Invest Like the Best with Patrick O'Shaughnessy - Leigh Drogen - Quant vs Traditional Investors and How Alphas Become Betas - [Invest Like the Best, EP.41]

Starting point is 00:00:00 This podcast is sponsored by CFA Institute, the Global Association of Investment Professionals whose mission is to lead the investment profession by promoting the highest standards of ethics, education, and professional excellence for the ultimate benefit of society. CFA Institute serves a global community of investment professionals working to build an investment industry where investors' interests come first, financial markets function at their best, and economies grow. The chartered financial analyst credential is the most respected and recognized investment management designation in the world.

Starting point is 00:00:30 The views expressed in this podcast do not necessarily represent the views of CFA Institute. Hello and welcome, everyone. I'm Patrick O'Shaughnessy and this is Invest Like the Best. This show is an open-ended exploration of markets, ideas, methods, stories, and of strategies that will help you better invest both your time and your money. You can learn more and stay up to date at investorfieldguide.com. Patrick O'Shaunisee is a principal and portfolio manager at O'Shaunicee Asset Management. All opinions expressed by Patrick and podcast guests are solely their own opinions and do not reflect the opinion of O'Shaughnessy asset management.

Starting point is 00:01:09 This podcast is for informational purposes only and should not be relied upon as a basis for investment decisions. Clients of O'Shaunity Asset Management may maintain positions in the securities discussed in this podcast. I've often joked that this show should be called This Is Who You Are Up Against, because I am so often having conversations with brilliant people across the investment landscape who are effectively my competition and yours. This week's conversation fits that description because it gives you an inside view into how things work among some of Wall Street's most competitive investment firms. My guest is Lee Drogan, who has worked as a statistical arbitrage portfolio manager and who founded and now runs Estimise, a data company which works with some of the world's largest hedge funds. Our conversation centers on the massive shift from what we call discretionary portfolio management, basically stock picking, to a landscape that is increasingly dominated by quantitative investors of various types. We talk about how any investor might hope to earn Alpha and how doing so is harder and harder. There are so many great stories in this episode told by someone with the perfect career experience to know how the system actually works.

Starting point is 00:02:15 After many episodes where I've been learning on the fly about topics like venture capital, permanent equity, or even health, this episode marks a return to my world of quantitative investing. I think you'll learn a lot and that you'll likely finish with an even deeper appreciation of just the type of investors that we are all up against. You can find show notes for this episode at investorfieldguide.com forward slash drogan. And now please enjoy my conversation with Lee Drogan. All right, Lee, well, I've been really excited for this one because I have intentionally avoided most discussion of quantitative investing on the podcast just because I've spent most of my 10 years of my career doing this. And I thought it'd be fun to branch out a little bit.

Starting point is 00:02:58 But I'm not ready to dip my toes back into the water. And we met very briefly at a, I think it was a, Quant Conference, actually. I was standing talking to Bev Fabor and we started chatting about something very specific, which we'll get to in a minute. But then I also came across a piece that you wrote on LinkedIn, which I thought was one of the most thoughtful pieces on kind of the current state of the investing spectrum between traditional deep fundamental work and quantitative investors and kind of how that balance might look in the future. And that's really where I'd like to focus today because I think it is a fascinating shift that's that we're watching real time and that a lot of

Starting point is 00:03:38 big people, even famous people are being left behind by this move towards quant. Or they're just simply opting out because of it. That's right. So I'd love to start with what we were talking about at when we first met, which was I said something to you like, yeah, you know, we used to use this this Ivis data set and look at like analyst revisions and it totally stopped working as a factor. And you said, yeah, yeah, that was me. I did that. I was the one that. That's one arving it. So can you use that story as a way of describing kind of your early career and background and investing and then we'll get into all the details?

Starting point is 00:04:12 Yeah. So I actually, my educational background is in behavioral economics and war theory. I actually thought I was going to like Rand Corp, CIA, like State Department, that route. And it was interesting. I got an opportunity to do an internship at a quantitative hedge fund between my sophomore in junior years and loved it. And it was incredible. And at the end of the summer, basically, I stayed in New York instead of going back to school and finished up at night at Hunter College here. And that world was so interesting to me mostly because it's all just numbers and letters, right?

Starting point is 00:04:47 You don't have to deal with the irrationality of people. You just get to deal with like, there's a variable and it's either correlated or it's not. And if you can find that correlation and you backtest it and the out-of-sample works and the P-values are good, like you have a strategy and you can make money and it's all very kind of straightforward. Look, it's science. And it's the difference between science and what politics is, which is not science. And the not science part kind of, you know, frustrated me. And so I started my career at this hedge fund, Geller Capital up in White Plains, doing the reverse commute from New York every day and fell in love with it, you know, just fell in love with the process. And then from there, and we were doing exactly that. We were doing

Starting point is 00:05:27 analyst estimate revision models, earnings acceleration, history surprise, beat rates. We were basically attempting to arbitragee inefficiency in that cell side analyst estimate data set. Can you describe in some detail what that inefficiency was? Because I think it's a really great example into how a Stad Arb strategy or an informational edge or some weird inefficiency in the markets was first identified and then Alpha sort of extracted from it. So if you could describe in some detail, because that's a different.

Starting point is 00:05:57 I think it's a great example of how this works. Yeah, so a lot of the stuff that we did was not necessarily cutting edge. It was just we did it well. And these strategies had been around for, I don't know, the academic research had probably been done 10, 15 years earlier. And so the basis of this is that the cell side analysts are inefficient in how they produce their estimates, largely coming about because of institutional biases, corporate access, investment banking pressures.

Starting point is 00:06:24 They also tend to not update their estimates in the last couple weeks before the earnings report, partially because they're lazy, partially just because compliance is really hard to get new research reports out the door. And they also tend to love to allow the companies to beat their estimates because they want to get on the earnings column and be like, ah, great quarter guys. But like, nobody pays attention to the fact that they move their estimates down 15% in the last like five or six weeks. So you can see all of the patterns in this data and the average movement in estimates up or down, what percentage of the time does the company beat or miss? And then just very simple linear regression models of like what happens if a company beats four quarters in a row

Starting point is 00:07:04 by more than a normalized 10%. The next six months, the stock is going to outperform on a momentum basis. Like, it absolutely will. And we've known this for a very long time. We know that companies which beat their consensus numbers by a significant amount will see post earnings drift. And you can collect, you know, residual return in the three to five days afterwards very easily. We know that revisions are highly correlated to equity direction on a normalized basis. So what we did was we basically attempted to normalize everything and then basically Z-score the entire universe and go along the top deciles and short the bottom deciles in a market neutral strategy. And this works incredibly well. And so I was there from basically 06 on through the crash and I saw a lot of everything.

Starting point is 00:07:53 And it was really fun seeing this thing work and eventually kind of ran one of the strategies and that was the start of my career. So one of the key points in the story is when things stopped working. At the turn, so at the turn in late 2007, when momentum all stopped working because we were, you know, we were coming up the crest of what was going on, everything stopped working for like six months. And then it all started working really well again. Yeah, we lived through in 2009, any quantitative manager that used momentum at all, especially

Starting point is 00:08:23 those that used it exclusively got absolutely destroyed in 2009 with a complete inversion whipsaw. And when we looked back, there had only been really five periods in the last 80 years that were as bad as 09 was for momentum investing of any kind. So it's amazing to watch stuff work really well and then stop completely. So the manager, so this was crazy. The manager at the fund went completely to cash about halfway down in 2008, halfway down the crash in 08. And we had made a bunch of money at the coming out of that turn on the short side he went completely the cash and I watched him trade index futures for the next six months incredibly well too like by hand just by feel and he caught he didn't catch so like we had that the next half of that first big drop and he caught

Starting point is 00:09:09 a decent portion of that then he was out he didn't catch the turn because he was completely out and then he caught the move back up and then he caught the move back down into march and then he was and then unfortunately shut the fun down and my mentor he eventually passed away from skin cancer, you know, shortly after that. And it was, you know, it was just, it was crazy watching him do that in the twilight of his life, like, just operating so well. So I'd like to get into this actual framework now and maybe hopefully have you kind of build it on the fly. So I always think in terms, or everyone always talks in terms of sources of edge being informational, analytical, behavioral. So first question before assuming you agree with that framework is,

Starting point is 00:09:48 do you think that those are the proper dimensions to separate any sort of potential alpha edge into? Yeah, I mean, I think the informational edge is becoming more arbitrage these days. Although it's interesting, with more information out there, new data sets like ours, new data sets all across the universe of different types. And we're at a new inflection where you need to be more aggressive about arbitraging that informational edge because, I mean, look, Rentec buys everything. And so if you're not buying some things, at least, you know, you're going to not perform well relative to them. The analytical part, there are not enough quants out there to do the research simply.

Starting point is 00:10:29 Like we just literally don't have enough of them out there, especially when the discretionary world is attempting to hire them now as well. So, yeah, you can have an analytical edge by having a really good team. And then I think the behavioral one is the most important one, especially for the discretionary guys. Quants, we do see some irrationality on their part sometimes where they'll test something. They will look at the out of sample and they'll look at P values and the P values will be really good. But then maybe the last three or six months of the performance of a 10-year track record will be like not quite as good. And they'll be like, well, we want to wait a little bit to put something into production. Like, no, you shouldn't do that. So there is some irrationality on their part when they

Starting point is 00:11:09 turn things on, when they turn things off. But really, it's the behavioral. on the discretionary side is probably the most important thing right now. Can you dig into that a little bit more? So what do you mean by the most important thing, meaning for a firm to have an edge, they need to behave well, meaning they don't have style drift, they don't shift in and out of strategies, things like that? God, I mean, that too, but I think it's more just how do you make decisions? Like, what is your, what is your decision flow process? And I like to use this really specific story. It's a very famous behavioral study, and it's in Michael Lewis's new book, which is really good if people haven't read it,

Starting point is 00:11:44 the undoing project. And the story is basically about a study that Thaler did with a colleague where they give a set of x-rays to these very famous oncologists and then give the same set of x-rays to first-year medical students. And they asked the oncologist, how do you determine whether it's cancer or not? And they give them basically a 10-point rubric of like how they determine and very quantitative structure of things. And they take that rubric and they give it to the first year of medical students and they say, tell us whether it's cancer. And then they give the same slides to the oncologists.

Starting point is 00:12:19 And the oncologists end up being no better than random at guessing whether it's cancer. And the students are like 70% hit rate. Why? Because the oncologists who are experts don't use their own damn rubric. And we find this is the exact same way in the discretionary trading world where if you ask any of these guys who run a billion dollars in discretionary long short money, they will tell you exactly how they're supposed to be making decisions, and then they just simply won't follow it. So I think firms need to seriously codify the flow of decision making from stock selection

Starting point is 00:12:54 to position sizing, to market timing, to risk management, to just the whole thing. And I don't see many firms that have actually codified this very well. I'm actually waiting for somebody to put together a good piece of software and sell it to them to, like, modify the behavioral heuristics that they operate against. Because, man, I mean, I could pick like half a dozen off the top of my head that these guys are just so exposed to. And most of them are availability heuristic. They're going to operate on whatever information is, like, readily in front of their face the most that day. And they just shouldn't be doing that. And that's the difference between a good quantitative strategy based on science and what some of these guys are doing, which simply,

Starting point is 00:13:33 isn't working anymore. The Michael Lewis story reminds me of, you know, a tool gawande and the checklist manifesto and, and my belief that, you know, as an investor, the only interesting strategies are ones which are repeatable and therefore follow a pretty specific process. We'll get into it a little while how I think you could find really interesting advantages where it's almost a quantitative-like process with more subjective fundamental inputs. So fundamental company models or whatever, but still very strictly follow. following a decision process. Or even doing the thing that we do on this, we have this app called force rank, right?

Starting point is 00:14:08 And the whole point is it doesn't matter how you get to the decision of saying on a quarterly basis you believe amongst these 20 stocks, we do 10, but let's just say 20 in a specific industry, these 10 will outperform the other 10 and you force rank them in order. It doesn't matter what your fundamental inputs or technical inputs to that decision making. It's just that you ranked them. And if you can rank them correctly, well, then just go along, the top five and short the bottom five and you're going to generate alpha. But it's just like that portfolio construction and adherence to a model is so hard for some of these guys because they want to be correct more than

Starting point is 00:14:43 they want to make money sometimes. And that causes all sorts of perverse things. You mentioned some terms and a process earlier that I'd love to have you flesh out a little bit more just so everyone can be on the same page. So viewing the quant research process as sort of the scientific method, basically, I mentioned z score, p value, all these things. But the basic idea. Sorry, I'm incredibly nerdy. I think you're preaching to the choir in many cases here. But to define this process a little more cleanly because I want to talk about data sets would be time well spent. So if the basic idea is you've got some smart people who know how to look at data, you've got a data set, we'll call it a generic data set, and you've got some set of future returns, basically. That's what

Starting point is 00:15:24 a back test setup looks like. So from your perspective, how is good research accomplished? So if you've got those really basic three inputs, which is usually the basis for all this. What is the actual process? So first, you need to start with an ex ante hypothesis for something. We can't just go and say, well, let's run a bunch of regressions and see what works. Because if you don't know why it works, you won't know why it stops working. So I think this is one of the biggest mistakes that I see everybody make and why there are certain products in the market right now for discretionary managers, which I will not name that I don't like because they skip the two biggest steps that I think are really important. And the first is the ex ante hypothesis. Let's just say you want to see what happens when

Starting point is 00:16:12 oil goes up 10 percent, right? And you just start running regressions across the board to different assets and stuff. And you say, oh, wow, look, if oil goes up 10 percent, then these things go down or whatever. Well, if you didn't start with a, I think if oil goes up, then these things will go down. You won't know why that happens and why it works. And it'll then it'll then. eventually stop working and you won't get out of it. So you need to have a hypothesis for why a data set is actually has some kind of causality. So we start with that. Then we simply take our time series and we split it into two or we can divide it up into chunks. Yeah. Simply we like to, if you have enough data, you just divide it up into two halves. If you have a 10 year data set,

Starting point is 00:16:52 five and five run the regression on the first five, then run the regression on the second five. And then look at the P values and the P values are basically like, does it? the performance from the first half equal the second half roughly. And if those P values are good, you know that you have not overfit your strategy in your in sample and your out of sample works. And if you're out of sample works, then you can be pretty sure that going forward it's actually going to perform well. If you don't run that out of sample and you're just like, wow, look, the regression works on the first five years. Let's go put it into production. You have zero clue whether this thing's actually going to hold up at all. And I see on the discretionary side again,

Starting point is 00:17:27 that step is getting skipped a lot. And that's why the efficacy is simply not there for some of this stuff. This may be a little too in the weeds. And if we get there, we can always just gut it out. But I'm fascinated by this stuff. So as you pointed out, there's lots of ways you can do this. If you've got a long enough time period or enough data, you can literally just chop it in half.

Starting point is 00:17:45 Yeah. But what do you think about the order of those time chunks? You know, is it always an older half for the newest half? Second, do you do kind of random sampling, bootstrapping, that kind of stuff? Can do that, yeah. So what are your thoughts on? You know, we keep it, at Estimides, we keep it pretty simple with our quant research team and our process. And we have enough data now on the core data set about five and a half years to just do first half, second half.

Starting point is 00:18:10 When we were early on and we wanted to figure this stuff out in 14 when we wrote our first white paper, we did chop it up into quarters. And then we took every other quarter and then we looked back at the other every other quarter. And that worked fine as well. But with some of these data sets, and I think this is stuff that people, need to think about these days in crowd-tourcing, the panel changes. It's going to grow, it's going to morph, different people, different reasons they're there, different types of people. And so it actually is good to go first half, second half, to see if, like, the regressions that were run in the first half. And I'm talking not just about, like, alpha-generating models,

Starting point is 00:18:49 but our select consensus model, how do we overweight and underweight certain analysts? And behaviors might have changed. Our platform has changed. The interface has changed. The heuristics associated with how they make the estimates might have changed. And so you want to see if the same things there hold up because if they do, you know that there's some inherent quality of their decision making that is at play instead of just some heuristic that we're affecting negatively or positively. So there's a lot of different ways you can do it. We like now to just go for first half, second half.

Starting point is 00:19:18 Talk about what Estimize is and does. So it's a relatively new business, I think 2011 founded it. So just describe the dataset, describe the idea. described the idea. It's related to the story you told earlier about analyst estimates. So just a quick history lesson on, and Estimized itself would be great. Yeah, the concept is basically that if you crowdsource all these expectations from a broad community of by side, independent, non-professional individuals, industry experts, corporate finance professionals, you get a more diverse data set, less bias, less hurting. You get a larger data set relative to

Starting point is 00:19:53 the cell side stuff. And it takes out a lot of those inefficiencies and bias associated with that one. And so we collect specifically EPS and revenue estimates for publicly traded U.S. companies. We also do economic reports, so GDP, CPI, oil inventory, stuff like that. And then we do we do force rank as well, which I talked about a little bit before. And then what we do with that is we run a bunch of models through it. We overweight certain analysts. We score and rank everybody. And then on the back end, we output all the raw data to institutional clients. We have a big front-end data visualization platform, a bunch of emails. and then we turn the raw expectations data through our own quantitative research process into factor models

Starting point is 00:20:34 so that both discretionary and quantitative managers that don't want to use the raw data can get the direct alpha out of it. And those factor models are relatively shorter term kind of days to weeks kind of models. And there are some that are going to be coming out that are more like earnings yield focus, which are longer, you know, quarterly rebalance type stuff. how do you incentivize the people providing the estimates? Yeah, so that's the billion dollar question here and why I think my background of behavioral economics is specifically relevant to what we do.

Starting point is 00:21:05 So historically, the by-side discretionary guys will call around to each other before an earnings report. And, you know, like this is known as story that is the whisper number, right? And then they'll call their equity research sales guy Goldman and Morgan Stanley and say, hey, what are you hearing from your by-side clients? And this game is just like,

Starting point is 00:21:23 I knew guys before I started this thing that would share Google Docs amongst each other at different funds. And all of this is totally against compliance, by the way, of all these funds. It's certainly, it's not illegal, but it's completely against compliance. You're not supposed to be sharing information outside the firm. But they have to. And the reason they have to is the fundamental premise of fundamentals-based trading is, I believe that a company is going to earn X over the next year.

Starting point is 00:21:51 It trades at some Y multiple now. I believe it's going to trade at some different multiple then. I multiply my fundamental expectation by the multiple, and that's my stock target price. My alpha is the delta between whatever the market is pricing in in terms of consensus and some kind of terminal multiple. But the problem historically is you don't know what that true market consensus is because the sell side stuff, especially going a year out, is like not representative at all. So they come to our platform, and the deal we basically make with everybody is we're going to cannibalize the industry by allowing, you for free to see everybody else's data if you contribute to that specific stocks,

Starting point is 00:22:29 specific earnings report. And then obviously people contribute to like four forward quarters or eight forward quarters and then they can see all that data. And that's the deal we make with them is like maybe there's $400 million of revenue a year in our industry total earnings estimates. And maybe we've eventually cannibalized that to $100 million because we're giving away a lot of it for free. And then we get to own that data obviously on the back end and sell it everybody. But for them, they get to suit anonymously without putting their real name up there, just under like an anonymous account, analyst 4-5, 9, 2, 3, 6, put their estimates up, get scored and ranked, use it as like a utility for themselves to understand where they are relative to

Starting point is 00:23:07 consensus. And yeah, that tends to work pretty well once you have a critical mass of data. And so, you know, we started this thing in January of 12, we're five and a half years in. the thing really tipped sometime in like mid-14 when it when we had enough data and the system was well enough known that people just like flooded through the door and then it's you know just grown exponentially from there how many companies do you cover for a typical company let's pick a large cap one how many estimates are there quarterly can what give me a sense of scale yeah so we cover with three or more estimates about 2100 companies now and we have about 1400 companies with with 10 plus estimates, and for a name like Apple,

Starting point is 00:23:51 we'll have 1,000 estimates. But I think more importantly, so there's this interesting effect that takes place. For large-cap names, the cell side will have 30 or 40 analysts covering a name, but for like a mid or small-cap name, you might have seven or 10. The relative difference between 40 and 500 is,

Starting point is 00:24:11 like, it matters, but it doesn't matter. It's just totally diminishing returns there. And our data set is far more accurate. 70% of the time, our consensus numbers are going to be more accurate than the street. But really where it matters is the street will have 7 or 10 for like a mid-cap growth name, and we'll have like 60. And that is really where the gold is. And there we're going to be like 75, 80% of the time more accurate.

Starting point is 00:24:34 And just so much more representative of that expectation. And we also find in all the models more alpha in the mid and small-cap stuff. And that's the case for most stuff for most stuff. Because the large-cap stuff tends to be more efficient. but the coverage is basically, it's like 97% of the market cap of the U.S. universe. We don't do REITs and we don't do community banks and those are the only two things we really don't do. Could you tell that story about, in your piece, it was about cars and data on cars as an example of informational edge and why you need to be a part of this arms race if you hope to survive?

Starting point is 00:25:09 Yeah, I mean, look, everybody's just like running to catch up to each other at this point. So I was at a conference in London back in March, and one of these data sets that's up there, it actually wasn't presented by the company itself. It was presented by Tamara from Kwandle, being one of these platforms that has a whole bunch of data sets that you can buy. And they have this data set where they're getting the new insurance registrations on a daily basis for like all cars in the U.S. from this one insurer. And this one insurer has a big enough panel that they represent like a good enough sample

Starting point is 00:25:44 of how many cars are being sold. And so now you go from getting the car sales number once a quarter or once a month or something like that from the companies themselves to having a daily look at how many cars are sold. Well, I mean, if you're a car analyst or you're trading cars, like if you don't have this data set now, you're screwed.

Starting point is 00:26:03 And there's these kind of data sets are coming out all the time, especially with location. You've got all these apps now. And this is crazy. And a lot of people don't know this. It's just one of those things that like data nerds are getting into now. So there are these companies that basically have an SDK that they will put into all these different apps. And they will pay the app to put the SDK into the app. And I'm talking like thousands and hundreds of thousands of apps. And so the likelihood that you have one of these apps on your phone is very, very high. And what this SDK does is it sends your location data. back to this one company. And they now have a large enough panel of everybody to know how many people are walking into urban outfitters every month or whatever. And it's just, we're getting to the point where the data is outstripping the ability

Starting point is 00:26:55 for funds to actually just do the science and correlate it to outcomes in the market. But man, like next 10 years, this is where the stuff is. So the car data one's an interesting one because probably, started as a huge source of alpha, so the first batch of people that bought this data set and started incorporating in their trading strategy were crushing those that didn't. And then it becomes this arms race where, well, everyone has to buy it. So it's like table stakes. But the alpha can go away very quickly in the informational game. I get asked often, isn't all the informational ads just going to be gone? And then you look back at market history and it's just always progressing,

Starting point is 00:27:34 right? There's always the next level of age. It's like, are we all not going to have jobs? No, we're going to have jobs. Like, there's going to be jobs. We're not all going to get put out of work. And there's always going to be another data set. My favorite story from a past episode was short sellers that would hire recent college grads to go sit at the Library of Congress that got the 10Ks earlier than everyone else and just run to a payphone and call and you'd be a week ahead of everyone. I have a good friend that runs a company called Reargue Research. And his whole company is predicated on having these people in the distress debt courts, like when they do the filings, in the actual court, right?

Starting point is 00:28:08 And then they have this product that like pushes out a feed of this stuff. And this dude went from like $0 to $10 million from revenue in his business in like a couple years. It was incredible how fast this company grew. And it's, yeah, there will always be new data sets. And there will always be new heuristics and inefficiencies that the humans operate on relative to the data sets and the bad decisions that they make with the information available. So it would obviously be a very good problem for you to have as the owner of Estimized. but what about the same scenario playing out here where everyone's buying estimated data and any alpha that's in it is armed away? So I like to, so we obviously got asked this question at basically every meeting.

Starting point is 00:28:49 The some data sets have obviously different capacities than others. So that card, you know, the car insurance, you know, stuff, it's just operating on cars. And there's like a small set of companies. And certainly that thing will be arbed out like within a couple of years. IBS, the earnings estimate data set, took, and still is not arbed out completely, but it took like 30 years to ARB. Ours should go faster than that, but is it 15 years? Is it 20 years? Is it 10 years?

Starting point is 00:29:18 It'll happen eventually if we're successful enough at distributing our data everywhere. Now, it's interesting. Companies, and I think traders and investors should know what they're using in this frame when they look at a data set. where is that company in the progression of that distribution model, which is why we always get asked, how many clients do you have and all that stuff? Because they're trying to figure out where we are. So as a data set gathers more data and becomes more useful,

Starting point is 00:29:48 it has more and more alpha. And then at some point there's this crest where enough people are using it that the alpha kind of tops out. And then you go down that slope on the back end. By that point in time, if you're really getting arbed, the company's making a lot of money. Like you're a very successful company. I said, good problem. Yeah, it's a good problem to have. So I will say, we are still on the

Starting point is 00:30:09 upward sloping part of that curve where there's more alpha today than there was yesterday. How long that lasts? I can't, I can't say maybe it's another four or five years. Then the value in the data set starts to come down and that company has to go through if they want to instead of selling. And a lot of these data companies get to like $15, $20 million in revenue a year. And then they sell because it's hard to get past that. The reason being you have to go through this trough. And the trough can take years to get through. But if you get through that trough and out the other side, you become a must-have arbitrage data set that becomes table-stakes, and that's where IBIS is. And that's where a data set, like the short-interest data set that

Starting point is 00:30:49 market owns now called, and I'm forgetting the name, that is, it's table stakes to have that data set, the short-interest stuff. And if you can get through that trough, you can, you can become a $100 million revenue company, really, really large. Not many companies, obviously, make get through that. What's the most interesting data set that you've got your eyes on? Or maybe before you even answer that question, set the stage for how this world works. So

Starting point is 00:31:13 I always joke that what this show really should be called is this is who you're up against to try to discourage people that think they have some insight on Apple stock or something from ever doing anything about that. So you've got a good sense

Starting point is 00:31:29 into the world of firms like Millennium and Citadel and now points 72 and Ballyazni and these extremely sophisticated firms that are a part of this arms race, some of which are purely quantitative, maybe a Rentech, some of which came up as discretionary, but now realized the importance of quantitative, whatever you want to call the hybrid, come up with some name for it. I keep hearing this term quantum. Well, the point 72 guys, Matthew Granade, who runs their big data group, likes to call it systemmental, which I actually think is the best term for it, frankly.

Starting point is 00:32:03 it just sounds weird. Yeah, so we'll roll with Systemental. So maybe set the stage before I ask the question about some data sets for how this world works. So you could pick a firm, you could describe it however you want. But what has the evolution been in these very sophisticated sort of hedge fund platform type company? Yeah, they're shifting a lot right now. So let's just take Millennium, for example. Historically, it was known as like a hedge fund hotel where they gave you some technology

Starting point is 00:32:31 and basically said, here's some cash. We're going to manage risk on you really tight if you have one bad, terrible quarter, like you're out. But you have to go do all your own research. You have to buy all your own data. You have to do all your own stuff. So that's how it kind of used to be. Then you had firms that were more kind of like one book or discretionary firms that were

Starting point is 00:32:52 tighter together. And, you know, they would buy data and have an infrastructure and have a process, you know, all together. And then you had kind of the quant firms that were very much the same. way. You've got a firm like AQR, which is just like one research team that runs a whole bunch of different portfolios, both short-term and long-term stuff. And you have teams like Ballyazni, which runs pods of quants or Paloma, which runs pods of quants. They're all separate, but the firms will give some kind of resources to those pods. And then you've got the third kind, which is the most

Starting point is 00:33:24 recent, which I think is really interesting, which is WorldQuant. And WorldQuant is unique amongst the entire industry where they decided they're... going to have a centralized risk management and portfolio management team, and they're going to centralize the data purchase and infrastructure part. But then all the analysis in alpha generation is going to be done by a group of like 500 analysts all around the world that they basically contract to do this. And it's incredible. WorldQuant was one of our first customers, and I can actually say this publicly because they allow us to, and they use our data. I went to their conference in Puerto Rico a couple years ago. And I'm walking around and meeting all these people.

Starting point is 00:34:01 And all the analysts know about our data set. And then I meet some of the PMs and they're like, never heard of you. And I ask the head of data over there, I was like, what's up with this? And he's like, the PMs don't know what's actually in the alphas that the analysts put into this bucket that the PMs just pick out and create a portfolio out of. It's an incredible way to do it because it's like infinitely scalable and which is why they've been so successful. So there's a lot of different ways to kind of set up the firm. What's changing now, I think, is a millennium is, is, is, realizing they have to build a infrastructure to get more people onto their platform and support

Starting point is 00:34:38 them. So they're building a big data thing to support people. They're bringing in new data sets. You're seeing discretionary firms attempt to build these overarching kind of infrastructures as well, but it's moving slowly on the discretionary side. I think there's still people are trying to figure out like which steps to take before they start running. And some of them have taken some false, you know, steps forward. So do you think that the best way, if you're a discretionary firm, or even a new firm, right? Because we were talking before we started recording how there's this funny dynamic, as you see everywhere, it's almost like Thomas Kuhn, like progress happens, one retirement or death at a time, that some of these older firms are just not going to

Starting point is 00:35:17 adapt. Even if they were enormously successful and super smart and had an edge, that the nature of edge changes through time. So if you were designing from scratch a firm, Would it look like, basically like WorldQuant, where you've got this centralized data kind of team and infrastructure that's then spit out? It makes me think of Numeri. Maybe you could describe what Numeri is. Yeah, that's a really, well, so I can't even really describe accurately what Numeri is. I don't think many people can, frankly. WorldQuant, I think is the, yeah, I mean, I think they got it right.

Starting point is 00:35:51 And it was a crazy idea. Let's hire 500 people in India and Hanoi and all around the world. Some of these guys are like rice farmers that had a mechanical engineering degree. And it's incredible, but it's very much along the lines of our philosophy, which is crowdsource all this stuff, pick the best out of the haystack and go with that. And if you have enough people giving you models, then you'll find the models that work. I don't think many firms can re, you know, engineer what they've done. It's a difficult thing. And there's a data procurement thing that feeds the heart of that engine that is just incredible.

Starting point is 00:36:27 it's a machine. There's another company called Quantopian that I think is really interesting, run by a friend of mine, John Fawcett, which basically attempts to go even further than WorldQuant in crowdsourcing and disrupting the whole idea of, okay, they built this massive infrastructure platform that they just put on the web instead of keeping proprietary and allowed anybody to go in there and run regressions and build quantitative models. And then they built a hedge fund on top of it that 0.72 invested in where if you run your model out of it, of sample for six months and it works out really well, they may pick it out of the bucket, put money towards it and give you a percentage of the returns or something like that. I don't

Starting point is 00:37:04 know exactly how it's set up with how you get compensated, but that's really interesting. That can scale incredibly large. And then this numerai thing, I honestly, like I've met the guy a couple of times. I think he's super interesting. He was at the conference in London that I was talking about before. This is Richard Craig. Yeah. And I can't even explain, like there's a bunch of machine learning involved in it. It's also crowdsourced. It's also models. There's a cryptocurrency involved there. So can I try your explain? Please explain because I would love to understand exactly what's going on. So I probably don't know any better than you, but here's, here's where I am. This is by the simplest explanation. So the basic idea is at firms like,

Starting point is 00:37:42 like Rentech, to use the example of everyone uses, because they've been so incredibly successful, you've got incredibly smart, talented data scientists who are doing the process that we've described, which is finding data sets, understanding the relationship of that data set to future returns and some asset class and building trading strategies around them. And I want to come back to how you know when something stops working because that's so important. We didn't get as deep into that as I'd like to. But that's the basic process. And I think numerize interesting insight, just like yours, is this idea of crowdsourcing, which is that, okay, so you can have how many, even if Renaissance can have, I don't know, a thousand of these brilliant people, what if you

Starting point is 00:38:22 had 100,000. And so what Numeri has done is create a data set, which frankly, I can't understand because the number of data points in it is something like 240,000. I also didn't understand this. What is the training data set? So it looks to me, just because I work with some of these data sets, like CompiStat. Like the number of observations and forward returns. What is it observing?

Starting point is 00:38:48 We don't know. We don't know. Yeah. But so here's the, it keeps it secret. So the interesting thing is that, so you've got your variables. And you don't know what they are. They're normalized. So it's like zero to one or zero one hundred or decimalized.

Starting point is 00:39:01 I can't remember what it is. So it's a factor. Right. So it's some factor of something. It's some factor, but it's not 12 where you might think it's a P.E. It's some number that they've scaled. Yeah. So it's kind of blind to you. And then this is the thing that I don't really understand.

Starting point is 00:39:13 The outcome, the thing that you're trying to correlate it to is binary. So you've got, whatever, stock, identifiable. fire one, two, three. You've got factor score X, let's say it's 0.5, whatever that means. And then the outcome, which normally in our world would be returns, is not returns, it's 01, which makes no sense to me because the magnitude of the return is everything. Right. So I don't know what they're doing. Well, wait. So no, no, so it's interesting. It could be event-based stuff. No, no, but it's actually interesting because that corresponds a bit to the way that we run force rank, right? It doesn't matter how much more accurate you are with your rankings. It just matters that your rankings are

Starting point is 00:39:55 correct. I don't care if your number one stock went up 20%, and the number two stock only went up 10, just as long as you got them in the right order. And if you get them in the right order, supposedly over time, and this is the case with this doesn't have to be, is that if you get enough of these right, the returns will work out correctly if you have the yes or no. But it doesn't have to. So that makes more sense. Yeah. Anyway, there's a data set. You build basically an algorithm that relates the inputs to the output. And if you do that really well and you're near the top of the ranking or whatever, you get compensated.

Starting point is 00:40:30 Now, the funny thing here is... So this is the part that I didn't understand. Like, there's some kind of cryptocurrency involved? Yeah, so now we get into a whole second world. I don't want to get into Bitcoin or too much. But the basic idea is that there's all of these new cryptocurrencies being created and then used as compensation for whatever. and the value of that cryptocurrency is related.

Starting point is 00:40:53 Oh, I get it. So you get paid in what's called numerare. Oh, Jesus. Do I have to go buy this too now in my Coinbase account? I don't think they supported that Coinbase. But anyway, it's an interesting idea where you're not even being compensated in cash. You're being compensated in this new cryptocurrency, where again, there's a network effect where if more and more people do this, the cryptocurrency becomes more valuable and so on and so on.

Starting point is 00:41:18 So I actually find this really interesting, and it's awesome that I learned this today. The idea behind, not to get too deep into it, obviously, but Bitcoin is like there's an amount of work that gets done to verify the ledger. So I guess this is the work. So the work is, did you produce an algorithm that is correlated to outcomes, good outcomes, right? That's really interesting. Yeah. So it's fascinating. And you can see how this is evolving, right?

Starting point is 00:41:43 That there are new firms coming out that are taking advantage of this crowdsourcing idea. there's this hunt for new data sets. I'll finally get back to that question in a minute. But you mentioned Ballyasini and I came across a great quote. I think it's public. It's from his letter, but it came out in Bloomberg. So this is from Dmitri Ballyasne. A long short manager 10 years ago might have been okay with a couple of good analysts and a chief operating officer. Competing today requires a significant investment in technology, infrastructure, data, recruiting, corporate access, portfolio finance, compliance, investment relations, trading, and more. So this is a guy who's obviously been super, super,

Starting point is 00:42:16 successful. I think Ballyazini manages $13 billion or something like that, maybe more than that. And I think that's spot on, that it has become so incredibly competitive, which finally brings me back to my original question, which is, what's an example of a data set that you've come across recently, whether you use it or not, that's really interesting? And where did you find it? Because it seems like the search for data sets versus individual pieces of information is a new source of alpha. So maybe tell a story or two about. about an interesting data set. Yeah, and Dimitri's right on.

Starting point is 00:42:50 It's actually hilarious. So, Balliizing, I can mention this publicly because they allow us to, their clients of ours. And I was floored when Dimitri, in a meeting, said, make sure you directly tell me about anything because he realizes, like, he totally gets all of this. And normally, like, the CIO of a massive, you know, whatever billion dollar fund would not be like,

Starting point is 00:43:12 you data company guy, like, make sure you email me directly. instead of going through my CTO Sankad or somebody else. So he's like right on the ball. Data sets that are really interesting today. So, you know, some data sets are directly derived and some of them are derivative of other stuff. I really think that the satellite data is interesting from a lot of different levels,

Starting point is 00:43:35 mostly because it's so hard to use. But if you can use it well, it's so valuable. And many people won't be able to use it well because it's really hard to normalize it. and there's this company called Orbital Insight that just raised an enormous amount of money mostly because they're selling a lot of stuff to the government but they're also selling stuff to hedge funds

Starting point is 00:43:52 and look if you can figure out like where all the where the levels of inventory for every oil storage tank in the world is you should be able to do better so that one's pretty cool but it's going to be difficult to parse it and I've talked to a bunch of these quant funds that are trying and they're moving slowly

Starting point is 00:44:11 and the universe of things you can use it on is not incredibly large. The location data is just... That's fascinating. Just really fast. Because it's everything. The credit card data is being used just across the board now. That seems table stakes at this point.

Starting point is 00:44:26 And I actually have... I question how much value the discretionary guys are actually getting out of it, though. Because I feel like... It's funny. Like, you talk to one firm and they're making a different inference on the stock based on the credit card data than another firm is making, which means none of them are actually doing science. They're all just kind of saying like,

Starting point is 00:44:43 well the credit card data for Chipotle is going in this direction so I'm going to trade in that direction and the other firm's doing the exact opposite thing so yeah is there any efficacy to it I don't know so the credit card is like table stakes at this point the location data is really cool I really want to see somebody put together like a better front end for the location data because it's all just raw stuff right now and most people don't even know it exists yeah so that would that would be pretty cool and then there's I think even price and volume now are getting a new with non-linear methods where there is a company, and I'm forgetting the name of it right now, that has run some really interesting non-linear models through just basically price and volume,

Starting point is 00:45:24 and they're coming out with some really interesting factor models. And so even stuff that we think has been arbed or classic stat-arb stuff has too crowded, I think these non-linear methods are going to end up working pretty well. In the piece, you talked quite a bit about the spectrum that we referenced, earlier, which I guess on one end would be pure, like shoot from the hip, discretionary stock picker, which is a breed that's dying. On the other end is pure quant, meaning you run models only. There's no PM override, meaning if something comes through the model, you're buying it, no matter what the PM might think about it. And in many cases, those, the people running those

Starting point is 00:46:03 kinds of strategies don't have, we'll call it domain expertise, whether it be a sector or an individual set of names or an industry or an asset class. They are, data people, not consumer discretionary stock people or not material stock people. You talked about the potential for a hybrid, which can outperform or has better return prospects than either of the extreme camps. Could you talk about kind of what your thoughts are there and why, what might make that true? Why would there be a case when some discretionary input improves on a just purely quantitative process. Yeah, and we're actually at Estimized up against this attempting to make this shift ourselves actually from being purely quantitative to that mix. And it takes more experience,

Starting point is 00:46:53 it takes more industry expertise. So the basic premise of this actually comes out of, you know, a really good example that happened the other day where the AlphaGo team from Google that produced this incredible machine learning, artificial intelligence, very non-lawful. linear algorithm to play the game Go has been crushing every human for like a year now. And nobody can touch the damn thing. But they recently gave AlphaGo to like a mediocre Go player and put it up against AlphaGo. And the mediocre go player crushed the AlphaGo algorithm in and of itself. I think this is an interesting and useful, you know, anecdote because it shows that a human with a machine can beat another machine. And the reason is because the basis of a linear algorithm is to figure out how to

Starting point is 00:47:42 fit something to an entire universe of stocks. And we know that not every stock and every market cap and every sector performs the same way relative to a given data set. And so I'll take something that we do specifically. So the post-earnings drift model, which we have a factor model for, is basically if a company beats their estimated consensus number by a significant amount, which is relatively normalized to the average variance in beat size. We find that over the following three days after the earnings report, it will drift in the direction of that beat or miss. But the thing is that not all sectors perform the same way. So like industrials don't care about the earnings report really because they operate off of peak earnings. They don't operate off of like what's the next quarter or even

Starting point is 00:48:26 the next year. And so we know that in the industrial sector, the model doesn't work as well. But our factor model doesn't do that. It's like it puts it on a Z score across everything. And so what we could do is say, well, de-weight the algorithm in the industrial sector or the utility sector, which also doesn't really matter, so that you're not using that signal as much in your Z score. So you're not going to put those into your portfolio on either direction. They may just fall out, like, right in the middle and you'll never actually use those scores. This is a really good example of how, as a pure quant, it's very hard to, like, normalize for all those given little, like, characteristics that you need the industry expertise

Starting point is 00:49:06 to understand what are the variables impacting whether a model will work or not, because most quants need to trade across S&P 500, Russell 1000, Russell 3,000, like, whatever it is, and they don't literally have the time to go in there and, like, cherry-pick the variables that you deweight or overweight for certain sectors, industries, or other pharma French factors. So I think the discretionary guys do have the time because they're mostly operating in one specific sector normally, or maybe a couple sectors, tech and consumer or stuff like that, energy materials, utilities. And so they can use the expertise of the analyst and the PM to understand, well, I know that industrials operate off of peak earnings, so I'm not going to run a post-earnings

Starting point is 00:49:49 drift model on this thing, even if I have a factor that says that. Or I'm going to go create factors that are, I'm going to go do the research specifically on variables that I have an ex-ante hypothesis for that are strong. And I think they can, in their niches, over time, they can beat the systematic quants, because the systematic quants are hitting for singles, and they can hit for doubles and triples, you know, on a regular basis instead. So the way that that manifests the singles versus doubles and triples, I think, is portfolio construction. So if a quant, you know, has a whatever, hundreds of positions or wants to keep the max position below a fairly low threshold because they're not trying to place bets on just one name,

Starting point is 00:50:31 a discretionary manager might have a 10% position in a single stock. If all the lights line up green. Right. Because of that sort of more context oriented experience that they know, you know, if there's an industrial analyst, they have so, they've got some sort of. set of understanding of how these names trade to know which data points are relevant and irrelevant so they can make bigger bets is that the kind of the basic idea is exactly yeah yeah for sure and then they can manage risk around those positions at different times during the quarter

Starting point is 00:50:59 so the systematic algos will do this so some of our models will only be used in the couple weeks before and week after earnings but the majority of these factors that the systematic wants use they need to work on a regular basis but man i mean if a discretionary guy has a factor that only works in the post earnings drift, right? And he sees that that thing is like bright green. Well, I don't know, go overweight your position by a lot, right? Like for that one stock. And you can do that because you don't have 500 positions on.

Starting point is 00:51:29 You might have 40 and you can pay attention to each one and all the signals coming out of it. Back to when something stops working. So I'm going to use a big broad generic example, which is value investing. So one of the original factors, maybe the original factor probably, something that over decades and decades and every geography. By the way, value investor is the original quant. Right. For sure.

Starting point is 00:51:52 Ben Graham was the original. The original factor. So here is something where, and I like to think in terms of half-lifes of a signal. So how long can you expect something to have alpha? You mentioned ibis is being arbed, but it's taken maybe longer than you might have expected. So value is one that's fascinating, right? Because it's not an informational edge. There can be an informational component to it.

Starting point is 00:52:15 you can measure value better. You can have an analytical edge there too. But the basic idea is that at extremes, markets overreact. So they get too excited about glamor stocks. They get overly despondent about cheap stocks. And this seems to be much more human nature than anything else. So arguably something that has an indefinite half-life that's just going to work forever. But you go through periods where anyone right now where value investing stinks for a long period.

Starting point is 00:52:43 It could be, I think, in this case, seven, eight years. of underperformance that that wears even the most disciplined people out. And so the question is, and we've started to get it, we're value investors at heart, is how do you know if something is irrevocably busted and broken? And my answer to that question for value is the reason why, is the ex ante reason why it works, which is, well, this is a behavioral phenomenon. And as long as people are people and are to some degree pricing securities, it will exist. It may shrink.

Starting point is 00:53:13 It may get lumpier. but if you have like a real long-term horizon and you buy a basket of very cheap stocks, I'm very confident that you will outperform over some decently long horizon. So how do you think about that question of when something is broken? What are the markers? Because I've got the ex ante idea behind value, but it hasn't worked in seven years. That's seven years a long time. So how would you think about something super simple like value?

Starting point is 00:53:39 I think you have to, we look at it as there are intrinsic properties and there are are properties that come about because of an informational edge. Value, momentum, growth, these are intrinsic properties of the market, and they will never go away. They will come in and out of fashion. And I think the difference is one is behavioral and the other is informational. And they can change. They can move back and forth sometimes, but some of these have been around forever. In fact, at this point, we just call, I mean, Fauma friends, these are betas.

Starting point is 00:54:15 And so stuff moves from Alpha to Beta. And I guess one of the questions that's going on in the industry right now is, is Beta levering Alpha if you get it right? And I think there's a big debate going on right now because 90-something percent of people have just been leveraging beta. And if you can do that at the right time, great. You have a great outcome. But more often than not, people end up blowing up eventually. doing that. I think factor timing in general is, or you call it beta time. If you think about these different things as betas, which I don't know if I agree with, but I actually, I think factors are the

Starting point is 00:54:50 better words. So people call it like smart beta. Nothing is smart beta, right? This is all just factor investing. So the timing of those things is it's a great interesting thought exercise, right? Because it's this kind of game theory, competitive edge, contrarian. There's all these interesting components into how might you time factors. Whenever we've looked at it, the answer is you can't. You eventually blow up and get caught on the wrong side of it. It caught on the wrong side. You end up overtrading versus like a stupid naive. It's amazing how often we find if you have things that work, if you just equal weight those things, you end up doing and just leave it alone, you end up doing better than the fanciest, most overfitting tactics. One of the questions I'd

Starting point is 00:55:32 love to ask everyone in kind of different contexts is for the most memorable individual day of your career in finance, investing and markets, a day that stands out? Flash crash. Can you talk about why? So I was actually working at StockTwits at the time. So my career went from working as a PM to running my own small fund at the same time that StockTwist was getting built. I was running product and then Bizdev over there.

Starting point is 00:56:00 And then I eventually put the fund down and started Estimized. I remember sitting at my desk. And I just remember the couple of days. days before, you see the market leaking, leaking, leaking, leaking, all the distributive signs were there, like the market was in distribution, thing was going back and forth for a while, the volume on the downside was much higher

Starting point is 00:56:21 than the volume on the upside. It didn't look good, and you could tell, like, something was going on, and then you come in that morning and just like, there's no bids. And it was incredible just watching this thing drop and drive, in the whole office, like we got no work done that day, and we have no, nothing to do with the market. We weren't allowed to own any stocks at the time, like in our

Starting point is 00:56:42 personal accounts. And the thing just keeps leaking and all of a sudden just gives way. And the thing that I remember is one of the lessons that my mentor, the first PM that I worked for taught me is when something crashes, it always retests. Always retests the bottom. And so the thing ends up bouncing, you know, over the next couple of days. Everybody's like, oh, okay, it was just like a one-time thing. I'm like, no, no, no, no, no, this thing's coming all the way back down, and it eventually does come all the way back down. And, like, roughly kind of bottoms out, like, right around where the low point that day. Many of the other stocks, which had just, like, massive liquidity issues that day didn't eventually come all the way down, but the index did. And it was just really interesting because I remember on that day, I'm, like, I'm going to get to confirm a thesis that I was taught, like, years ago over the next couple of months.

Starting point is 00:57:33 And I'm pretty sure that he's going to be right. And lo and behold, like, absolutely correct. These things always retest. And it's because, like, that's where support was. Fundamentally, like, that's where Fidelity, the CIO ran to the PMs that day. And he's just like, at that level, you have to buy everything. No matter what it is, just buy it. And it's not like the random person, like the random retail trader out there.

Starting point is 00:57:56 It's like fidelity and Wellington and those massive long only funds were just like, you have, I don't care what's going on. Buy it. And that's where they're, that's where they're, what, limited. orders were or buy stop orders. And you know that when it comes back down there, they're going to buy it again because that's where they feel the intrinsic value of the companies are. So yeah, that was interesting. Maybe we've kind of already answered this because of our discussion on kind of man plus machine versus machine only. But looking forward, given how much more of market

Starting point is 00:58:25 orders are based on algorithms, on quant traders, on systematic processes, do you think that we'll see a lot more unforeseeable dislocations like that? And I guess the ultimate stabilizing measures, there will always be someone that says, this is insane. Go buy everything. I mean, so this is back to the like intrinsic versus, you know, informational edge. Liquidity will always be, you know, an intrinsic property of the market. And that will change over time, like what is adding or removing liquidity. I actually think the stat arb guys in many ways add liquidity.

Starting point is 00:59:00 If you look at many of the models that they're running, they're liquidity takers. and that actually negatively affects their models, but they are. The market makers will shut their things off whenever they can. So the more systematic guys we can put in that are not simply making markets and trying to jump in front of each other, I think the better it will be for the market and the less of these things you'll get, because they tend to don't, they tend not to shut their things off, you know,

Starting point is 00:59:25 when things get volatile, because they actually make a lot of money when things get volatile. That's what they like, whereas the market makers just don't want to be involved at all. So I think there's a shift going on right now where the high frequency guys are going out. In fact, many of these old high frequency firms like Tradeworks are actually, or in Jump, not Jump still makes a ton of money in high frequency, but they're turning into Stadar desks because the guys there know how to do. It's all the same work, basically, just different time frames. And so I think the market is actually going to have more liquidity.

Starting point is 00:59:54 You're going to see less of these dislocations going forward. The better question or the thing I'm thinking about is, are there going to be any stock pickers left at fidelity to say, no mass at some point if it does happen relative to like just all the passive strategies and a lot of people talk about well will passive strategies cause more volatility because nobody's actually there to like bring stuff back into line on a fundamental basis I don't think that's an issue honestly I think the issue is really in a dislocation is there a human there to actually just be like I want to buy a million shares of this stock because there's no way like this is overvalued

Starting point is 01:00:33 It's an interesting question where the, as we approach whatever, maybe there isn't an actual equilibrium point, but there's some active passive balance that will settle out in some range, right, where you need a certain amount of active. You need price discovery. You need liquidity. So the interesting question is how will, let's say we're at 40% passive today, something like that. 33, 35, 40, whatever it is. There was a Bloomberg article the other day or yesterday that was just like, and they literally timed it out on the current. growth rate, when will it be the whole thing? Right, which it can't be.

Starting point is 01:01:08 But let's say it gets to 75% passive or something like that. And there's some massive disruption, some breakdown, there's a total lack of bids. And everyone isn't, you know, selling their GM or selling their IBM. They're just hitting sell SPY, sell vanguard, sell, whatever. It's really interesting to think about what those little short-term periods of time and who will be the stabilizers, right? And maybe that could maybe something like that happens and that results in a resurgence of more active money and maybe it's cheaper or whatever. But, but it's an interesting, interesting problem to think about it. I'm really interested to see, and all of my

Starting point is 01:01:46 money personally is embedermint right now because one, I'm not allowed to own individual names, given what I can see on the back end of our platforms. And two, I literally just, I don't think anybody should trade individual names if they can't put their whole like emotional effort towards it, because it's such an emotional process, unless you're running a systematic strategy and I just don't have the setup for that. I'm really interested to see what happens to betterment and the behavioral things that they've built into that product

Starting point is 01:02:13 during the next crash or whatever. Go and drag that little wrist heater down. Yeah, I'm dragging the thing to zero, right? Am I going to go in there and drag the thing to zero? Now, personally, like my whole philosophy on passive investing is that it shouldn't be completely, passive. We know that if you use some very easy like trend following algorithms with a long only strategy, you can get yourself out of massive drawdowns when basically the index drops

Starting point is 01:02:42 would below a falling 200 day moving average. Just get out. Super simple. Just super simple, right? Like, I would love it of betterment put in just some super simple trend following algos or or allowed me to allocate towards a smart beta, whatever, you know, quotations. They use value to us now. They do use, yeah, they do use value. It's funny, I've discussed this. It's not your choice, though. I've discussed this with John, the CEO over there,

Starting point is 01:03:10 and our company also uses them for our 401K, which is an incredible product, by the way. I've discussed this, and he doesn't want to break the glass box. The glass box being they don't want to give all these options because then they don't want to be responsible for your returns because they feel like their responsibility is your behavior, and this is what I'm actually really interested in. in the next crash, do people go in there and, like, against their own good judgment,

Starting point is 01:03:34 against their own best interests, do they, like, move that slider to zero? Or can they prevent these people from doing that? And I actually think the dislocation in the market, and obviously, like, these platforms are a tiny percentage of assets. But the behavior that takes place on these platforms will roughly represent the behavior that takes place in all the passive money. And so the next one will be the canary in the coal mine. And I hope they release some kind of.

Starting point is 01:03:59 report on like the market dropped 20% over three months and we only saw an outflow of like 1% of our assets or something. And if that happens, we're safe. If everybody takes their money out, we're all completely screwed. My opinion on this, because everyone asks these questions of the various automated advisor platforms, you know, will, how will they manage behavior like the best financial advisor in the world might manage their client's behavior in the next downturn? My answer is it's impossible to it because if there is a real downturn, meaning like a, a, a, a, a, a crash, a big market drawdown. 20, 30, 40%, something really catastrophic.

Starting point is 01:04:36 People always think, well, everyone sells at a bottom. I kind of think of the other way around. Bottoms happen because everyone's selling. So I think by definition, if we're in a really bad situation, people are going in and dragging the wrist meter down to zero. Well, or it was a flash crash kind of thing that had to do with liquidity, but that should be relatively quick. That's right.

Starting point is 01:04:56 Yeah, that'll be, that's actually a really interesting way to look at it. bottoms happen because people sell. Yeah. Yeah. So that's going to be an interesting thing to watch is how behavior is managed. And then you could, if Betterment does better than wealth front, then all of a sudden they're loaded with material to win customers for the next, you know, several years. It's interesting that you bring that up because one of the things I talk about in that piece

Starting point is 01:05:21 that I wrote, which is on LinkedIn, is I think one of the reasons that there's just generally less alpha in the market today is because you have less of these just retail investors and you have less. Muppets. Yeah, I mean, that's the golden term for it. And I'm parroting them. You have less Muppets in the market and you have less by side Muppets too, right? Like, and so if you have less behavioral, less bad behavior in the market, will you ever actually get that selling, right? But if everything's passive and nobody panics, will you ever actually get another crash? Or, will the market become much less volatile?

Starting point is 01:05:59 And I actually, this is a hypothesis that I have that is not in any way backed up by any long data set at all. But it's starting to seem like one of the reasons why you're seeing lower volatility in the market against everybody's better judgment and intuition is because you're seeing more data sets that allow people to manage risk that allow you to predict what's going on

Starting point is 01:06:20 in the individual companies on a smaller interval, right, instead of waiting for the earnings report, being surprised and then having to rejigger expectations, you have less active money in the market, more passive money, so you have less of these behavioral dislocations. All of these things should add up to higher equity valuations because of lower volatility, and you should get a lower vol market in general. There will also be less alpha, because that comes from the vol and the dislocations. That's what it seems like right now. I could be completely wrong here, and in like three years, We could just see just massive whipsaw all over the place because like some other variable is we're not accounting for here.

Starting point is 01:07:01 But it's like that's what it feels like right now. I think that the other variable here is lumpiness. So if there is on a 20 year period, you know, average amount of all, how that's distributed within that period is interesting. Same thing with factor returns. So if value investing is going to work in the future, maybe in the past the, the, you know, three year base rate was 80%. meaning value outperformed to 80% of three-year periods. Maybe the magnitude of excess return is similar, but the base rate falls to 60% or something like that.

Starting point is 01:07:34 And maybe the same is true of VAL. Obviously, I have no idea what's going to happen. We're just going to have these very discreet, like, one-week periods where VAL goes nuts and then outperforms by 20%. And then everything just goes back to normal. Everything just goes back to baseline. Yeah, that would be crazy. So what has you most excited about the future?

Starting point is 01:07:53 Obviously, you're a market junkie like I am and just kind of love this whole game. What parts of the market? What has you most excited? Look, our whole space is interesting because I think we're just at the beginning of it, honestly, with the number of data sets that we can crowdsource, which is, you know, that's just simply fun for me. Whether it's at Estimized going forward the next 10 years or whether it's in another vehicle or whatever it is, I feel like my career will probably be there.

Starting point is 01:08:20 I want to get back at some point to running money because I love. love it. I just like this discrete period of time and history with stuff changing and the opportunity in my specific industry put me, you know, in a position to do these kind of things. But I miss running money. It's just the exercise of it to me is fun and intellectually interesting and emotionally interesting because you have to keep your emotions in check all the time. And the emotions associated with running a, you know, a startup technology company are just completely different across the board. So that's cool. Josh Brown, one of the One of my friends, I think you know well, writes about this. And I've written about it for a while.

Starting point is 01:08:59 I think there's a mispricing and disruption risk across the board in everything in our economy right now. Retail companies is a good example. Well, right now, that's the, yeah, that is the example that's taking place. But, I mean, you're going to see this in industrials. You're seeing it kind of in energy right now what's going on with oil prices. Maybe, who knows what oil prices are correlated to. But I think there's massive disruption risk, and I think you're seeing the mispricing of that in the market, both on the upside and the downside. And everything is shifting towards technology. Every business is shifting towards technology, and that's going to cause these massive dislocations in industries. One of my favorite shorts right now is AutoZone.

Starting point is 01:09:42 And for a lot of different reasons, but like I would, I can't, I can't buy any individual names, but I would be buying leaps on this thing. because this thing's a zero. Like, who can understand how to fix a car these days? It's just a massive computer. Not only that, but in, like, you know, five or ten years, you're not going to be driving a car. You're going to be, you know, rolling around in one of these driverless things that Tesla licenses out to you, and they're going to be fixing it

Starting point is 01:10:05 because how do you fix a Tesla? Like, good luck figuring that out. So, like, AutoZone's dead eventually, and the question is, what are the inflection points for when to get, you know, short this thing? Is it now? Is it later? Is it whatever? And there's a lot of those kind of interesting anecdotes that I have a lot of fun trying to figure out. And then, you know, you place all the different like market timing aspects on top of that general thesis and then you got to go build a portfolio.

Starting point is 01:10:32 And these are the things that kind of, you know, these are the things I nerd out on. The AutoZone's a funny one because it's actually one of the classic examples of a name that screens incredibly well in a lot of systematic strategies, especially those that look at value, at capital allocation, sherry purchases, all these things. and it's done ridiculously well in a lot of the periods that these models have owned that stock. And it is the big question, right? Is how do you price disruption risk? And how does that affect value investing? Because the period over which we formed our belief that value investing has worked, not the ex ante, but the actual evidence is, let's call it 1926 to now.

Starting point is 01:11:13 That's really the, really it's 1963 to now, but we do have data on value going back to 1926. The question is, with an accelerating pace of change in technology, and the fact that we tend to extrapolate in the wrong, we're not good at exponential extrapolation as human beings. The question is, does that change the story for value investing? Because value, if you look at a, if you go look at a best decile portfolio of value stocks, you're going to see retail, you're going to see AutoZone, you're going to see some of these, a lot of these stocks and industries where your concern is most manifest. So it's a, I don't know the answer. Even look at the technology. Like I even think that there's there's mispricing and disruption risk of the disruptive technology companies. Look at Zinga.

Starting point is 01:11:59 So Zinga comes out. It's generating a ton of revenue growth. It's doing really well. And people don't understand that like there's another whole set of things behind Zinga that will disrupt Zinga because the innovation cycle in the valley and just in general has sped up so much that business models come in and out. a favor within a matter of five years. And so you can't even count on the high growth momentum names continuing to disrupt the old industry because there's one right behind it. And that may take a whole new view on what is value. Look at IBM. Like, good luck getting IBM to like innovate out of this cycle. I just don't see it happening. And that would be a classic value stock right now.

Starting point is 01:12:39 One of my favorite charts is that rolling lifespan of a company in the S&P 500 that shows this kind of Yeah. There's some cyclical moves, but the secular trend is very clear where it was, you know, 50, 60 years. Now it's 12 or 15 years. It's pretty amazing. And then you get, and then you get Apple, which everybody's been kind of on the multiple basis betting against forever because they're like, well, they're not going to sell another set of iPhones this season because eventually every, and lo and behold, they sell more iPhones like every quarter and it's incredible. My closing question for everyone is to ask what the kindest thing that anyone's ever done for you. Oh, man. Well, I mean, I'll, I'll, it's very simple. My co-founder, Brian Smith, I think it's the luckiest thing that ever happened to me was finding him. So a lot of people don't know this. Some people know it, but a lot of people don't know that the first year of this company was just terrible. Like, we did everything we could possibly do wrong. And the company almost died. In fact, there's a, there's an article at Business Insider from our launch day in January of 12. where I have shingles because I have not eaten or slept and I'm just like freaking out

Starting point is 01:13:50 because we have to get this thing out the door and I fired my original co-founder who was a very nice guy but we just did not work well together at all and we just barely got the platform out the door and I was just a wreck emotionally, mentally, physically and they had to airbrush my face because one side of my face is like all puffy

Starting point is 01:14:12 and everything so the, you know, The article is great. The image is ridiculous because one of my eyes is like almost closed. So we get the platform out the door and our intern at the time was an incredible engineer, held the thing together for three months. And I find this guy who lives in San Diego where I went to school who I met on Angelist, you know, of all places, which we've hired a ton of people from since and it's an incredible platform. I meet Brian and on a whim he flies out to New York for a week and stays with me in my apartment. And we go at this, you know, problem, okay, so we have this platform out the door, but it's like, is this going to work?

Starting point is 01:14:51 It's very much just still the experiment. And he spends a week and he goes home and he emails me the next day and he's just like, I'm in. And like we paid him basically nothing for the first like six months while we did this experiment. And this guy could have gone to work anywhere. He was former Google cryptography expert, incredible guy, just did this on a whim, like, like the idea, me, whatever, he bet on, he just bet on all of it. And without him, like, this thing never would have worked. And I wouldn't have gotten to do any of this. I would have gone back to running money or something. And I think it's the luckiest thing that happened in my life. And I wouldn't have taken that bet. Like, it was a disaster. Like, it was really, the operational

Starting point is 01:15:31 aspect of this thing was a disaster up until then. And he comes in and the thing goes, nothing, nothing goes perfectly well. But, like, you know, we've been on a very upward trajectory. since then with very minimal kind of hiccups. Awesome. Great place to end. Thank you for your time. This has been a blast and a sign to me that I need to do more episodes on Quant because this is really fun. Yeah.

Starting point is 01:15:54 Hey, everyone. Patrick here again. To find more episodes of Investors like the best, go to investorfieldguide.com forward slash podcast. If you're a book lover, you can also sign up for my book club at investorfieldguide.com forward slash book club. After you sign up, you'll receive a full investor curriculum right away and then three to four suggestions of new books every month. You can also follow me on Twitter at Patrick underscore

Starting point is 01:16:19 Oshag, OSHAG. If you enjoy the show, please leave a quick review for us on iTunes, which will help more people discover Invest Like the Best. Thanks so much for listening.

Invest Like the Best with Patrick O'Shaughnessy - Leigh Drogen - Quant vs Traditional Investors and How Alphas Become Betas - [Invest Like the Best, EP.41]

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.