A case on whether to launch a momentum mutual fund, tested against the Fama-French momentum factor from 1927 to 2023.
The case is whether to launch a momentum fund as a mutual fund product. The factor evidence is strong enough that the answer is yes, but the product constraints are what decide whether it survives.
Momentum runs 0.62% a month, 7.48% annualised, against 16.29% annualised volatility for a Sharpe of 0.46. The t-statistic on the mean is 4.52 under OLS and 4.62 with Newey-West at six lags, so autocorrelation and heteroskedasticity do not explain it away.
The obvious objection is that the premium was arbitraged once Jegadeesh and Titman published. Splitting the sample says otherwise: 8.83% a year pre-1993 over 792 months, 11.09% a year post-1993 over 192 months, 7.48% over the full 1,164 months.
The post-publication window is stronger, not weaker. Independent industry-level evidence corroborates it: ranking the 49 Fama-French industry portfolios on prior 12-1 month returns and going long the top quintile against the bottom gives a t-statistic of 4.81, with no individual-stock effects available to explain it.
A 63% monthly win rate is not what the investor experiences. The factor has lost 52.05% in a single month, August 1932, and 45.02% in July 1932. April 2009 cost 34.30%. Maximum cumulative drawdown reached -77.24%.
Every one of the worst months is a sharp rebound: the market rose 37.06% in August 1932 while momentum lost half its value. Momentum crashes are what happens when short-leg losers rally hardest off a bottom. That is a structural feature of the strategy, not bad luck, and it is not diversifiable within the factor.
The correlation between momentum and value is -0.406 over 1927 to 2023, and 91% of decades show a negative correlation. Correlation with the market is -0.017 and with size near zero.
That is the whole argument for the product. Momentum is a mediocre stand-alone asset and an excellent complement to value, because the periods when value is being punished are disproportionately the periods when momentum works.
Two constraints degrade it in a mutual fund wrapper. Long-only removes the short leg, where a meaningful share of the historical alpha lives. Tracking-error limits pull the portfolio back toward the benchmark and dilute the factor exposure investors are buying. My conclusion was launch it, but as an explicitly diversifying sleeve next to value rather than as a stand-alone return product, and size it so that a 1932-style month is survivable.