballs ripped out when those trends reverse. And so the only thing better than doing really well in a trend strategy, when markets are suffering is keeping most of it when those markets inevitably recover. So that’s kind of the objective function, we want to do well, but we also want to keep most of the benefit, once markets start doing what they do normally.

Corey Hoffstein  41:14

What about the potential interaction effects between trend and carry? So you’re using this mixed portfolio construction approach? And I’m thinking of a scenario where trend and carry both push you strongly into the same asset classes. Is that a risk that should be moderated at the portfolio construction level? Or is that a conviction that you should lean into?

Rob Croce  41:39

We get this conviction question all the time. So all of our models right now hate a lot of stuff, and they hate bonds, for example. And the question is, Well, should we really be? I mean, from a firm perspective, should we really be leaning against? And how our portfolios, like all one way? Like, is that a good business decision? As well as it’s not an investment decision? Should we be moderating these things, even though, our signals are telling us in very convicted way that this is their view, like we believe our signals, mostly, they’ve worked historically. So I don’t think it’s as much of a concern in trend plus carry, we have other mechanisms for managing extreme risks, position limits, in single markets, position limits in groups of markets, stuff like that. So there’s a bunch of different ways that we can manage that type of risk. But I don’t think it’s as much of a concern here, because the carry measure for an asset class tends to get bigger as that asset class gets hit. So if a market is falling, the carry that I perceive in that market will typically go up. So I will be getting longer and carry and shorter and Trent. And that’s the dynamic. And so actually, they’re also working at cross purposes, or they are temporary. And this combination is effective, particularly effective because of this exact characteristic. They’re really betting on the opposite things, in many cases.

Corey Hoffstein  43:00

I want to go into a bit of the world of conspiracy theories, perhaps someone tweeted about the impact of risk parity and managed futures on the market as a whole to which you responded, oh, no, not you, too. I know you have some views here around the way, maybe banks have have stoked fear about the potential impact of risk parity strategies, and their ability to move markets match future strategies and their ability to move markets sort of as a boogeyman. I’d love for you to expand on those thoughts. Why do you think that risk parity and managed futures are not having an impact?

Rob Croce  43:41

Oh, no, not you to look, anyone who trades any market participant that trades a market is having some impact? The question is, how big is that market participants volume relative to the market. And so when we look at the let’s look, specifically, when we look at the specific models that are being used to make these assertions about X number of billions of dollars of volume coming out from his parody strategies over the coming X number of days Ba ba ba, the models that are being used to do that are exceptionally unrelated to the models that the risk parity managers are actually trading. They just looked nothing like it. And so for that exact reason, it’s clear that these models have been put together in such a way that they can be used to make these sort of spectacular assertions about systematic traders. And assertions, at least based on these models have no foundation. Now, that’s not to say that risk targeted investors that manage risk and have drawdown controls aren’t training when markets are falling. I think that that’s a totally different question. But to say that systematic investor flow is going to be x over the coming number of days based on a model that sucks is not a great idea. And so I think it’s indefensible. And so I think it’s natural for investors to blame strategies that they don’t have fully understand. And so you know, and it’s not anyone’s job to fully understand risk parity unless they operate in that space and are interested in it. So it’s, you know, it’s a boogeyman thing. And so, but I’m liable to believe the same thing about strategies that I don’t fully understand, like, banks, delta hedging their option books, right. So I don’t know anything about what banks option books look like. And so when someone asserts that markets are going to be pegged at their current level, because banks have a tremendous amount of options that are struck near current levels and our delta hedging as we move away from these levels, I’m like, Oh, that makes sense. Sure. Could it be that I’m believing the same boogeyman type of scenario that others are blaming risk parity for? Possibly, one thing I know for sure, is that markets have always been risky. Markets have always had big fat left tails, that does not seem to have changed materially. What does seem to have changed is the recent performance of markets, you know, post global financial crisis markets, the average riskiness fell significantly because of monetary policy, the central bank put, and humans have a recency bias. And so we remember what markets were like the last couple of years, but we really don’t remember, you know, the sort of gut wrenching level of volatility that was normal. Before 2008. Markets are risky. Markets have always had big downside risks, markets have always had much bigger down days than they should, given what their sort of statistical distribution of returns looks like. That’s normal. And so Google might say that some big down days due to some boogeyman, I think it’s less likely then. The other thing I’ll say is that systematic investors are typically much more disciplined in terms of managing transactions costs than non systematic investors, right? Because you’re trading a lot on a regular basis. And you’re able to measure how much you expect to be relative to volume. And you’re thinking a lot about trading costs. When you trade a lot, you have to think a lot about trading costs. And so investors that are systematic and trading a ton, are likely also the ones that are thinking the most about what trading costs are, in a way that maybe a manager that trades less often, but as you know, tapping out, may not right, like get me out of these positions now is something that a systematic investor typically doesn’t say. And so they’re typically not going to be paying an unusually large trading costs. In a way that’s not expected.

Corey Hoffstein  47:17

I want to end the podcast by circling around back to the beginning talking a little bit about your history. I think a lot of the themes in which you built your career within this industry remain the same. I don’t want to speak for you. But it certainly seems that themes that have risk parity continue to hold throughout your portfolio construction. But other areas that seem to have changed is perhaps this idea that you don’t as an investor need to check all the boxes, and take extreme comfort and allocating across all these different factors, that perhaps a more concentrated set of factors is something that is superior. And I was wondering sort of as maybe one of the final points of the podcast, you could discuss a little bit your conviction in that new perspective, this idea that maybe fewer is actually better and focusing on the Super factors is a superior approach in the long run, versus I think the phrase you used in our pre call was allocating constant risk across different cells in a matrix.

Rob Croce  48:21

It used to be the case that banks would come through our offices pitching their quantitative investment strategies. And in front of their deck somewhere, it would be a matrix, and across the top would be asset classes, stocks, bonds, FX, commodities, and down the side would be different factor names that you’ve heard of carry value, momentum, low, Vol, whatever. And within each cell would be an X saying that I as bank offer you exposure to every single possible combination of asset classes and factors. So if it was really true that I could have exposure to 20 unrelated factors that all had a point five Sharpe ratio, it’s basically impossible lose money. And what we’ve seen in the year since then, is that it is very possible to lose money as a factor investor. And so something about our assumptions 10 years ago was wrong. Some of those factors were I think some of the literature calls them false discoveries. And so what can we do to avoid the potential of false discoveries? Well, we can invest in factors that we are 100% convicted are the real deal. And I think trial and carry are the most obvious examples of that in multi asset space. And they tend to complement each other really well. And so rather than taking comfort in investing in a huge number of factors, why not invest in a small number that complement each other well, and also allow us to do sort of higher level portfolio construction. So that’s been our philosophy. Now there’s stuff that doesn’t fit neatly in that box, right. So commodity term structure commodities, short spreads that doesn’t fit neatly in a box, we just layer it in as an alpha return string. There’s nothing in here to say that you can’t do this. And other alphas we believe that there are other alphas that are really hard to beat like commodities for spreads is, I mean, the one thing I would not skip on any given day, right? Even in my PA, I find a way to do commercial spreads. So we think that there are other alphas that really make it into this space that don’t fit neatly within this sort of carry and trend box. And we find a way to get them in, we layer them in as uncorrelated alphas, I don’t want to sound like I’m myopic, and only looking at Karen trend, but boy, are they good complements for each other. And do they, they create a very nice tight sort of macro exposure. And then we can dial in individual alphas that we have very high conviction and think we understand pretty well, on top of that, and they’re not correlated to it on average. And we end up with something that’s better than than any of the individual pieces. So I’ve learned that, you know, I can torture data as well as anyone or as well as a lot of folks. And I can find something that looks pretty good in the rearview mirror. And that’s not always going to operate well on a go forward basis. And so, you know, I think that some of the factors that we see investors using today that probably aren’t actually factors, a commodity value is one of them. There was a good paper a number of years ago, value momentum everywhere, and they found a way to get commodity value in there. And it’s like a mean reversion measure. We do not find that to be particularly compelling. And after the paper was published, if you look at the data that came since then, not particularly compelling, that’s not to say there’s not a valuation measure that you can use in commodities. But that’s not the one for us. We think that there’s more fundamental relationships between some commodities that you can use to glean relative value among related commodity markets. But you know, just applying a broad brush, hey, let’s back test something until we find something that works, and then have it populate one cell, the matrix, let’s torture data until we have every cell of the matrix populated and invest in that whole thing. We find that that doesn’t work.

Corey Hoffstein  51:47

All right, Rob, last question of the episode for you. It’s the same question. I’m asking everyone. The end of this season, which was making every guest pick a tarot card that resonates with them that will inform the design of the cover art of their episode, you chose the two of pentacles, which I believe is also known as the two of coins. What stood out to you about this card? Why did you pick it?

Rob Croce  52:10

Well, the description of it says balancing of decisions, priorities, adapting to change. And those are the things that, you know, I started out as a risk parity investor, first and foremost, balancing decisions. And then adapting to changes what I think a lot of us have done, as we’ve operated in this space over the last, it’s been 12 or 13 years for me, but we’re adapting to change and the thing that I’ve learned about not adapting. So if I look at strategies and managers that have only had one strategy, and I’ve continued to focus on that one strategy and sort of never lose focus, those are managers that when a strategy went out of favor, I’m thinking about trend following here, and then came back into favor, they had those live track records that have extended over the entire time period, they were the ones that were in a position to be very successful, and pick up on what’s been working recently. And that is different from over adapting to change, which would be offering whatever strategies worked well, recently killing strategies that have gone out of favor and have been out of favor for a few years. Because if you do that too much, I think it’s, you know, there’s a chance that you won’t have a lot of track record and an attractive background. Next time the strategies come around these strategies do come around, I think risk parity is the one I would look at right now, you know, it didn’t do see last year, risk parity tends to come ripping out of drawdowns like this quite aggressively. And the reason why is because it starts on the bond side, right? Expectations for policy change, bonds start to appreciate, then those lower discounts, rates start to get factored into equity prices, and then risk assets recover as well. So I think that to say, risk parity that favor right now when might be a good time to load up on it, as opposed to reduce exposure to it and exactly when it’s likely to have its best performance on a risk adjusted basis. It’s been a tough environment for something like risk parity for a while, right, like it’s the single best performing assets. Equities is the same thing that everyone already owns. It’s really hard to look good relative to that when you’re balanced across a bunch of stuff. But in every other state of the world balances what works better. And it certainly over, you know, decade long periods balances work better than concentration over virtually every horizon.

Corey Hoffstein  54:16

Rob, this has been absolutely fantastic. Thank you so much for joining me.

Rob Croce  54:21

Thanks for it. Really appreciate it.