Key takeaways:
- The Quant Winter of 2018 to 2020 exposed how heavily quant managers were relying on the same well-known market patterns, or ‘factors’
- Since the winter, the industry has broadly reduced its reliance on generic factors, though not all managers did so equally, and those differences in the investment approach and positioning carry very different performance implications
- Our analysis shows that managers most dependent on those patterns earned less than their factor exposures alone would have delivered, in every period measured
Introduction
This paper is the third in our Quant Renaissance series. The first documented the early signs of recovery in systematic equity investing after the 2018–2020 Quant Winter, and the second turned to the causes of that winter, identifying macroeconomic sensitivity and factor crowding as its two primary sources of vulnerability.
This paper takes a different vantage point. Rather than examining what managers are doing differently, as our earlier work did, our analysis looks for the evidence in the returns themselves: how dependence on generic factors evolved through the Quant Winter and its aftermath, which managers proved resilient and which did not, and where Man Numeric stands within that distribution.
We answer these questions by splitting each manager’s monthly active returns into the part a standard factor model explains and the part it cannot. The managers are the systematic equity products reported to eVestment, the database allocators use to compare managers. The regression, the factor set and the universe screen are set out in the Appendix.
Our three key findings are:
- Reliance on generic factors has fallen from its Quant Winter peak, and the fall is larger still once we hold the factor environment constant, so it reflects managers repositioning rather than a calmer backdrop
- The managers whose risk is most factor-driven have carried negative residual alpha in every period we measure, earning less than their factor exposures alone would have delivered
- The larger managers became disproportionately more factor-driven through the winter and have stayed that way
Skill or style?
The question of resilience comes down to a simpler one, about where a manager’s returns actually come from. Part of a quant manager’s performance is skill, the proprietary insight that sets one manager apart from another. The rest is style, the return earned by tilting towards the same well-known factors that much of the industry already harvests. While those factors are performing, skill and style are hard to tell apart, because both may produce attractive returns. They separate only when conditions turn, and it is in episodes like the last Quant Winter that the difference between a resilient manager and a vulnerable one becomes clear.
How far dependence fell
Managers' reliance on generic factors follows a clear trajectory across this universe (Figure 1). Reliance was high in the early years. Through the mid-2000s, the rolling three-year median R2 (the share of a manager's active return variation that generic factors explain) sat around 30%. This was the industry’s emerging phase, when systematic equity was a young and narrow field and a small number of managers largely harvested the same well-known premia, so that generic factors explained much of the variation in their active returns.
As the industry matured and diversified, and as the universe itself grew, that reliance fell, reaching its low of around 15% in the mid-2010s. It then rose steadily again into the Quant Winter, and the rolling three-year median R2 climbed above 40% as generic factor movements came to dominate manager returns.
Once the Quant Winter cleared the three-year lookback at the end of 2023, a clean post-winter reading emerged: the median stood near 32% and has since eased to around 28%. That is a meaningful decline from the peak, and a regime in which generic factors explain considerably less of the variation in manager returns.
Figure 1: Rolling three-year R2 distribution and number of managers in eVestment global quant universe
Problems loading this infographic? - Please click here
Source: eVestment, as at 31 March 2026.
Past performance is not indicative of future returns.
The median, however, conceals a substantial change in the shape of the distribution itself (Figure 2). Before the Quant Winter the distribution was narrow and right-skewed where 56% of managers fell in the 10% to 30% range, and only around 3% exceeded 50%, leaving a small number of systematic outliers in an otherwise idiosyncratic universe.
The winter transformed that picture. Generic factor returns moved sharply and erratically, and because those moves drove so much of the variation in performance, even moderate factor exposures became highly explanatory of a manager’s returns. The distribution widened and flattened: the dominant bin shifted from 10 to 30% up to 40 to 50%, and the share of managers above 50% jumped to around 25%. In effect, the episode sorted the universe by factor style. Managers with heavy generic exposures saw their R2 spike as factor moves overwhelmed everything else, while the few genuinely idiosyncratic managers were left largely untouched.
Since the winter, the distribution has pulled back. The share of managers above 50% has fallen to 5%, and the bulk of the universe now sits in the 20 to 40% range, well below the winter’s peak. At the other end, the share below 10% has recovered from 9.3% during the winter to 15%, so a meaningfully larger group of managers now earns returns that generic Barra factors leave largely unexplained. Therefore, generic factors play a smaller role in driving manager returns than they did at the height of the dislocation.
Figure 2: R2 distribution by period, global quant universe
Problems loading this infographic? - Please click here
Source: eVestment, as at 31 March 2026.
Repositioning, or a calmer market?
The shift in Figure 2 raises a natural question. Does the lower post-winter R2 reflect genuine repositioning, or merely a calmer factor environment? Answering it requires looking at both what managers did and what happened to the factor environment, and Figure 3 covers both. Figures 3a and 3b address the manager behaviour side, with the collective factor betas from a rolling panel fixed-effects regression.
Of the nine Barra factors in the model, Value and Momentum are the most revealing. Value shows the most pronounced shift. It was already a crowded, universe-wide tilt in the mid-2000s, when its collective beta stood above 1. That crowding then unwound. By 2015 the collective Value beta had fallen to around zero, and briefly below it, as the universe diversified away from the factor. It rose again into the Quant Winter, and as the rolling three-year lookback came to span the dislocation the beta returned to around 1 in late 2021, a sign of renewed, universe-wide crowding in the factor that drove the winter. It has fallen back since. The Value beta entered 2024 at 0.63 and had declined to 0.42 by early 2026. It remained statistically significant, but the collective tilt has been more than halved from its winter peak.
Momentum tells a different story. Its collective tilt is far smaller than Value’s and shows none of the same crowding and unwinding, holding mostly between roughly 0.10 and 0.25. Yet despite its smaller size, it has remained statistically significant for most of the period, with t-statistics typically around 3 to 4. Momentum is less a position the universe crowds into and out of than a steady, ever-present tilt.
This steadiness may seem at odds with how the factor is commonly perceived. Momentum is widely regarded as one of the more crowded factors in equity markets, so its consistently lower beta may appear surprising. A large part of that gap, however, is mechanical rather than behavioural. Beta coefficients are not comparable across factors without normalising for each factor’s variance, so any given level of true economic exposure will produce a larger estimated beta for a less volatile factor. The analysis in this section therefore focuses on whether each factor’s beta is rising or falling relative to its own history, rather than on cross-factor comparisons in beta space.
Figure 3: Rolling collective factor betas and t-statistics
Figure 3a. Regression Betas
Problems loading this infographic? - Please click here
Figure 3b. Regression T-statistics
Problems loading this infographic? - Please click here
Source: eVestment, MSCI Barra, as at 31 March 2026.
Figure 3c turns to the other side of the question raised earlier, examining whether the factor environment itself has changed. It plots the trace of the Barra factor covariance matrix, a summary measure of the total variation and co-variation in factor returns. Through the mid-2010s and into 2019, the trace held within a narrow, low range. It then rose sharply as the rolling three-year window absorbed the 2020 shock and the turbulence that followed, peaking in early 2023 at more than three times its pre-winter level. It has since moderated as the most volatile months have progressively dropped out of the window, but even after the window clears the winter years entirely, from around 2025 onward, it remained above its pre-winter baseline. This persistent elevation partly explains why the median R2 in Figure 2 has not fully retraced to its pre-winter trough despite genuine manager repositioning.
To isolate the effect of this shifting environment on R2, Figure 3c also plots a counterfactual, the median R2 recomputed with the factor covariance matrix held fixed at its full-sample average. By removing variation in factor variance from the equation, any remaining movement in this fixed-covariance R2 reflects changes in manager positioning alone. Two features stand out. First, the fixed-covariance R2 moves inversely with the trace over time, rising when factor variance was subdued and falling as managers repositioned. Second, and more importantly, its decline in recent years is larger than that of the standard R2 in Figure 2, precisely because it is no longer partially offset by the elevated factor environment. Together, these observations indicate that the downward trend in R2 is not a statistical artefact of a calmer factor backdrop. It reflects a meaningful reduction in managers’ collective reliance on generic factor exposures.
Figure 3c. Rolling three-year R2 with fixed factor variance-covariance versus rolling three-year trace of variance-covariance matrix
Problems loading this infographic? - Please click here
Source: MSCI Barra, as at 31 March 2026.
Dependence and skill
The universe has reduced its reliance on generic factors. Whether that reliance bears on skill at all is a separate question. In principle the two are unrelated. Factor dependence is a property of risk, and a heavily factor-driven manager could still generate strong returns of its own. Nor does a high R2 rule out skill of a subtler kind: a manager that profitably times its factor exposures might exhibit a high R2 and should still be considered skillful. Across the universe, though, the two prove inversely related (Figure 4). The managers whose risk is most dominated by generic factors carry negative residual alpha, the return left once every factor exposure is stripped out, in every period. They earn less than their factor exposures alone would have delivered, while the least factor-dependent are the ones with positive residual alpha. In this universe, factor dependence has gone hand in hand with weaker residual alpha.
Across the three periods, the high-dependence group’s residual alpha stayed negative throughout, at roughly -1.2% before the winter, -2.4% during it, and -1.2% after. Its total active return, by contrast, swung with the factor environment. The group lagged the least-dependent managers before and during the winter, and pulled ahead only afterwards, earning 2.3% a year against 1.8% as Value and Momentum delivered. That single spell of outperformance rested entirely on the factor tailwind, with no skill beneath it.
This is why we believe residual alpha, not total active return, is the durable measure. As Quant Renaissance Part II established, generic factor returns are highly sensitive to the macro environment, and the tailwinds that flatter factor-dependent managers in one regime can reverse in the next. Residual alpha, earned independently of every factor, carries no such conditionality.
Figure 4: Return decomposition by factor dependence group and period
Figure 4a. High R2 Managers
Problems loading this infographic? - Please click here
Figure 4b. Low R2 Managers
Problems loading this infographic? - Please click here
Source: eVestment, MSCI Barra, as at 31 March 2026.
Size and capacity
The split by factor dependence invites a further question. Beyond their returns, do the high and low R² groups differ in observable characteristics? Tracking error does not distinguish them. Across all three periods the two groups run broadly similar tracking error, within a range of roughly 2.4% to 3.2%, and show no consistent gap between them (Figure 5a). The amount of active risk a manager takes bears little relation to how much of that risk comes from generic factors.
Size shows a clearer difference, and one that changed at the winter (Figure 5b). Before the winter the more factor-dependent managers were the smaller ones, with median assets of around $210 million against $590 million for the least dependent. During the winter that relationship inverted. The high-dependence group’s median assets rose to about $545 million while the low-dependence group’s fell to roughly $200 million, and the gap has persisted since, standing at about $1.8 billion against $1.1 billion after the winter. Larger managers became disproportionately more factor-driven through the crisis and have stayed that way. This suggests that scale itself may act as a constraint on a manager’s ability to maintain an idiosyncratic approach. As assets grow, capacity pushes a manager towards the liquid, high-capacity generic factors and away from the smaller, more differentiated positions that are harder to sustain at scale. Capital discipline may therefore be an underappreciated condition for preserving differentiation.
Figure 5: Characteristics of High R2 managers versus low R2 managers
Figure 5a. Median tracking error by R2 group
Problems loading this infographic? - Please click here
Figure 5b. Median AUM($M) by R2 group
Problems loading this infographic? - Please click here
Source: eVestment, as at 31 March 2026.
An uneven recovery
The trajectory traced so far is a universe-level one, where collective factor dependence rose into the winter, peaked and has receded since. What an average cannot show is how far individual managers diverged along the way. At the manager level the response was bifurcated. Some reduced their reliance on generic factors materially after the winter, so that less of their risk comes from generic exposure and more of their alpha is genuinely their own. Others held onto, or rebuilt, the factor-heavy positioning the winter had exposed. The renaissance did not accrue equally across the universe.
Our own returns show the same shift. As we set out in our earlier work, we reshaped our investment process in the years around the winter. We diversified our alpha sources through alternative data and machine learning, moved from static factor weights to dynamic, macro-aware factor selection, and deliberately reduced our reliance on generic factor signals. Those papers described the change in approach. The question here is whether it shows up in the returns themselves, in how much of our performance a generic factor model can explain and how much it cannot. Figures 6 and 7 trace Man Numeric’s returns through that lens, before, during and after the winter, measured against the 20 largest managers by AUM rather than the universe as a whole, that being the peer group most comparable to us in scale.
In the years before the Quant Winter, our residual alpha ran above the large-peer median while our R2 sat broadly in line with it. A meaningful share of our risk was driven by generic factors, and we earned solid idiosyncratic alpha on top of it. The Quant Winter exposed the weakness in that position. Our R2 climbed with the group and to the top of it, peaking around 55%, so our risk remained tied to generic factors. Our residual alpha, meanwhile, fell to the bottom of the peer group, reaching roughly negative 4% a year at its worst against a large-peer median only modestly below zero. The alpha we had treated as our edge proved less idiosyncratic than it appeared. It was crowded with the rest of the market and unwound alongside it.
After the Quant Winter, the two measures moved sharply, and in opposite directions. Our R² fell below the peer group median, dropping towards zero recently, so that very little of our risk is now driven by generic factors. Our residual alpha rose above the peer median. This is the transformation described in our earlier work, now visible in the returns. We rebuilt our alpha around more diversified and less correlated sources, so that the return we generate is genuinely idiosyncratic rather than a crowded position shared with the rest of the market. At the same time, we reduced the generic factor exposure that had kept our risk tied to the broad market. That the two improved together matters. A manager that had simply become more passive would show a lower R², but stepping back from the market does not raise the alpha earned independently of it. Our risk and our alpha moved together because the source of our returns changed.
Figure 6: R2 and residual alpha, Man Numeric versus top 20 largest managers
Figure 6a. Rolling three-year R2
Problems loading this infographic? - Please click here
Figure 6b. Rolling three-year residual alpha
Problems loading this infographic? - Please click here
Source: eVestment, as at 31 March 2026.
The same shift is visible directly in our factor exposures. Figure 7 traces Man Numeric’s Barra factor betas across the three periods, set against the median of the 20 largest managers. In the years before the Quant Winter we already carried sizeable factor tilts, led by a Value beta of around 1.4, well above the large-manager median of roughly 0.3. Those tilts did not ease as the downturn approached. They rose further, and our Value beta reached about 1.8 during the Quant Winter. After the winter our exposures came down. Our Value beta fell to roughly 0.5, still a positive tilt but much closer to the top-20 median of around 0.35. The rest of the profile moderated in the same fashion. This is what the fall in R2 looks like in the portfolio: with smaller loadings on any single generic factor, less of our risk depends on how those factors behave.
Figure 7: Barra factor betas, Man Numeric vs top 20 largest managers, by period
Source: eVestment, MSCI Barra, as at 31 March 2026.
Parting thoughts: a renaissance, unevenly shared
Reliance on generic factors has receded from its Quant Winter peak, but the average conceals a wide divergence beneath it. Some managers used the winter as a reason to change. Others rebuilt the crowded, factor-heavy positioning it had exposed, and were rewarded for doing so by a few strong years of factor performance.
That divergence is what matters, because the diversification investors look for when they hold several systematic managers depends entirely on those managers being different from one another. When the industry crowds into the same factors, that diversification quietly disappears at the moment it is needed most, as it did during the Quant Winter. Scale may make it harder to escape, as the largest managers became the most factor-driven through the crisis and have stayed that way.
A return-based lens separates managers whose performance is their own from managers whose performance is a shared position on the factor environment, and it does so in advance rather than after the fact. Allocators holding more than one systematic manager have good reason to ask for it. The Quant Renaissance is real, but it is not universal, and it belongs to the managers who did the work to earn it.
Appendix: Our methodology
Skill and style affect returns in different ways, which makes the balance between them measurable. For each manager we estimate the following time-series regression of active return on a standard set of Barra factors.
where rᵢ,ₜ is manager i’s active return in month t, fₖ,ₜ is the return to Barra factor k, βi,k is the corresponding factor loading, αᵢ is the idiosyncratic intercept, and εᵢ,ₜ is the residual. The adjusted R2 of this regression, referred to throughout as R2, is our primary measure of factor dependence, how much of a manager’s active return variation is explained by generic factor premia rather than proprietary insight. The intercept αᵢ is our measure of residual alpha: the return that persists after stripping out all factor exposures.
The regressions are estimated across a universe of systematic equity managers drawn from eVestment, screened so that managers within the group have comparable profiles. We begin with eVestment’s Global Large Cap Core and All Cap Core quantitative universes and retain only products that report gross returns and are benchmarked to the MSCI World Net Total Return Index. We keep core strategies alone, removing any product carrying an explicit style tilt in its name so that differences in factor dependence reflect how managers invest rather than the style bucket they were sold into. Finally, to match the peer group on risk and scale, we require a historical tracking error between 1% and 6% and a peak historical AUM above $100 million. Within this universe, a manager enters a given period’s regression wherever it has at least 24 monthly return observations in that window. The analysis compares three periods: pre-winter, 2010 to 2017; the Quant Winter, 2018 to 2020; and post-winter, 2021 to March 2026. Manager returns are gross throughout, sourced from eVestment. Factor returns and exposures are taken from the MSCI Barra GEMLT model.
A single manager’s R2 can be high even when managers load on entirely different factors, since each simply fits its own model well in isolation. The panel approach strips out manager-specific effects and estimates one set of betas, capturing the exposures common to the universe as a whole. When managers crowd into the same factors at once, that synchronisation shows up directly in the panel betas, making them a more precise gauge of crowding than an average of individual regressions.
Caveats
The pre-winter window in Figure 2 is not uniform. As Figure 1 shows, the rolling median sat at its lowest around 2015 and was climbing by 2016 to 2017, so the single pre-winter distribution in Figure 2 averages over a period in which reliance on generic factors was already building.
One qualification applies to the low end of the distribution in Figure 2. A small number of managers register an R2 close to zero, and in a few cases negative, meaning generic factors explain almost none of the variation in their active return. On inspection, the managers at this extreme are predominantly sustainability-focused strategies, and for them a low R² reflects a deliberate style objective rather than genuinely differentiated skill.
You are now leaving Man Group’s website
You are leaving Man Group’s website and entering a third-party website that is not controlled, maintained, or monitored by Man Group. Man Group is not responsible for the content or availability of the third-party website. By leaving Man Group’s website, you will be subject to the third-party website’s terms, policies and/or notices, including those related to privacy and security, as applicable.