Decision brief · Factor investing
Do factor ETFs deliver what their labels promise?
Factor regressions test whether six style ETFs exhibit their advertised exposures and whether unexplained return remains after multiple-testing correction.
Depth 1 · Answer
The advertised exposures were recovered; no alpha survived multiple-testing correction.
MTUM loaded on momentum at β = 0.34 (t = 49), and VLUE loaded on value at β = 0.31 (t = 29). QQQ's estimated +3.29% annualized alpha had a raw HAC p-value of 0.024, but its Benjamini–Hochberg q-value was 0.147. Across the six ETFs, no alpha survived the 5% false-discovery threshold.
Published 8 August 2026 · Underlying investigation last updated 3 August 2026
- MTUM momentum β
- 0.34
- VLUE value β
- 0.31
- QQQ raw p
- 0.024
- QQQ adjusted q
- 0.147
Why it matters
ETF allocators, product researchers, and exposure owners
The result matters when a branded strategy is selected for a specific systematic exposure, or when a portfolio report attributes historical return to manager skill rather than priced factors.
Decision affected
What evidence is required before calling a return “alpha”
First verify the intended exposure, then test residual return with robust uncertainty, multiplicity correction, and a chronological holdout. A raw p-value from one of several regressions is not the conclusion.
Depth 2 · Evidence
The regression found product identity more clearly than unexplained return
The study aligned 2,867 daily observations from 5 January 2015 through 29 May 2026. It estimated six-factor time-series regressions for MTUM, VLUE, QUAL, IWM, USMV, and QQQ using the Fama–French five factors plus momentum, with Newey–West HAC(5) intervals.
The named style loadings were economically meaningful and precisely estimated. The alpha tests told a different story: after Holm and Benjamini–Hochberg corrections across six simultaneous hypotheses, none met a 5% false-discovery threshold. This does not mean the funds delivered nothing; it means the measured return was explained primarily as exposure under this factor model.
A chronological 50/50 holdout froze first-half coefficients and applied them to the second half. Holdout R² fell by 0.4 to 29.6 percentage points across the ETFs, with USMV showing the largest decay. Exposure estimates can remain useful while still being state-dependent.
Depth 3 · Application
Evaluate the exposure, inference, and stability as separate questions
- Test whether a branded ETF has the intended sign and economically meaningful magnitude on its named factor.
- Correct alpha inference for the number of products or strategies searched.
- Use heteroskedasticity- and autocorrelation-consistent intervals for daily return regressions.
- Freeze coefficients into a chronological holdout instead of reporting full-sample fit alone.
- Monitor rolling betas because a full-sample coefficient is a time average, not a permanent product identity.
Depth 4 · Limits
Alpha is always conditional on the model used to explain it
- The selected factor menu defines the alpha left behind; omitted technology, industry, or intangible-capital effects can matter.
- Fama–French factor portfolios are paper long–short portfolios, not costless tradable products.
- Eleven years remains one regime, including an unusual growth-over-value period.
- The chronological 50/50 split is one holdout path rather than repeated rolling-origin evaluation.
- HAC lag choice and vendor adjustments to ETF prices can move inference.
Depth 5 · Method and code
Inspect every loading, interval, holdout result, and adjusted test
Related research