Momentum and the Five-Factor Model

So far you’ve met the three classic Fama–French factors — market, size (SMB), and value (HML) — and learned that most of what looks like manager “skill” is really just rented exposure to those factors. This lesson adds the factor that wouldn’t stay in the box. It’s the strongest standalone anomaly anyone has ever documented, it embarrasses the three-factor model, it quietly explains why “hot” funds stay hot — and roughly once a decade it detonates and gives back years of gains in a couple of months.

Then we’ll meet the modern descendants of Fama–French: the five-factor model, which adds profitability and investment factors (and, awkwardly, makes value redundant), and the q-factor model, a rival four-factor framework built from corporate-finance theory rather than empirical sorting. By the end you’ll see the factor zoo for what it is — a competition to span the return space with the fewest, cleanest building blocks, not a race to pile on more factors.

Momentum: winners keep winning

Analogy. Markets are supposed to be efficient, so a stock’s past return shouldn’t predict its future. Momentum is the rude exception — like discovering that the horse that won last month’s races is genuinely more likely to win this month’s, and the perennial loser keeps losing. It shouldn’t work in a tidy theory. It works anyway, across countries, asset classes, and centuries of data.

Definition. Momentum is the tendency for assets that have outperformed recently to keep outperforming, and laggards to keep lagging, over horizons of a few months to a year. The canonical implementation comes from Jegadeesh and Titman (1993): rank every stock on its return over the past 12 months while skipping the most recent month — the “12–1” lookback — then go long the winners and short the losers, and hold the portfolio for roughly 1 to 6 months before re-ranking.

Why skip the most recent month? Because at the one-month horizon stocks tend to reverse (short-term mean reversion from bid–ask bounce and liquidity effects). If you included last month, the most recent losers would partly bounce back and contaminate your winner/loser sort. Skipping it isolates the genuine intermediate-horizon momentum from short-term noise.

The factor. Packaged as a long–short portfolio, momentum becomes a tradable factor with two interchangeable names you’ll see in papers: UMD (Up Minus Down) and WML (Winners Minus Losers). Same thing — the return of the past-winners portfolio minus the return of the past-losers portfolio.

What makes momentum special is its sheer strength. Its long-run premium runs around 7–9% per year, among the largest of any documented factor, and crucially it carries a big positive alpha against the Fama–French three-factor model — market, size, and value do not explain it. That’s the whole reason it forced its way into the next model.

Carhart’s four factors

If a factor produces a big alpha against your model, the model is incomplete. So Carhart (1997) simply bolted momentum onto Fama–French and created the four-factor model, which became the standard yardstick for evaluating mutual funds for two decades:

$R_i - R_f = \alpha_i + \beta_i\,\text{MKT} + s_i\,\text{SMB} + h_i\,\text{HML} + w_i\,\text{UMD} + \varepsilon_i.$

Here $w_i$ is the fund’s momentum loading — how much it tilts toward recent winners. Carhart’s famous finding wasn’t the equation; it was what the equation revealed. Mutual funds that had been hot tended to stay hot for a while — apparent “persistence” of skill. When Carhart added UMD, most of that persistence vanished into the momentum loading. Hot funds weren’t run by geniuses; they were passively riding the momentum factor, often by accident, because their recent winners had simply drifted up the rankings. This is the canonical “alpha is just hidden beta” result: a return that looked like skill was really uncompensated exposure to a known factor you could rent for free.

Worked example A — a fund’s “skill” evaporates. Suppose a fund posts excess returns that, run against the three-factor model, leave a fat $\alpha = +4\%$ per year. Looks like a star manager. Now run it against Carhart’s four-factor model and you discover the fund has a momentum loading of $w = 0.5$, and over the period UMD earned $8\%$ per year. The momentum contribution to its return is therefore $0.5 \times 8\% = 4\%$. Subtract that newly-explained piece and the four-factor alpha is $4\% - 4\% = 0\%$. The “skill” was momentum all along.

The dark side: momentum crashes

Momentum’s average looks like a gift. Its distribution looks like a trap. Unlike the market factor, momentum returns are negatively skewed with fat tails: most months it earns a steady positive premium, and then — rarely — it suffers a sudden, brutal loss that swallows years of accumulated gains.

Where the crash comes from. A momentum portfolio is short the recent losers. During a market panic, those losers get hammered (good for the short leg), so momentum often looks great mid-crisis. The danger is the rebound. When a beaten-down market sharply reverses, the most violently-rebounding stocks are exactly the trampled losers on your short leg — they can double or triple off the bottom. Your shorts explode upward, your former-winner longs lag, and momentum takes a catastrophic loss precisely when everything else is recovering.

The historical scars. The textbook example is mid-2009: after the March 2009 bottom, the most beaten-down “loser” stocks (banks, financials, junk) rebounded 100% or more in a few months. Momentum’s short leg detonated and the UMD factor lost tens of percent in a matter of months. The same pattern struck in 1932, during the violent bounce off the Great Depression lows. Both times the underlying bet was sound on average — and both times it lost a decade’s premium in a quarter.

And it’s expensive to run. Momentum’s signal decays fast, so the portfolio must be rebalanced monthly, churning a large fraction of holdings every period. That high turnover means high transaction costs — bid–ask spreads, market impact, financing the short leg — which eat heavily into the gross premium, especially at large scale. The 8% on paper is meaningfully smaller after you actually trade it.

Fama–French 5: profitability and investment

Fama and French watched momentum (and a pile of other anomalies) beat their three-factor model and responded — not by adopting momentum, but by adding two fundamentals-based factors. The five-factor model (Fama–French, 2015) keeps market, size, and value, and adds:

• RMW — Robust Minus Weak (profitability). Long firms with robust (high) operating profitability, short firms with weak profitability. Highly profitable companies have historically outperformed by roughly 3% per year. Intuition: profitability is a proxy for the expected cash flows that should command a return.
• CMA — Conservative Minus Aggressive (investment). Long firms that invest conservatively (low asset growth), short firms that invest aggressively (high asset growth). The conservative-investment side has historically beaten the aggressive side by about 3% per year. Intuition: firms that plow capital into rapid expansion tend to over-invest and disappoint.

Both new factors are built the same way as SMB and HML: from 2×3 sorts, double-sorting stocks on size and then on profitability (for RMW) and on size and then on investment (for CMA), then differencing the corner portfolios.

The awkward twist: value becomes redundant. Once RMW and CMA are in the model, the value factor HML loses its explanatory power — its average return is largely spanned by a combination of the profitability and investment factors (cheap stocks tend to be profitable and conservative). In the five-factor model, HML is essentially redundant: you can drop it and barely change the model’s fit. And notice what’s conspicuously absent: the five-factor model omits momentum entirely, despite momentum’s large unexplained alpha. Fama and French preferred two new fundamentals-based factors over the high-turnover, crash-prone momentum factor — a deliberate, much-debated choice.

The q-factor challenger

Fama–French build factors by finding patterns in the data and sorting on them. A rival camp argues you should derive the factors from economic theory first. The q-factor model (Hou, Xue, and Zhang, 2015 — “Digesting Anomalies”) does exactly that, building four factors straight out of the q-theory of investment from corporate finance:

• Market (the same MKT).
• Size (ME) — a market-equity factor, like SMB.
• Investment (I/A) — investment-to-assets, the q-theory analogue of CMA.
• Profitability (ROE) — return on equity, the q-theory analogue of RMW.

The motivation is theoretical: q-theory predicts that firms invest more when their cost of capital (expected return) is low, and that more profitable firms with the same investment must have higher expected returns. So investment and profitability should price the cross-section — and empirically they do. The q-factor model explains the cross-section of returns about as well as, or better than, the three-factor and Carhart models, and Hou–Xue–Zhang famously find that roughly half of about 80 documented anomalies become statistically insignificant once you control for their four factors. Many “anomalies” were just repackaged investment and profitability effects.

So the modern landscape has two competing four-factor successors to the three-factor model: the five-factor model essentially minus its redundant HML (leaving market, size, profitability, investment), versus the theory-first q-factors (market, size, investment, profitability). They’re strikingly similar — both lean on profitability and investment — but they disagree on construction, on whether value survives, and on whether momentum belongs at all.

Putting it together

Momentum is the factor that broke the three-factor model: a ~7–9% long-run premium and a fat positive alpha that market, size, and value can’t touch, so Carhart (1997) added UMD to make the four-factor model — and in doing so showed that fund “persistence” is mostly hidden momentum exposure, not skill. But momentum is negatively skewed: rare crashes (mid-2009, 1932) erase years of gains when beaten-down losers on the short leg rebound, and its monthly turnover bleeds the premium through transaction costs. Fama–French (2015) went a different way, adding RMW (profitability) and CMA (investment) — which together make HML redundant and which omit momentum on purpose. The q-factor model reaches almost the same place from corporate-finance theory, with market, size, investment, and profitability, killing off half the anomaly zoo. The contest between these models isn’t about who has the most factors — it’s about who spans the return space most cleanly.

Next up, we’ll turn from which factors exist to how you actually invest in them — building factor portfolios, the costs and capacity limits that shrink paper premia in practice, and why factor investing is a discipline of patience, not a free lunch.