What this covers
Cliff Asness is one of the most influential quantitative investors of the last 30 years — and one of the most candid.
In this conversation, Cliff joins Infinite Loops to talk about why losses hurt more than wins, how bubbles form, why modern investing increasingly resembles gambling, and what the dot-com era can teach us about today’s markets.
TIMESTAMPS: 0:00 Intro 1:10 Losses Hurt More than the Wins 6:20 Dot-Com Bubble vs. the 2020 Tech Bubble 13:30 Meme Stocks, Robinhood, and Gamified Investing 21:40 Why Asness Doesn’t Know Which Stocks He Owns 29:10 Quant Discipline, Models, and Looking Stupid 38:30 Market Crises: GFC, Asian Debt Crisis, and Volatility 52:10 Human Nature as the Last Investing Edge 1:04:40 Machine Learning, AI, and “Surrendering to the Machines” 1:23:50 Bubbles, Forecasting, and Betting Against the U.S. 1:37:40 Private Equity, Volatility Laundering, and Risk Illusions 1:55:30 What Asness Would Change If He Ran the World
Source description (no synthesized summary yet).
Asness and O'Shaughnessy argue that quantitative investing remains viable despite crowding, but requires discipline to avoid behavioral biases, acceptance of opacity in machine learning, and realistic expectations about expected returns in increasingly efficient markets.
- Disciplined quants outperform by resisting emotional overrides and maintaining systematic processes during market stress, as demonstrated in GFC and COVID periods
- Machine learning creates opacity that forces quants to accept statistical improvements they cannot fully explain, shifting from 50/50 intuition-evidence balance toward evidence-weighted decisions
- Expected returns are compressing across asset classes as strategies become crowded and fees erode illiquidity premiums; mean reversion trades offer low Sharpe ratios and private equity returns are normalizing
This asset isn't compiled yet
You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.
Over time, quant edges erode as more people adopt the same strategies and market efficiency improves; this means that to maintain the same edge, a quant firm must improve execution, reduce trading costs, and refine models more rapidly than competitors—it's not enough to stand still and expect the original edge to persist.
“Edges erode over time, and more people are out there. So it quite possibly I'd actually argue it was was actually true was not as important to be super optimal on the trading when you were the first person doing a certain That's right. strategy. That's right. When something is part of everyone's toolkit, and you got to measure it a little better than other people, and you got to trade more efficiently than other people, the One of the We could say the world's improving. It is, but it doesn't always make you better cuz it's improving for your opponent your opponents at the same time.”
A machine learning specialist who previously worked at a major tech firm told O'Shaughnessy that humans cannot examine millions of data points and find patterns, but ML can find patterns; however, ML can tell you 'what and when' but never 'why,' creating a fundamental interpretability limitation.
“He was their machine learning guy. And so I nailing with all these questions and and finally he says Jim, here's the thing that might be really hard for you to accept. And I went what's that? And he goes no human can look at arrays vector arrays of hundreds of millions of data points and find the five that go together, right?”
AQR and O'Shaughnessy both studied whether beta is actually good long-term and believe it is, contrary to some thinking that hedging is always optimal; the question is whether you should access beta through alternatives or just hold beta directly.
“I actually think I cost myself a bit on this cuz for so many years I talked about this that I almost forgot that beta is actually good long term... if expected return minus a little bit of diversification and compounding effects, we're going to ignore that... if the expected return on the fully hedged good hedge fund is less than the naked equity market, you have improved your Sharpe ratio, but you have lowered your average return.”
Natural language processing (NLP) of corporate earnings statements and news—as used in modern ML approaches—outperforms the old-school quant method of scoring good and bad word lists because NLP learns non-linear combinations of semantic vectors that capture meaning better than human-designed keywords; however, like all ML, it's a black box and must be validated against simpler measures to confirm the model isn't overfitting.
“natural language processing where you probably remember for a lot of years some quants used it some didn't but the quant way to process textual information was a table of good and bad words and phrases. And it was bespoke. You made it up as you as you as as you went and you know if the word increasing is there plus one. And we know the downside of that, right? If the actual sentence was massive embezzlement is increasing. You know, our bad on that one.”
Both Asness and O'Shaughnessy agree that leveraged basis trades (long off-the-run Treasuries, short on-the-run Treasuries) are the closest financial instruments get to true statistical arbitrage because the cash flows are identical and default risk is the same; yet these trades still blow up regularly at the worst times, usually when basis has widened because of stress, proving that even 'safe' leverage is deadly.
“we've never done the so-called basis trade. Right? Buy off-the-run Treasury, sell on-the-run Treasuries. It's It's the best example because you literally can cash flow match the thing. And it's the same government cash flows and as as as nervous you might be about the budget deficit, the notion that they will default on only selected bonds, it's a little weird. So it's it's about as close as you can get into a true arbitrage in that sense. And it still rises up to kill people all the time and exactly at the times you don't want to be killed. Right.”
A strategist at Prudential in 2000 stated that 'there are a set of stocks you need to own at any price,' representing the logical endpoint of bubble thinking where price becomes irrelevant to investment decisions.
“I also used to carry around in my wallet. I'm sad I lost this eventually. Probably disintegrated. A quote from 2000. It was from a strategist at I think Prudential saying there are a set of stocks you need to own at any price.”
Retail traders on social media predominantly share their winners and rarely discuss losses, creating a selection bias where the market narratives are dominated by outlier success stories rather than the average negative outcome.
“And it's it's such typical stuff where you'll only hear about people's winners... Anyone who's describing this is I bought MicroStrategy when it was here and I got out here. I'm like, I promise you that Citadel and Jane Street are taking your money.”
Prospect theory—the phenomenon that losses hurt more than equivalent gains feel good—is empirically real and emotionally powerful even for professional investors with discipline and long-term track records.
“People debate whether the prospect theory is real that losses hurt more than gains. It's very real to me.”
Private equity and hedge funds are inherently levered active equity portfolios that obscure their actual risk through accounting practices that show low volatility (because positions aren't marked-to-market daily); this 'volatility laundering' makes their betas appear lower than they actually are, typically 0.8-1.2 depending on the fund.
“I coined the term volatility laundering Yeah, I love that term by the way. of clients general media articles that that will do the equivalent...Here's an efficient frontier...Private equity, 6.5 percent. And I'm like, no, it's not.”
O'Shaughnessy founded his first company as a quantitative consultant building normal portfolios for pension plans in the late 1970s / early 1980s, and found extensive closet indexing (managers claiming to be active while mimicking the index).
“back in the late '70s, early '80s, I started O'Shaughnessy Capital, my first company, as a consultant. And it was a quantitative consultant... How many closet indexers I found back then? Like I won't name them... And that's only gone up over time.”
Asness does not believe small-cap stocks deserve a premium allocation in quant portfolios if that premium requires excessive leverage on illiquid names; microcap strategies can work if they carefully adjust for liquidity, but index-based microcap strategies that ignore the implementation side are fundamentally broken.
“We trade small. We like small. But ignoring the implementation side of small is an utter disaster. And the small firm effect, almost all of it came from the first decile. Which is just crazy.”
In mid-to-late 1990s academic research, many papers studying stock market anomalies used small-cap stocks, some of which had minimal to no trading volume; when Asness challenged academics on this, they responded that you could theoretically capture the premium if you had enough capital to move markets or infinite time, but these responses avoided addressing the real constraint: practical tradability.
“I would get into fights with academics who would say, 'Yeah, but look at the smallest stock decile.' And I'm like, do you do understand that there's no liquidity there and if you tried to put a bid in that it would skyrocket? And you know, it's like the old joke about economists on a desert island...And the economist says, 'Let's just assume we had a can opener.' And and so, that always bothered me.”
In August 2007, quantitative equity strategies experienced a 'harrowing week' with significant drawdowns, but this was a pre-crisis phenomenon distinct from the broader 2008 Global Financial Crisis; the event was painful and a standard-deviation shock but not fundamentally life-threatening to well-capitalized quant firms.
“There was no global financial crisis at that point. That was a little adumbration of what was going to what was going to come.”
Backtesting is essential to quantitative investing because it provides 45 years of data showing base rates, worst-case drawdowns, and probability distributions; while backtests can be overengineered and data-mined, comparing a backtest to making investment decisions based on gut feeling and anecdotes ('I talked to my buddy') is not a fair tradeoff.
“because that is that's one that gets me hot. It's like yes, of course you can overengineer a backtest and do data snooping, even honest-minded monkeying with the backtest. But the value of a backtest to me it's just like, what other It's like I used to get pissed off, like when the when the 'Oh, I do it what I feel like. You know, I talked to my buddy and he hooks me up with a great stock.' And I'm like, 'I'm sitting here with 45 years of data. Not only will I show you the 10 worst drawdowns, I will show you the base rates. I will show you the fact that this does not win in every 3, 5, 7, and 10-year period.'”
Leverage is a powerful tool for diversified portfolios but should never be used on concentrated bets or illiquid assets because the combination of concentrated risk and illiquidity creates uncontrollable downside when liquidity dries up, a lesson taught repeatedly by history (LTCM, private equity drawdowns, 2008).
“That is the death combination. It really is. And I was at a club in Greenwich and talking to somebody who had been at Long-Term during their heyday. So this is way back. And he got really mad at me when I pointed out that that amount of leverage it isn't if you're going to die, it's when you're going to die.”
The small-cap effect (that small-cap stocks outperform large-cap) documented in academic studies is largely driven by illiquidity and the smallest-decile stocks, which are highly illiquid and have high true betas when adjusted for infrequent trading; simple liquidity adjustments make most of the small-cap premium disappear.
“First, we don't believe there actually is a small firm effect, period. Um if you adjust for like 17 other factors, we wrote a paper on that. But if you just adjust for market beta, it doesn't outperform. The original small cap studies...our big thing is the betas were always woefully underestimated cuz they don't trade all the time. Right. Simple adjustments for that. Yeah. Just just make it go away.”
The Martingale strategy (doubling bets after losses) is being resold to retail traders as a novel discovery despite existing for 400 years, and advocates omit the requirement for infinite bankroll and infinite willingness, making it equivalent to 'pennies in front of a steamroller.'
“I was reading about some some some person who gives their financial advice out for for money and had a system that worked. And this thing is And they blew up. And they're saying they're selling it, right? They lost They This one actually around Christmas um um lost a a whole bunch of money. Um and they're like explaining, 'Well, it was a Martingale system.' And I'm like, I'm reading this and I'm having this internal dialogue. I mean, you mean the thing that's been around for like 400 years where you just keep doubling the bet?”
A person with a concentrated single-stock position (their child's college fund in one name) is extremely undiversified and taking concentrated risk, which is a bad investment idea regardless of the specific stock, and this is a statistical fact, not opinion.
“you know, if you put your kids whole education fund in any one stock, that doesn't strike me as, you know, That's a bad idea. Yeah. So but then they they scared me a bit.”
Retail investors consistently lose money long-term according to the published literature; high-frequency individual stock trading and options trading on apps like Robinhood transfer wealth from retail to professional traders and market makers, not because professionals are smarter but because retail has structural disadvantages (speed, information, transaction costs).
“retail loses long term. It's in the literature. I know. And that was like my big crusade when I was younger, right? Like I I did What Works on Wall Street and published it because I wanted people to have access.”
Sharpe's arithmetic—that the average active manager cannot beat the average because they are the average—is logically powerful and implies active management must overcome fees and transaction costs to create value for clients.
“Sharpe's arithmetic is diabolically powerful... Sharpe's arithmetic says the average can't beat the average. So the idea that active management on average loses Yeah. Or best breaks even.”
The distinction between gambling and investing is that gambling has negative expected value while investing has positive expected value; positive expected return is a necessary (though not sufficient) part of the definition of investing.
“I don't exactly know where a bet becomes an investment, but positive expected return Bingo. is is a small part of it.”
In April 1999, at the height of the dot-com bubble, Asness wrote a piece predicting that 95% of internet companies would fail or fall 90% from peak valuations, even winners like Amazon; the piece generated hostile emails from investors who saw his skepticism as ignorant old-fogyism despite his being only 39 years old.
“I wrote a piece called the Internet Contrarian in April of 1999, in which I said, 95% of these companies are either going to be carried out feet first or be 90% lower, even the winners.”
Bear markets are when stocks return to their rightful owners, and retail investors who bought in during meme stock frenzies were not rightful owners because they had no underlying thesis or diversification.
“Bear markets are when stocks return to their rightful owners. And those guys were not rightful owners.”
ESG investment constraints are costly (they reduce expected returns), and the only mechanism by which ESG can improve the world is by raising the cost of capital for companies engaging in bad practices; therefore, ESG investors should expect to make less money, not more, and claims that ESG delivers both social good and financial outperformance are double-dipping.
“constraints have a max are the best they can be is zero expected value, and that's unlikely. Very unlikely. And I give the example that two different clients give us the utterly same mandate, but one says there's a third of the stocks you're not allowed to own. How could I look at the one who lets me own those and tell them, 'Why you buying those?' Cuz cuz I I think it's going to make the portfolio better.”
Quantitative strategies historically required roughly a 50/50 balance between empirical backtest evidence and theoretical/intuitive reasoning about why a factor should work; machine learning shifts this balance because modern ML techniques can improve performance in ways that are theoretically opaque, forcing quants to accept approximately 2/3 evidence and 1/3 story rather than demanding equal weight.
“we always were very proud and we still are but little less so that we demand roughly and it's impossible to know you know this is just a way of explaining it. Roughly 50% each. You know, do we think this makes sense and how good of the historical results been?”
Black swans defined as unpredictable tail events cannot be predicted, but the question is whether a black swan can be *confirmed* or detected in real time; using oil's crash to negative prices as an example, could ML have detected the black swan approaching before it fully occurred.
“if we take the whole idea of the black swan can't be predicted, right? I ask the question, well okay. Let's assume that's true... But can a black swan be confirmed? Do you mean after the fact or Let's let's use oil when it went negative, right? So, black swan. Um but was there a point in that series of oil on its way to going negative where there could be some machine learning algorithm that could figure out oh, this is a black swan.”
When experiencing good years, investors rationally attribute gains to randomness and luck rather than skill; but in bad years, they emotionally reject the model itself rather than accepting normal variance, revealing an asymmetry in how they process positive versus negative outcomes.
“In good years, I'm like, well, there's a lot of randomness to this. I'm just you know, I I think we'll make money long term, but but that was probably luck. In bad years, I'm like, I don't invest based on this.”
The four main psychological forces that drive poor investment decisions are fear, greed, hope, and ignorance, with ignorance being the only one that can theoretically be addressed through education.
“The four horsemen of the investment apocalypse are fear, greed, hope, and ignorance. And ignorance you can deal with or not.”
Asness believes that overriding a model due to losses is the worst decision he's made, and he has intentionally tested whether those decisions had negative returns; his hypothesis is that loss-motivated model changes produce negative Sharpe ratio strategies.
“I will I have done it in my career. And I I am fond of saying, which I've never actually tested, cuz it's hard to formally test, but I'm pretty sure I'm at least directionally right, the times we've made a change that I think were motivated by losing money is the only negative five sharp ratio strategy I've ever developed.”
O'Shaughnessy witnessed a Cisco message board discussion in the late 1990s where he asked investors at what price they would sell, receiving 30 mostly irrelevant answers and only one person answering with a price—but that answer was momentum-based ('if it goes down 50%') rather than valuation-based, showing how manic investors lose the ability to think in terms of fundamental value.
“I said, all right, you know, we all love the stock, but at what price would you sell it?... I got like 30 irrelevant answers, but my favorite one, the only one who actually answered in terms of price Yeah. was one person said well, if if it went down by 50% I'd be out. And I'm like, I was asking a value question and I got a momentum answer.”
During the 1997 Asian debt crisis, Asness witnessed the equity market down 7% in a day, causing Goldman Sachs' P&L system to show AQR up 6% because short US positions had closed while long European positions had not yet opened, creating a false appearance of gains during losses—a form of 'volatility laundering.'
“Um and the S&P was down about 7% on in a day. And I have other stories about this day. Um, our P&L system said we were like up 6%. We knew that was not true. We were we at the time we were short the US and long Europe.”
During periods of underperformance driven by valuation extremes (rather than factor breakdown or momentum shifts), quantitative managers should increase public advocacy and aggressiveness in explaining their views, because the market has become irrational on a measurable dimension.
“If we lose for reasons that I think are for instance, valuation based, um where I think things are getting a little stupid, then I get more aggressive.”
Quantitative strategies can serve as a 'canary in the coal mine' for market dislocations because mispricings, if real, will show up first in factor-based portfolios before they manifest in obvious ways; O'Shaughnessy observed tiny steel companies appearing in growth portfolios circa 2004-2005, presaging the China commodity boom months before it became public.
“One of my theories is that quant also can be the canary in the coal mine. Um because in my research and I'm sure you've seen it in yours”
Even if AQR had to pick among investments, going long their own arbitrage strategy while shorting old-school factor indices would be difficult because (a) the trade is small-capacity and (b) you'd have to lever the idiosyncratic alpha extremely to make it meaningful, increasing the risk of the bet.
“We have the arrogance and I'm sure you share this arrogance to think that in the last 30 years since the early '90s our measures of of so-called factors have gotten better than they were back then. And one thing that comes up is there are still a fair amount of indices for amount of people who trade kind of the old school ones. So why not go long ours and short the old school ones? Um That You can do that to some extent, but not much. Because you got to lever the bejeezus out of that trade. Because for instance, you know, if you're doing a momentum or a value strategy, you've taken out that part. And you're leveraging up the idiosyncratic part, which may in fact be fairly high risk-adjusted return if your measures are better. And beautifully like uncorrelated with things in the world cuz you're value on both sides, momentum on both sides. You just have to lever it an uncomfortable amount.”
Putting private equity into ETFs would increase trading and price discovery, forcing real-time marking that would destroy the 'low volatility' story and likely kill the illiquidity premium by revealing true price movements.
“I actually risk think they risk killing the golden goose a little bit, because the more they make it tradeable, the more it's going to trade to real-life prices... And again, I think we're just starting on this road, Yep. but also the 401(k), the next time there's like an equity disaster, someone's going to sue saying, "You're still marking that private here in my 401(k) when it's not”
A portable alpha strategy (fully hedged positive return asset, equitized to beta-1) should be run at high volatility with low dollar amount rather than low volatility with high dollars, because both approaches have equivalent risk but the high-vol/low-dollar approach is more capital efficient.
“you should do two things. It should be run aggressively, and it should be equitized often... It's It's not that it's always going to work... But if you now imagine investing out of equities into this thing, you are replacing the equity beta. You're not actually lowering it. You're simply adding the positive expected return... they should give us less money at a higher risk level. It's just more capital efficient for them.”
Bubble identification is subjective and difficult even in hindsight; a proper bubble analysis requires assuming the most favorable case (maximum growth, ideal conditions) and seeing if valuations still make sense; if you cannot come close even with generous assumptions, it's a bubble.
“There's no bright line test... To me, a bubble is is still subjective. But it's I have leaned over backwards to put in every assumption that would make this price be too high or too low in negative bubble. Make this price reasonable and I can't come close. Um I remember the things I wrote very similar when you were writing during the dot-com bubble. I was just saying, all right, let's take Cisco Systems. Let's assume it grows more than any other company has ever grown ever for the next 10 years and then gradually comes back to just growing fast. It still sucks.”
If private equity and other alternatives are better to hold (less volatile) because they're not marked-to-market, this means illiquidity is now a feature rather than a bug; consequently, you should expect lower expected returns on these assets going forward because features (not bugs) are priced, meaning investors won't pay an illiquidity premium.
“That means illiquidity is not a bug anymore, it's a feature. Mhm. And in the markets, all else equal, you get paid for putting up with a bug that's undiversifiable, you know, you know all that. Um and you pay for a feature that everyone would want.”
If machine learning were completely transparent and Asness could understand every step, it would paradoxically be reason to distrust it, because if he could fully grasp the improvement with simpler intuitive methods, the opacity itself suggests the improvement is real and not something he's already discovered through linear statistical thinking.
“if if artificial intelligence machine learning if I understood every step of what it's doing, what's it doing? What's it adding to my world? There almost has to be and this is not a mathematical theorem but there almost has to be some opacity to it it's hard to believe we weren't finding it with simple intuitive linear statistics before.”
Quantitative hedge funds responded to Asness's criticism of their AMC short by noting that even if they were wrong on AMC, the position was so tiny that losses on it couldn't hurt the overall portfolio; however, Asness's real concern is that retail investors had concentrated portions of their life savings in a single meme stock after being taught by Robinhood that this is investing.
“if if if if if I actually shorted, you know, all of our money and it went down about 98% since that point. So 12 basis points is actually quite disappointing. Sometimes it hurts to be a quant.”
The meme stock phenomenon (GameStop, AMC) represents a generational failure of capital allocation where young retail investors believe trading behavior—aided by gamified apps like Robinhood with celebratory animations—teaches them how to make money when it actually teaches them to confuse investing with gambling.
“this is going to [ __ ] up a generation of young investors. They're going to think that this is the way you make money.”
Long-term US equity returns are unlikely to be dramatically lower than history, but expected returns going forward are lower than what occurred in the past 25 years due to starting valuation levels; therefore, long-term forecasts based on current yields/valuations should not be used as trading signals but rather as reality checks for long-term planning (retirement, pension contributions).
“we think valuation over 10, 15-year horizons is about the only thing that can guide you to to how expected returns change. It's not perfect. It doesn't have an R squared of one at your favorite horizon. Um, but it's not it's not a very high risk-adjusted return at any reasonable time horizon that that you can live with in this actual world. Right.”
The correlation of AQR's multi-factor quantitative strategies to traditional value (Fama-French HML) is only about 0.2 (20%), indicating that AQR's approach captures something different from simple value, though the correlation has been negative over rolling 3-year periods during value bubbles.
“Long term, we we just did this. We wrote a blog on this about two years ago. The correlation of our strategies to, you know, what people think of as basic value, long short, you know, Fama-French HML, or or or or somewhat more sophisticated versions than that, it's about 0.2. It's not that large. Uh it's been negative for rolling three-year periods.”
The US equity market has outperformed the rest of the world by approximately 75-85% due to valuation expansion (multiple increases), with only 15-25% attributable to better fundamental performance; therefore, betting on international mean reversion (world catching up to US) requires assuming either that US fundamentals won't keep improving or that valuations will compress.
“How much of the the US depending on you want to measure it is probably a 25-year drubbing of the the world. Um, it's probably since the Japan peak in like '89 or '90. Yep. Partly because of the the Japan debacle, but Europe whatever US has beaten pretty much everything. Yeah. Something like it all it depends on your favorite valuation measure, but pick your favorite. It's going to be 75 to 85% the US getting more expensive against the world.”
AQR launched in 2000 with only one product: a 22% volatility, fully-hedged quant strategy; the theoretical math was correct (a lower-vol version would require a quarter of the money), but practically the firm lost accounts when it experienced a two-standard-deviation drawdown because corporate treasurers couldn't be convinced that variance is manageable without observing calm years.
“When we originally launched our firm, this is a story I probably told you before, we only launched with one north of 20% vol product.”
Experts are right on average, meaning it's a disaster both to assume they're always right and to assume they're always wrong; the current environment tilts toward distrust of expertise, which is symmetric in its badness to over-credentialism.
“you know, I like to say experts are on average right. It's a disaster to assume they're always right, but it's also a disaster if you need your think that that the pointy-headed expert over there is just always wrong.”
Asness agrees with Nassim Taleb's core insight in 'Fooled by Randomness' that people systematically overattribute apparent patterns to skill/causation when they're often just random outcomes; however, Asness has critiques of Taleb's other work and thinks Taleb sometimes misrepresents prior researchers (like Gene Fama) as not understanding fat tails when they did.
“I have my issues with Nassim Taleb. I won't go through them here. And when I say issues, not a fan. But his first book, Fooled by Randomness, It was a good book. It's a good book, massively pretentiously written.”
Academic papers sometimes make significant contributions (like Fama-French HML factor) but the discoverers don't always receive proper credit when practitioners publish versions of the same idea in practitioner journals, creating a citation gap between academia and practice.
“Right. But even me who's quite close to to academia, came from, intended to be an academic, you know, close to a fair amount of people there, things that we've written, not just me, in the more practitioner journals, the FAJ, the JPM, there've been several times the academics just rewrite the freaking paper. Yep. And they don't feel they have to cite the practitioner. No, I know.”
Asness believes if someone can't distinguish between FanDuel (sports betting app) and Robinhood (trading app), they shouldn't be trusted to make independent investment decisions without guardrails.
“I am I'm fairly convinced there are a lot of I'll use the term investors, but there should be air quotes around this. Um who can't distinguish FanDuel's from Robinhood. It's just an app.”
The end of 2020 saw value spreads that were even crazier and wider than the dot-com bubble in 1999-2000, which Asness did not think would happen in his career after witnessing the dot-com peak 20 years earlier.
“I still think uh literally on the value spreads that we measure, um the end of 2020 got a little bit crazier, which I never saw coming. I you know, and I admit this. You know, a lot of my thinking is has evolved based on saying, we saw the craziest thing in 50, 60 years of good data in '99, 2000. If someone had asked me then, are you going to see something crazier in your career? I definitely I [6:48] you know, I don't know if I'd say definitely no. That's a stupid thing in our business to ever say. But I'd say it's highly unlikely, right?”
O'Shaughnessy ran a microcap strategy that adjusted for liquidity and was successful, distinguishing it from the index-level small cap effect because the strategy filtered for tradeable names and quality adjustments.
“I remember this goes back to '94, '95 when we were building our first First models we built at Goldman Sachs were actually macro models. Um I had written or was still writing a dissertation on quant equity... we had a back test and it was still based on simple value and momentum stuff. Um One of my favorite strategies was a microcap strategy. I'm still Oh, even though I've sold the company, I still let O'Shaughnessy manage all my public equities. Um and and I love the I just love our microcap strategy. But it's adjusted for liquidity. It's adjusted for all of the things that would destroy you. Because the index itself is dog [ __ ]”
O'Shaughnessy was shut down by academics on his ability to pitch to their endowment when he admitted he only had a BA, illustrating how credentialism (PhD requirement) can override intellectual merit in institutional settings.
“And one of the guys says to me, "Where'd you get your PhD?" And I said, "Oh, I I barely have a BA." ... of hard knocks. Boom. Shut it down... Like literally over here, boy, this guy's got a lot of really good ideas. Stupidest person on earth.”
AQR has a brilliant partner who did not graduate from college, which illustrates that credentials can be less important than ability; however, the market trend has moved toward distrust of experts, which is as problematic as excessive credentialism.
“one of our partners at AQR didn't graduate from college... he's a brilliant guy uh he's a polymath. He's great at so many different things... Nor am I. He'll spend his own money to keep people from going to college... but the fact that someone is a brilliant guy uh he's a polymath. He's great at so many different things. And he I think I I don't even know his story about this, but I I think he just probably would have found college boring.”
Asness and O'Shaughnessy are both concerned about democratizing private equity to retail 401(k)s through new offerings, because retail investors lack domain expertise to evaluate private equity, and the plan sponsors who actually approve the offering may not understand volatility laundering or the true risks involved.
“The move to quote democratize private investing, to bring it to the 401(k)s near you, really bothers me. Me, too... I'm a little bit of a hypocrite on this, cuz I get all paternalistic... just saying it's just not a very good idea. I mean, there are rules in 401(k)s... I think it's a pretty bad idea.”
Approximately 68% of quant managers overrode their quantitative models during the Global Financial Crisis, meaning they stopped following their systematic process in response to panic, whereas AQR and O'Shaughnessy Asset Management did not override their models despite significant losses.
“We had a guy from who had What had been Lehman, but Barclays took it over. He was a consultant to quants, right? And I'm sure you've met with this guy. I don't remember his name, but he came out right after and basically said to us that 68 and probably that's probably [ __ ] wrong, but 68% of quants overrode their model.”
O'Shaughnessy attempted to remove stock names from trading systems, replacing them with ticker numbers to reduce emotional attachment, but faced rebellion from traders citing logistical communication costs, showing the conflict between theoretical dispassionateness and practical information flow.
“When when I was reestablishing O'Shaughnessy Asset Management after rolling out of Bear um I I suggested to my team you know what? Let's let's take the names away. Let's give them the ticker has a number. And I got rebellion.”
When Asness was younger and building his first quantitative strategies at Goldman Sachs, he and his team were naive about trading costs; they would send a single list of currency trades to a broker and accept the fills at face value, not realizing the broker was optimally fleecing them to the edge of where they'd stay as clients.
“we would go to one broker with a list of currency trades we want to do. And we'd get fills from them...We would be watching Bloomberg. And we literally were print screen on Bloomberg as they were doing it. And we'd get some estimate of how far off of what we were seeing on the screens at the beginning of the trade we were getting executed at.”
Even well-intentioned quants can improve their discipline by waiting 6 months before implementing model changes made during losing periods, because diversified models have no urgency—nothing breaks overnight—and allowing a cooling-off period reduces panic-driven decisions.
“If you make a change and we've gotten really good, this is probably 20 years now, at saying, "All right, we will consider that change, but in 6 months." Um because... There's nothing in a model that's diversified that you have to do this second, right?”
The valuation spreads (price-to-fundamental ratios) between growth and value stocks are still at the 80th percentile of historical levels, meaning value is cheaper than normal but nowhere near the extremity of the tech bubble or end-2020 peak; the dot-com bubble may have been slightly more extreme in that crap companies were concentrated among the worst-performing value picks.
“Um it's still high. It's nowhere near as high as it was”
Asness previously believed AQR should offer ESG-focused products as a strategic expansion, but he now doubts this because ESG returns have underperformed and he's unwilling to tell clients that restricting the investable universe will improve their returns, which he doesn't believe.
“Um and it wasn't that I had a strong belief in it, but I believed that directionally it was important for our firm to offer it. And that was when we were building out Canvas and putting a lot of emphasis on ESG.”
Currency quoting conventions (some as foreign per dollar, some as dollar per foreign) create unnecessary cognitive friction that persists despite being 'dumb'; O'Shaughnessy still has to think about which direction he wants even after 35 years, and firms won't change because they'd eventually make errors due to muscle memory.
“I'm explaining that some are quoted as the foreign currency per dollar and some as dollar per the foreign currency. And I was admitting that doing this for 35 years, still have to occasionally think which way do I want this to go? Um and they were like, isn't that a dumb system? And I'm like yeah. So, and and and she was like, why don't you change it at your firm and you just do it the right way? I'm like, CUZ WE'RE GOING TO TRADE IT and do the opposite of what we intend at some point.”
Asness had his first fight about volatility laundering in 1997 during the Asian debt crisis, illustrating that the problem of comparing levered illiquid assets to unleveraged public markets on risk-adjusted metrics has persisted for 25+ years.
“I had my first fight about this in 1997. Um, private equity is a much smaller part of the world, but but I was at Goldman Sachs at the time and we had a private equity area and again, this is something that only you and I on the podcast will will remember, but the Asian debt crisis.”
Asness was short AMC (a major meme stock) at 12 basis points of the portfolio (0.12%) when he went on CNBC in 2023; despite the stock being bad on every metric he cares about (expensive, unprofitable, issuing shares, high beta, no fundamental momentum), he received substantial hate mail and threats from retail investors offended that he called their investment 'crazy'.
“we're short a whopping 12 basis points of the portfolio in this thing I just described that it that we hate.”
During the dot-com bubble, Asness wrote 'Bubble Logic' arguing the entire Nasdaq 100 looked like one big Cisco, and this prediction proved correct within about 26 years (Cisco peaked around 2000 in the bubble and just surpassed that peak recently).
“in this thing called bubble logic I wrote then I I said, and I'm only using Cisco as an example, the entire Nasdaq 100 looks like one big Cisco... Cisco was like I think I just saw a headline just past its tech bubble. Not the last bubble. This is the 2000 just past its its peak there... 26 years.”
Brokers responding to quant traders seeking better execution explicitly modeled the optimal amount to overcharge clients (slippage) while keeping them satisfied enough to maintain the relationship, treating it as a deliberate revenue extraction problem rather than cost reduction.
“You know what the broker did? They said, "We're going to screw these guys. How much can we What's the optimal amount to screw them?" How much can we screw them and still keep them happy?... So they they were just literally saying, "How much can we fleece Goldman Sachs and and Cliff's poor clients for?"”
Asness would use a magic microphone (if granted emperor-level power) to instill two universal beliefs: (1) government policy quality matters more than deficit/surplus status, and (2) basic statistics should be taught to every student in junior high school instead of advanced calculus.
“All of our issues with government spending and taxation. They matter a lot. Um and the debt problems are real problem, but I wish people understood most of the issue is what government actually does, whether it's funded as a deficit or surplus. If it's a good idea, we should do it. If it's a bad idea, we shouldn't do it.”
Private equity investors in Connecticut increasingly admit that they are paying much closer to public market multiples than they used to, which is consistent with the theory that private equity is becoming more crowded and multiples are compressing.
“if you get a few cocktails into one of these people, they will tell you, "Yeah, we're paying much closer to public multiples than we than we used to."”
Momentum strategies became a standard part of academics' factor models (Fama-French) in a form that lagged returns by one month specifically because Asness demonstrated in his dissertation that this lag reduced the contrarian effect at short horizons; Asness's specific methodological contribution (the lagged form) is sometimes credited to Mark Carhart or others instead.
“The way the simple way on Fama-French's website they they describe momentum is the last year returns leaving off a month, which tends to be more contrarian. I did that in my dissertation. You probably did it 10 years before me. But I did that in my dissertation. That's the version I and I'm the one who got them to adopt that.”
AQR runs roughly 1,000 long positions against 1,000 short positions globally in a relatively industry-neutral and micro-granular way, making their portfolio boring to describe; they often do not know if they own specific individual stocks.
“For us, we try to be industry neutral mostly, not entirely. Um and we'll have a thousand longs against a thousand shorts around the world, but in more micro small ways. Um we're a really boring portfolio to actually talk about. I mean, when people ask me, "Do you own this stock?" which happens all the time cuz they're like, "This You're a professional investor, we're at some party. I'm like, I don't know.”
Asness has sent approximately 8 emails in his career to academics or practitioners who failed to cite his work on a particular topic, which he acknowledges is petty but he feels is justified when the citation is clearly warranted.
“So, I'm not above not too often, but probably about eight times in my career have I sent an email to someone saying, "You know, you might want to cite the 12 papers I wrote on that."”
Asness's initial claim that if AQR lost more than 100%, he would 'run to Bolivia,' was a bad example to use publicly because it trivializes the risk (clients think theft/fraud is unlikely) and implied they should accept losses up to 100%, which he now regrets mentioning.
“Oh, I made a huge error. This is a 30-year-old 20 25-year-old error. Occasionally, when making this argument, I would point out to people that if we lost more than 100% due to utter incompetence on our part or malfeasance, I ran off to Bolivia with it. I don't even know how I would do that.”
AQR has improved significantly at not responding to social media provocations, but Asness still occasionally engages, trying to be nice initially and explain statistics, though people often become sad or defensive rather than receptive.
“I have a bit of a problem in that I will at least initially respond to the crazy people on social media. Yeah. And I You You and I have talked about that. Well, I know. You've tried to You've one of my many good friends who've tried to have an intervention with me on it. And I have gotten much better... I try to be nice the first time.”
When O'Shaughnessy pointed out to a large pension plan that a specific active manager was essentially the S&P 500 index, the manager was doing something very expensive and active, but market conditions happened to make their active bets invisible due to correlation with the index.
“I was a kid. I was 27 when I founded this company. And and I'm like in front of this big pension plan. And I'm like, um manager A uh is essentially the S&P 500. But but you asked about beta.”