Re-Run: Cliff Asness | Money Stuff: The Podcast

That's the thing. You look like you're on a podcast when you got the headphones on.

Part of it is like with Bloomberg podcasts behind us.

That's true. Part of the reason this has to be on video is because Matt shaved. Matt has had a beard for the past... I've had a winter beard from like Christmas break through Memorial Day.

Okay.

Yeah.

Yeah. My kids didn't care that much, but one of my sons said, 'It's a different daddy.'

Yeah. Does Mr. Clean not have... He doesn't.

Mr. Clean. You know, clean. Come on.

Mr. Clean Shaving.

Thanks for having me. I always like to ask hedge fund managers like what do you do for a living? Like what is the like economic function of like the business that you run?

Even if it's about AQR, what you do for a living is, and people might not like this phraseology, but you're trying to predict what happens to securities. You're trying to buy ones that go up either in the absolute or more than some benchmark and sell ones that do the opposite. The broader question, which I think is behind what you're saying, is what do you do for the world by doing that? And they overlap, but they're not exactly the same thing. Something can have a net positive effect on the world, even if you're not waking up, and I'll admit this, I don't think most people in their jobs wake up and always think that way. And certainly not active managers just go 'I am just making the world a better place.' They're thinking is Nvidia undervalued, is it overvalued. The things that I think you do for the world is first take the other side of positions other people disagree with you on or don't want to bear. That can take two forms. That can mean one of you is biased and wrong, you hope it's not you, but in that case what you do for the world is you move prices back towards not necessarily all the way to in the abstract the correct price. That's hard to define actually, but something is mispriced and you take the other side of that. It could be a risk premium other people don't want to bear. In a lot of strategies, it's not necessarily that they're making an error. If a merger's announced and 3/4 of the pop you should get if it closes happens, a lot of people might not want to stick around for that last quarter. And if you're willing to take the other side of that, maybe you get paid a little for doing that. That is a service to the market. People want to get out and you're helping. That's kind of the positive side, but again, it's not what you're thinking about when you do the trade. There are positions active managers will take that are not about that. Let me put it this way. If what you're trying to do is predict returns, you can predict returns because the price is moving towards truth. But you can also make money if you predict the price moves further away from truth. You know, if you're a momentum meme stock investor, and doesn't mean you can't get that right. You know, I think of that a little more as trading than investing, but they all come together. And it even gets complicated within some famous quantitative factors. One famous quant factor is the momentum factor. I asked a finance professor what should I ask the thousands and he said you should ask him if momentum trading makes markets more or less efficient.

Is the explanation just that like there's a correct price and momentum is trading from below to above the correct price?

So, did you make money on GameStop?

You never went back and checked?

What did your Twitter look like after that?

We'll keep going.

Oh well, we'll find out.

I had to discover that for myself. So my Twitter got ugly for a while. You may have noticed. I've gotten, I think, a fair amount better at this, but I used to be pretty bad about responding to ugly, which you learned your lesson on that.

Less so than I used to. At least I think I'm better. Maybe I'm wrong, but I became public enemy number three to the meme stock crowd for a while. I did not really... Ken Griffin was number one. I don't believe Ken did this, but it's the whole pull the buy button thing. And oddly enough, Gary Gensler was clearly number two. Cuz they thought he was covering for the manipulators and the naked shorts and whatnot. I've met both. I know Ken a little better than Gary. I don't think there are any two people on earth less likely to be in cahoots than those two. I think they're on the opposite side of most issues, but that was the theory. But both Ken and Gary are too smart to respond to them on Twitter. So, I certainly became the most actively engaged. And then I never did that again before today when I've accidentally done it.

Yeah, a year ago.

The same crowd or just like...

Oh yeah. Yeah. You do innocuous. I was obnoxious, so I kind of deserved it. You didn't deserve it.

I haven't seen Dune Two. I saw Dune one. It was good.

No, you didn't, actually.

You were asleep.

Um it would have been better for me if we didn't have the call because then I could have just said I don't know. I rarely know individual stocks and if I do know I'm probably not happy. I know. And even then it's like we lost 20 bips on that today which is a giant number for us to lose on one stock. And even then I probably don't notice.

I might be told about it. For us it's whether these 750 stocks beat these 750 stocks. I don't memorize 1500 stocks. Now we take a fair amount of risk. Some funds more, some funds designed to be less. And that's about the size of these two positions. So when I say small, I'm not saying we're not both taking risk and trying to generate pretty decent returns. But if you just think about it, a quant is playing the odds. They're saying a firm, a company with these characteristics, and this can be old school factor quants from the 1990s. These can be modern machine learning, but with these characteristics tend to beat these characteristics. If that's all you know, and it is all we know, why on earth would you take a lot of risk in any one company? AMC really could have done well. It could have been bad on every single thing that on average doesn't work. And it could be a special situation that we don't understand. Something could be good on everything and the CEO can embezzle all the money. We don't want to take a lot of risk on any one thing cuz we have no insight in that. It's risk for no return.

I think it's moved closer. There are still differences and maybe some of the like momentum if it's overreaction if you're riding momentum. I don't think a Graham and Dodd manager does that. So I don't want to push the analogy but this came out very very early in my career. This is like 30 years ago. This is like my Goldman Sachs days. I started hearing a lot of active stock pickers. Some I'm still friends with. One guy in particular. I was laughing at them and I was telling my friend, 'You all say the same thing. You all say you're looking for valuation plus a catalyst.' That's like a... I don't know if everyone says it, but I've heard it many times. And I'm making fun of him and at some point he looks at me and goes, 'You do value and momentum.' And it was a gotcha. He won that round because I'm like, okay, I see your point. So even back then, you can think of those two together. We literally add them up. But if you think of them as this holistic system, we're looking for cheap things that are starting to get better in price or fundamentals over time and not... now I'm talking I'm not talking the very modern stuff, alternative data, machine learning. I'm talking just classic quant stuff that's been academia and then ports over to applied. Profitability is a factor. Robert Novy-Marx wrote a great paper that we all incorporated where all else equal given valuation and momentum if a company is more profitable on some famous scales gross profitability ROA ROE that's some degree of positive. Low beta investing, two of my colleagues Andrea Frazzini and Lasse Heje Pedersen, resurrecting stuff Fisher Black did and they're very good about saying Fisher did it first, that lower beta stocks if you're a famously a capital asset pricing model person. They're supposed to on average underperform higher beta stocks. That's the main output of that model. It doesn't work. I mean, it's one of the largest empirical failures ever. It doesn't work in any kind even outside of stocks. People test it in other places. Therefore, it kind of makes low beta stocks a little bit of a free lunch because they are lower risk and they keep up. If you actually go read Graham and Dodd, they're not just buying low multiples. They're much more holistic than that. You know, high quality companies that have a moat that have some kind of margin of safety, I think was the term they use. Margin of safety and looking for low risk doesn't sound so different. So over time, I've thought at least a core amount of what quants and academics if you take them as a whole are finding is the full panoply of stuff that a Graham and Dodd investor... we do it very differently. Again, we're betting on the concept working on average. They are using it in a soft or a hard sense as a screen to look for candidates and then they're trying to learn a lot about that situation. Their upside is if they learn a lot about that situation, they could be more reliable than my, you know, hey, we could be wrong about AMC. And their downside is they better be right. The concept can work and they can still lose if they're wrong about the specifics. So, over time, I've gotten a little less hubris about this. I think quants caused the problem. By the way, I think when Gene Fama and Ken French started looking at like price to book in the late '80s and early '90s, and there were other people who did it, too. I'm just a Chicago guy, so I'm going to just go with Fama and French. They did it best in my very biased opinion. I don't think I'd have to go back and check, but I don't think the first few papers used the word value investing. Over time, low multiple investing in the quant world came to be called value investing. And in the Graham and Dodd world, they'd get kind of mad at that and they'd be like, 'It's not value investing. There are plenty of low multiple companies that deserve to be low multiple and there are plenty of high multiple companies that deserve it.' And I think over time, the quantitative process agreed with them more and more. So, I still think it's a communication problem because they'll still talk about the value factor. In the quant world, that's just low multiples. And if a Graham and Dodd manager gets mad or any old school active stock picker gets mad at that, I'll just say you're right cuz value implies a more holistic thing.

If you look at machine learning where either to construct factors, one of the best uses of ML we found is a subset of ML called natural language processing where you take textual data and you try to say is this good news or bad news. Quants have kind of done this forever. You get a transcript of an earnings call and the old school way to do this for a long time was you count up good words and bad words, good phrases and bad phrases. Increasing plus one and you parse the whole thing and I'm sure you guys immediately see the problem. If the actual sentence was massive embezzlement is increasing then you were off on that plus one. Quantitative stuff can survive doing some horrifically stupid things in isolation. If 53% of the time increasing is a good word then 47% of the time you were stupid. Turns out natural language processing or NLP if we want to sound like the cool kids that is taking that same data and training a ML model to say what predicts and what doesn't predict. It of course never gets near perfect. At the end of the day though we believe it does a lot better than the word count methods and is additive to a model. But importantly and I think this was your point Matt. It's not qualitatively different. We've looked at both price and fundamental momentum forever. Fundamental momentum, the classic measures of things like are earnings being revised and you want the revision. You want the new news up faster or slower? Are earning surprises coming in positive or negative? And this is this anchoring and adjustment idea that if that's good or bad news, you can make money trading on it because it's not fully incorporated in the stock price. Parsing an earnings call poorly as in the past or better now. We think of it as just another form of are good things happening in the near term and if so they're probably under appreciated. So yes, I don't think it's changed dramatically. It's much more of an evolution rather than a spirit change but we got bigger tools in the tool chest now.

Yeah. As someone with four kids in or around college age, my wife and I, and she's done a lot more of this, have spent a lot of time trying to figure out what careers will not be utterly destroyed by ML and it's not an easy question.

Terrible answer. Um, so we'll all be podcasters. But yeah, it's another example. I always say these useless statements like on a sufficiently long horizon, we're all replaced by machine learning. It's a question of what...

We're getting way too philosophical. But the transition to that can be very painful and weird. Yeah.

I'll take it.

I will say he may be one of the only hosts of a podcast to read that paper. This is not a simple paper, right? Though he writes very clearly.

If you can do it.

It has changed a bit. We do trend following on macro assets, old school CTA stuff. We think we've made it new school by incorporating fundamental momentum by doing a lot of more esoteric markets. So we think even that's had a march of progress to it. But all else equal, I wrote my dissertation on momentum in individual stocks for some reason that I cannot explain. If you're using past returns to predict the future, just what's going up will keep going up and vice versa. To pick individual stocks, the whole industry calls it momentum. And if you think markets tend to keep going the same direction, everyone calls it trend following. Same thing. Starting in I think 2008, we started, we always had it in our macro models, but we started formally offering separate trend following products. It doesn't take hero bets on any one market. It's not Gazarelli selling all the stocks a minute before October 19th of 87. I'm dating myself. My wife's birthday is October 19th, which always gets a little amusement. Crash 87.

There have been some jokes over time. So trend following is essentially market timing, but it's highly diverse. Many small bets, whatever's been happening tends to keep happening. That's the main way we'll do market timing.

That was kind of a toy model to illustrate a point, right? I know we're not taking a lot of risk on that model. Market timing I still think is quite hard. Might there be advances in it in the future? Absolutely. But are we taking significant risk in it now? Aside from trend following, not really.

Oh, there's so much to talk about here. First, there are a lot of reasons we do it. One is just personal consumption. We grew up in this world. We like being part of it. We were interested in this stuff in a purely academic sense before we got seduced into making money.

There are people including our two Yale professors I don't know if would have come to AQR if we said you can never write about this. We do a crass calculation though. If we think there is something that we are relatively unique on entirely unique or a very small handful of people know this we won't publish it. The optimal time...

The optimal time to publish a paper is after you've made money from something for 11 years and an hour and a half before someone else is going to publish the paper. And we cannot get that right. I can think of one example of momentum and factors. The fact that factors themselves like value, profitability also exhibit momentum is something that we've traded on for many years that we've always refrained from writing a paper on cuz no one else had. And again, we knew we couldn't be the only people to do this, but we didn't think the cat was out of the bag. And then someone else wrote the paper. And I'm sure we've done this to other people too because you never know what they're doing internally. So life, you know, you do it long enough, life works out kind of fair. But that is the goal.

You could call it childish, but it's human. It's fun to discover something.

Maybe a little bit. Maybe a little bit. So far, it's still worked wonderfully. The trend is still your friend when it comes to factors, but it would be a fireable offense for AQR to publish something that we thought was truly proprietary. Let me give you my favorite example of this cuz it came up recently. One of the things that we've gotten into in a fairly big way, as have some other quants, is what's called alternative data. Those are new data sets that people put together with sweat equity. The classic example and the only one, and this is the point that I'm allowed to talk about, are credit card receivable data. But where are credit cards? Cuz that's been discussed a million times.

Yeah. Well, that's the point is this stuff is arbitrageable. I mean, it can go away. Value, be it the narrow quant sense of value or the more broad Graham and Dodd sense of value, is trying to take advantage of, I think, basic human nature. I think spreads between cheap and expensive are wider still than the historical average, not tighter. I don't think there's a lot of evidence that so much capital is in that that it's making it go away. If you get a short-term information advantage because you have put the time in or paid someone who's put the time in to create a new data set that other people don't have, they're going to have it eventually. So I am talking to the Australian Financial Review. I don't even know if we had discussed alternative data beforehand, but he asked me about it. And I said to him that our head of stock selection, a fellow named Andrea Frazzini, has asked me, I might have said told me, but has asked me not to talk about these things. There are things I'll talk about, but there are things we think are, you know, overused word, but are true alpha that the world doesn't know. And I think it's reasonable. So, I really can't talk about them. He writes the article and what it says is mostly that, but instead of saying he's asked me not to talk about it, it says Andrea Frazzini won't tell me what we're doing in alternative data, which it has a slightly different connotation. It has a connotation of adult old man, don't worry about it. We're doing some stuff here. So again, it's an example where there certainly are things we won't talk about.

I read most of the papers. I don't read all of the papers. I at least skim all the papers. I know the gist. We do a lot of different things at this point. Once you're a quant, you want more factors and you want to trade it in more places. And I will admit that there a week doesn't go by where I don't ask someone for a review of what we're doing somewhere. At some point I approved it.

Yeah. But also just, you know, some of this is pretty decent math and the old mathematicians do their best work in their 20s. I can tell you I'm probably still decent in math, but I'm not what I was when I was 22.

Yeah, I think there's some truth to that.

First I think there's going to be a correlation that the closer you get to the high frequency world the more you're going to be more in the pure math realm. My very tortured analogy is quantum mechanics versus Newtonian physics. When you get into the high frequency world, first you have a ton of data by nature. You just have a lot more instances and a lot less theory. So turning yourself over to the data more so than having an economic rationale is far more rational and that becomes a more mathematical exercise. We tend to think in our long 700, short 700 stock portfolios, average holding period is maybe 9 months, maybe closer to a year at times. That's nowhere near high frequency. Yeah, high frequency when I wrote my dissertation and I did have this in there was a monthly contrarian strategy. Now we're talking, you know, subsecond kind of thing. I think when you're going to have a medium holding period strategy, as I think of us, we're not Warren Buffett, but we're not HFT. Even the machine learning stuff needs some economics too. You simply don't have enough data and there's too much of a dimensionality. Even with Brian's virtue of complexity, you need to give it some structure or else it's going to overfit and go mad. So I think that's the reason. Not saying we'll never do something in a higher frequency world in which case we'd probably have to shift more, right? But in our world, being a mathematician, being an excellent programmer, but also having the economics behind it, it's kind of what we're looking for, right?

I think that's accurate. I do have a couple of Renaissance observations.