Photo by Cesar La Rosa on Unsplash
The short version
- Documented return predictors decay after they are published — roughly 58% lower post-publication than in the original sample, per McLean & Pontiff (2016) — but they rarely disappear entirely.
- The decay is diagnostic: premiums that compensate for genuine bad-times risk should survive publication; the ones that vanish were largely mispricing that arbitrage capital corrected.
- Bottom line: for a long-horizon investor, the practical question is not "is the factor dead?" but "what net premium survives after crowding, fees, and tracking error — and can I hold it through the stretch where it doesn't work?"
There is a familiar arc to every factor premium. A researcher documents an anomaly — small stocks, cheap stocks, high-momentum stocks, low-volatility stocks. A working paper circulates. A few funds package it. And then, in the years after the paper is widely read, the premium seems to shrink. The central question is uncomfortable for anyone building a portfolio around factors: once everyone can read the paper, is there anything left to earn?
The honest answer is more interesting than either "factors are dead" or "factors are free money." The data shows premiums decay after publication — meaningfully — but not uniformly, and not to zero. Understanding which premiums decay, and by how much, tells you something the raw backtest cannot.
Context: what "crowding" and "alpha decay" actually mean
A factor premium is the extra return historically associated with holding stocks that score high on some characteristic. Crowding is what happens when capital flows toward that characteristic: as more money buys the cheap or high-quality or low-volatility names, their prices rise relative to the rest of the market, the valuation spread between the long and short legs compresses, and the forward-looking premium falls. Alpha decay is the observed result — the premium earned after the strategy becomes known is lower than the premium in the original study window.
Two mechanisms sit underneath this. The first is statistical: some published anomalies were never real. With thousands of variables tested against the same few decades of U.S. stock returns, a fraction will clear conventional significance by chance alone. This is the data-mining and multiple-testing problem that Harvey, Liu & Zhu (2016) confronted when they argued the appropriate t-statistic hurdle for a "discovered" factor is closer to 3.0 than the textbook 2.0. The second mechanism is economic: some anomalies were real mispricings, and publication invited arbitrage capital that partially corrected them. These two mechanisms decay for different reasons, and telling them apart matters.
The reference the whole debate rests on
The cleanest evidence comes from McLean & Pontiff (2016), "Does Academic Research Destroy Stock Return Predictability?" They took 97 predictors from published studies and tracked each one's return in three windows: the original in-sample period, the post-sample-but-pre-publication period, and the post-publication period. The decomposition is the useful part.
| Window measured | What it isolates | Reported effect on predictor returns |
|---|---|---|
| Out-of-sample, pre-publication | Statistical bias (overfitting, luck) | ~26% lower than in-sample |
| Post-publication | Statistical bias plus arbitrage response | ~58% lower than in-sample |
| Publication effect (the gap) | Attributable to publication itself | ~32 percentage points of the decline |
| Source: McLean & Pontiff, "Does Academic Research Destroy Stock Return Predictability?", Journal of Finance, 2016. | ||
Read carefully, this is not a death notice. A premium that falls by 58% still retains roughly 42% of its original magnitude. The 26% out-of-sample decline that appears before anyone could have traded on the paper is the part that should worry a backtester most — it is the tax on overfitting, and no amount of patience recovers it. The additional ~32 points that arrive after publication are the market doing its job: capital chased a documented edge and shrank it.
A premium that keeps working after everyone can read the paper is, by that very survival, more likely to be compensation for real risk than a free lunch waiting to be arbitraged away.
Why some premiums survive and others evaporate
This is the non-obvious part, and it inverts the usual anxiety. Decay is often treated as bad news. But the pattern of decay is a diagnostic. Financial economics distinguishes two explanations for any factor: it is either compensation for bearing a genuine risk that hurts in bad times (a risk-based premium) or it is a behavioral mispricing that persists only while limits to arbitrage keep smart money out.
If a premium is risk-based, publication should change little. Investors who dislike the risk still dislike it; the compensation for holding it persists because it is not a mistake being corrected, it is a price for discomfort. If a premium is purely a mispricing, publication is corrosive — it directs arbitrage capital straight at the anomaly, and the edge compresses toward zero as the valuation spread closes.
So the fact that value and profitability-based premiums have degraded but not vanished, while some narrow, exotic signals have collapsed post-publication, is itself informative. Fama & French's later work adding profitability and investment factors, and Asness, Frazzini & Pedersen's quality framework, lean on characteristics that plausibly track real risk or durable behavioral frictions rather than a single-regime statistical fluke. Chen & Zimmermann's open-source replication of the anomaly zoo made the broader point measurable: many published predictors replicate, but with materially smaller magnitudes than the headlines claimed. The premiums that survive tend to be the ones with an economic story that does not depend on other investors staying asleep.
What this means once you hold it through a fund
An academic premium is gross, frictionless, and long-short. What reaches your account is net, frictional, and usually long-only. The gap between the two is where a lot of the surviving premium goes to die, and it is entirely within an investor's control to measure.
Three frictions deserve explicit attention. First, expense ratio and the fund's tracking error to its target index — a factor fund that charges more and tracks loosely can hand back the entire surviving premium before you notice. Second, capacity and AUM scale: a small-cap or deep-value strategy that works on paper can face real bid-ask spread and market-impact costs at scale, and a fund that has gathered a great deal of assets may be forced toward more liquid, less "pure" exposure — a quiet form of self-inflicted crowding. Third, the long-only constraint: most retail factor ETFs capture only the long leg of the academic long-short spread, which mechanically dilutes the premium and adds market beta you may already own elsewhere.
This is why the same factor label can describe quite different products. The distinction between selecting wide-moat businesses and mechanically loading on the quality factor is a live example, one worth reading closely in MOAT vs QUAL: Wide-Moat Selection vs the Quality Factor. The dividend-versus-total-return question runs on the same logic — a yield tilt is a value-and-quality bet in disguise, examined with the realized numbers in SCHD vs VOO. And the temptation to amplify a surviving premium with leverage runs straight into path dependency, which is unforgiving over long holds — the arithmetic is laid out in The Honest Math of Leveraged ETFs.
The behavioral tax, and the regime you are reading this in
Initially I assumed the main enemy of a factor premium was crowding. Then I looked at how factor funds actually get held, and the larger leak is behavioral. A premium that shows up over a full cycle can underperform the broad market for stretches long enough — five years, sometimes more — that most holders sell before the payoff. The surviving 42% is only available to the investor who stays in the seat during the drawdown when the factor is out of favor. Crowding compresses the premium; impatience forfeits what remains.
Regime matters for framing here. This is written with the VIX at 16.5 and the 10-year Treasury at 4.58% (FRED, asof 2026-07-14). Calm markets and a positive real risk-free rate are precisely the environment in which patience with an underperforming factor feels most expensive — a T-bill pays something, and the tilt that is lagging looks like dead money. That is exactly when the behavioral tax is highest, and exactly when the surviving premium is being handed from impatient holders to patient ones.
Winner by category
| Question | What the evidence favors |
|---|---|
| Does publication kill premiums? | No — it shrinks them (~58% lower post-publication), but survivors persist |
| Which decay most? | Purely behavioral mispricings with weak economic stories |
| Which survive best? | Premiums plausibly tied to real bad-times risk (value, profitability/quality) |
| Biggest controllable leak | Fees, tracking error, and holder impatience — not crowding itself |
Frequently asked questions
Is the small-cap or value premium "dead"? The precise claim the data supports is that documented premiums are smaller post-publication, not that they are zero. McLean & Pontiff put the average post-publication decline near 58% — a large haircut, but not a full erasure. Whether the residual survives your fees and your patience is the operative question.
If crowding shrinks premiums, why hold factor funds at all? Because a shrunken risk-based premium is still a premium, and because factor tilts change a portfolio's risk profile, not just its expected return. The reason to hold a tilt should be a view about the risk you are being paid to bear, not a backtest number you expect to repeat.
How do I tell a real premium from a data-mined one? You cannot do it perfectly, but Harvey, Liu & Zhu (2016) offer a practical filter: demand a higher statistical hurdle (a t-stat near 3.0) and, more importantly, an economic story for why the premium should persist. Premiums that survive out-of-sample and across markets are more credible than ones that appear only in the original sample.
Does adding more factors reduce the crowding problem? Diversifying across several weakly correlated premiums can smooth the ride, but it does not manufacture return that crowding has removed. It also multiplies fees and tracking error, so the net benefit depends heavily on implementation cost.
Where does this leave a plain index investor? In a defensible spot. A broad, low-cost total-market position captures whatever premiums exist in proportion to their market weight, without paying up for a tilt that may or may not survive. The case for that baseline is walked through in The Three-Fund Portfolio in 2026.
What this analysis can and can't tell you
The McLean & Pontiff evidence is drawn largely from U.S. equities over a specific multi-decade sample; it cannot tell you how a given premium behaves in a future regime it has never seen, nor how any single fund's implementation will track its target. The decay figures are averages across many predictors — individual factors vary widely around them. And "risk-based versus behavioral" is a useful lens, not a settled classification; reasonable researchers disagree about which premiums belong where.
Key takeaways
- Published premiums decay — roughly 58% post-publication on average (McLean & Pontiff 2016) — but survivors retain a meaningful, tradeable residual.
- About a quarter of the decline shows up before publication and reflects overfitting; that portion is a warning about backtests, not about markets.
- The premiums most likely to persist are those compensating for genuine bad-times risk, because publication does not fix a risk that investors still dislike.
- Fees, tracking error, capacity, and holder impatience typically leak more of the surviving premium than crowding itself.
- The advantage, where one remains, accrues to the investor disciplined enough to hold the tilt through the years it underperforms.
Editor's read
If forced to a single stance, the editor treats factor tilts as satellite positions justified by a risk story, not by a backtested number — and sizes them so the multi-year stretch of underperformance is survivable without selling. The most durable edge available to a long-horizon investor is not a clever factor; it is a low-cost, broadly diversified core held through cycles, with tilts added only where the fee and tracking-error budget genuinely earns its keep. Crowding is real, but impatience and cost do more damage.
Holdings disclosure: the editor holds broad market-cap-weighted index exposure as the long-horizon core and small factor tilts as satellites; no position is taken in any specific fund named for illustration in this article.
Methodology & sources: academic findings cited from McLean & Pontiff (2016, Journal of Finance), Harvey, Liu & Zhu (2016), Fama & French, Asness, Frazzini & Pedersen, and the Chen & Zimmermann open-source cross-section. Macroeconomic figures (10-year Treasury 4.58%, fed funds 3.63%, VIX 16.5, CPI 3.7% YoY) from FRED, asof 2026-07-14/06-01. No fund-level price series were required for this conceptual analysis. Window analyzed reflects the samples reported in the cited studies.
This article is for educational purposes and does not constitute personalized financial advice. See our full Disclaimer.