Key takeaways
- The Fama/French momentum factor compounded at 6.19% a year from January 1927 to June 2026. Value managed 3.59% and size 1.50% over exactly those months.
- Momentum fell 78.5% from its June 1932 peak to its September 1939 trough, and didn't regain the 1932 level until December 1956.
- April 2009 alone cost the factor 34.36%. Across calendar 2009 it lost 52.91%, in a year the US market factor returned 28.52%.
- Novy-Marx and Velikov measured 34.52% one-sided monthly turnover on a decile momentum strategy, and costs that cut its gross spread from 1.33% a month to 0.68%.
- Momentum and value have correlated -0.41 month to month since 1927. A 50/50 blend compounded at 5.72% a year with a worst drawdown of 40.8%, against 78.5% for momentum on its own.
The biggest premium of the three classic factors, and the deepest hole
You want to know whether buying what's already going up is a real edge or a story people tell afterwards. The answer sits in the same free data library that publishes the value and size factors, so here it is up front, before the caveats.
Momentum — the return on stocks that rose most over the past year minus the return on stocks that fell most, with the portfolios rebuilt every month — compounded at 6.19% a year from January 1927 to June 2026. Over exactly those months the value factor compounded at 3.59% and the size factor at 1.50%. The t-statistic on momentum's monthly mean is 4.63. That number measures how large the average is next to how much it jumps around, and above about three, chance alone becomes a strained explanation.
Now the part that doesn't fit on a factsheet. The same series lost 78.5% from peak to trough between June 1932 and September 1939. Its monthly returns carry a skewness of -3.04, against +2.04 for value and +0.15 for the market. Skewness only says which tail is longer, and momentum's long tail points down. On top of that, the strategy rebuilds roughly a third of each side of its book every month. That last fact is where a large slice of the 6.19% goes.
What Jegadeesh and Titman measured in 1993, including the turnover
The paper that started this is Narasimhan Jegadeesh and Sheridan Titman's, published in the Journal of Finance in March 1993. They sorted US stocks on past returns and held the winners against the losers.
Their headline result: the version formed on six months and held for six "realizes a compounded excess return of 12.01% per year on average" over the 1965 to 1989 period. Their strongest variant, formed on the past twelve months and held three, "yields 1.31% per month". Those are large numbers, and they survived enough replication that momentum became the fourth factor in the standard model.
What gets quoted far less often is the same paper's own accounting for trading. "On average, the relative strength trading rule results in a turnover of 84.8% semiannually." They then charged the strategy half a percent each way: "The risk-adjusted return of the relative strength trading rule after considering a 0.5% one-way transaction cost is 9.29% per year, which is reliably different from zero." The people who found the anomaly published a gross figure and a net figure in the same section, and they were not the same number.
Run French's momentum factor over their exact window and it compounded at 9.49% a year from January 1965 to December 1989, with a t-statistic of 4.02. French's construction is milder than theirs: the low leg is everything "below the 30th NYSE percentile" and the high leg everything "above the 70th NYSE percentile", rather than the extreme deciles. That's why the number comes in lower. Out of sample since, from January 1993 to June 2026, the factor compounded at 3.98% with a t-statistic of 1.88. Value's post-publication record broke the same way, and the value premium after 1992 is computed from the same library by the same method.
The premium by decade, computed from the series itself
The chart above compounds monthly momentum returns within each calendar decade. The source is the Fama/French momentum file from Kenneth French's data library, built from the June 2026 CRSP database, so this isn't an estimate of the series. It's the series.
Left to right: -6.99% a year in the 1930s, 6.37% in the 1940s, 10.78% in the 1950s, 11.09% in the 1960s, 9.34% in the 1970s, 8.60% in the 1980s, 13.63% in the 1990s, -2.03% in the 2000s, 2.62% in the 2010s, and 4.63% a year so far in the 2020s.
Eight of the ten are positive, and the best of them is the 1990s — the decade the discovery was published. That's awkward for a simple story about publication killing the trade. The two negative decades are the 1930s and the 2000s. Each contains a crash, and the crashes are the whole character of this factor.
The decade split also hides a sharper break. From January 1993 to December 2008 momentum compounded at 9.89% a year. From January 2009 to June 2026 it compounded at -1.14%. Nothing about the publication date explains a split placed there. One crash does.
The 1930s crash: down 78.5%, and 24 years to climb out
Momentum's worst episode is the oldest one. Run the factor as a cumulative index and it peaked in June 1932, then bottomed in September 1939 at 78.5% below that peak — seven years and three months down. July 1932 took -45.63% and August 1932 took -52.61%. Back to back, that's -74.23% in two months. The index didn't regain its June 1932 level until December 1956.
A 78.5% fall needs a 365% gain to get back to level, which is the arithmetic behind that 24-year wait and the reason deep drawdowns aren't symmetrical with the gains that repair them.
Kent Daniel and Tobias Moskowitz put decile portfolios on the same months. In their 1927 to 2013 US sample, "the two worst months for a momentum strategy that buys the top decile of past 12-month winners and shorts the bottom decile of losers are consecutive: July and August of 1932. Over this short period, the past-loser decile portfolio returned 232% and the past-winner decile portfolio had a gain of only 32%." The losers didn't perform in 1932 so much as violently stop collapsing.
April 2009 is the case worth studying, because the setup is recognisable
The modern crash is easier to picture, because everyone remembers the market it happened in. Momentum lost 34.36% in April 2009, its worst month outside 1932. From March to May 2009 it fell 49.37%. Across calendar 2009 it lost 52.91%, in a year the US market factor returned 28.52%. A strategy that halved while the market rose by a quarter is not a diversifier having a bad year. It's a mechanism firing.
Daniel and Moskowitz, on the decile version: "over the three-month period from March to May of 2009, the past-loser decile rose by 163% and the decile portfolio of past winners gained only 8%." Their portfolios use the top and bottom tenth while French's factor cuts at the 30th and 70th percentiles, which is why their swings are wider than the ones computed here from the factor file.
This wasn't a US-only event. French's developed-markets-ex-US momentum factor lost 22.52% in April 2009 and 36.79% across that calendar year. Its deepest drawdown in the entire file, which starts in November 1990, is 40.88%, running from February to September 2009. Two independent universes, the same month.
The US index peaked in November 2008 and bottomed in September 2009, 57.85% lower. As of June 2026 it was still 22.32% below that November 2008 level, having spent every one of the 211 months since underwater. That's the gap between a factor's average return and a person's experience of holding it, and it's the reason drawdown and volatility measure different things.
Why the crashes land where they do
Daniel and Moskowitz's real contribution isn't the crash list. It's the mechanism. From their abstract: "These momentum crashes are partly forecastable. They occur in panic states, following market declines and when market volatility is high, and are contemporaneous with market rebounds."
The reason sits in what a momentum book holds after a bear market. The loser leg fills up with whatever has been destroyed, and by then those names are high-beta, distressed and priced for failure. When the market turns, they snap back hardest — and the momentum book is short them. Daniel and Moskowitz measure the asymmetry directly: in a bear market "a momentum portfolio's up-market beta is more than double its down-market beta (-1.51 versus -0.70 with a t-statistic of the difference = 4.5). Outside of bear markets, there is no statistically reliable difference in betas."
Their summary of the pattern is the sentence to remember: "Fourteen of the 15 worst momentum returns occur when the lagged two-year market return is negative. All occur in months in which the market rose contemporaneously, often in a dramatic fashion." A momentum book behaves like a written call option on the market, but only once the market has already fallen. You collect the premium for years, then pay it back in a single rebound.
None of that stops the long-run numbers from being large. Over their 1927 to 2013 sample the winner-minus-loser portfolio ran a Sharpe ratio of 0.71 against 0.40 for the market, with a CAPM alpha of 22.3% a year and a t-statistic of 8.5. The crash and the premium are the same object seen from two ends.
Turnover is the third fact, and it's the one that eats the premium
French's own file states the construction: "The portfolios are constructed monthly", ranking stocks on the previous year's return with the most recent month left out. A stock's rank on last year's return changes constantly, so the book has to be rebuilt constantly. Book-to-market doesn't behave that way; a value book barely moves in a year.
Robert Novy-Marx and Mihail Velikov priced that difference. Over July 1963 to December 2012, their value-weighted decile momentum strategy earned 1.33% a month gross, turned over 34.52% of each side per month, paid 0.65% a month in trading costs and netted 0.68% a month with a t-statistic of 2.45. Costs took 49% of the gross spread — not a rounding error, roughly half the thing.
The contrast with the low-turnover factors is in the same paper. For size and value, "the time-series average cost of trading these factors is 5.66 and 5.45 basis points per month". For momentum, "its time-series average cost of trading is 48.39 basis points per month". Around nine times the drag, for a portfolio rebuilt twelve times a year instead of once. If you've read what turnover costs an ordinary trading account, this is the same arithmetic with an institution's execution.
David Lesmond, Michael Schill and Chunsheng Zhou pushed it further. Replicating the Jegadeesh-Titman strategy over 1980 to 1998, they found a gross semi-annual spread of 7.826%. After trading costs at the strategy's actual turnover, that spread came to 0.128% on their headline estimate, with a t-statistic of 0.07. Their verdict: momentum's returns create "an illusion of profit opportunity when, in fact, none exists".
The strongest case against the cost argument
That verdict is disputed, and by people with better data. Andrea Frazzini, Ronen Israel and Tobias Moskowitz used "nearly a trillion dollars of live trades from a large institutional money manager from 1998 to 2011 across 19 developed equity markets". Real fills, not spread estimates. What they found: "actual trading costs are less than a tenth as large as, and therefore the potential scale of these strategies is more than an order of magnitude larger than, previous studies suggest."
On momentum specifically: "UMD has an annual drag of 3.51 percent per year from trading costs, due to its 127% turnover and an average market impact cost of 23 basis points." At the size their trade data implies, "the net expected return at its implied fund size of $3.7 billion is 4.34 percent per year (t-statistic of 8.42) and its break-even fund size is $52 billion." Not an illusion — a real premium, at real institutional scale. The drag is still the point of comparison: 3.51% a year for momentum against 1.54% for value, same desk, same years.
There's a sting inside their own table, and it cuts towards the critics. Over 1998 to 2011 the realised gross return on US UMD was 2.56% a year and the realised net return was -0.95%. The 4.34% figure uses a long-run expected gross, because the years they held trade data for happened to be poor ones for momentum. French's factor agrees on the period: over 1998 to 2011 it compounded at 2.77% a year. Both numbers are honest, and which one you quote settles the argument on its own.
Four after-cost estimates that disagree about the level, not the direction
Side by side, the disagreement is narrower than the headlines suggest. Jegadeesh and Titman, on their own 1965 to 1989 sample with a 0.5% one-way charge: 9.29% a year, reliably positive. Novy-Marx and Velikov, 1963 to 2012, small-trader effective spreads: costs take 49% of the gross, and what's left still carries a t-statistic of 2.45. Frazzini, Israel and Moskowitz, institutional fills, 1998 to 2011: a 3.51% annual drag and a 4.34% net expectation. Lesmond, Schill and Zhou, 1980 to 1998, retail-style costs including the short leg: nothing left.
Most of the spread between those four is a question of who is doing the trading, not what momentum earns. A large arbitrageur with an execution desk pays roughly a tenth of what a retail-style cost model assumes. Every one of the four agrees that momentum's costs run several times value's, and every one agrees the gross premium is larger. Where the net lands depends on which of those two facts is bigger in your account, and that isn't something the literature can settle in general.
What pairing momentum with value does to the arithmetic
Momentum and value are natural opposites. One buys what has risen; the other buys what is cheap, which is often what has fallen. Since January 1927 their monthly returns have correlated -0.41. Cliff Asness, Tobias Moskowitz and Lasse Pedersen measured -0.53 between stock value and stock momentum across their global sample, which runs from January 1972 to July 2011.
The arithmetic of that negative number is worth seeing. A 50/50 monthly blend of the two French factors compounded at 5.72% a year from January 1927 to June 2026, with annualised volatility of 7.93% and a return-to-volatility ratio of 0.74. Momentum alone scored 0.46 on the same measure and value alone 0.35. The blend's worst drawdown was 40.8%, against momentum's 78.5%.
Frazzini and his co-authors found the effect in costs too: "A ValMom combination has a realized cost of 2.18 percent per year with turnover of only 79% since value and momentum trades tend to offset each other". When one sleeve wants to sell what the other wants to buy, the trade nets out inside the book and never reaches the market.
What this data can't tell you
Four things, and each of them matters.
The French factor is a costless paper portfolio. It pays no spread, no borrow fee on the short leg, no commission and no tax. Every net figure above comes from a separate study with its own cost model, and those models disagree with each other by a factor of ten.
It's also long-short by construction. Being short the loser decile produces both the crash and a good share of the premium. A long-only momentum tilt is a different instrument with a different risk profile, and nothing measured here describes one.
Ninety-nine years of monthly data sounds like plenty. It contains about two genuine crashes. Any claim about how often momentum crashes rests on a sample of two, and a t-statistic computed on returns that cluster in time overstates its own confidence.
Finally, decade boundaries are an accident of the calendar rather than a feature of markets, and the record here is US except for one developed-ex-US series that begins in November 1990. This is a backtest of a rule applied to history. It is not a forecast.
What would change the conclusion
Three things would, and they're worth watching for.
The first is a crash without the setup. Daniel and Moskowitz's claim is conditional: crashes arrive after market declines, in high volatility, during rebounds. A 30% momentum loss in a calm bull market would break that mechanism, and with it the case for every conditional version of the strategy. Their own dynamic version, which scales exposure using forecast volatility, reported an annualised Sharpe ratio of 1.19 across markets and asset classes. That figure depends entirely on the forecastability holding.
The second is costs moving against it. Momentum's net premium is a difference between two numbers of similar size. If spreads widen, or if crowding lifts market impact for everyone at once, the residual can vanish without the gross premium changing at all.
The third is simply more time. From January 2009 to June 2026 the factor compounded at -1.14% a year and never recovered its November 2008 high. Another decade of that, and "momentum has the largest documented factor premium" turns into a statement about the twentieth century.