Momentum investing: what it is and its S&P 500 record

Stocks that did best over the last year tend to keep doing somewhat better than the rest for a while. That is momentum. Academic research documented it with data — the reference paper is Jegadeesh and Titman, 1993 — and it has been one of the most studied factors ever since.

Two things momentum is not. It is not buying whatever is up today: the most recent month is left out on purpose, because at a one-month horizon the effect reverses. And it is not a forecast about any particular company: it is a rule that ranks five hundred companies the same way every month, with no opinions involved.

In words: the stock's rise from twelve months ago to one month ago, divided by its annualized volatility. Dividing by volatility penalizes jerky rallies: a stock that rises 50% with low volatility scores higher than one that rises 80% with high volatility.

Each company's score is then compared with everyone else's (a Z-score, clipped at ±3 so one extreme case cannot dominate), the ~500 index members are ranked, and the portfolio holds the top 20 at 5% each. The list is rebuilt every month. A company needs 253 closes: with less history it has no momentum score, and the site says why instead of inventing one.

The full recipe, thresholds included, is in the methodology. Today's ranking, company by company, is in the rankings.

The site publishes the same rule in four sizes. Held in equal parts, the fair benchmark is the S&P 500 in equal parts too:

The honest reading of this table is uncomfortable for the factor: the more concentrated the portfolio, the worse. The 100-stock version beats the index by six tenths of a point a year; the 20-stock one trails it by four tenths; the 10-stock one trails by almost two points, with a 77.58% drawdown. Drawdowns follow the same pattern: the 50 and 100 fell somewhat less than the index, the 10 and 20 a good deal more.

Weight the same stocks by company size instead — like the regular S&P 500 — and the fair comparison is SPY, the fund that tracks that index. There, all four come out ahead:

The site does not publish SPY's daily max drawdown over this window. Measured on the monthly points of the curve (which always come out a little shallower than daily ones), SPY fell 52.2% and the cap-weighted momentum 20 fell 58.5%, both between late 2007 and March 2009.

It wins three periods out of eight. What mattered happened at the start: in the 14 months of the dot-com bubble the portfolio more than doubled while the index went nowhere. Starting both at 100 in December 1998, on March 2, 2000 the portfolio was worth 2.38 times the equal-weight index; on August 21, 2026, 0.90 times. Against its benchmark, the portfolio has not been back to that high in 318 months.

The factor's known weakness shows too: it falls harder when the market turns suddenly. In the financial crisis it lost 11.3 points a year more than the index, and in the COVID rebound, when what rallied was what had just fallen, it trailed by 15.7 points. Its worst full year was 2008, −57.09% against −43.61% for the index; its best, 2013, +39.96% against +34.30%. It beats the index in 14 of 28 years.

On its own curve, the portfolio spent 81% of the history below its previous peak (the index, 63%). Its worst episode started on December 7, 2007, bottomed on March 11, 2009 at −63.74% on monthly data, and did not regain the peak until March 13, 2014: 75 months without a new high.

The standard test is the information coefficient (IC): each month, the rank correlation between the momentum score and what each stock did the following month. Over 331 months, with every company that passed through the index:

It ranks in the right direction a little more than half the months, but only barely, and the decile ladder is almost flat: it is not that the top decile beats the second and the second beats the third; it is a weak, concentrated effect. The site's methodology puts it this way: positive IC, flat ladder. With these numbers it cannot be separated from noise.

And it is expensive to follow. In the site's turnover measurement, momentum 20 replaces on average 35.9% of its positions at each rebalance, 431% a year: the highest of the 20-stock portfolios measured. That measurement covers a different window (59 rebalances in 2021-2026, without a point-in-time universe), so it is good for the order of magnitude, not the decimal.

• They are simulated. Nobody traded this portfolio in 1998. It is the rule applied forward, month by month, with the data that existed at each date.
• They carry no costs. No commissions, no bid-ask spread, no taxes, no currency conversion. With momentum's turnover, that matters.
• Survivorship bias: nearly closed, not fully. Up to 8/21/2026 the history comes from Sharadar, including the prices of companies that later left the index; around 0.1% of membership slots are missing.
• The start date matters. Starting just before the dot-com bubble hands this portfolio its best period. Another start date would give another number.
• It is a single market. The 500 largest US companies, in dollars, over 27.6 years. What happened here need not happen elsewhere or in another era.

The portfolio, its curve, its holdings and its monthly ins and outs are on the momentum 20 page, with the curve downloadable as CSV. To try another size or weighting, use the simulator. All twenty portfolios side by side are in Model Portfolios.