Introduction
Private equity (“PE”) returns have weakened over the past several years, and distributions in particular have remained depressed. The funds formed between 2020 and 2022, a period that absorbed a substantial share of all the capital committed to the asset class this century, now rank among the weakest vintages of the past two decades.[1] Faced with these figures, many limited partners (“LPs”) have begun to reconsider the pace at which they committed to that cycle, and some have started to pull back on new commitments.
The historical record offers a reason for caution before acting on that instinct. Weak vintages are typically the ones that deployed capital into crowded, expensive markets. The vintages that follow them deploy into the aftermath, when pricing has softened and opportunities in the dealmaking environment have improved. Retrenching after a disappointing cohort therefore tends to mean pulling back at precisely the point in the cycle when conditions are turning more favorable. This is the central pattern we examine in this article: a tendency for PE returns to mean revert at the vintage level.[2]
The record also points to a second distinction. A fundraising boom does not carry the same meaning in venture capital (“VC”) as it does in the buyout industry. When VC fundraising surges, data suggest the returns of the following vintage compress in a fairly direct way; but when buyout fundraising surges, the same signal is weaker. This asymmetry has practical implications for LPs.
We examined a version of this question in a recent essay for State Street.[3] Here, we extend that work, validate the results against a second data source (Cambridge Associates), and focus on the distinction that matters most for allocation decisions: which fundraising signals seem more trustworthy than others, and how much weight to place on them.
The cyclical pattern in vintage year returns
Our primary performance measure is pooled total value to paid-in capital (“TVPI”) by vintage year.[4] TVPI is a multiple-of-money ratio that divides the distributions a vintage’s funds have returned, plus the remaining unrealized value still held, by the capital those funds have called. It captures both realized and unrealized returns in a single figure. Plotted across vintages, TVPI clearly exhibits cyclical features.
VC TVPIs peak twice: 5–6x in the 1994–1996 cohorts during the dot-com run-up, followed by a collapse through 1999–2002; and a lower second crest near 2010–2012 before normalizing. The buyout series is cyclical as well, but its cycle does not move in step with VC’s. Through the dot-com era, the two ran in nearly opposite directions. Buyout TVPIs sagged to roughly 1.3–1.7x across the 1995–1997 vintages, just as VC was cresting, and then recovered in the 1999–2003 cohorts, which deployed into the post-bubble environment that was crushing the same years’ VC vintages. From 2000 through the mid-2010s, buyout vintage TVPIs held within a narrow band of roughly 1.6 to 2.4x, before decreasing after 2017, falling to about 1.3x by the 2021 and 2022 cohorts.
These patterns have also been noted in the academic literature. Harris, Jenkinson, Kaplan, and Stucke (2023) report average multiples of invested capital by vintage year for Burgiss funds, and their figures show the same shape: VC averages 5.4–6.7x across the 1993–1996 vintages and below 1.4x across 1999–2002, while buyout stays within a 1.25–2.07x band over the whole 1994–2015 window. They likewise observe that VC vintage-year performance has been considerably more variable than buyout’s.[5] The lack of synchronization between the two series is itself consistent with the mechanism we describe below: each asset class cycles around its own capital flows and entry conditions, and the capital that crowded into venture in the late 1990s was not crowding into buyout to the same degree at the same time. What the two share is a common structure, and in both asset classes the performance spread across vintages is far wider than a stable long-run mean would imply.
Put another way, if vintage TVPIs were independent draws around a fixed (long-term) mean performance, there would be no visible performance patterns, and the correlation between vintages would be statistically indistinguishable from zero at every horizon. A mean-reverting cycle, on the other hand, leaves a distinctive statistical signature: returns exhibit short-run momentum, then reverse over a roughly four-to-five-year horizon as capital conditions shift.
To test for mean-reversion directly, Figure 2 plots autocorrelation, or the degree to which the returns of one vintage are correlated with the returns of vintages that came before it.
At a one-year lag the correlation is strongly positive in both asset classes, at roughly 0.75 to 0.8. This means that neighboring vintages tend to land in similar territory in terms of performance. The correlation fades to near zero by the third year and turns negative once the gap reaches four or five years. This is the signature of mean reversion, not the flat line of low correlations that independent draws would produce. The pattern is clean and statistically significant in buyout, and directional (if less decisive) in VC.
Why it happens
The mechanism behind this pattern is established in the academic literature, and it is worth covering briefly. In any given vintage, the supply of investable opportunities is approximately fixed in the short run, while demand scales with the capital available to deploy across the cohort. When recent vintages have performed well, LP appetite expands, fund sizes grow, and the additional capital competes for a finite set of deals. Entry multiples rise, leverage stretches, and general partners (“GPs”) are pushed toward deals they would ordinarily have passed over in order to deploy capital. The following cohort inherits that entry environment regardless of any individual manager’s skill, and pooled returns compress. The reverse holds when capital retrenches: pricing softens, the marginal deal improves, and the next cohort deploys into a more favorable set of opportunities.
This is the “money chasing deals” phenomenon, and each step of that sequence has empirical support. Kaplan and Strömberg observe that buyout activity tends to surge and contract in recurring cycles.[6] Gompers and Lerner tie inflated VC entry valuations directly to surges in fund inflows, and Harris, Jenkinson, and Kaplan identify an inverse link between the capital committed in a given vintage and the returns that follow.[7] More recently, Bhardwaj, Gupta, Howell, and Zimmerschied trace the causal link from capital supply to performance using donation windfalls at private universities, which flow through endowments into larger fund commitments. The funds that grow larger as a result go on to deliver lower returns, suggesting that the expansion of capital itself can depress performance.[8]
The mechanism runs differently in VC and buyout
“Money chasing deals” is at work in both asset classes, but it shows up more clearly in the VC data. We tested it by examining whether capital raised in a given vintage predicts the following vintage’s returns.
Here the two asset classes diverge. VC shows a strong negative relationship: more fundraising in a given year reliably predicts compressed returns the following year (slope of roughly −0.68, explaining about a third of the variation in next-vintage TVPI). The buyout relationship runs in the same direction but is both weaker and less significant. It has a slope of roughly −0.11, accounting for about a sixth of the variation (R² = 0.16), with a confidence interval running from −0.18 to just above zero. The point estimate is negative and non-trivial, but because the interval still includes zero, the buyout signal is suggestive rather than statistically reliable. The contrast is one of magnitude and confidence: the VC effect is roughly six times larger and cleanly distinguishable from zero, while the buyout effect is directionally consistent but cannot be pinned down with the same precision.
The reason is perhaps structural. VC funds bid primarily against other VC funds for a finite supply of early-stage equity. The marginal source of demand on any given deal is the dry powder held by peers, and that dry powder is close to a direct function of recent fundraising. A fundraising surge therefore translates mechanically into entry-valuation inflation in the subsequent vintage.[9] Buyout funds bid not only against other buyout firms but also against strategic acquirers and, at times, public market investors; and buyout pricing reflects credit conditions at least as much as fundraising. Axelson, Jenkinson, Strömberg, and Weisbach (2013) find that economy-wide credit conditions, rather than firm characteristics, are the primary driver of buyout leverage, and that more accessible credit feeds directly into higher prices paid and weaker subsequent fund returns.[10] A megafund deploying into a high-rate, tight-credit environment cannot lever up the way it could in 2006, regardless of how much committed capital it holds, because buyout leverage is procyclical and contracts when debt markets tighten. As the Kaplan-Strömberg characterization of buyout cyclicality suggests, the buyout cycle mixes fundraising with debt market conditions, public market valuations, and broader macro factors, and that mix dilutes the fundraising signal in the data.[11]
The practical implication for LPs is concrete: a VC fundraising surge carries more information about subsequent-vintage returns than a buyout fundraising surge of comparable magnitude. The same record fundraising headline may warrant more caution in one asset class than the other.
How robust is the cross-sectional relationship?
The question remains whether the contemporaneous relationship between capital raised in a vintage and that same vintage’s returns reflects something real or is merely a statistical distortion caused by a few unusual years. Across the full 1990–2022 sample it is negative and statistically significant in both asset classes (buyout slope ≈ −0.19, R² = 0.38; VC slope ≈ −0.63, R² = 0.31), in the direction the mechanism predicts. Note that this concurrent cross-section is a different test from the lagged predictive relationship above; it serves as a robustness check on the contemporaneous link, and its significance across asset classes does not need to match the predictive result.
One concern is that the dot-com era of the 1990s, with thin fund universes and extreme VC outcomes, may drive the result. To test this, we re-estimate on the 1998-onward subsample.
The relationship holds. For buyout it is essentially unchanged and remains significant (slope ≈ −0.21, confidence interval still excluding zero). For VC it attenuates but stays significant, if only marginally, with a confidence interval that nearly touches zero (slope ≈ −0.48). The negative association is therefore not merely a feature of the 1990s, but survives in the modern sample.
What the cross-section cannot do is tell an LP how much to commit to any single vintage. The VC relationship is the noisier of the two outside the 1990s, consistent with the small number of vintage observations and the large idiosyncratic, deal-level variation that Korteweg and Sorensen (2017) document.[12] The cross-section therefore establishes the direction of an effect with reasonable confidence, but it does not support fine calibration of commitment size.
Two historical precedents
The current cycle has a reasonable historical analogue in each asset class. For buyout, the clearest precedent is the 2005–2008 cohort, which deployed at the tail end of a record fundraising surge and into entry multiples enabled by exceptionally loose credit conditions. The Axelson et al. dynamic described previously played out in this period: the credit environment set both leverage and entry prices, and the cohorts that borrowed most went on to return the least.[13] For VC the analogue is the 1999–2001 cohort, which produced the deep trough visible in Figure 1 by combining record fundraising with collapsing exit conditions.
The 2020–2022 vintages share features of both precedents: a sustained low-rate environment that supported buyout multiples on the credit side, and a tech-led valuation surge that pushed VC entry pricing higher into 2021.[14] Their early performance is tracking accordingly. Realized distributions, measured as distributions to paid-in capital (“DPI”), are already running below age-matched benchmarks in both asset classes, as demonstrated in Figure 6[15] and reinforced by prior research.[16]
As in 2005–2008 and 1999–2001, the entry conditions and the resulting drag were not fully visible in real time, and neither precedent cleared within a single subsequent vintage.
Implications
Vintage cohort returns in US PE show a cyclical structure rather than random-walk behavior. Returns are positively autocorrelated in the short run and negatively autocorrelated at four-to-five-year horizons. The shape matches the capital-flow mechanism: crowded cohorts compress the returns that follow, and leaner subsequent cohorts set up the recovery.
The cross-sectional link between vintage year capital and returns is negative and significant across the full sample, and it is not merely a product of the extreme dot-com years. When the 1990s are excluded, the buyout estimate is nearly unchanged, and the VC estimate weakens but keeps its sign and remains significant. The slopes, however, explain only a minority of the variation in vintage returns. The relationship is therefore useful for reading where the cycle sits, but it cannot calibrate the size of any single commitment, and it moves expected returns by only a fraction of a turn against a much noisier backdrop. Over the cycle, the case for cutting exposure to a disappointing vintage is weak, because those cohorts are the ones deploying as conditions improve. A pacing plan that holds steady regardless of how the latest vintage has printed should outperform one that chases recent results.
The 2020–2022 vintages are a case in point. They now fall inside the estimation sample, at TVPI levels consistent with the negative autocorrelation that follows the strong 2016–2018 cohorts, and their realized distributions are already running below precedent. A turn in relative performance, if it comes, is more likely several years out than imminent.
Appendix: data and methodology
This analysis uses Cambridge Associates’ Q3 2025 benchmark calculator, with pooled TVPI since inception (capital-weighted, net to LPs) as the performance metric and Total Capitalization (aggregated committed capital across the benchmark universe) as the vintage-capital measure. The sample covers US buyout and US VC funds vintaged between 1990 and 2022. The 1990 floor removes the thin and noisy fund counts of the asset class’s early years, and the 2022 ceiling reflects TVPI immaturity for funds still in the first several years of life.
Vintage capital is entered in logs for two reasons. First, total capital across the VC and buyout funds in the dataset grew from roughly $1.5 billion in 1990 to roughly $212 billion in 2022, so a linear specification would be dominated by the largest recent vintages and would impose a constant dollar-for-dollar effect with no economic interpretation across such different scales. Second, the proposed mechanism of capital chasing a finite set of opportunities operates on relative rather than absolute scale, so a doubling of vintage size is the economically meaningful comparison. On that basis, a doubling of vintage capital is associated with roughly a 0.13x decline in buyout TVPI and a 0.44x decline in VC TVPI in the full sample, comparable to the State Street estimates (approximately 0.10x and 0.27x). The fact that a different benchmark universe and a different vintage window produce similar results suggests the pattern is not an artifact of any single data provider’s sample.
Three relationships are estimated: the serial correlation of vintage pooled TVPI at lags one through five; the concurrent cross-sectional relationship between log capital raised and same-vintage TVPI; and the lagged relationship between log capital raised in vintage t and TVPI of vintage t+1, which more directly tests the “money chasing deals” mechanism. For each regression of pooled TVPI on log vintage capital, point estimates of the slope, intercept, and R² are obtained from a single OLS fit on the full sample. Considering the roughly 33 vintage observations per asset class, sampling uncertainty in the slope is characterized by a pairs bootstrap: in each of 1,000 iterations, n vintage-year observations are drawn with replacement from the cleaned sample, with (log capital, TVPI) pairs kept together to preserve their joint distribution, and an OLS slope is computed on the resampled data. The reported 95% confidence interval is the 2.5th–97.5th percentile range of the resulting empirical distribution of slopes.
Because the bootstrap resamples vintages independently, it does not account for the serial dependence documented in the autocorrelation results (Figure 2). The reported intervals should therefore be interpreted as a lower bound on the sampling uncertainty in the slope.
[1] Global Private Equity Report 2025, 2025. https://www.bain.com/insights/topics/global-private-equity-report/2025/.
[2] David T. Robinson & Berk A. Sensoy, 2016. “Cyclicality, performance measurement, and cash flow liquidity in private equity,” Journal of Financial Economics, vol 122(3), pages 521-543.
[3] Lerner, Josh. “Cycles in Capital: Vintage-Level Mean Reversion in US Private Equity.” Private Capital Indices | Publications, June 22, 2026. https://globalmarkets.statestreet.com/portal/peindex/publications/4d99a1f4-ceb8-4d83-9949-1b275506f193.
[4] Except where noted, performance is measured as of September 30, 2025, the most recent Cambridge Associates benchmark date with data for buyout and venture. Figure 1 displays the 2023 and 2024 vintages for context, but these cohorts are excluded from the statistical analysis (which ends with the 2022 vintage) because their TVPIs are not yet mature. See the appendix for details.
[5] Robert S. Harris et al., “Has Persistence Persisted in Private Equity? Evidence from Buyout and Venture Capital Funds,” Journal of Corporate Finance 81 (August 2023): 102361, https://doi.org/10.1016/j.jcorpfin.2023.102361.
[6] Kaplan, Steven N., and Per Strömberg. 2009. “Leveraged Buyouts and Private Equity.” Journal of Economic Perspectives, 23 (1): 121–46.
[7] Gompers, Paul, and Josh Lerner. “Money Chasing Deals?: The Impact of Fund Inflows on Private Equity Valuations.” Journal of Financial Economics 55, no. 2 (February 2000): 281–325; Harris, R.S., Jenkinson, T. and Kaplan, S.N. (2014), Private Equity Performance: What Do We Know?. The Journal of Finance, 69: 1851-1882. https://doi.org/10.1111/jofi.12154
[8] Bhardwaj, Abhishek, Abhinav Gupta, Sabrina T. Howell, and Kyle Zimmerschied. 2025. “Does Fund Size Affect Private Equity Performance? Evidence from Donation Inflows to Private Universities.” Working paper.
[9] Gompers, Paul A., and Josh Lerner. 2000. “Money Chasing Deals? The Impact of Fund Inflows on Private Equity Valuations.” Journal of Financial Economics 55 (2): 281–325.
[10] Axelson, Ulf, Tim Jenkinson, Per Strömberg, and Michael S. Weisbach. 2013. “Borrow Cheap, Buy High? The Determinants of Leverage and Pricing in Buyouts.” Journal of Finance 68 (6): 2223–67. https://doi.org/10.1111/jofi.12082.
[11] Kaplan, Steven N., and Per Strömberg. 2009. “Leveraged Buyouts and Private Equity.” Journal of Economic Perspectives 23 (1): 121–46.
[12] Korteweg, Arthur & Sorensen, Morten, 2017. “Skill and luck in private equity performance,” Journal of Financial Economics, Elsevier, vol. 124(3), pages 535-562.
[13] Axelson et al. (2013), cited above.
[14] Fredrik Dahlqvist, Alastair Green, Paul Maia, Connor Mangan, Alexandra Nee, David Quigley, Aditya Sanghvi, Rahel Schneider, John Spivey, and Brian Vickery. 2024. “Global Private Markets Report 2024: Private Markets in a Slower Era.” McKinsey & Company. March 28, 2024. https://www.mckinsey.com/industries/private-capital/our-insights/global-private-markets-report-2024.
[15] The benchmark DPI for the 2020, 2021, 2022 vintage funds is calculated as the average pooled DPI for 2010 to 2019 vintage funds of the same age. For example, 2020 vintage funds are currently six years old as of the most recent data published by Cambridge Associates (and we mark their age in 2020 as 1). To determine the benchmark DPI for 2020 vintage funds at 6 years old, we calculated the average pooled DPI for six-year-old funds using annual historical pooled DPIs of six-year-old funds from vintages 2010 to 2019. Calculations performed using Cambridge Associates data as of December 31, 2025 (accessed July 14, 2026), one quarter later than the data used elsewhere in this article.
[16] Lerner, Josh. “Cycles in Capital: Vintage-Level Mean Reversion in US Private Equity.” Private Capital Indices | Publications, June 22, 2026. https://globalmarkets.statestreet.com/portal/peindex/publications; PEI Staff. “New Report: Private Markets Performance Data.” Private Equity International. February 9, 2026. https://www.privateequityinternational.com/new-report-fund-performance-data/.
