A landmark new study 1 reaching back to 1866 delivers what may be the most compelling out-of-sample test in the history of factor investing. Researchers Guido Baltussen, Bart P. Van Vliet, and Pim Van Vliet constructed a novel database of 3,269 U.S. stocks spanning the pre-CRSP era from 1866 to 1926, a 61-year period entirely untouched by the data sets that gave rise to modern factor models. Their verdict is unambiguous: the equity risk premia that define today's smart beta landscape were already present in markets a century before the research that discovered them.
The central concern motivating the study is data snooping. If the factors that generate excess returns were identified using CRSP data from 1926 onward, how confident can investors be that those premia are real and not artifacts of the sample? The pre-CRSP era provides a definitive answer. "Conducting tests in the pre-CRSP era provides a powerful ground for independent out-of-sample tests," the authors note, "while at the same time negating the arbitrage hypothesis for factor decay, since investors at that time could not have traded on insights from future research."
What the Data Reveals
Baltussen, Van Vliet, and Van Vliet report factor premia averaging 4.0% in the pre-CRSP sample versus 4.7% during the CRSP period, a difference that is not statistically significant. "We find no statistically significant evidence of data-snooping-driven decay," they write, adding that "the studied equity factor premia do not materially decay out-of-sample when unaffected by post-publication arbitrage."
The results are striking in their specificity. Value, measured by dividend yield (HML), produces a CAPM alpha of 9.16% with a t-statistic of 7.41. Momentum (UMD) delivers a CAPM alpha of 9.41% with a t-statistic of 5.00. Low-risk (BETA) earns 7.04% (t=4.66), and seasonality contributes 6.87% (t=5.45). The overall factor CAPM alpha across 159 years of data averages 4.96% with a t-statistic of 15.27. Size, by contrast, shows a premium of -0.53%, indistinguishable from zero.
The authors are direct about the significance of their dataset: "We are the first to create an extensive dataset for this period that also includes market capitalization," they state, enabling value-weighted portfolio construction that removes the small-cap bias that has long complicated factor research.
Why Standard Explanations Fall Short
The study puts four prominent risk-based explanations to the test across the full 159-year span and finds each wanting. Macroeconomic risk, using the Chen, Roll and Ross factor model, cannot account for the premia. "Macroeconomic risks do not materially explain stock factor premia," the authors conclude. The delegated asset management hypothesis, which attributes factor behavior to institutional incentive structures, also fails: "the presence of significant value, momentum, and low-risk premia in this setting therefore challenges explanations that rely solely on delegated management structures," they write, noting that institutional delegation looked very different in 1880 than it does today.
Crash risk and downside risk face the same verdict. On crash risk, the authors observe that "crashes are an inherent feature of momentum but are unlikely to explain its existence." On downside risk: "downside risk does not materially explain stock factor premia over 159 years of data." Summing up, Baltussen, Van Vliet, and Van Vliet state that "time variation in factor premia is difficult to reconcile with explanations based on macroeconomic risk, delegated management, crash risk, or downside risk."
Their overall conclusion is measured but firm. "Overall, these findings indicate that equity factor premia are robust empirical regularities that persist out-of-sample," they write, adding: "we conclude that these factors stand out as persuasive empirical equity factors in out-of-sample analysis."
5 Key Takeaways for Advisors and Investors
- Factor premia are not statistical accidents. Value, momentum, low-risk, and seasonality all survive a 61-year out-of-sample test that predates the research that named them.
- Post-publication arbitrage is not eroding returns. Pre-CRSP and CRSP-era factor premia are statistically indistinguishable, undermining the thesis that smart beta has been "arbitraged away."
- Size is the notable exception. The size premium is absent in both the pre-CRSP and full-sample data, a caution for strategies that rely heavily on small-cap tilts.
- Risk-based explanations remain unsettled. None of the four leading economic explanations (macro, delegation, crash risk, or downside risk) can account for the full 159-year pattern of factor returns.
- The 159-year record raises the confidence bar. Advisors can position factor-based strategies with greater conviction knowing the evidence base now spans conditions far removed from modern markets.
Footnote:
1 Baltussen, Guido, Bart P. Van Vliet, and Pim Van Vliet. "The Cross-Section of Stock Returns before CRSP." SSRN, September 2026, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3969743.
