🎙️Philip and Alan Maymin, co-heads of global asset allocation at Janus Henderson, joined Corey Hoffstein on Flirting with Models (S7E35, 65 min).
This is an outstanding interview. I’ll let the quotes carry, but I add my own comments and emphasis to direct your attention to what I found especially important from the parts of the transcript I curated.
It’s fitting to add that another great dork gift this season would be my favorite finance book, which is by Philip Maymin, Financial Hacking.
Corey’s intro on the guests
Corey: In this episode, I speak with Philip and Alan Maymin, co heads of global asset allocation at Janus Henderson. Philip and Alan run the asset allocation group originally founded at Janus Henderson by Nobel Laureate Myron Scholes more than a decade ago, And they’ve arrived there by wonderfully different routes. Philip through LTCM, academia and behavioral analytics. Alan through high frequency trading derivatives and prop desks. At the heart of their process is a simple but subversive idea. As Philip put it on our pre call quote, people live in an average return world and they ought not. What matters to investors is compound growth. And the brothers believe that the surest way to improve compounding isn’t chasing higher returns, but managing risk. Drawdowns destroy compounding and avoiding them is the more tractable lever. They connect theory to practice through the options market calls and puts embed the markets collective forward looking views of risk, making them one of the richest sources of information available to an asset allocator. We discuss how they extract those signals and propagate them across more than 4,000 assets to build portfolios around client objectives rather than benchmarks.
why risk is the easier lever
Corey: I know this idea of compounding is really critical to the way that you all think about portfolio construction and our pre call, you sort of laid out the chain of logic as being, well, everyone wants more money, more money comes from better compounding. And while you could try to improve expected returns, I think your quote was risk is easier. You can talk a little bit about why risk is more tractable as a lever for improving compounding versus expected returns.
Philip: This is awesome because there’s quite a few reasons. First is this anomaly or puzzle. If you have a positive average return scenario, right, you have some alpha, that’s your average return. What you really want is to make more money later. So you want the compound return to be higher, and the compound return differs from the average by the risk drag…Fine. If you think about that equation, compound return is average return minus risk. Average return, you know, I don’t use this word a lot, and I’ve never heard this word used a lot. Whenever I hear it, it’s always related to alpha. Alpha is fickle. You never hear about fickle. Right? It’s not a very common word, but it’s perfectly encapsulates what alpha is. You could spend your entire career, and it could be a short career, it could be a long career, and suddenly you wake up one day and what you thought was your source of alpha is gone. The world has changed. It’s moved on. Your contacts are gone. The sort of information or knowledge you thought you were extracting just suddenly gone. That’s not the case for risk management. If you’re managing risk, you’re expecting things to change and you can build not a fickle source of knowledge, but a compound base of knowledge that grows on itself. You know how to manage in different environments. That’s not gonna go away. We’re gonna suddenly live in a world where there’s only one environment. No. Environments will always change. The other reason is Bob Merton in 1980. He wrote a paper and one of the things most memorable… It was a great paper, but in… Even in the appendix, he had a really important insight. The insight was this, purely on a statistical basis. How confident, how far back do you have to be just looking at data to distinguish an alpha, an average return, an average from being a zero? And you have to go pretty far back. All it depends on is the length of time. It doesn’t matter if you slice it up into weekly or daily returns. It doesn’t matter. Just the starting and ending and how long a period you have. So you have to go back pretty far, sometimes decades, to be sure that what you’re measuring just statistically is different from zero. So at some point, you’re one is IBM, does it have a positive alpha? And you’re comparing returns from the nineteen seventies. Like that’s not the IBM of today. What is the point? So alpha is fickle, it’s hard to measure. Fine. But risk? Risk as he shows in that appendix and as you well know, the finer you cut the returns, the better your measurement of risk. It’s a stronger, more secure. You don’t have to go further back in time. You just cut it from monthly to weekly to daily to hourly and you have a better and better estimates of risk. So those things combined, risk is easier to measure, it’s reflecting the environment that things can change, you’re able to compound your knowledge, you’re able to build things for the long run.
In short, alpha is unreliable, but the principles of risk management are stable.
This point deserves an extra comment: All it depends on is the length of time. It doesn’t matter if you slice it up into weekly or daily returns. It doesn’t matter. Just the starting and ending and how long a period you have.
This idea is echoed in Benoit Mandelbrot’s Misbehavior of Markets, where the scientist explains that price behavior is fractal and non-Gaussian. Market shocks are power-law distributed. Since power-law functions are less steep than exponential functions (the normal distribution curves are exponentials), the tail probabilities do not decay as quickly as bell curves predict. Mandelbrot shows that cotton future price charts are statistically self-similar. If you zoom in on a daily chart of cotton prices, the jagged pattern looks similar to a monthly or yearly chart despite the difference in scale.
My post Know Nothing Sizing discusses how risk is the sturdier foundation for sizing because volatility is far more predictable than returns.
Go here if you’re interested in following up on power laws.
reading the options market
Corey: I know one of the key inputs for your process is looking at options markets, information embedded in puts and calls to inform your risk estimates. And this is across not just major asset classes, but correct me if I’m wrong, I think it’s thousands of assets and securities that you’re grabbing this data for to help inform your portfolio construction. Can you talk a little bit about how you utilize the options chain and the information embedded in there to help you measure both upside and downside risk and how it informs your process?
Alan: The best thing about the options market is that they provide a forward looking estimate of the upside and downside of a particular asset. Further, they require an exchange of premium in order for a trade to occur. So truly, people need to put their money where their mouth is in order for the options market to clear. As a result, by reading the information, by reading the trades, by reading the prices, we can see where people are expecting the asset to move over a certain period. This information is extremely rich. It transcends multiple different time horizons and multiple different strikes. In other words, you can see across an entire distribution of possible outcomes over a variety of time periods, what the market is saying is the probability of the underlying to achieve that particular outcome. For us, we can take that information and we can translate it into what is the probability or what is the likelihood of this asset moving up or what is the upside risk? Similarly, what is the downside risk or the left tail risk of a particular asset? By keeping track of that information over time and importantly, their ratio, we can see how a asset changes its upside versus downside over time.
the next best thing to an oracle
Corey: One of the potential constraints of looking at something like the options market is that the options prices can only reflect what market participants can inherently imagine. So how do you think about the unknown unknowns? The risks that by definition aren’t being priced in anywhere?
Philip: The options don’t reflect, like Alan’s example before, just fire or just earthquake. If you’re buying insurance on a house, it’s anything can happen. Similarly with option markets, it’s not that I’m buying a call option in case their earnings are up or in case of something else. It’s in any of the possible cases. So in a sense, a lot of the unknown unknowns are already in the options as much as they possibly can be. Even if you don’t know what might drive it, you can have a sense of foreboding. That’s what investors will draw on to set their prices and put their money where their mouth is. Of course, there’s always unknown about the future. Right? But what is the alternative? We don’t have a perfect oracle where we’re saying the options are not as good. Yes. They’re not as good as a perfect oracle. What’s the next best thing? Right? Just look at past performance. Look at the rearview mirror. Look at earnings or other news items that come out that’s already in the past as well. What other alternatives are there? One way that Myron puts it, you know, sometimes people ask us, are we fundamental or quantitative? And it feels like we should say we’re quantitative, like everything we’ve been talking about. Right? We’re not, like, visiting companies and going undercover and checking their stockrooms or anything. On the other hand, Myron puts it this way, we have thousands of the best fundamental analysts working for us. Suppose you really were, you had an infinite budget or maybe you’re using OpenAI’s agents or whatever. You have hundreds, thousands of human and AI agents investigating everything they can. How would you incorporate all of that information into a single signal that you could then manage your portfolio off of? Wouldn’t it be nice to use something like prediction markets or option markets where they could express their confidence through some sort of reward mechanism like the market? That’s what the options are doing. That’s what they provide us as a way of unifying information signals confidence from thousands of the smartest people on earth, tracking what they believe to be the most honest true thing, whether it’s known knowns, known unknowns, whatever. And that provides a signal. So it’s not gonna be the Oracle of Delphi, but it is something that you can look at and use to your benefit not to necessarily trade against them. That’s another important caveat… We’re not thinking like there’s a mistake in the option pricing in the market and we could take advantage of it. On the contrary, there’s just all this free information that people are shouting from the rooftops. Here’s what we see. Here’s where we’re putting our money. And almost everyone on Earth is covering their ears, their mouth, their throat, their eyes, everything and ignoring it. Why? Why not at least take in that information? There may be times when you deviate from it. It’s like using GPS. Right? Traffic enabled GPS. Every time I deviate, I regret it. I’m sure this happened to you too. It knows better than you almost all the time because it has forward looking information. It has traffic. It has people reporting stuff. There may be a few exceptions. When I’m really close to my house or a place I’m very familiar with and it tells me to go some way, I know I can go a different way. I can cut through the neighbor’s yard or whatever the heck I do. So, similarly, if you’re taking the options information overall, but you have some specific view on a particular thing and you like to deviate from it, so be it.
Philip is preaching to the choir (me). This idea of taking the options market’s collective intelligence, parameterizing its features, calling the liquid ones “fair”, then looking for departures from that is the foundation of moontower.ai’s approach.
risk-neutral vs real world
Corey: Anyone who’s classically trained as a quant or certainly in derivatives is gonna immediately jump on the fact that options are priced under what we call the risk neutral measure and you’re trying to use them to make statements about real world risk. The P measure, those are different measures and they’re not necessarily always in alignment with each other. So I’m curious how you think about sort of this p verse q problem when using option implied signals.
Philip: What is the biggest difference in the two measures? It’s of course the drift, right? And that’s what Fischer and Myron showed in 1973. The drift is completely different. It’s risk neutral in one versus whatever arbitrary thing it is in reality. The rest of the distribution can also be different. However, you can try simulation of assuming what the real is and then figuring out what the risk neutral is. Almost all of that difference is coming from the change in the drift. The shape of the distribution is fairly robust across the changes in measure. Like Alan was saying, we have the upside risk. We have the downside risk. It’s not a coincidence that we’re ignoring the middle. The entire philosophy is built on the idea that middles don’t matter. It’s the tails that matter. You can throw away… We have a paper showing this. You could throw away the middle 80% of a distribution, and your risk is basically the same. You’re gonna have an 80 to 90% correlation, and 80, 90 of the risk is still there. The tails are what matter, and the tails, that’s what the options price very, very well. And that’s the shapes that we take as signals. And bear in mind, like Alan was saying, it’s not that we’re taking one number on one day and that’s some God given truth. We’re looking at the changes in that signal relative to itself, relative to others. Things don’t always work 100%. But the risk we’re not so much worried about is the p versus q, the risk neutral versus real world because it would have to differ in a way that goes against all of these different parts of our process that translated into a reasonable signal. And at the end of the day, it’s a very reasonable thing. Look. What is Alan saying that we’re doing? Ultimately, boiling it down. If suddenly and for an extended period of time, people are buying a lot more calls and a lot less puts, that’s interesting information. P versus q notwithstanding.
The emphasized parts were amazing to hear after I recently published are the options market’s odds any good? which showed that insofar as the market’s pricing of option spreads is “wrong” compared to what was realized, the bulk of the edge is coming from the risk-free rate drift assumption that is required to ensure “no-arbitrage”. If you use a more empirical, markets-tend-to-go-up drift assumption, the mispricing goes away, but my claim is that someone trying to exploit options because of the risk-neutral assumption is just underwriting equity risk premia and conflating an option edge with harvesting beta.
correlations in the tails
Philip: Correlations are tricky. When? And where? In the middles. Like equity and fixed income for the past few years and decades. It’s tricky. Right? It can change from minus point three to plus point three, and that ruins people. So estimating that is really hard. It’s really slippery. But what about in the tails? In the tails, it’s much easier, isn’t it? We have a much simpler challenge in the tails. In the tails, we know the correlations are one. The only real challenge is it gonna be plus one or minus one. We can talk about some tools we’ve developed for that kind of thing using regime based correlation and stuff like that. But already, we’ve simplified it because we know that the correlations will be very, very high, and we’re managing to the tails. It does rhyme with expected return over variance. The way we look at it is skewness or expected tail gain divided by expected tail loss, and then that whole thing divided by expected tail loss again. So you get exactly what you were talking about where you have expected tail gain divided by expected tail loss square, just like you have mean variance optimization instead of mean standard deviation. So the downside plays a much more oversized role. That’s the most important thing is on the downside. So if we can get estimates of correlation for… On the tails, the downside 10% tails on the downside, that essentially answers the questions that we need. We don’t have to worry about the middle in correlation. We don’t even have to worry about the upside correlations.
Corey: I would love to hear you talk a little bit about that regime dependence because I can certainly think of regimes where certain assets might have positive correlations and other regimes where they would almost certainly have negative correlations. Can you talk a little bit about how that flows through the optimization process?
Philip: Again, remember that we’ve simplified it to just thinking about tails. So one of the things we can do is look at the current levels of volatility, let’s say equities and fixed income, relative to their own history. If they are both in the top quartile, the top 25% of their history, equities are elevated and fixed income is elevated. What do you think that tells you about the tail correlation in the current environment? If there’s a problem, they’re probably positively correlated. It’s probably a plus one correlation. They’re both very elevated. People are worried about both things. What if one of them is above… It’s in the top quartile and the other one’s in the bottom quartile? So people are terrified in one asset class and not at all in the other. Then maybe we’re thinking about, like, a minus one tail correlation environment. And if neither of those extremes are in the situation, we’re both kind of around fifty fifty or something, then we’re closer to a zero tail correlation. That’s a simple way of doing it. We have more complicated things where we would also blend it with recent or very modern, like high frequency correlation measures just to ground it towards something more dynamic. But that’s the essential approach is first simplify the problem and then use simple proxies to get at what we’re looking for.
You should realize by now just how rich in substance this interview is. Philip is telling you that in a market shock, the correlation of the vols themselves as assets move around is a clue as to what the tail correlations are in the tails. This is why those macro risk radar charts are interesting…you have events where all assets get smoked (gold, equities, bonds, etc) and some where the correlations diverge sharply (during covid bonds ripped higher as equities fell). Vol surfaces gauge stress, so by considering not just the asset moves but how the vols change given those moves you can learn what the market expects if the stress becomes more acute (ie in the tails).
It looks like Philip gave you enough to point an LLM at the transcript to work out the math to initiate your own monitoring framework.
what risk means
Corey: Alan, the process that you and the team have built is around what you guys call your dynamic risk management process. When most of the industry hears risk, when we talk about academic papers being published around risk, they’re traditionally talking about volatility. What does risk mean to you when you think about it both conceptually and from an actual measurement perspective?
Alan: For us, what we believe is that risk is the chance of falling short of the client’s objective, not necessarily just how much the portfolio moves. For us, if we take an objective based approach, and we can define various scenarios for a particular client or a mandate. What we like to look at in risk spaces, the chance of not delivering what the client is looking to achieve. When people think of risk and they think of volatility, the other problem that they will experience is that volatility treats gains and losses alike. What we do is we try to separate out the upside risk, which for all intents and purposes, could be something that you want. Downside risk, on the other hand, is something that we look to tilt away from. From our perspective, we think of risk not necessarily as a volatility measure, but more so a probability measure, something along the lines of what are the various outcomes available, and how can we tilt those outcomes in the direction upside of what the client wants and away from the downside of what the client would not want.
when dynamic risk goes mainstream
Corey: One of my favorite questions to ask people who are in the portfolio manager seat is, if I’m doing due diligence, what questions should I be asking you? Because no one knows where the skeletons are buried better than the people who are actually running the money. And when I asked you that question on the pre call, you said, well, the question naturally that should be asked every time is, well, if this is so great, if everybody starts doing it, can the opportunity still continue to exist? … If all of a sudden every asset allocator everywhere said, oh, you know what, you’re right, we should be taking these ideas and these measures from the options market. These are great implied return and risk measures and we’re going to build our portfolio that way. Can this still work for you as a means of managing money?
Philip: I think there’s three ways of answering that. One is it’s not the current situation. Why is that not the current situation? Many of the biggest investors are still very constrained. They’re benchmark relative. They can’t deviate too much. When things go up, have to mechanically rebalance. So we live in a world of constraints that prevents that from happening. Imagine you’re driving on the street, you have GPS traffic enabled, but nobody else is allowed to use it. There’s a law. You cannot use GPS. You can’t use any traffic enabled navigation system. What would happen if suddenly more people did? Well, traffic would work better. Right? There’d be fewer traffic jams. We would live in a better world. I remember this is a question I asked Myron early when I first joined because I was as amazed as you are. It’s a reasonable thing to focus on dynamic risk management, take information from option market, reallocate. Why doesn’t everybody do that? The real question is, how can we know when it’s done? … How would we see in the world when this has become so commonly accepted that it probably has gone away? And he gave one of the most amazing answers I could imagine. … He said, we will know that this approach has come to be accepted, and this is his vision. He wants this to happen. I agree. Wouldn’t you? It’s wonderful to be the only person with GPS traffic enabled navigation when everyone else doesn’t have it. But, actually, your life overall would be better if everybody had it. You would be less stuck in traffic. You don’t have to go into side roads all the time. The world would be a better place. That’s Myron’s vision. It’s a very inspiring one. How would we know? One of the ways we would know is look at any fund, any investor. They all have benchmarks, don’t they? And those benchmarks are static. It’s S&P 500 or sixty forty or something else. It’s almost unimaginable now, but what he imagines, and I believe it will happen some day, people should have risk managed dynamic benchmarks. They should have benchmarks say down 60%, down 30%, drawdown or whatever it is, right, not just a sixty forty equity. When that happens, it’ll be obvious just as it would be obvious when everybody suddenly has nav enabled on their cars. The institutions would evaluate how they’re doing business, not to beat a static benchmark, but to manage their risk. And one of the problems with a static benchmark is it feels passive. Passive is a beautiful word. It sounds gorgeous. I’m passive. I’m not doing anything. I believe in the market. But you’re not actually passive. If you have a passive dollar investment and we know that the risks fluctuate just like they did this year, there’s some months that are just far more volatile than other months, then your risk exposure in those months is far greater component of your overall risk than the other months. Just like you don’t want to have too much of your cross sectional risk in one particular asset class, you don’t wanna have too much of your time series risk in one particular period of time. And we know that goes in regimes. We know that risks change. So a passive dollar investment like into the S&P or a 60/40 will actually have wildly changing risks that you’re not controlling. Same as when you’re driving. You suddenly have lots and lots of traffic. You’re not doing anything at all. I’m just being passive. I’m just following the guy in front of me. I’m not banging into him. Great. Good. You’re beating the benchmark by not dying, but it could be a much better world if we all manage our risk in the same way dynamically with this vision that Myron has. That’s when we’ll know that it’s taken over and we’ll all celebrate it.
I won’t belabor the point, but you can appreciate Philip’s answer by remebering that there is such a thing as wall time (named for the clock on the wall) and vol time. Not all 15 minutes are created equal. Just like we acknowledge the difference between time-weighted returns and asset-weighted returns, we acknowledge that how and when returns happen change our interpretation of our exposures.