Factor investing has long promised diversification. What it has too often delivered is a different kind of concentration. A new research paper1 from STOXX, co-authored by Anthony Renshaw, Yurong Gu, Matt Candar and Andrew Ang, argues that the problem is structural, and that the solution is to apply diversification not once, but at every level of portfolio construction simultaneously.
The paper, published July 2026, describes a multifactor portfolio benchmarked to the STOXX US Universal Index that diversifies within factors, across factors and across time. That three-layer architecture is the paper's central contribution, and the empirical record behind it is compelling.
The First Problem: Single-Signal Factors Are Noisy
The foundational insight is deceptively simple. No single metric measures a factor premium without error. Averaging across signals, therefore, does for measurement error what diversification across assets does for idiosyncratic risk. The paper's most distinctive construction choice is splitting value into two separate sleeves. Cyclical value, built on book-to-price and earnings yield, behaves very differently from enhanced value, which draws on economic profit, intangible capital and free cash flow. The correlation between the two sleeves' active returns is just 0.13. Over the 2007 to 2011 period, enhanced value returned a cumulative 6.91% against a benchmark loss of 1.23% and a cyclical value loss of 9.49%. During the growth-led 2017 to 2020 market, enhanced value kept pace with the benchmark while cyclical value fell more than 20 percentage points behind. Renshaw's team notes that neither sleeve dominates in all environments, and that holding both provides more complete exposure to the value premium. The implication for the construction is that value still accounts for one-third of the multifactor score overall, but is diversified internally.
Momentum is similarly broadened. Beyond standard price momentum, the composite adds idiosyncratic momentum, analyst revisions, economic profit momentum, short interest signals and hedge fund holdings. The composite reduces the portfolio's average active exposure to the systematic momentum risk factor from 20.1% to 13.0%, while maintaining a comparable information ratio. As Renshaw puts it, "the composite momentum portfolio has less systematic momentum risk because the diversifying signals contribute along dimensions that are not spanned by the momentum risk factor."
The Second Problem: Factor Leadership Rotates
The second layer of diversification addresses a harder problem: no single factor outperforms in every environment. The paper documents this plainly. Cyclical value was the best performer in 2025 and 2026 to date, but the worst in 2023 and 2024. Quality led from 2020 to 2023, then lagged. Momentum outperformed in 2024 and 2025, then stumbled early in 2026. Each factor has, in different years, been both the best and worst annual contributor.
The portfolio-level evidence is the cleanest argument for staying diversified. The multifactor portfolio earns an information ratio of 1.00 over the full March 2004 to May 2026 backtest, exceeding every individual factor sleeve. The correlations across active returns are low, ranging from negative 0.21 between quality and cyclical value to just 0.17 between quality and enhanced value. Renshaw makes the logic plain: "the composite does not depend on any single sleeve's cycle."
The Third Problem: Timing Is a Low-Breadth Problem
The third layer is the most structurally honest. The paper does not claim robust factor timing is easy. With only four factors and slow-moving cycles, Renshaw's team acknowledges directly that "factor timing is a low-breadth problem." The response is deliberately modest: three timing models, each bounded so that no factor's weight can be more than doubled or halved, and the final tilt averaged across all three. The models draw on different information sets: market similarity based on technical conditions, factor momentum based on recent information ratios, and a factor-of-factors model tracking exposure drift. The averaging of models is itself the third application of the diversification principle. Over the full 21-year sample, the timing overlay improves the information ratio from 0.99 to 1.00, a negligible amount. Over the last three years, the improvement is more visible, rising from 1.83 to 1.92. Crucially, the effective number of factors in the combined model is 3.39, higher than any individual model. Timing can tilt the portfolio, but it cannot concentrate it.
Five Key Takeaways for Advisors and Investors
- Single-factor products carry hidden concentration risk. A portfolio labeled "value" or "momentum" may derive most of its active exposure from a single metric or regime.
- Splitting value into cyclical and enhanced sleeves is not a cosmetic choice. The two have a return correlation of just 0.13 and behave differently enough to justify treating them as separate factors.
- Factor leadership rotates in ways that cannot be reliably predicted. Staying diversified across factors eliminates the need to forecast which one will lag next.
- Factor timing can add value, but only modestly and only when bounded. Timing should diversify the allocation, not concentrate it.
- Diversification applied at every level, within signals, across factors and across timing models, produces a portfolio that consistently earns a higher information ratio than any of its components.
Footnote:
1 Renshaw, Anthony A., Yurong Gu, Matt Candar and Andrew Ang. "Diversification All the Way Down: Multifactor Investing Within Factors, Across Factors and Across Time." STOXX / ISS-STOXX, July 2026. Available at SSRN: ssrn.com/abstract=7156278.
