Value and Momentum Everywhere

ELI5 / TLDR

The two most famous return anomalies — buying cheap things (value) and buying things that have been going up (momentum) — were studied for thirty years almost entirely in US stocks, and almost entirely in isolation. Asness, Moskowitz, and Pedersen run them across eight different markets — stocks in four regions, country index futures, currencies, government bonds, and commodities — and find both work everywhere. More importantly, value strategies are correlated with each other across totally unrelated assets, momentum strategies are correlated with each other across totally unrelated assets, and value and momentum are strongly negatively correlated everywhere. That common structure shouldn’t exist if the standard explanations were right, and a 50/50 combination of the two — which is largely immune to funding-liquidity shocks — produces a Sharpe ratio around 1.45.

The Full Story

The setup

The literature treated value and momentum like two unrelated curiosities of the US stock market. Fama and French built a value factor in 1992. Jegadeesh and Titman documented momentum in 1993. Outside US equities the evidence existed but was scattered. Nobody had built the same two strategies, with the same two simple definitions, side by side across eight asset classes at once. That is what this paper does, and the act of doing it changes what the data say.

The eight test grounds are individual stocks in the US, UK, continental Europe, and Japan; country equity index futures across 18 developed markets; ten currencies; ten-year government bonds in ten countries; and 27 commodity futures. The samples run from 1972–2011 in the longest cases. They use the largest, most liquid 17% of stocks in each equity market — a more conservative set than Russell 1000 — so the returns understate what’s available in smaller, less liquid names.

The signals are deliberately crude. Value for stocks is book-to-market, with no clever cleanups. For non-equity assets they use the negative of the past five-year return — long-term reversal as a value proxy — except for bonds, where they also use yield change and term-spread variants. Momentum is the past 12-month return skipping the most recent month, the standard MOM2-12 from Jegadeesh and Titman. The point of the simple measures is to avoid data-snooping. Better measures exist; using them would only strengthen the results.

What the strategies do

Value works in every asset class. Momentum works in every asset class except Japanese stocks (where it’s positive but insignificant — Asness has a separate paper on this). Sharpe ratios for the long-short signal-weighted factors run roughly 0.4 to 1.0 in individual markets, which is decent but not magical.

The interesting numbers come from combinations. Inside each market, value and momentum are negatively correlated — averaging around –0.60 in equities and –0.49 across the non-equity classes. Combining two positive-Sharpe strategies that move opposite to each other is the closest thing finance has to free lunch. Globally, the 50/50 value/momentum combo earns a Sharpe of 1.45 — roughly three times the US equity premium that asset pricing models already strain to explain.

The striking thing — common factor structure

Treat each value strategy in each asset class as its own time series and stack them up. Value strategies are positively correlated with each other globally — US stock value with European stock value with currency value with commodity value. Same for momentum. Same negative correlation between value and momentum, within and across asset classes.

The first principal component of the covariance matrix of value and momentum strategies across all asset classes is essentially “long momentum, short value” in every asset class. It explains 22.7% of the all-asset-class covariance and 53.6% of the within-equity covariance. There is one global factor pulling on momentum strategies in commodities and momentum strategies in Japanese equities at the same time.

This is hard to dismiss. The strategies are dollar-neutral long-short portfolios, so the common factor isn’t just exposure to the underlying asset class. The asset classes have different investors, different microstructure, different information environments. Whatever is moving them together isn’t local.

What doesn’t explain it

Standard macro variables — long-run consumption growth, recessions, GDP growth, the term spread, default spread, the market itself — capture only modest fractions of the variation. The R² values are in the low single digits to mid-teens.

Behavioral theories that work in equities (Daniel-Hirshleifer-Subrahmanyam, Barberis-Shleifer-Vishny, Hong-Stein) predict idiosyncratic mispricings driven by individual-investor cognitive biases. They don’t naturally produce coordinated global structure across currencies and government bonds. Rational investment-based theories (Berk-Green-Naik, Zhang, Liu-Whited-Zhang) predict value premia in equities tied to firm-level investment options. They don’t naturally extend to currencies or commodities, which have neither.

What partially explains it — funding liquidity risk

The story that does fit, partially, is funding liquidity. The TED spread, LIBOR-repo spread, and swap-Tbill spread are signed so a wider spread means worse liquidity. Value strategies load negatively on funding liquidity shocks. Momentum strategies load positively. The pattern shows up across asset classes and gets statistically much stronger when you average across markets — the average individual-strategy t-stat on liquidity beta is unimpressive (–0.95 for value, 1.81 for momentum), but the t-stat on the average return series across all asset classes is –3.25 for value and 4.43 for momentum.

The intuition: momentum portfolios hold whatever has run up most recently, which tends to be the crowded popular trades. Value portfolios hold the contrarian unloved positions. When funding tightens — investors face cash needs, redemptions, risk-management constraints — the crowded trades suffer disproportionately because everyone runs for the same exit. Pedersen calls this dynamic “when everyone runs for the exit.” Funding risk loads positively on momentum because momentum gets compensated for being on the wrong side of liquidations; it loads negatively on value because value gets a tailwind from being on the right side.

This makes momentum’s positive premium partly intelligible as compensation for funding risk. It makes value’s positive premium more puzzling, since value loads the wrong way on a priced risk and still earns positive returns. And it cannot explain the combination premium at all — the 50/50 portfolio’s funding-liquidity exposure is roughly zero by construction, and yet that’s the portfolio with the Sharpe of 1.45.

The funding-risk pattern is much stronger after August 1998 (LTCM). Pre-1998 funding shocks barely register; post-1998 they are the dominant story. Correlation among value strategies globally rose from 0.16 pre-1998 to 0.64 post-1998. Among momentum strategies, 0.43 to 0.71. The negative correlation between value and momentum became more negative. The natural reading is that arbitrage capital chasing these styles globally has become large enough to move correlation structure, and that this capital is sensitive to funding conditions.

A three-factor global model

The empirical answer they propose is a three-factor model: a global market index, a value-everywhere factor (equal-volatility-weighted average of value across all eight asset classes), and a momentum-everywhere factor. They show this model:

Prices their own 48 test portfolios with a cross-sectional R² of 0.71 and average alpha of 18 bps/month.
Prices the Fama-French 25 size-value and 25 size-momentum US portfolios about as well as the Fama-French US factors do — even though the global factors contain mostly non-US assets.
Prices a panel of 13 hedge fund indices better than US-only models do, suggesting hedge fund returns are partly a global value-and-momentum trade.

GRS tests still reject the model in absolute terms — pricing isn’t perfect. But it’s better than CAPM, Fama-French three-factor, Carhart four-factor, and a six-factor extension with TERM and DEF, both globally and inside US equities.

Implementation footnotes

The paper sticks to gross returns. Frazzini-Israel-Moskowitz separately show that real-world trading costs for these strategies at large institutional scale are an order of magnitude lower than the calibrated estimates that earlier papers used to argue momentum was uneconomic. Value contributes at least as much from the long side as the short, and Israel-Moskowitz show long-only versions still produce abnormal returns, so shorting costs aren’t load-bearing. The eight asset classes were intentionally restricted to the most liquid 17% of equities and futures contracts, so the implementable Sharpe is closer to the reported gross Sharpe than a typical academic anomaly study.

Claude’s Take

This is one of those papers that quietly reframes a thirty-year literature by doing one thing — running the same two simple strategies in eight places at once — and looking at the residual. The residual turns out to be a global factor structure that almost nobody had documented and that almost no existing theory predicts.

The score is 9, not 10, because the paper is honest about what it doesn’t do: the GRS test rejects the three-factor model in every specification, the funding-liquidity story is partial, and the deepest puzzle — why does an equal combination of value and momentum, immune to funding shocks, earn 1.45 Sharpe — is left explicitly unsolved. Asness, Moskowitz, and Pedersen are not trying to close the question. They’re trying to enlarge it past US equities, and they succeed.

The political subtext is also worth noting. AQR runs value and momentum strategies for a living, which the paper discloses cleanly. The findings are flattering to AQR’s business — the message “value and momentum work everywhere” is good marketing. But the construction is conservative on every margin: simplest possible signals, most liquid possible securities, gross returns that understate the strategies’ edge. If anything they’re putting a thumb on the scale against themselves and still finding the effects. That’s the right way to do it.

What ages well: the framework. The factor model has been re-tested by other researchers, mostly held up, and the underlying empirical finding — that value-everywhere and momentum-everywhere are real and persistent — has survived another decade of out-of-sample data. What aged less well: momentum specifically went through a long drought in the 2010s in US equities, and the funding-liquidity story has been complicated by the post-2008 zero-rate environment where TED spreads were compressed for years. The paper’s claim that “after 1998, funding liquidity risk dominates the dynamics of these strategies” was made in 2013; whether 2013–2024 confirms or muddies it depends on which sub-period you slice.

For an investor, the practical implication is unchanged from the day this came out: if you’re going to run a single style, run momentum and value together; the negative correlation is the dominant feature of the joint distribution and it dwarfs the gain from being clever about the individual signals. The combo also stops being an anomaly and starts looking like a global risk factor, which means a portfolio without it is making an active bet that other people’s funding constraints won’t matter to your returns.