Define the clock before the ranking
Cross-sectional momentum ranks stocks by their own prior returns relative to other eligible stocks. It differs from a time-series trend rule that compares a security's return with zero or another absolute threshold. The French monthly momentum-factor documentation uses prior months two through twelve, omitting the most recent month. That is a specific formation rule, not a generic instruction to use a year of prices.
Write out the endpoint dates for at least one formation month. For a portfolio held in January, this convention uses returns from the preceding January through November and excludes December. The return observed in January cannot enter January's initial signal. A unit test should verify these boundaries around year changes, holidays and missing observations rather than relying only on a rolling-window label.
Compound the included observations
Assume a fictional stock earns 10% over January through October and another 5% in November. The included-window return is 1.10 multiplied by 1.05 minus one, or 15.5%. If December then loses 12%, the full January-through-December return is 1.155 multiplied by 0.88 minus one, or 1.64%. The two windows tell markedly different stories.
Do not add the 10% and 5% components or treat a missing month as a zero return. Use a return series with a declared dividend and corporate-action convention. When evaluating several lookbacks, save each trial and its exact boundary rule. Choosing whichever window best avoids a past reversal after seeing the results turns the skip decision into hidden optimization.
Turn ranks into investable decisions
A ranking needs an eligible universe, a rule for ties, and enough valid price history. A new listing should not receive an invented full-year return. Stale prices and trading halts require a policy distinct from ordinary missing data. Applying today's constituents to the old window can remove the very companies whose weak outcomes a momentum study is supposed to include.
Formation at month-end also does not establish a fill at that closing price. If the final input arrives only after the auction, an order using it cannot have participated in that auction. Preserve signal time, order time and execution assumption separately. A monthly signal may still create substantial trading when names cross a cutoff, so report both rank persistence and portfolio turnover.
Understand the exposures you are accepting
Momentum can concentrate in industries that have recently outperformed, and rapid reversals can affect many selected names together. A sector-constrained version changes that exposure rather than universally improving the original. Compare variants using the same calendar and information set. A volatility-scaled version also introduces a second estimated input, whose lookback and timing must be included in the research specification.
The skipped-month convention is a starting definition, not evidence that a strategy works after costs or across all markets. Test stability across predeclared periods and liquid subuniverses, with realistic treatment of turnover and missing prices. The numerical example demonstrates compounding and endpoint sensitivity only. It contains no actual market history, momentum premium estimate or claim of executable historical profits.
Inspect the example
| Window component | Assumed return | Growth factor |
|---|---|---|
| January through October | 10.0% | 1.1000 |
| November | 5.0% | 1.0500 |
| Included January-November window | 15.5% | 1.1550 |
| December, excluded from signal | -12.0% | 0.8800 |
| Full January-December window | 1.64% | 1.0164 |
Sources & reproduction
Primary references support the definitions and method. Numerical examples are illustrative unless a source record is explicitly identified.