Challenging the Foundation of Asset Pricing Theory with Andrew Chen and Alejandro Lopez-Lira

[00:00:00] Welcome to excess returns where we focus on what works over the long term in the markets.

[00:00:04] Join us as we talk about the strategies and tactics that can help you become a better

[00:00:07] long term investor.

[00:00:08] Justin Carbon, now in Jack Forehand, are principles at the Lydia Capital Management?

[00:00:11] The opinions expressed in this podcast do not necessarily reflect the opinions of the

[00:00:13] ability of capital. No information on this podcast should be construed as investment advice.

[00:00:16] Securities discussed in the podcast may be holdings of clients at the Lydia Capital.

[00:00:19] Hey guys, this is Justin. In this episode of excess returns, Jack and I sit down without a

[00:00:23] hundred Lopez Lira of the University of Florida podcasts, but Jack and I are the furthest thing from people with PhDs in empirical finance. But we are involved with the fact of investing. And so, you know, even when we initially saw this paper, we found it very interesting.

[00:01:40] The title of the paper is peer review theory does not help predict the cross section of

[00:01:43] stock returns. And we wanted to have you both. So we're going to try to keep it at the level of not PhD level, but we'll get into some of this. And I think this will be a really great learning experience for everyone listening. To start, let me ask you, and if one of you or Andrew, if you want to sort of give your little disclaimer here

[00:03:01] about this ad, that's fine.

[00:03:01] You can just drop that in.

[00:03:03] But maybe just start what gave you guys the idea

[00:03:06] to tackle this subject.

[00:03:07] Justin, thanks for prompting me. And we normally think there's something special about academic strategies. So we were kind of surprised and just tried to make that the focus of the paper. Yeah, there was another project that kind of got merged together that we're interested in how to do data mining correctly. So we just started mining data. And yeah, then Alejandro just described the result.

[00:04:22] We just they look really similar data mining and the peer reviewed stuff. and explain some specific portfolio with those returns, then that became an anomaly. You know, the general thing that I would say is like, it's whatever we don't know how to explain. So Alejandro, you just sort of use the word factor, which is one that our audience will be pretty familiar with because you see it all over investing these days. Like what is the difference between an anomaly and a factor?

[00:05:41] Or is there one?

[00:05:44] Andrew, do you want to start? Oh no, you should go ahead with this.

[00:05:46] You know, you're thinking about this all,

[00:05:47] you have papers on this. that has like some expector return and doesn't have like zero variance, that can be like proxying for a factor. So it's a little bit complicated, but basically factor is what we treat as explanatory variables. It's almost a matter of convention. And this just leads to a lot of confusion. Yeah. And I can pretty much guarantee you I'm going to interchangeably use those words and screw

[00:07:00] it up throughout the rest of this podcast. So this area, leading up to where we are today? I mean, the body of work is huge. Let's see. I suppose it's really started with

[00:09:22] But sometimes the world is more complicated than that. And I think that's where we are now.

[00:09:25] Yeah, no, I think that was a very good summary.

[00:09:28] So I would just said that, basically, an anomaly

[00:09:32] depends on what kind of model you're using to price.

[00:09:35] For the history of anomaly, it's just related

[00:09:36] to the history of factor models.

[00:09:38] So before, again, anything that was beating the market

[00:09:41] was basically an anomaly.

[00:09:43] And after a final French, well, at least you

[00:09:45] had to show that your strategy was a little bit different

[00:09:47] from what those in the JFE, they provide a rational perspective. My advisor also has a rational paper, a momentum. But I think you're right, in general, people think of momentum as a behavioral issue. Like people pile on to returns, they extrapolate from past returns to future, and that causes the return to the brain.

[00:12:22] So I'll take a tackle of the risk because it was my class of days,

[00:12:24] so I'm a little bit of compensation for that, although historical industries do not get that much of a of our return. It is one of those considered stronger from a persistent standpoint. So if I expect something

[00:13:43] to persist in the future, like, do I do I can come up with a rational model for it Like I'm sure like I can just spend like a day or two just working at the mouth and giving you the math and giving you like a Very nice rational story, right? But the point and I guess later we'll talk about it is like do these two stories matter at all, right?

[00:15:01] Like is this like you know, it's me like writing an explanation when I have like any impact on that

[00:16:02] define what data mining is. Data mining is just looking through data

[00:16:04] for interesting patterns.

[00:16:06] That's all it is.

[00:16:08] I guess maybe the key aspects to it

[00:16:09] is that it's usually unguided by any theory.

[00:16:15] I would say the same thing.

[00:16:16] Like, you know, the defining explanation

[00:16:18] is that you do not have any structure on the data.

[00:16:21] You're just looking for patterns.

[00:16:22] And I think historically, it was seeing with bad eyes

[00:16:26] because if you just focus on your current sample, Did we figure out they worked and then try to come up with a reason? Or I'm not sure, like the chicken and egg type thing. How did that work? I think it's a chicken and an egg issue. It's unclear. You know, people write down some things in the paper, but they don't write everything. I'm thinking about Satman's paper. And I think he motivates it by pointing to kind of industry

[00:17:44] talk that this is an idea. So you could think of there's four, at least four ways to come up with an anomaly or trading strategy with unusual returns. One is to think about risk and try to think about the sources of risk and then to identify exposure to that risk. Another one is to think about mispricing

[00:19:00] and behavioral biases and think about which stocks

[00:19:02] will be underpriced and overpriced.

[00:19:04] Another one is to kind of do a combination of both.

[00:19:08] You're not really sure which it is, family of French's famous papers in 1992 will compare the data from 1963 to 1990 to the anomaly returns after 1990 until today. And you would think that if you do all that work, you would get better out of sample performance. But it turns out that the answer is no. It's actually an empirical question whether these two studies would perform different. We would really expect that the theory one should do better, but it's something that we're trying to test in the paper. It's interesting in our world too, if you're going to a client

[00:21:40] with an investment strategy and you say, we're buying cheap stocks with a PE ratio, that's very

[00:21:44] different than saying, we divide an inventory by the average return as a measure of that. And you know, the first thing to do is like note that that average return is going to have like some variation, some uncertainty surrounding it, right? You don't really know if maybe you just got lucky this period and you would just expect

[00:23:02] by chance that these kind of returns like sometimes happen.

[00:23:05] Like you kind of want to know if that's is above 2. So you're kind of like, you're kind of like twice, you know, it takes two levels of unlucky draws in order for this to be zero, something like that. And then I think in the literature,

[00:24:21] you usually find T stats of about 3.5 is typically quite common. combination and that yeah, just doing this will lead to like 29,000 different trading signals. So were there any like restrictions in terms of like what could be in the numerator, the number, or you basically just tested everything? We basically tested everything. We don't want to divide by zero because that's very confusing, not really sure what to do with that.

[00:25:40] So aside from kind of avoiding by zeroy, make a table that compares strategies that we found to famous strategies. It was originally just be snarky, but then when we started presenting it, going to be some groups that are going to look like value, some groups that are going to be like leverage, some groups that are going to look like investment. But in general, the interesting thing is that all of these have the same average returns in some cases. So right now I was reading from the studies

[00:28:22] that look like size, for example.

[00:28:25] So let's actually, you reference these charts in the paper.

[00:28:27] And they're really cool.

[00:28:27] We're going to put them in the video. performance during that sample period, like between 67 basis points per month to 115 basis points. It's like roughly comparable. And then we just list like 15 or so of these variables that have similar returns. And as we were just describing, a lot of these pickup

[00:29:40] themes that people have talked about in a literature since family French 1992, like a like, different synesthetics divided by a lot current assets. That's going to be something that has similar performance to momentum in sample, or property planning equipment and machinery divided by current liabilities. So there's going to be a bunch of things that look with very, very high average returns, but have a completely different explanation,

[00:31:01] I would say, than this momentum strategy.

[00:31:04] I think the kicker to these tables

[00:31:06] is if you look at the rightmost column that that has good returns in sample and you would have outperformed all that human capital. It's so surprising because we allocate so much time in academia to discuss some very specific variables, like Book to Market or Momentum, but there seems to be... Well, it's not infinite, but it's a very, very large amount of things that we could have started

[00:32:22] instead if some researchers decided to look at a couple of different things instead.

[00:33:24] a testing standpoint, can you like randomize?

[00:33:29] It would never make sense to like randomize the months so that you're not

[00:33:37] going consecutive years, but you're more randomly looking at these anomalies, you know, across just a randomized set of months.

[00:33:40] And I wonder if that's something that even would make sense or a lot.

[00:34:44] benchmark is to say like they knew everything that happened in the world up until 19, oh, not everything, but they knew what happened in the world up until 1990. They don't know what

[00:34:47] happens afterwards. So we need to see what actually happened. And this is, you know, it's not just

[00:34:52] something that we're thinking about. This is in the key PhD textbook. It's a wonderful textbook,

[00:34:58] but a key message of the asset pricing book by John capital expenditure ETF, we're probably not going to get too many couldn't probably not get too many potential investors are going to look at that thing. But realistically, it might do just as well as if we went with a momentum ETF. Well, I could I could come up with a strategy based off this. That's what the table says, right?

[00:36:22] Maybe we could get a bunch of academics to invest in the PPE and machinery.

[00:36:25] They would understand behind the scenes that there's really no reason to get the less crowded You did also, in addition to the accounting variables, you also did do a look at some ticker-based data mining strategies in the paper. What did you find with those? So we found that if you mind tickers filing the same kind of idea that you're looking to pass, look for ticker-based strategies that perform well, on average, they do nothing in the future.

[00:37:41] So that means there's something really special

[00:37:44] about mining accounting variables,

[00:37:46] or perhaps something very unspecial about tickers. that those researchers had in 1992 or 1991, and we looked through the past in mine accounting variables, and we come up with that red line, stuff that performs similarly in sample. So in sample, they're similar by construction. And then the big question, you know, the question of our paper is what happens out of sample, what happens in the future when you're actually trying to predict instead of like predict with quotes around it.

[00:39:01] And what you get is the same performance, roughly speaking,

[00:39:05] that the blue line declines I'll just start with the simplest one, the simplest one is that the anomaly's literature really peaked around the mid-2000s. And it's, you know, after that time, that information technology really blew up.

[00:40:24] I asked this pretty senior professor, like, how did people trade book the market in the of these variables are, they're just kind of proxying for more micro stuff that's, that I think, for lack of a better word, real life investors are working on. So to, even though someone who is doing, doing equity is research and finding some particular

[00:41:42] story about why this, this one firm is going to exploit it. So I'm pretty sure there's a lot of cash flows just constantly looking up for new and interesting signals to exploit. You know, once if these signals are public enough that academics can write about it, then academics are

[00:43:03] also constantly looking for things that have like high t-stas and average returns and are to publish stuff in the data mine stuff. So it looks as if evidence starts building and then people trade on it and it goes away. I just wanted to read before I handed back to Justin just to get into conclusions of this. I wanted to read one quote from the paper and get you guys to comment on it because it was kind of eye opening for me. The quote is economic theory helps predict returns if it provides information about expected returns.

[00:44:21] Our empirical results imply unfortunately

[00:44:22] that theory does not provide such information.

[00:44:25] Though these findings are negative for economic theory

[00:44:27] they are positive for data mining. So the bright side of it though is that I think we're seeing in a lot of other fields like in linguistics and in protein folding. So in linguistics, for example, you don't, Alejandro should probably talk about this one, but you don't feed known Chomsky's theories into the neural network, do you?

[00:45:40] No, no, no.

[00:45:42] All of this.

[00:45:43] I think what Andrew was saying, like models like Chaji Petit do not have any underlying frame or work without sample returns where you let the data speak by itself. You don't want to impose any kind of big strong assumptions that are done. Well, you know what? We're going to look at accounting data because accounting variables are informative. But after that, do you really think better if past returns divided by lag

[00:47:01] expenses is better than property plan equipment divided by a lag accruals? but the frictions are reduced, that the predictability will become weaker. These kind of ideas that are probably still extremely helpful, they're really not the core of the paper. The core of the paper is about the stuff that gets published, the stuff that will help you get tenure at a major university.

[00:48:21] Yeah, so we're focusing on a very specific subset of signals,

[00:48:24] and we're focusing on signals that you can construct easily make the absolute least sense. And I was thinking, like, is that going to start applying to our world now in the longer-term world? You know, like it is implied in their short-term world, you know, in the past. Yeah, I discussed it at the recent AFA meetings. I pointed out a quote from the man who solved the market where, you know, they basically not, it's not a hundred basis points per month. It's more like 80 and they'll actually give you that exact adjustment. And we have a paper, I have a paper with Chukuma Dim of George Washington showing you how to, how to apply those kinds of methods. Yeah. And I think looking forward, uh, we want to explore more about the kind of like

[00:51:01] the ideas that get generated in, in these journals.

[00:51:03] So we applied kind of like it are very, very basic, uh,

[00:52:02] Andrew, thank you guys very much for coming on. I think it was a great discussion.

[00:52:04] I think we kept it pretty high level.

[00:52:05] And I think our audience is going to learn from this.

[00:52:09] So really appreciate it.

[00:52:10] And the door is always open for you guys to come on

[00:52:12] and talk the research and the things that you're doing.

[00:52:16] So thank you.

[00:52:17] Yeah, thanks so much for having us on.

[00:52:19] That's fun.

[00:52:20] Thank you so much.

[00:52:21] I love this podcast.

[00:52:22] It's always nice to speak about asset pricing.

[00:52:26] I love this type of conversations.

[00:52:28] Thank you.

[00:52:28] This is Justin again.