The Past, The Present & Future of Quant
What this covers
This is a conversation between host Patrick O'Shaughnessy and Cliff Asness of AQR about how quantitative investing works, how it has held up, and where its limits lie. Asness spends much of the time on the question of what makes a strategy real rather than data-mined. A simple value strategy with a three Sharpe Ratio would already have been arbitraged away, he says. The persistent ones pass tests across many markets and periods and come with a plausible economic story. He also treats economics as a field that can't run controlled experiments, so out-of-sample tests and logic have to stand in for them.
The middle of the conversation turns to the investor's side. Asness says clients and managers overestimate how much loss they can tolerate, so a strategy should never be pushed to the edge of someone's comfort. He also argues that worst-case models built from history are flawed, because the true worst case is the one that hasn't happened yet. His preferred "first-best" portfolio is a multi-factor long-short book across every applicable market, with each sleeve sized by risk rather than dollars.
The range is wide, and several positions are contested. On crowding, he says too much capital does not automatically produce losses, and he points to the value spread as a way to measure it. He gives the Shiller CAPE a narrow, weak edge as a timing tool, and he argues that twelve-month momentum is a poor result for efficient-market theory. He defends the social usefulness of active price-setting while suggesting fewer active managers may be needed. He holds that factor returns survived a century of technological change, from steamboats to railroads. Smaller threads include AQR taking value risk within industries and why it would not go public.
Asness argues that quantitative factor strategies (value, momentum, quality, carry) are robust across assets, time, and technology, but that their returns come with risks, behavioral tolerance limits, and crowding dynamics that investors and managers routinely misjudge.
- Factor premia have held for roughly 100 years across asset classes and geographies, so a bad stretch like 2017-2018 is not evidence the factors have broken
- Worst-case models built from history are flawed because the true worst case is the one that has not happened yet
- Risk-adjusted return only matters insofar as it produces dollar returns that clients can tolerate and stick with
Value and momentum are robust factors that carry across markets, asset classes, and timing tilts.
- Treating countries as individual stocks by comparing aggregate market-cap-to-book ratios across country indices (for example the DAX versus the FTSE) produced a value and momentum signal, and this holds as long as the comparison is like-to-like among about 20 equity markets.
“We found it did hold up, but there as long as it's like to like, even among say 20 equity markets”
- Market timing is a sin that will probably hurt investors if overdone, but very small timing tilts, with valuation and trend moving in the direction a cheap but improving market suggests, can be justified because the basics hold at a weak level.
“we basically are telling people market timing is a sin. It will probably hurt you if you do too much.”
- For pure market timing, trend or momentum is more powerful than valuation because valuation measures are harder to compare through time, while one-year price momentum is a very robust measure, and the two are negatively correlated so combining them is useful.
“Trend or momentum as it applies to direction for pure market timing is a little more powerful.”
- The same ideas of value, momentum, quality, and carry apply at the asset allocation level, with momentum called trend, but the links between factors break across asset classes, so carry and value in currencies are largely unrelated to their stock-market counterparts.
“I think it is mostly the same ideas repeating.”
- Value and momentum work well together, and a value investor's momentum tilt is a complement, so a single momentum factor can be added as a compliment to someone who already has a value tilt.
“some even single factor on the momentum side where we think, maybe that's a compliment to people who already have a value tilt.”
- AQR is most excited about applying value, momentum, and quality factors to fixed income and credit, including benchmarks-relative trading, because the same ideas hold in credit as in equities, such as credit spread as a value indicator and spread on its own as carry.
“I'm very excited about fixed income... So I am intellectually excited to see the same ideas hold up”
Crowding and capacity erosion eliminate premiums, so strategies should be closed when expected returns disappear.
- The fee an investor can charge for a strategy depends on its secrecy: once a premium is known, supply and demand mean you simply cannot charge alpha fees for it, which is why many factor strategies are now priced as styles rather than alpha.
“You simply can't charge for something that's not a secret.”
- A useful way to detect too much money in a value strategy is to measure how cheap the long side is versus the short side on valuation (the value spread), since crowding would compress that spread.
“One thing you might expect if way too many people were doing this would be great compression in these numbers.”
- Alpha that is proprietary today tends to become a known style over time, since the arc of history runs toward proprietary strategies becoming styles, and what was alpha to clients twenty-five years ago (like systematic value and momentum) is now a commodity-like factor.
“the ark of history runs towards proprietary becoming a style”
- If a strategy is arbitraged away by too much capital, it does not automatically lose money going forward; it simply looks more like a random number, so too much money is a poor explanation for a specific bad year.
“if a strategy was arbitraged away... it would just look more like a random number, going forward.”
- The most major change in how institutions use factors is that factor investing has become much more widespread, whereas when Asness started there was no such thing as factor investing and the work was just testing whether academic ideas could make real clients money.
“By far, the most major change is of course them just being much more widespread. I'm old enough to remember when there was no such thing as factor investing.”
- If transaction costs and spreads collapsed so that a value spread stayed flat near a low threshold (for example one and a half times rather than three to six) for an extended period, AQR would stop running products that only work in that compressed state, as an ethical obligation.
“We've got to stop doing this. And if we have any products that only do that, we're going to stop doing those”
- Investors would sleep better and earn better results if they could not see their portfolio value for ten years, because illiquidity lets them take more risk, though this is not the world investors live in.
“if you knew they didn't look at it for 10 years... You'd sleep better at night, and you'd probably end up with a net better result.”
- If the strategy's expected return disappeared, continuing to run it would be an ethical failing, and Asness would close even a product with a zero Sharpe ratio rather than keep selling it.
“if we saw the opposite and we don't do it, that would be a ethical failing”
- The most valid client question is whether too much money in the strategy could arbitrage it away, and this is plausible because no strategy can run unlimited capital without affecting its risk-adjusted return.
“there is no strategy that can run any amount of money it chooses without affecting its risk adjusted return.”
- A client redeemed from AQR's strategy in its early years, even though they found it the most attractive holding in their portfolio, because they judged that recovery from the drawdown would take about four years, illustrating that investor patience limits strategy use.
“we think you are the most attractive thing in our portfolio right now, but we think... it will be four years until you make the money back and we can't wait that long.”
Factor premia need economic explanations and cross-context testing to avoid spurious findings.
- Given 100-plus years of evidence across asset classes, geographies, and periods of technological change like rural electrification and the shift from steamboats to railroads, factor returns are reasonably robust to technological change.
“across asset classes, across geographies, across other times of modest technological change, like rural electrification and going from steamboats to railroads, that the stuff is reasonably robust.”
- AQR will not deploy a machine learning process that lacks an economic explanation, because with limited data (only one observed history of the economy) you cannot rely on pure statistical discovery and need interpretable, plausible stories.
“it is unlikely at AQR you will see a machine learning process that has no economic explanation.”
- Quant investors can attribute performance to eleven decimal places and still be unsatisfied, because they have strong economic stories for why factors work on average but cannot explain why in one particular year momentum failed to offset value.
“why in one particular year, momentum didn't catch it right and more than make up for for value”
- Testing an academic author's factor story across the asset classes, geographies, and time periods where it should apply raises the hurdle for data mining, since a finding that holds in similar spirit across contexts is less likely to be spurious.
“all you're creating now is a higher hurdle to data mining. It could still be data mining, but when you find some similar spirit story, works in different asset classes and different geographies and different time periods”
- Economics cannot use the scientific method's randomized A/B experiments because the world is observed only once, so each event like Pandora's box cannot be unrung, and plausible economic logic plus out-of-sample tests must substitute for controlled trials.
“We get to observe the world once, often it's Pandora's box. Somebody observed something and you can't unring that bell.”
Scalable strategies with modest Sharpe ratios serve clients better than rare high-Sharpe ones.
- Strategies with a three Sharpe ratio are usually a pejorative in the industry because true ones are very rare and hard to scale; even when they exist, managers like Renaissance Technologies keep their best capacity closed and sell open products that look more like ordinary strategies.
“Three sharp ratio strategy at AQR is generally, as you said, used as a pejorative.”
- Lower fees are always better for the same product, but most of the consumer welfare gain from lower costs has already been captured, and obsessing over two basis points versus five is low-value except when two products are truly equal.
“If there are two things you're absolutely convinced are equal... take the three.”
- A simple value strategy with a three Sharpe ratio would have already been arbitraged away, so the real persistent strategies are those that pass tests across many places and time periods with a good story, while data-mined ones that look excellent will not repeat.
“if value ever, it never was a three sharp ratio simple value strategy, but if it was, it wouldn't be now. It would have been arbitraged away.”
- AQR targets a range of about 0.5 to 1.0 Sharpe ratio for real-life strategies, which is reasonable and realistic for institutional capacity, and any strategy with a Sharpe far above the S&P 500's is an outlier that is hard to sustain.
“we tend to use a range of about 0.5 to to one as a sharp ratio that we're targeting”
- Investors require a somewhat higher Sharpe ratio than the S&P 500's from unconventional strategies, but the equity risk premium's strong theoretical story means even a low Sharpe strategy can survive, since a 0.4 to 0.5 Sharpe held for 20 years almost never loses money.
“if you really have a 0.4 to 0.5, and you stick with it for 20 years, it's a very small chance you're going to lose money.”
- Risk-adjusted return serves total return, so a 0.5 Sharpe strategy that can absorb $100 billion and diversifies the S&P 500 can make far more clients better off than a three Sharpe strategy that can only absorb $100 million.
“Dollars are what counts. A 0.5 sharp ratio strategy which is actually better than the S&P500... that you can apply to a $100 billion versus a three sharp ratio strategy that you can apply to a $100 million”
- The equity risk premium has a stronger theoretical explanation than any factor story, which is why investors can justify holding a low-Sharpe strategy such as the stock market itself, since investors should be compensated for bearing risk.
“the story for why there should be a positive equity risk premium is better than any story you or I have.”
Investors overestimate their loss tolerance, so strategies must fit what they can actually stick with.
- The tough period for systematic value has run mostly since the global financial crisis, not just this year, and over the post-GFC period value has lost most of the time even though the overall multi-factor models have done well.
“the tough period for systematic value has been going on mostly since the GFC.”
- Investors consistently overestimate their comfort with losses, so when tough times arrive they discover they are a bigger wimp than they thought, which is why strategies should never be pushed to the edge of someone's comfort point.
“I think we all over estimate our comfort point a little bit. When the tough times hit, you discover ... you're a bigger wuss than you thought you were.”
- Planning for a worst case much worse than the Great Depression for the economy and stocks is unproductive, and such a scenario would call for the proper assets to be firearms and canned goods.
“If you think the worst case is much worse than that, and the proper assets were firearms and canned goods.”
- The most common worst-case model, the worst thing in a backtest or real history, is inherently flawed because the true worst case is by definition the one that has not yet happened, so investors must haircut history and stress beyond it.
“you know that's a bad model, because it missed the one that happened.”
- A strategy that looks better on paper is not necessarily better for investors, because the strategy an investor can actually stick with is the one that matters, and managers and clients overestimate their tolerance for fluctuation.
“The better strategy you can stick with is not the better strategy.”
- Private equity's ability to avoid marking to market provides a behavioral and volatility-smoothing benefit that lets investors hold more risk and tolerate their positions, which is a rational reason for its popularity even though it is a form of fooling oneself with open eyes.
“a lot of people pursue it with very open eyes, knowing that that inability to market allows them to be better investors.”
- Long-run negative autocorrelation in stock returns adds only a little to the long-term safety of equities; the main reason long-horizon holders rarely lose is that even a modest Sharpe ratio compounds to a very small chance of losing over two decades.
“So a little bit of negative order of correlations in stock returns makes the long term a little bit better, that's not most of it.”
Credible research rests on curiosity, skeptical bottom-up backtesting, and AQR's data-honest, rigor-driven culture.
- AQR's research process combines bottom-up work, where researchers read the latest papers and test them everywhere the theory should apply, with a more formal top-down structure of scheduled seminars in which someone presents a favorite recent paper, modeled on the University of Chicago finance seminar.
“We do a fair amount of top-down in that... and this has increased over the years.”
- Asness's dissertation trading on momentum ignored transaction costs and used equal-weighted portfolios, which was a common flaw in the literature of the time, so the strategy's survival after adding real-world frictions owed substantial luck.
“the trading I did, ignored transactions costs and worse was trading equal-weighted portfolios.”
- The top quantitative researchers are driven first by genuine curiosity about finding new things and secondarily by economic motives, and raw talent without that curiosity, focused only on making money tomorrow, does not get there.
“a genuine curiosity, the top researchers here... are excited about finding out new stuff first.”
- The persistence of twelve-month price momentum as an effective strategy is not a good result for the efficient market hypothesis, since it appears to produce excess returns that a risk-based explanation has not yet fully accounted for.
“it is not a wonderful result for the efficient market hypothesis that 12 month price momentum is an effective strategy.”
- Eugene Fama's acceptance of Asness's dissertation on price momentum was an act of intellectual honesty, because the result challenged efficient markets, and Fama's guidance was: if it is in the data, write the paper.
“If it's in the data, write the paper.”
- AQR's founders were asked at Goldman Sachs to pick countries because a long-only product had been in the wrong countries for about three years, and the team's first attempt was largely luck-based and admittedly guilty of minimal rigor.
“we were asked to do this because the current long-only equity... were in the wrong countries for about three years.”
- On a 1997 Asian debt crisis day, Goldman's P&L screen showed AQR's book up about 3 percent while the team's best estimate was flat, and the team's real-time mark-to-model view was the only one that was accurate.
“that day, our PNL system said we were up. In reality, we guessed we were flat.”
- Machine learning is best viewed as an evolutionary tool for finding nonlinearities and interactions that ordinary linear regression misses, rather than a revolution, and AQR will test it over the coming years before judging its value.
“One is to sound trite, evolutionary not revolutionary.”
- AQR's internal culture models the University of Chicago finance seminar, where presenters are challenged with 'I don't buy it', which forces researchers to convince peers rather than accept ideas on authority.
“people here have to kind of convince other people of their idea.”
Whole-portfolio diversification across factors, industries, and value measures limits single-factor risk.
- Describing a bad year as the worst factor's loss misleads, because a multi-factor portfolio losing on value may be offset by others; flipping the sign on the losing factor would make the year look like a win, so the goal is the whole portfolio rising, not any one factor.
“I don't really care which factor goes up, I care desperately the whole goes up.”
- The first-best approach for an investor who can stick with it is to run a multi-factor long-short portfolio across every market where quantitative factors apply, scaling each sleeve by its risk (levering low-risk ones and de-levering high-risk ones) so each contributes roughly comparable risk rather than equal dollars.
“In theory, and I believe in practice, the first best use if you are an absolute Vulcan ... is to go through every place you think quantitative factors apply, do a multi-factor long-short portfolio, lever the ones that are very low risk.”
- AQR takes most of its value risk within industries rather than across them, a design choice supported by an unpublished 1995 paper with Asness, Porter, and Stevens, and this reduces exposure to a few dominant stocks reshaping an industry.
“We've always taken most of our value risk, not across industries, within industries.”
- Any single valuation measure can be told as a story where it is broken, and price-to-book takes the most heat because of intangible-asset businesses like Netflix and Facebook, but AQR has never made price-to-book dominant and diversifies across many value measures.
“For one thing, we've always given some weight to price to book but never literally since we started in late '94 at Goldman Sachs. It's never been a dominant factor for us.”
Public markets and active price-setting remain viable despite indexing growth and declining listings.
- Indexing has risen to the mid-30 percent range of mutual fund assets, up dramatically from about 10 years earlier, but further growth in indexing is plausible while a world with no active price discovery, the singularity, is unknowable and unlikely.
“The last numbers I saw from mutual funds were somewhere in the mid 30% were indexed”
- Quarterly earnings calls and public reporting pressure would make a long-term firm like AQR behave more short-term, which is one reason it would not want to go public; Sarbanes-Oxley is probably only a partial explanation for the decline in IPOs.
“Now having to do quarterly earnings calls... does not seem like a good strategy to me.”
- Even with more companies staying private, there are still more than enough public stocks with a giant amount of market cap, and public markets would only be constrained if they shrank to around a tenth of their prior size.
“I still think there are more than enough stocks that are public.”
- Markets need some active managers to set prices, because if everyone indexed the last bit of mispricing is worth nearly nothing, but historically there were probably too many people trying to beat the market, so fewer may be needed for an efficient market.
“we've probably had too much for the last 50 years. We've probably had more people trying to beat the market than is necessary”
- A market economy's advantage over a planned economy is that many smart people competing to set prices creates a large positive externality, so the activity of pricing securities is socially useful rather than mere betting.
“there's a tremendous positive externality from people doing it.”
- Sarbanes-Oxley is often blamed for the decline in public listings, but Asness thinks it is generally overdone as an explanation and he cannot determine how much it contributes, so the decline reflects both supply and demand factors.
“I think it's generally overdone as an explanation. I don't have an alternative one.”
Valuation signals carry only modest, fragile information about future returns.
- Price-to-book looks cheaper than other value factors and value overall is probably a little cheaper than average, while low beta is the only factor that looks expensive versus history, though still within historical bounds.
“The only one really at all expensive versus history is the low beta factor.”
- Historically the expensive stocks on price-to-book have traded in a band of three to six times the price-to-book of the cheap stocks, except during the 1999-2000 tech bubble when the ratio reached roughly 12 to 14 times, and it has stayed in that band since.
“the expensive stocks have always with one exception traded in a band between three and six times the price to book of the cheap stocks.”
- The Shiller CAPE predicts 10-year equity returns and should lower expected returns when high (e.g. 35 vs an average of 16), but using it to time the market out of sample yields only a narrow, weak edge across about 120-140 years and is not a career-making strategy.
“if the Shiller Cape is 35 versus an average of 16, I would use a lower expected return for the equity market going forward. That doesn't mean it's time-able.”
- Testing CAPE timing by bucketing returns by Shiller P/E over the full period is cheating because it uses future information, so a fair test must look only backwards on rolling windows, and under that test CAPE still shows a positive but narrow edge.
“if the Shiller Cape is 35 versus an average of 16, I would use a lower expected return for the equity market going forward. That doesn't mean it's time-able.”