Peter Hecht helps run AQR Capital Management's $242 billion in assets, and he says the academic paper that made small-cap stocks famous in finance had the data signs backward.
Most quant shops now face the same worry: if every fund has access to the same large language models, does the edge disappear. Hecht argues the model itself is a commodity and always was — the edge was never in the tool.
"All of the special sauce is not in the large language model."
Hecht was one of two teaching assistants Eugene Fama picked for his University of Chicago PhD class, later taught at Harvard Business School, and now co-heads the Portfolio Solutions Group that designs how AQR's institutional and high-net-worth clients actually hold its funds.
I listened to the full interview so you can skip it. 65 minutes of audio, 23 minutes of reading.
Here are the 17 takeaways that matter.
👤 Guest: Peter Hecht, Managing Director and Co-Head of AQR's North America Portfolio Solutions Group, who holds a Ph.D. in finance from the University of Chicago and previously taught at Harvard Business School
🎙️ Host: David Weisburd, an investor who has raised over $2 billion in institutional capital and hosts How I Invest, which interviews the world's leading institutional investors
📰 Published: 9 September 2026
🔴 YouTube | 🟢 Spotify | 🟣 Apple Podcasts | ⏱️ 1 hr 5 min | ✅ Time saved: 42 min
Key Takeaways
A 3% tracking error means a manager could just as easily beat or lag the index by 3%, and refusing to take any of it means giving up any chance of beating a benchmark
AQR treats diversification like a casino running thousands of small, uncorrelated bets rather than one big one
Alpha has no single definition — it depends entirely on which risk model you measure it against
The only test that matters to a client is whether adding a strategy improves the portfolio they already hold
The academic paper that made small-cap investing famous had positive and negative signs reversed in its own data
AQR found no small-cap effect once it adjusted for beta
AQR won't converge with rivals just because everyone has access to Claude and ChatGPT
It skips the prompt-based approach of asking a chatbot if a stock is cheap, because that's too easy to copy
AQR now spends as much time worrying about underfitting its models as overfitting them
In stock selection, AQR is right only 51% to 55% of the time — not 80%
Trend following has zero correlation to stocks and has outperformed in every major downturn since 2008, but it isn't insurance
It bled through the calm 2010s, and investors who quit right before 2022 missed a 30% to 40% year
Roughly 90% of a diversified portfolio's risk usually comes down to one factor, because private credit behaves like equities in a downturn
Bonds stopped protecting stock portfolios in 2022 because they only hedge growth shocks, not inflation shocks
Portable Alpha lets an investor keep full S&P 500 exposure while porting in a separate hedge-fund strategy on top of it
Hecht's biggest career regret is being too humble to state his own credentials
1. Tracking Error, Defined
Weisburd opened on one of his own pet peeves: the term "tracking error," and whether it's a misnomer.
Tracking error isn't a mistake — it's the volatility of your return relative to a benchmark, not your return in isolation. "Tracking error is the volatility, but it's volatility of the relative return, your return versus a benchmark." A tracking error of 3% means, assuming a normal distribution, a roughly 66% chance a manager finishes within 3 percentage points of the S&P 500 in either direction
Tracking error can be good, and it's required if a manager wants to beat a benchmark at all. "Yes, because if you want to beat the benchmark, if you want to try to add value, you have to take tracking error." Hecht separates it into bad tracking error, taken by a manager with no real process for picking securities, and tracking error taken by a manager with a good one
More tracking error raises the ceiling on outperformance and the floor on underperformance in the same move — the number itself says nothing about whether the risk is worth taking
2. Does Passive Help Active?
Weisburd asked whether the growth of passive investing has made it easier for active managers to find alpha, and whether passive flows are making markets more volatile.
Whether passive helps or hurts active managers depends entirely on which active managers are the ones getting fired into it. If the managers going passive were bad, the remaining active field gets stronger on average; if some of them were good, competition for the same alpha shrinks
A body of early academic research argues passive investing is making markets more inelastic, meaning flows move prices even without any new information behind them. "There have been some academic papers, and it's early innings, and there's people who disagree with it, that believe that the rise of passive have made markets more inelastic." Hecht said that would mean more volatility for a given flow than markets saw before, but cautioned that in academia it can take "twenty years to figure out who was right"
3. AQR's Casino Diversification
Asked to describe AQR's core philosophy across all of its strategies, Hecht went straight to diversification.
Most systematic managers have only a small edge on any single trade, and AQR's answer is volume, not precision. "Well, like a casino, a casino would never exist if it could only play blackjack against you, one person. They bring in thousands of uncorrelated blackjack customers. That's like bringing in thousands of uncorrelated trades. So we believe in diversification. We're militant about thinking about are these trades correlated with each other?"
That discipline is what lets AQR run the same underlying process in very different wrappers — a market-neutral fund with no exposure to the broader market, or a fully invested one with a beta of 1 — because the risk management benefit of diversification (a handful of bad positions can't sink the fund) holds either way
4. What Alpha Actually Means
Weisburd asked how AQR answers the question every prospective client asks: does it actually generate alpha? Hecht said the honest answer starts with admitting the term is contested.
Alpha isn't physics, and it isn't one number. "So first off, finance is a social science. It's not physics. It's not math." What counts as alpha depends entirely on the risk model used to define it — Gene Fama taught Hecht at the University of Chicago that a factor like value (cheap stocks by book-to-price) isn't alpha at all if it's a compensated risk factor in your model, though other economists would call the same return a behavioral mispricing rather than a risk premium
The models have stacked up over decades: Bill Sharpe's Capital Asset Pricing Model measured everything against market beta alone; Fama and Ken French then added a value factor; other managers add ten more factors they believe are real risk premia. Whichever model you pick determines whether the same return counts as alpha or not
The only definition that matters in practice is whether a strategy improves the portfolio a client already owns. "If some theoretical exercise says it has alpha but doesn't improve your portfolio, it's dead to you as far as I'm concerned." AQR calculates this with what's called an appraisal ratio — running a new investment idea in a regression against the client's current holdings to see whether it genuinely helps, rather than relying on an abstract academic definition of alpha
5. Doubting Small-Cap and Value
Weisburd raised a mutual friend's tip that Hecht was one of two teaching assistants for Eugene Fama, the Nobel laureate who helped popularize the "small in value" style of investing — and asked what Hecht makes of today's challenges to whether that effect still exists.
AQR was skeptical of the small-cap effect before it was fashionable to be. The early academic paper that brought the small-cap effect to light had a basic data problem. "There were like positive signs that should have been negative signs on returns." Once you control for the fact that small caps carry more beta than large caps, Hecht said, AQR found no small-cap effect at all — a levered large-cap portfolio at the same beta performed about the same
What made the Fama-French value paper important wasn't the specific ratio, but the underlying idea. "And what made the Fama French insight amazing wasn't because it was just book to price. Was the larger underlying guiding principle that something that's fundamentals scaled by price might be informative about future returns."
One flaw in the original paper Hecht flags: it never controlled for industry, so it compared a cheap-looking tech stock's price-earnings ratio directly against an expensive-looking utility's, even though the two have completely different growth profiles. Serious managers today measure value stock-against-stock within the same industry, and increasingly use machine learning to build better value measures than a single ratio
False ideas can survive in markets for decades, even among smart people, and diversification is the best defense against betting on one that turns out to be wrong. "These memes in the market, even when they're untrue, persist, could persist for decades, even among some of the most intelligent people." A fund holding a thousand small, uncorrelated positions survives being wrong about one theme; a concentrated manager long 20 names and short 20 can have an entire year ruined by one bad call
6. AI, Convergence and Judgment
Weisburd relayed a worry he's heard from other quant investors: that everyone loading the same information into Claude and following its output is a risk to AQR's edge. Hecht disagreed, and drew a sharp line between two different ways of using a large language model.
AQR doesn't use AI the way a retail investor might, and that's the whole point. "You literally ask ChatGPT or Claude, is Apple cheap or expensive? Give me your analysis and give me the details why you think it's cheaper or expensive. That's prompt, okay?" Hecht calls that the prompt-based approach, and says it's "very inefficient" as an investment process because every competitor can ask the identical question and get a similar answer
AQR instead feeds financial documents, broker reports and earnings calls into a model and extracts a word embedding — a numerical representation of the text — then treats that number like any other signal it has to test for whether it actually predicts returns
The commodity is the embedding. The edge is proving it works. "All of the special sauce is not in the large language model." Getting a numerical representation of text is now cheap and available to everyone; figuring out whether that number forecasts future returns is the hard part: "a ton of skill and it's highly subjective" is how Hecht put it — which is why Hecht doesn't expect competitors to converge on the same trades just because they use the same underlying models
What separates a good systematic manager from a bad one is judgment that doesn't show up on a résumé. Two people with identical academic backgrounds can build very different processes — one who correctly judges which signals genuinely predict returns, and one who is either overusing quantitative tools or leaving money on the table by underusing them
7. Recruiting After ChatGPT
Weisburd asked whether recruiting has changed since large language models arrived, citing Ken Griffin's line that Citadel's biggest asset is the people it recruits, and the well-worn story that Renaissance Technologies' black box is really a few hundred PhDs.
AQR is still hiring for the same qualities it always has, chiefly intellectual curiosity — the trait that stops someone from running a backtest that works by chance and declaring victory, rather than testing it out of sample until they find out it doesn't hold up
Communication is not a soft skill at a quant fund — it's the difference between having an idea and having an idea anyone can use. Hecht credits Harvard Business School with teaching him this the hard way. "Isn't it about the engine? And they said actually the paint job's important." He had assumed presentation was superficial until colleagues there convinced him that an idea trapped in one person's head, however good, does nothing for the firm
AI doesn't erase the need for people. "But human capital is where it's all at, right?" Innovation is not optional at an asset manager, in Hecht's framing — a firm that stops innovating isn't necessarily doomed immediately, but a client should already be looking elsewhere
8. A Signal Earns Its Way In
Asked how AQR decides what goes into its models, Hecht described a high bar built around economic reasoning first, empirical testing second.
A candidate signal typically starts with an economic thesis for why it should predict returns — a behavioral under-reaction story, for example — rather than starting from a data-mining exercise that risks finding patterns in randomness
Once a thesis holds in one market, AQR keeps testing it everywhere it should apply if the thesis is true: US large caps, then US small caps, then European, Japanese, Australian and emerging markets, then a freshly purchased dataset treated as an out-of-sample check, then macro assets like country indices if the underlying logic says it should transfer there too
The industry's traditional fear is overfitting. AQR now worries just as much about the opposite mistake. "Everyone's always worried about the risk of overfitting. We also now spend time worrying about the risk of underfitting." Machine learning has shown AQR that some signals have non-linear relationships with future returns that older, simpler linear models were leaving on the table
Once a signal is in, it isn't cut after one bad month — AQR now uses machine learning to systematically up-weight and down-weight signals based on confidence in how well each is currently working, rather than removing one on a single rough stretch
9. Conviction and Hit Rate
Weisburd asked whether conviction in a trade's underlying thesis is what lets an investor hold through drawdowns, and how often AQR is actually right.
Staying rooted in why a trade was made in the first place is what determines whether an investor sells at the right time, not whether the original call was correct. Hecht called it a theme he hears "from podcast to podcast": a lack of conviction on the way in can lead someone to sell at the wrong moment even if they made the right purchase
AQR tries to balance conviction with genuine open-mindedness, and founder Cliff Asness has a line for the limit of that. "We want to be open minded, but not so open minded that our brain falls out, right?" The hard part, Hecht said, is telling the difference between someone with rigorous evidence for changing their mind and someone quietly bailing on a losing position and calling it flexibility
AQR is not right nearly as often as most people assume. "But yes, it's not like we're right 80% of the time. It's more like low fifties, mid fifties for our stock." A run of 55% turns into a good year at target returns; a bad year isn't a collapse to 37%, it's closer to 49% — a difference of a few percentage points either way
AQR deliberately stays out of trading for its own sake. Unlike a market maker such as Citadel, which earns a liquidity premium providing it, AQR is a fundamental, systematic manager focused on horizons of months to years. "That's not our business. Our business model is we're fundamental systematic manager. We're thinking about the next month, the next quarter, the next year, not the next tick or the next five minutes."
10. What Clients Buy Now
Asked what institutional and high-net-worth clients are actually asking for, Hecht pointed to a shift toward multi-strategy products, and used the moment to take a swing at recency bias.
The dominant request is a single multi-strategy line item that bundles market-neutral stock selection, market-neutral macro, directional macro (trend following) and corporate arbitrage, because clients have realized picking which single hedge fund strategy will win next year is close to impossible
The follow-up question is always the same: do you want more protective properties, or not — and the answer determines how much trend following gets added on top of the base multi-strategy allocation
Weisburd used the conversation to flag a recurring mistake he tracks: smart investors chasing whatever just worked. "You see these tweets about, well, if you had just invested in the Mag Seven the last seven months, you would be up two x. Why even buy the market?" — before noting the Magnificent Seven has since underperformed the broader index this year
Hecht's answer generalizes the point: no single theme, including momentum, should dominate a portfolio. AQR builds positions from multiple signals so that they're never all flashing the same color at once, which he said also protects against overweighting a single theme like momentum even when that theme is genuinely worth holding over the long run
11. Trend Following, Explained
Weisburd asked Hecht to explain trend following simply, since it's a strategy many sophisticated investors believe in without necessarily being able to describe.
Trend following bets that markets are slow to fully price in news. "Trend following is taking advantage of the fact that market participants tend to under react to news in the short run." If prices moved down on bad news but not far enough, a trend follower shorts on the expectation the market keeps catching up; the same logic applies going long after good news
It has close to zero correlation to equities, which is exactly the point. A large move in the S&P 500, up or down, carries no information about how trend following will do in the same period — which is why it doesn't reliably underperform or outperform alongside stocks
It also has genuine protective properties, which are a different thing from diversification. "Protective properties are, do you tend to have an above average return during challenging market environments like an 'eight, a 'twenty two, the tech bubble burst, etc." In a falling market, a trend follower is already short and extending the position as the decline continues
It is not, however, a hedge in the way an insurance policy or a put option is. "So even though it's like your home insurance, like when your house burns down, you make a lot of money, right?" Buying protection that pays off big in a crisis but costs a small, steady premium the rest of the time is a negatively correlated asset; trend following isn't that, and unlike a put option, which "bleeds and bleeds and bleeds" the rest of the time, trend following can post ordinary positive or negative returns in calm periods rather than a guaranteed steady loss
The strategy's own history is why some investors have soured on it at exactly the wrong moment. It did well through the 2008 financial crisis, drawing in new money afterward; the 2010s were unusually calm, so a strategy built on under-reaction to news had little to react to and delivered mediocre returns for a decade. Many investors gave up on it around 2021, just before the strategy returned roughly 30% to 40% in 2022
12. Secretly All Equity
Weisburd asked how much market exposure the smartest investors he knows carry in their own personal portfolios. Hecht's answer was that most diversified portfolios aren't diversified at all.
Private credit and other alternative assets often behave like disguised equity risk. "Credit is sort of like mini equities, right? They don't do well when there's massive recessions, and they can do well in an expansionary period, for example."
The number that surprises clients is how much of their risk boils down to one factor. "So most people's portfolios, 90% of the risk can be explained by some type of an equity factor, no matter how diversified it looks like when you look at all the line items." That's true no matter how many line items a portfolio statement shows
For clients unwilling to give up equity exposure, Hecht points to two fixes. A market-neutral multi-strategy fund or trend-following allocation supplements the equity-heavy core; for clients who "just love their equities" and won't sell any down, a structure called Portable Alpha lets them add a separate hedge-fund return stream without selling a share of stock, which Hecht returned to in detail later in the conversation
13. Bonds Broke in 2022
Referencing a comment from the former chief executive of a trillion-dollar-plus asset manager, Weisburd raised how correlated stocks and bonds became in 2022, and asked Hecht to unpack why.
Before 2022, most investors believed central bankers had already solved inflation, leaving only recession risk to worry about. "Most people before '22 thought central bankers got rid of inflation risk. The only risk out there are recessions." Under that assumption, a recession simply gets central banks to cut rates, and Treasuries do their job of rising in price as growth-shock protection
2022 broke that assumption directly. "'22 comes around and people are like, Oh, actually if it's an inflation shock, stocks and bonds both get crushed." When inflation news dominates the macro backdrop, a bond's fixed nominal coupon becomes worth less in real terms exactly as stocks are also being hurt by the same inflation, and the two asset classes fall together rather than offsetting each other
Hecht's own preferred long-only core still includes bonds, alongside inflation-sensitive assets. Global equities and global bonds for diversification, plus commodities and TIPS — Treasury bonds indexed to inflation — specifically as protection in an inflation shock that a regular nominal bond can't provide
14. Leverage to a Vol Target
Weisburd asked how AQR thinks about applying leverage once a long-only portfolio is built.
Leverage is sized to hit a target volatility level a client is comfortable with, most often benchmarked against a traditional 60/40 portfolio (60% stocks, 40% bonds). "So a lot of times people say, well, if sixty-forty was acceptable, and sixty-forty has roughly a 10% volatility, we would use enough leverage to get that portfolio to 10% volatility."
The payoff of that leverage is a higher expected return at the same risk level as 60/40, because a more diversified unlevered base portfolio — global equities, bonds and commodities together — has a better risk-adjusted return than stocks and bonds alone, but too little raw volatility to be useful without being levered up to the same target
15. Volatility Isn't One Number
Weisburd pushed on whether volatility is really the right way to define risk, given that some investors care specifically about permanently losing money rather than short-term swings. An investor at Marc Andreessen's family office had told him that a high enough expected return can be worth taking on with real volatility, using the analogy of a three-sided coin where two sides win and only one loses. Hecht's answer turned on the difference between horizons.
Volatility measured over the wrong horizon produces a misleading picture of risk. A ten-year investor should measure ten-year volatility, not annualize daily or monthly numbers the way most people do — which is the specific error Hecht says critics like Howard Marks are really pointing at when they call volatility a bad measure of permanent loss of capital
Long-horizon investors usually aren't picturing the expected return when they think about a ten-year outcome — they're picturing the median, and the two aren't the same number. "The ten year base case fiftieth percentile return will be that expected return that you talked about minus one half volatility squared." This is what's known as volatility drag: a stock that goes down 20% and then up 20% doesn't get back to even, because of the gap between an arithmetic mean and a geometric one, and that gap always cuts against the investor
A higher expected return with higher volatility is not a guarantee of eventually winning, even over decades. "No one ever thought like the Japan equities would be in this malaise for so long." Stocks are expected to beat bonds over a 30- or 50-year horizon, Hecht said, but expected isn't a promise — there are real scenarios in which the higher-volatility asset simply underperforms for the length of an investing career
16. Portable Alpha, Explained
Weisburd asked what questions Hecht is asking himself right now, which turned into a return to Portable Alpha — the structure he'd introduced earlier as a fix for clients who won't give up equity exposure.
Top of mind is where inflation risk could come from next, whether the Middle East, Ukraine or tariff policy, since Hecht says most portfolios are built pro-growth and pro-disinflation by default because they're dominated by equities. Commodities, TIPS and market-neutral strategies like trend following are his answer, because they don't carry stock or bond beta and so aren't exposed to the same macro shock
The complaint Hecht hears most from clients is that a diversifier improved their risk-adjusted return but not their actual total return. "A lot of times people will come to me and say, you improved my risk adjusted return, but you didn't improve my total return." He calls that a capital-efficiency problem rather than a flaw in the underlying hedge funds themselves — the fix is how the strategy is packaged, not the strategy
The mechanics, in dollar terms: put $1 million into a Portable Alpha fund targeting, say, 7% volatility of trend-following exposure, and the fund holds a million dollars of S&P 500 futures exposure alongside a million dollars of trend-following exposure on the same capital. "So for your million dollars, you are getting a million dollars of S and P exposure plus a million dollars of managed futures trend following exposure." The investor earns the S&P 500's return, plus the alpha strategy's return above cash, minus a modest financing cost — not both returns added together in full
The original idea, from twenty years ago, was to stop constraining alpha to the same asset class as the benchmark. "Portable alpha, the original insight twenty years ago was, should you, given that Alpha is so hard to harvest, why would you constrain yourself to looking for Alpha only in U.S. Large Cap?" An investor who needs US large-cap beta doesn't have to source the alpha from US large-cap stock-picking — it can come from anywhere, and get ported back by pairing it with an S&P 500 futures contract
Past Portable Alpha blowups, including before the 2008 financial crisis, were an implementation problem, not a strategy problem. Splitting the alpha manager and the derivatives overlay between two separate firms worked "98% of the time," in Hecht's telling, but introduced cash-management and operational risk that a single manager running both functions avoids. Today's integrated, single-manager turnkey solutions take most of that off the table, leaving what Hecht considers the only risk that should remain. "Which is the risk you signed up for."
Portable Alpha also solves a behavioral problem with diversifiers that have mediocre standalone returns. A trend-following allocation returning 3% on its own invites tough questions from a CIO or investors; the same allocation packaged with S&P 500 beta on top can make an investor a top-decile active manager even in a year when trend following itself was unremarkable
17. Hiding His Own Credentials
Asked what advice he'd give his own younger self right after finishing his PhD, Hecht didn't pick a market call — he picked a regret about talking about himself.
Hecht spent years refusing to use his own credentials, and says it hurt his career. He never put "Ph.D." after his name on business cards, avoided mentioning he was in honors classes in high school, and generally tried to present as an ordinary person rather than someone with a distinguished academic record
A colleague at his old allocator job finally called it out directly. The colleague pointed out that being a University of Chicago finance PhD, one of Eugene Fama's best students that year, a tenure-track Harvard Business School professor and an allocator who'd managed multi-billion-dollar portfolios was exactly the kind of proof a stranger needs. Sitting across the table from someone deciding whether to trust him: "Is this guy the real deal? Is he ChatGPT? Is he Claude?"
Even his wife has called out the habit. When someone at Harvard once asked what he did for a living, Hecht said "I'm a teacher" — technically true, but a deliberate understatement of a tenure-track business school professorship. "What are you doing? What's the big deal? What are you ashamed of?"
His advice now is blunt: use real credentials, because plenty of people you're competing against don't have any. "If you really have bonafide credentials that signal real talent, and you know you're competing against people who have resumes that are inflated, that are sitting in meetings claiming to be knowledgeable about stuff that they're not that knowledgeable about, use these hard credentials, concrete credentials that are bonafide verifiable to signal that you're the real deal."
Bonus Insights
An investor at Marc Andreessen's family office gave Weisburd a coin analogy for why a real edge is worth holding through volatility. Picture a coin with three sides, where two are a win and only one is a loss: "If you flip enough time, you're going to be doing pretty well." Hecht's answer was that people conflate that kind of expected return with the median outcome, which volatility drag pulls below it — the point behind the fiftieth-percentile math in the volatility section above
Hecht's framing for good risk management is imagination, not prediction. On thinking through scenarios like a pandemic before they happen: "You're not saying it's going to happen. You're not saying it's probable. You're just saying it could happen." He added that treating a scenario as merely possible rather than likely isn't false modesty — "And I would say it's not even noble, it's Bayesian, meaning nobody could definitively know that COVID spread."
Hecht's throughline across an hour on tracking error, factor investing, AI and portfolio construction is the same idea applied everywhere: a small, provable edge only compounds into something real once it's spread across enough uncorrelated bets, whether that's a thousand stock positions, a menu of hedge fund strategies, or simply being honest about the credentials that got you in the room.
Products, Companies & Tools Mentioned
AQR Capital Management (Hecht's own $242 billion firm, founded by Cliff Asness, whose open-mindedness line Hecht cites as a guiding principle)
ChatGPT and Claude (The two large language models Hecht says AQR does not use the prompt-based way — asking one directly whether a stock is cheap — because that approach is too easy for any competitor to copy)
Citadel (Cited twice: as the market maker earning a liquidity premium that AQR deliberately doesn't chase, and via Ken Griffin's claim that recruiting is his firm's number one asset)
Renaissance Technologies (Weisburd's example of a black box that's really built from the hundreds of PhDs the firm recruited)
S&P 500 (The benchmark behind Hecht's tracking-error example, and the futures contract that supplies the beta in a Portable Alpha structure)
Books & Resources Mentioned
Common Risk Factors in the Returns on Stocks and Bonds – Eugene Fama and Kenneth French (The 1993 paper that made book-to-price famous as a value signal; Hecht says its real insight was scaling fundamentals by price generally, not the specific ratio, and that it never controlled for industry)
If this was worth your time, send it to someone closer to the industry than you are.
Get the latest market chatter as it happens: