Which Inflation Are You Hedging?

In August 2026 the United States ran two different inflations at the same time. Headline CPI rose 3.4% over the year while core CPI, which strips out food and energy, rose 2.4%. The gap was gasoline, up 27.4%, inside an energy index up 16.3%. BLS, CPI news release. Long-horizon expectations disagree about what that means. The Philadelphia Fed’s Q3 2026 Survey of Professional Forecasters puts average CPI inflation over 2026 to 2035 at 2.30%, and the ten-year Treasury breakeven closed at 2.38pp on September 15, 2026. Households are less relaxed: the University of Michigan’s preliminary September reading has consumers expecting 3.4% a year over the next five to ten years.

That spread is a useful place to start, because it exposes a question most inflation-hedging advice skips. Which inflation are you hedging? The energy spike that shows up in a headline print, the core services drift that central banks watch, or the tail scenario where long-run expectations come unanchored? Those are different risks, and the assets sold as protection against them behave differently in each.

AQR’s Alternative Thinking 2026 Issue 3, “Inflation Redux? Real Solutions for Real Returns” takes that question seriously. It is the best-constructed piece of publicly available work on inflation sensitivity, and its framework is worth adopting. Its optimized portfolio and its active-strategy backtests deserve a good deal more skepticism than its measurement work, and the paper contains one result about TIPS that it passes over in a sentence and that should probably reorganize how a DIY investor thinks about the whole problem.

What the paper measures

Most inflation-hedging content correlates asset returns with the level of CPI. That is the wrong test. Markets already know what inflation has been; prices should respond to inflation news. AQR builds its news measure from two components: the change in the year-on-year CPI rate versus the previous year’s rate, and realized CPI versus the one-year forecast made a year earlier. The first treats expectations as a random walk, the second uses surveys, and combining them reduces the noise in either. It builds a matching measure for GDP growth and uses it as a control, so that a return earned during a growth shock is not credited to inflation. The sample runs from 1972 to June 2026, on rolling twelve-month windows sampled quarterly.

This is a real improvement over the usual analysis, and the limits are worth stating. These are contemporaneous correlations, not causal estimates. Overlapping twelve-month windows mean the effective number of independent observations is far smaller than the number of data points. U.S. inflation is used to characterize assets around the world. And one GDP control cannot fully separate inflation from growth, monetary policy, and risk aversion. AQR says so itself: discussing the non-linear response of equities, the paper notes that adverse growth shocks coincided with many of the disinflationary shocks in the sample, and that its growth proxy may be imperfect. A practitioner paper that flags the weakness in its own headline result is doing something right.

What the paper’s own data says about TIPS

Exhibit 6 maps each asset’s correlation with changes in headline CPI and with its core, food and energy components. The paper prints it as a bar chart with no accompanying table, so the figures below were measured off the chart against its axis. Two rows sit next to each other and tell opposite stories.

U.S. TIPS score +0.05 against headline inflation news. That is a rounding error. In the same exhibit, real estate scores +0.02 and non-U.S. developed equities +0.05. The security whose entire purpose is CPI linkage sits in the same neighborhood as assets with no CPI linkage at all. Breakevens, meanwhile, score +0.61.

A breakeven position is a long TIPS paired with a short position in a duration-matched nominal Treasury. It is the same CPI linkage with the interest-rate exposure removed. So the 0.56 gap between those two rows is a measurement of how much of a TIPS fund’s twelve-month behavior is interest rates rather than inflation. Over a one-year horizon, it is nearly all of it.

Exhibit 9 shows what that meant in practice. Across the 2022 inflation shock, which AQR dates from January 1 to September 30, 2022, Global TIPS fell about 12%, a slightly worse result than the roughly 11% loss on Global Treasuries over the same window. The instrument indexed to consumer prices, during the sharpest consumer-price shock in forty years, did not beat the instrument that is not.

Nothing about this is a defect in TIPS. The CPI adjustment arrived on schedule and did exactly what the contract promised. It was swamped by something larger. When real yields rise, the market value of an existing TIPS falls like any other bond, and over twelve months the price move on a seven-to-ten-year bond dwarfs a year of CPI accrual. Three things get called “TIPS” in ordinary conversation and they are not the same investment:

• A TIPS held to maturity. The real return is fixed at purchase. Real yields can do whatever they like in between; you collect the starting real yield above realized CPI. This is the only version where the CPI linkage is the dominant feature.
• A constant-duration TIPS fund. It never matures. Its price is a real-yield bet that pays a CPI-linked coupon on the side. Over a year it behaves mostly like a bond fund.
• A breakeven position. Long TIPS, short the matched nominal. This isolates the inflation exposure and removes the duration. It is also a leveraged derivatives trade that almost no individual investor should run directly, which is why it appears here as a measuring device rather than a recommendation.

Try it: the inflation shock decomposer

The calculator prices a par bond at a starting yield, holds it twelve months, and marks the remaining years at an ending yield, then splits the result into the CPI accrual and everything else. The 2022 preset uses the actual ten-year real yield on December 31, 2021 and December 30, 2022 from the Treasury’s real yield curve, with the 6.45% CPI-U change over those twelve months.

On the 2022 preset, a ten-year TIPS gains about 6.5 points from inflation accrual and loses roughly 22 points to the real-yield move. The hedge worked and lost anyway. Switch the real-yield change to zero and the same bond returns roughly its real yield plus realized CPI, the result people expect from TIPS in the first place. Both outcomes come from the same security. The difference is entirely the path of real yields and the length of the holding period.

For an investor with a dated real spending need, this points somewhere specific. See how to build a TIPS ladder for the mechanics, and whether locking in today’s real yields counts as market timing for why the current curve matters.

Commodities hedge which inflation?

AQR’s commodity result is strong and, on its face, broader than the usual claim. Commodities score +0.68 against headline inflation news and +0.43 against core, second only to breakevens. The paper says directly that breakevens delivered the most potent protection against core inflation shocks but that a diversified basket of commodities was also effective over the sample.

A 2026 Review of Financial Studies paper reaches close to the opposite conclusion. Xiang Fang, Yang Liu and Nikolai Roussanov use a VAR to extract unexpected innovations to headline, core, food and energy inflation separately, then estimate each asset’s beta to each. In their words, “currencies and commodity futures’ hedging properties against headline inflation mainly come from the energy component, while their core betas are negative and insignificant,” and “a long position in none of these 8 asset classes can hedge against the core inflation.”¹

Two credible teams, opposite answers on the same question. The differences that could produce this are worth naming rather than waving at. Fang, Liu and Roussanov run quarterly data from 1963Q2 to 2019Q4, so their sample ends before the 2021 to 2022 inflation episode entirely. AQR’s runs to June 2026 and includes it, on overlapping twelve-month windows. They also extract inflation surprises differently, one from a VAR and one from year-on-year changes plus survey forecasts. A reader is not in a position to declare a winner, and neither am I.

Two other pieces of independent work lean the same way as Fang, Liu and Roussanov, which is what moves my own view. Chunbo Liu, Xuan Zhang and Zhiping Zhou apply a Markov-switching model to 1983 through 2021 and find that aggregate commodity futures fail to hedge inflation over the full sample, with only industrial and precious metals showing capacity that itself varies by regime.² And Ke Tang and Wei Xiong supply a mechanism: after 2004, non-energy commodity futures became progressively more correlated with oil, and significantly more so for commodities inside the major indices than outside them.³ If an index-tracking commodity position keeps drifting toward a levered position on crude, its protection keeps concentrating in the energy component of inflation.

2026 is itself a reasonable live test. The Bloomberg Commodity Index total return was up 32.06% for the year through August 31, in a year when energy CPI ran at 16.3% and core cooled to 2.4%. That is exactly the pattern you would expect if commodity protection is mostly energy protection, though a single year settles nothing.

What survives all of this is narrower than “commodities hedge inflation” and still useful. A diversified commodity position has the strongest direct link of any liquid asset to inflation driven by oil, food, metals and supply disruption. Its protection against a wage-and-services inflation is contested, sample-dependent, and should not be assumed. It is also a directional bet in a way TIPS are not: in Exhibit 9, commodities fell about 41% through the Global Financial Crisis, the largest loss of any asset in the exhibit.

What trend following provides

Trend following is the part of the paper I find most interesting, partly because the label “inflation hedge” describes it badly. A trend strategy does not profit because prices rose. It profits when moves persist long enough to be captured, in either direction, because it can be short bonds in 2022 and long bonds in 2008.

Exhibit 9 is the cleanest evidence in the paper. Over the Global Financial Crisis the SG Trend Index gained roughly 22%, while global developed equities lost about 54%. Over the 2022 inflation shock it gained roughly 35%, while commodities gained about 13% and global Treasuries lost about 11%. Those two episodes had opposite inflation signs and trend was strongly positive in both. Nothing else in the exhibit managed that.

The independent literature supports the mechanism and complicates the marketing. Moskowitz, Ooi and Pedersen documented time-series momentum across 58 futures and forward contracts and found the strategy’s loading on squared market returns significantly positive, which is the statistical signature of doing well in extreme moves in either direction.⁴ They also found no significant relation to the VIX, the TED spread, or investor sentiment, so the effect is about large directional moves rather than stress as such.

Three caveats belong with that. Huang, Li, Wang and Zhou re-examined the same dataset and concluded the statistical evidence for return predictability is far weaker than originally reported, with bootstrapped 5% critical values of 12.53 and 4.83 against t-statistics that conventional tests treat as decisive.⁵ Their target is the predictability claim rather than whether trend strategies made money, and AQR published a reply, though four of its five authors work at AQR and no independent peer-reviewed rebuttal has appeared. Second, Hutchinson and O’Brien, academics with no managed-futures business, found that trend returns in the extended period after a financial crisis run at less than half their no-crisis level, staying weak for roughly four years.⁶ Third, trend is path-dependent in a way a CPI-linked coupon is not. A sharp inflation shock that reverses quickly is a poor environment for it, and AQR says as much.

For mechanics, fund selection, tax treatment and how large a sleeve makes sense, see our full guide to managed futures.

Gold, briefly

AQR’s data is not unkind to gold, which scores +0.49 against headline inflation news and +0.35 against core. Yet gold fell around 10% during the 2022 window, when inflation was at its highest in four decades. Both facts can hold at once because gold’s volatility is large relative to that correlation, so the average relationship tells you little about any particular episode. Erb and Harvey’s conclusion still stands: gold may hold purchasing power over horizons measured in centuries, and is unreliable as an inflation hedge over the horizons investors actually have.⁷ We covered this at length in the best inflation hedges are boring.

Inside the optimized portfolio

Exhibit 11 is the exhibit most likely to be screenshotted. AQR starts from a global 60/40, allows inflation-linked bonds, commodities and trend following, permits 15% leverage, and constrains the result to no more expected volatility than the starting portfolio, no more than 5% trailing twelve-month relative underperformance, and no worse a loss than 60/40 in either the Global Financial Crisis or the 2022 shock. The combined answer:

The obvious criticism is that the optimizer was handed generous return assumptions for the assets it then picked. Check the appendix and that criticism does not survive. AQR’s assumed Sharpe ratios as of June 30, 2026 are 0.25 for global equities, 0.27 for global IG bonds, 0.24 for TIPS, 0.27 for commodities and 0.25 for trend following. They are the same number to within rounding. No asset is being bribed into the portfolio with a better risk-adjusted return.

What does the work is the correlation matrix, and AQR estimates two of them: one from growth-dominant periods and one from inflation-dominant periods. In the inflation-dominant matrix, commodities correlate −0.53 with bonds and −0.19 with equities, and trend correlates −0.26 with equities. In the growth-dominant matrix, commodities correlate 0.00 with bonds and +0.52 with equities. Those are very different assets depending on which regime you are in, and the allocation follows almost mechanically from assuming both regimes recur with the estimated correlations.

That is the assumption to interrogate. Those correlations come from overlapping twelve-month windows carved out of a 54-year sample and sorted into two regimes, which is a small number of genuinely independent regime episodes. DeMiguel, Garlappi and Uppal tested 14 optimization models across 7 datasets and found none of them consistently beat an equal-weight portfolio out of sample, and estimated that roughly 3,000 months of data would be needed before sample-based mean-variance reliably beats 1/N with 25 assets.⁸ That finding does not invalidate AQR’s exercise, and AQR’s own text describes Exhibit 11 as one of several ways to approach the analysis. It does mean 56.1/22.8/11.0/9.1/16.2 is a sensitivity analysis showing how alternative diversifiers change a portfolio’s macro exposures, rather than a target allocation to copy.

Two more things a reader should notice. The portfolio gross exposure is 115% funded by a 15% cash short, so it is a leveraged portfolio; AQR notes this can be achieved through capital-efficient funds rather than borrowing, which is true and still a change in the kind of product being held. And the commodity assumption of 6.9% total return against 3.8% cash implies a 3.1% excess return, while the paper’s own Exhibit 14 reports that the Bloomberg Commodity Index delivered 1.8% excess since May 1990. AQR has a defensible answer, since the 139-year history in “Commodities for the Long Run” is more encouraging than the last 36 years. That paper’s working-paper version also states that “the evidence does not, however, support equity-like returns for commodities.”⁹

How to read the backtests

Exhibit 14 is the most striking chart in the paper and the one I trust least. A hypothetical risk-balanced commodities strategy with an active overlay returns 11.9% geometric from May 1990 to June 2026 against 3.6% for the Bloomberg Commodity Index, with a Sharpe ratio of 0.57 against 0.12. Three things qualify it.

First, it is not a commodity allocation. It combines risk balancing, volatility targeting and active signals spanning carry, momentum, seasonality, macro drivers, supply and demand, and value. AQR’s own attribution says that of the 8.2 percentage points of excess return, 6.4 came from active tilts and 1.7 from the strategic allocation. The exhibit is evidence about AQR’s quantitative process, not about owning commodities.

Second, it is net of estimated transaction costs and gross of fees, while the trend backtest in Exhibit 16 is net of a 1.25% management fee and a 20% performance fee. Applying the trend exhibit’s own fee schedule to the commodity exhibit, and charging the performance fee on excess return after the management fee, the reported 10.0% arithmetic excess return becomes about 7.0%, the Sharpe ratio falls from 0.57 to roughly 0.40, and the headline total return drops from 11.9% to roughly 8.9%. That is a Summitward estimate applied to AQR’s published figures under those stated assumptions, not a number AQR reports, and it understates the real drag, because performance fees are charged in winning years with no rebate in losing ones. Even so, roughly 8.9% would still beat the index by a wide margin. Three percentage points of annual return is the size of what the exhibit’s fee convention leaves out.

Third, it is hypothetical, and AQR is unusually direct about what that means. The disclosures state that hypothetical results are prepared with the benefit of hindsight, involve no financial risk, and cannot account for the difficulty of adhering to a trading program through losses. The same page states that the presentation “is not research and should not be treated as research.” Those are the paper’s words, and they should be taken at face value.

The strongest independent check on all of this comes from Bhardwaj, Gorton and Rouwenhorst, who studied commodity trading advisors from 1994 to 2012 and found excess returns net of fees that were statistically indistinguishable from zero, while gross excess returns ran at 6.1%.¹⁰ Managers captured essentially the entire gross performance. Whatever a backtest shows before fees, that is the gap an investor has to clear.

Three things deserve credit on the other side of the ledger. AQR inflates its estimated transaction costs for older periods, by a factor of two for 1993 to 2002 and six before 1993, which is the opposite of flattering. The trend backtest is presented net of a full hedge-fund fee load. And the predecessor analysis, “When Stock-Bond Diversification Fails,” was published in 2021, before the 2022 bear market, already arguing that stocks and nominal bonds could fall together on upside inflation surprises and already naming inflation-linked bonds, commodities and trend as the diversifiers. That is genuine out-of-sample evidence for the framework, and it is the part of the case most critiques of AQR skip.

Three jobs, three different instruments

The question worth asking is what job you need done. The three candidates do not compete for the same one, so a ranking of them answers nothing on its own.

My own reading, for what it is worth. Job one is the one most individuals really face, and it is the cheapest to solve well. A ten-year TIPS yielded 2.62% real and a twenty-year 2.90% on September 15, 2026, against a ten-year real yield of −1.04% at the end of 2021. A maturity-matched ladder against dated spending is a far better proposition at those levels than it was five years ago, and it requires no view on inflation at all.

Job two is where a broad TIPS fund falls short and where the case for commodities is real but narrower than the pitch. If you hold commodities for this reason, hold them understanding that you are mostly buying protection against energy and supply shocks and that the position will hurt in a disinflationary recession.

Job three is the most interesting and the hardest to execute. Trend following is the only thing in AQR’s own tail-risk exhibit that was strongly positive in both a deflationary crisis and an inflationary one. It also costs more than everything else on this list, it is supported by a literature where the most enthusiastic papers are written by people selling the product, and the independent evidence on delivered net-of-fee returns is sobering. It deserves a place in the conversation and a smaller allocation than its backtests suggest.

Bottom Line

AQR has written the most careful piece of inflation-sensitivity analysis a retail investor can read without a subscription, and the framework holds up better than the portfolio built on top of it. Take the measurement approach, take the insight that different assets protect against different components of inflation, and take the finding that a TIPS fund and a TIPS held to maturity are different investments. Treat the optimized weights as a sensitivity analysis, discount the active commodity backtest for its missing fees, and remember that the firm publishing the paper sells both of the strategies it recommends.

Key Takeaways

• A TIPS fund’s twelve-month return is mostly an interest-rate outcome. In AQR’s 1972 to 2026 sample, U.S. TIPS correlate +0.05 with headline inflation news while breakevens correlate +0.61. The gap is nominal duration.
• The 2022 shock is the clean illustration of that gap. Over AQR’s January to September 2022 window, Global TIPS fell about 12% and Global Treasuries about 11%, in the sharpest CPI shock in forty years.
• Commodity protection concentrates in energy-driven inflation. AQR finds meaningful core protection over its sample; a 2026 Review of Financial Studies study whose data stops in 2019 finds core betas that are negative and insignificant. Independent regime and financialization evidence leans toward the narrower claim.
• Trend following was positive in both of AQR’s tail episodes. Roughly +22% through the Global Financial Crisis and +35% through the 2022 inflation shock, per Exhibit 9. Independent work finds it weak for several years after a crisis and finds CTA returns net of fees historically close to zero.
• Exhibit 11 is a bet on correlations, not on return assumptions. Assumed Sharpe ratios run 0.24 to 0.27 across all five assets. The allocation comes from regime-conditional correlations estimated on overlapping windows, which is where the uncertainty sits.
• Fee treatment differs between the paper’s two backtests. Applying Exhibit 16’s own 1.25% and 20% schedule to Exhibit 14’s commodity results moves the reported 11.9% return to roughly 8.9% and the Sharpe ratio from 0.57 to about 0.40 on our estimate, which still clears the index comfortably.
• Today’s real yields change the calculus for the most common job. At 2.62% on the ten-year and 2.90% on the twenty-year as of September 15, 2026, a maturity-matched TIPS ladder funds a dated real expense on terms that were not available in 2021.

Frequently Asked Questions

Does this mean I should not own a TIPS fund?

It means you should know what you own. A TIPS fund is a sensible way to hold short-to-intermediate real duration and it will track CPI well over long holding periods. It is a poor instrument if your goal is to offset a specific inflation surprise over the next year, and it is not a substitute for maturity-matched bonds when you have a dated real liability. Short-duration TIPS funds sit much closer to the CPI linkage than long ones, because there is less duration to overwhelm it.

Why did TIPS fall in 2022 if inflation was high?

Real yields rose sharply while long-run inflation expectations stayed anchored. The ten-year real yield went from −1.04% at the end of 2021 to 1.58% at the end of 2022. A bond with seven to ten years of duration loses far more to a 2.6 percentage point yield move than it gains from a single year of CPI accrual. The inflation adjustment paid out in full; it was simply the smaller of the two effects.

Is a breakeven position something I can buy?

Not practically. It requires shorting nominal Treasuries against a long TIPS position, which means futures or swaps and ongoing margin management. It appears in this analysis as a way to isolate the pure inflation exposure that sits inside a TIPS, which is what makes the comparison with the TIPS fund informative.

Should I add commodities to a 60/40 portfolio?

Only if you have identified which job you want them to do and you can hold them through a disinflationary recession, where the same exposure that helps in an inflation shock works against you. AQR’s Exhibit 9 puts the commodity loss through the Global Financial Crisis at about 41%. Anyone who would sell at that point should not own the position.

Is AQR’s research biased because it sells these strategies?

AQR manages commodity and trend-following strategies, and its own managed-futures fund is a constituent of the SG Trend Index the paper uses to represent trend in Exhibits 9 and 11. That is disclosed in the appendix and it does not make the empirical work wrong. It is a reason to treat the paper as a source of hypotheses and check the important ones against research by people with no product to sell, which is what this review tries to do.

What about I Bonds?

I Bonds sidestep this problem entirely, because they are non-marketable. There is no market price to fall, so the CPI linkage is the only thing that happens. The constraint is the annual purchase limit, which makes them impractical for large allocations. See our guide to inflation hedges for the account-by-account comparison.

Related Guides

• The Best Inflation Hedges Are Boring for which instrument to hold in which account, and the liability-side hedge most people already own.
• How to Build a TIPS Ladder for the maturity-matching mechanics behind job one.
• Is Locking In Today’s TIPS Yields Market Timing? for what today’s real yields cost in terms of guaranteed real income.
• Do You Need Managed Futures? for the full treatment of trend following, fund options and taxes.
• Are Bonds Still Good Diversifiers? for why the stock-bond correlation depends on which shock arrives.
• HEDGEFUNDIE’s Excellent Adventure for a longer look at what it means to build a portfolio on a correlation estimate.
• AQR Funds for Bogleheads for whether the strategies in this paper are worth buying in practice.
• Tail-Risk Hedging for how trend following compares with puts and other crisis hedges.

Sources

1. Xiang Fang, Yang Liu and Nikolai Roussanov, “Getting to the Core: Inflation Risks Within and Across Asset Classes,” The Review of Financial Studies 39(3), 2026, 702–743. doi.org. Working-paper version: NBER 30169.
2. Chunbo Liu, Xuan Zhang and Zhiping Zhou, “Are commodity futures a hedge against inflation? A Markov-switching approach,” International Review of Financial Analysis 86, 2023, 102492. doi.org
3. Ke Tang and Wei Xiong, “Index Investment and the Financialization of Commodities,” Financial Analysts Journal 68(6), 2012, 54–74. doi.org
4. Tobias J. Moskowitz, Yao Hua Ooi and Lasse Heje Pedersen, “Time series momentum,” Journal of Financial Economics 104(2), 2012, 228–250. doi.org
5. Dashan Huang, Jiangyuan Li, Liyao Wang and Guofu Zhou, “Time series momentum: Is it there?,” Journal of Financial Economics 135(3), 2020, 774–794. doi.org. AQR’s reply is Babu, Levine, Ooi, Pedersen and Stamelos, “Trends Everywhere,” Journal of Investment Management 18(1), 2020, 52–68.
6. Mark C. Hutchinson and John O’Brien, “Is This Time Different? Trend-Following and Financial Crises,” The Journal of Alternative Investments 17(2), 2014, 82–102. doi.org
7. Claude B. Erb and Campbell R. Harvey, “The Golden Dilemma,” Financial Analysts Journal 69(4), 2013, 10–42. doi.org
8. Victor DeMiguel, Lorenzo Garlappi and Raman Uppal, “Optimal Versus Naive Diversification: How Inefficient Is the 1/N Portfolio Strategy?,” The Review of Financial Studies 22(5), 2009, 1915–1953. doi.org
9. Ari Levine, Yao Hua Ooi, Matthew Richardson and Caroline Sasseville, “Commodities for the Long Run,” Financial Analysts Journal 74(2), 2018, 55–68. doi.org. The quoted line appears in the working-paper version, NBER 22793.
10. Geetesh Bhardwaj, Gary B. Gorton and K. Geert Rouwenhorst, “Fooling Some of the People All of the Time: The Inefficient Performance and Persistence of Commodity Trading Advisors,” The Review of Financial Studies 27(11), 2014, 3099–3132. doi.org
11. Alfie Brixton, Jordan Brooks, Peter Hecht, Antti Ilmanen, Thomas Maloney and Nicholas McQuinn, “A Changing Stock–Bond Correlation: Drivers and Implications,” The Journal of Portfolio Management 49(4), 2023, 64–80. doi.org
12. Market and inflation data: U.S. Treasury daily real and nominal par yield curves; BLS Consumer Price Index; Federal Reserve Bank of Philadelphia Survey of Professional Forecasters, Q3 2026; University of Michigan Surveys of Consumers.

Author disclosure

Summitward has no business relationship with AQR Capital Management and receives no compensation from any firm mentioned here. AQR manages commodity and trend-following strategies of the kind its paper recommends, and its managed-futures fund is a constituent of the SG Trend Index used in two of the exhibits discussed above. Figures attributed to Exhibits 6 and 9 were measured from the published charts because the paper does not print them as tables; figures from Exhibits 11, 14 and 16 are printed values. The fee-adjusted estimates in the backtest section are ours, computed from AQR’s published figures under the assumptions stated there. Nothing here is investment advice.