In 2019–2026, 12-month factor momentum on 23 public equity factors gave no clear net gain over equal factor weights (gain +0.06%/month, 95% CI −0.30 to +0.44)
Question: after factor momentum became public in 2017–2018, did the past 12-month return of equity factors predict their next-month return well enough to beat a static factor portfolio after trading costs? Approach: we froze a protocol before the holdout test. We built 23 long-short factors (17 US factors from the French and AQR data libraries, 6 developed-ex-US factors from French) in the 202608 data vintage. The primary strategy holds each factor long or short with weight 1/N, as the sign of its past 12-month return sets. The baseline holds all factors long with weight 1/N. We charged 10 bp per dollar traded, with scenarios of 5 to 50 bp. We used stationary block-bootstrap intervals. The holdout is 2019-01 to 2026-08 (92 months). Results: the net timing gain was +0.06% per month (95% CI −0.30 to +0.44), so the pre-registered hypothesis is not supported. The gain was also not positive in 1964–2018 (−0.08%, CI −0.21 to 0.06). The equal-weight factor premium fell from 0.35% to 0.09% per month. The alpha of factor momentum on the baseline was 0.17% per month in both periods, with intervals that include zero. Seven other variants (lookbacks of 1 to 60 months, a volatility-scaled rule, a cross-sectional rule) and eight other universes also gave no holdout gain with an interval above zero. A post-hoc measure, the spread between past winner and past loser factors, stayed near its earlier level (0.38% versus 0.45% per month). Pooled machine-learning forecasts (ridge, LightGBM) had out-of-sample R² near 0% against each factor's expanding mean in the holdout, against +1.6% for ridge in 1985–2018. In archived data vintages, revisions changed 0–2% of the 12-month signals for six US factors. Main limitation: the minimum detectable gain in the holdout is 0.60% per month. This is larger than any earlier estimate, so the null result is weak evidence of no effect. Costs are a scenario model, not measured costs. The package has code, the frozen protocol, provenance hashes, forecasts, monthly strategy returns and figures.
factor momentum · factor timing · out-of-sample test · post-publication decay · trading costs · equity factors · pre-registered protocol · stationary bootstrap · Fama-French factors · return forecasting · null result · data vintages
References
11- challengesdoi:10.1111/jofi.13131
We test its time-series factor momentum after publication; the net timing gain over equal factor weights is not significant in 2019–2026.
- challengesdoi:10.3905/jpm.2019.45.3.013
We test its scaled factor momentum rule out of sample in 2019–2026; no significant gain over equal weights.
- doi:10.1093/rfs/hhad006
Earlier factor momentum evidence that sets the hypothesis.
- doi:10.1111/jofi.12365
Post-publication decay of anomaly returns, which agrees with the fall of the equal-weight factor premium we find.
- usesdoi:10.1016/j.jfineco.2014.10.010
Five-factor definitions and the French data library factors.
- usesdoi:10.1016/j.jfineco.2013.10.005
BAB factor data from AQR.
- usesdoi:10.1007/s11142-018-9470-2
QMJ factor data from AQR.
- doi:10.1093/rfs/hhv063
Trading cost and turnover context for anomaly portfolios.
- usesdoi:10.1080/01621459.1994.10476870
Stationary bootstrap for confidence intervals.
- usesdoi:10.1016/j.jeconom.2006.05.023
Clark-West test for nested forecast comparison.
- doi:10.1093/rfs/hhm014
Out-of-sample R-squared against the prevailing mean.
Cited by
12 publicationsPublic catalog entries on Sciscape.
- Past VIX values reduced weekly Nasdaq Composite variance forecast loss in 504 tests from 2016 to 2025
This SCI study reports a monthly factor timing test with uncertain net gains. The present study tests weekly index forecasts instead.
- New months changed six-month industry momentum weights more than revisions on average in ten archived U.S. releases from 2015 to 2026
Our sparse archive test also leaves return prediction and net trading gains uncertain. We use a different industry rule and separate source freshness from revisions; this is not an independent replication of its factor test.
- VXN cut QLIKE loss for monthly Nasdaq-100 daily-return variance forecasts by 34% in a 2017–2026 vintage test
Related recent out-of-sample return-prediction result; catalog abstract inspected.
- Past returns reduced errors in daily return size but not direction for five US indexes in CNNpred data, 2015–2017
This related monthly factor test motivates careful references, date-respecting evaluation, and explicit source-vintage limits.
- Archived seasonal industry forecasts lost 10.9% a year after assumed costs across five 12-month tests
Related work tests later factor returns and checks historical data versions.
- A downside term gave a small and uncertain Nasdaq risk forecast gain in a 2015–2025 archive test
Related chronological equity-return study with a later null result; its monthly factor momentum scope differs from our daily index-risk test.
- A 252-session trend input gave no clear monthly Nasdaq Composite return forecast gain in a 2018–2025 archive test
Related factor momentum test with weak later-sample forecasts; a different asset universe.
- Five-year return means gave no clear forecast gain with annual US market archives in 2005–2025
Related return-forecast and cost benchmark for public equity factors. Its negative result was known before this test.
- supportsIn 2019–2025, winner equity factors beat loser factors in 43 non-US countries, but 12-month factor momentum gave no net gain over equal weights at 10 bp costs (153 JKP factors)
Independent test of the same 2019+ factor-momentum question on 23 French/AQR factors. Our 153-factor US estimates agree (alpha 0.18 vs 0.17, spread 0.42 vs 0.38), and our 43-country pool reduces the minimum detectable effect that limited both studies.
- An input delay of 21 trading days cut the 5% tail forecast gain from 16.2% to 7.0% in revised US market data for 2000–2025
This earlier SCI study tested factor return forecasts and data revisions at a monthly horizon. It motivated a separate daily risk test with explicit input cutoffs.
- Adding prior overnight gaps raised ridge error for next-session open-to-close returns in SPY, QQQ and IWM in 2016–2025
Its catalog summary reports uncertain later gains for monthly factor momentum.
- usesFive years of returns gave higher risk forecast loss than ten years in 20 annual US market tests, 2005–2025
Its negative factor-timing test and vintage limits informed the question and baseline choices. This study tests broad-market second moments.