This interview covers the past, present, and future of quant investing (using computer models to pick stocks). Cliff Asness of AQR says once-secret quant strategies are now public knowledge, so the real challenge is sticking with them during bad times. He argues value investing (buying cheap stocks) isn't dead—its price gap is still normal. Machine learning in finance needs an economic story, and market timing is very hard. Key mentions: Fama-French HML (a value measure) is in its normal range; Renaissance's Medallion fund (great returns but closed to outsiders); Shiller CAPE (a market valuation gauge) has weak predictive power.
Cliff Asness, co-founder of AQR Capital Management, reflects on the past, present, and future of quantitative investing in a podcast. Since its founding in 1998, AQR has managed $226 billion in assets. Core insights include: factor portfolios have evolved from simple long-short strategies to more co
Cliff Asness, co-founder and managing principal of AQR Capital Management, which manages $226 billion in assets. This issue explores the past, present, and future of quantitative investing, with the core thesis being: Factor investing has evolved from a "secret weapon" into a "public tool," but what is truly scarce is not the strategy itself, but the client's ability to persevere through adversity. Asness offers a counterintuitive insight: "A better strategy you cannot stick with is not a better strategy" — the most critical challenge in quantitative investing lies not in the models, but in behavioral finance.
Cliff Asness believes the biggest change in factor investing is that "it has gone from alpha to style" — what was once known only to a few has now become public knowledge.
Asness describes the evolution of factor portfolios:
AQR's response:
1. Cross-asset, cross-factor, cross-geography — not just equity value, but "every factor you believe in, in every place you believe in it."
2. Risk parity — "If you put $1 into equity long-short and $1 into bond long-short, you're an equity manager, and bonds are just your hobby" — capital must be allocated based on risk contribution.
3. Leverage and deleverage — Leverage low-risk factors, deleverage high-risk factors, aiming for roughly equal risk contributions from each factor.
Asness emphasizes that the ideal "Vulcan" solution is a fully diversified portfolio of multiple factors, multiple assets, long-short, with leverage, but in reality, few can stick with it. Therefore, AQR offers a spectrum from "traditional long-only" to "fully unconventional," allowing clients to choose a position they can tolerate.
Asness argues that contrarian factor timing is extremely difficult, and the question clients most frequently ask — "is there too much money?" — is a reasonable one but often misapplied.
Core chain of reasoning:
1. Measuring whether a factor is "crowded": During the 1999 tech bubble, AQR developed the "value of value" metric — assessing how cheap the long side is relative to the short side.
2. Historical range of the Fama-French HML (book-to-market factor): The price-to-book ratio of expensive stocks relative to cheap stocks typically fluctuates between 3x and 6x; the sole exception was the 1999–2000 tech bubble, when it surged to 12x–14x; it has since returned to normal ranges and is currently "likely on the cheap side."
3. Other factors: The low-beta factor is slightly expensive but remains within historical bounds.
4. Conclusion: "If the factor were truly being arbitraged away, you would see the spread compressed — but we don't see that."
Asness's response to whether "value has been disrupted by tech stocks":
On timing, Asness cites his own paper title, "Contrarian Factor Timing Is Deceptively Difficult" — even if you believe factors have predictive power, precise timing is nearly impossible. "We would not short value just because it has underperformed for 6–7 consecutive years — because we believe no one can do this well."
Asness coined the "Three Sharpe Ratio Strategy" as an internal pejorative at AQR—referring to strategies that claim exceptionally high Sharpe ratios but operate at minuscule scale.
Core arguments:
1. Real-world Sharpe ratio range: AQR targets 0.5–1.0, which is "achievable at true institutional scale"
2. Scale vs. ratio: "A strategy with a 0.5 Sharpe ratio that can be applied to $100 billion makes more clients wealthier than a strategy with a 3.0 Sharpe ratio that can only be applied to $100 million"
3. The paradox of the "Three Sharpe Ratio Strategy": If such a strategy truly existed (e.g., Renaissance's Medallion), "they would kick you out"—no external capital is accepted
4. Economic natural Sharpe ratio: Asness argues that real strategies "sink" to an economically sustainable Sharpe ratio—"high enough to be worth doing, but not high enough to be arbitraged away"
On Liquid Alt Ragnarök (the "Ragnarök" of liquid alternatives): Asness warns that investor expectations for alternative strategies are too high, while actual Sharpe ratios are far lower than imagined. "If you could build a 1.0 Sharpe ratio, market-neutral long-short strategy, theoretically you should put more money into it than the S&P 500—but no one would do that"
Asness’s stance on machine learning: AQR has invested heavily (hiring Marcos Lopez de Prado), but adheres to two principles.
1. Evolution, not revolution: Machine learning is "a tool for finding nonlinearities and interactions," not a disruptive breakthrough.
2. Must have an economic explanation: "At AQR, you are unlikely to see a machine learning process without an economic explanation."
Asness acknowledges this is "an open question" — can machine learning improve performance while maintaining explainability? "We will know more in two years."
The core conclusion of the paper Sin a Little, co-authored by Asness and Antti Ilmanen: Timing the stock market is a "sin," but if you must do it, do it sparingly.
Key findings:
1. Value timing (e.g., Shiller CAPE): Shows positive returns in 120-year backtests, but with a "very thin edge" — "there is a 40-year flat period"
2. Trend/momentum timing: Stronger than value timing because "momentum indicators (e.g., 12-month price momentum) are more robust and do not require rolling windows"
3. Combining the two: Value and trend are negatively correlated in timing, so "doing a little of both" is better than doing just one
4. Recommendation: "When the market looks cheap but is improving, go slightly long; and vice versa — but these should be very small adjustments, as the signals are not strong enough"
Asness emphasizes that the "value + momentum + trend" framework is universal across asset classes — "there are parallels in the cross-sectional predictability of stock returns and country returns" (citing the 1995 paper The Parallels Between the Cross-Section of Expected Stock and Expected Country Returns)
Asness shares a story from the 1997 Asian debt crisis, illustrating how his view on private markets evolved.
Key points of the story:
Asness's reflection:
On the rise of private markets:
Asness argues that the major trend of fee declines is largely complete, and the current debate over "a few basis points" is a "strange macho slogan."
On the future of the active management industry:
| Position | Guest Sentiment | Key Data |
|---|---|---|
| Fama-French HML (Book-to-Market Factor) | Neutral (still valid, but not the sole indicator) | Historical spread 3-6x; peak of 12-14x in 1999; currently "likely on the cheap side" |
| Renaissance Technologies (Medallion) | Bullish (acknowledges its performance, but notes it does not accept external capital) | "They won't take your money" |
| Shiller CAPE | Neutral (has predictive power, but very weak) | Positive returns over 120-year backtest, but with a 40-year flat period |
1. "A better strategy you can't stick with is not a better strategy" (Cliff Asness) — Behavioral constraints matter more than model precision; AQR offers a spectrum from "traditional" to "non-traditional," allowing clients to choose a position they can tolerate.
2. "A strategy with a 0.5 Sharpe ratio that can be applied to $100 billion makes more clients richer than a strategy with a 3.0 Sharpe ratio that can only be applied to $100 million" (Cliff Asness) — Scale matters more than Sharpe ratio; the "Three Sharpe Ratio Strategy" is a pejorative term inside AQR.
3. "At AQR, you are unlikely to see a machine learning process without an economic explanation" (Cliff Asness) — Machine learning is "evolution, not revolution"; financial market data is insufficient (can only be observed once), and interpretability is a hard requirement.
4. "If factors were truly arbitraged away, you would see spreads compressed, but we don't see that" (Cliff Asness) — The spread of HML remains within its historical range (3–6 times); "People ask 'is there too much money' in bad years, but being arbitraged away means zero expected return, not negative return."
5. "Market timing is a sin — if you must do it, do it a little" (Cliff Asness, citing the paper Sin a Little) — Value timing (e.g., CAPE) has positive returns over 120 years but a "very narrow edge" (with a 40-year flat period); trend timing is slightly stronger; combining both is better than either alone.
6. "They are 'lying to themselves with eyes wide open' — using not marking to market to become better investors" (Cliff Asness, reflecting on private markets) — Illiquidity is an advantage at the behavioral level (allowing more risk premium to be taken), but it is not a "free lunch."
7. "If the whole world is doing index investing, who does the pricing?" (Cliff Asness) — Active management will not disappear, but historically "there were too many people"; the current indexation rate is about 35%, with room to grow but not to zero.
8. "If it's in your data, write the paper" (Eugene Fama to Asness) — Asness considers this the "ultimate ethical statement"; Fama himself is not a fan of momentum investing but allowed DFA to use it.