Something Ventured

Why We Are Reopening This File

In August 2020 my partner Adam Day published Nothing Ventured, a defense of venture capital written into the teeth of a pandemic recession. It argued that venture was less risky than investors believed, that it out-earned public equities on every horizon anyone cared to measure, that downturns were the right moment to commit, and that the celebrated barrier of access was largely a myth. Its advice was four words long: emerging, focused, small, early.

Six weeks after we published, the largest venture expansion in recorded history began. Eighteen months later, it ended. What followed was not the shock the paper had prepared for, but a slower and stranger one: a market that kept producing value on paper and stopped producing cash.

Rereading it now is uncomfortable, which is precisely why it is worth doing in public. Some of it has aged extremely well. One central exhibit has inverted outright. And the largest risk in venture investing today is one the paper never named.

The Short Version

• The public-market comparison has reversed. As of 31 December 2025 the Cambridge Associates US Venture Capital Index trails the S&P 500 public market equivalent at three, five, ten and twenty-five years, and trails the Nasdaq at every horizon published. In 2020 that table was positive in every cell.
• The bottom quartile now loses money. The original's most reassuring line — that even the worst funds produced positive returns — does not survive the 2021, 2022 and 2023 vintages, at −3.8%, −5.6% and −13.0%.
• Distributions, not marks, are the binding constraint. The 2008 vintage returned 2.22x in cash; the 2021 vintage has returned 0.04x. Fewer than one in five funds raised 2017–2024 has returned its called capital.
• Access inverted. You still do not need entry to brand-name firms. What changed is that limited partners decided otherwise: nine firms took half of all US venture capital raised in 2024, and first-time fund formation is at its lowest level since 2011.
• Small, focused and disciplined still wins. The structural recommendations survive the new data. What changed is that following them now means going where capital is leaving.
• The new risk is entry price. Seed valuations rose 4.5x between 2019 and 2024 while public software multiples fell by two-thirds — a squeeze from both ends of the trade.

The Timing Problem

The paper's thesis was that a recession was a good moment to commit capital to venture. On the narrow question, it was right almost immediately and more emphatically than we could have wanted. United States venture deal value went from $175.6 billion in 2020 to $358.2 billion in 2021. Exit value went from $340 billion to $864 billion. Limited partners committed $169.7 billion to US venture funds in 2021 and $222.9 billion in 2022 — the two largest fundraising years ever recorded.

And then the window shut. Exit value fell 83% in a single year, to $150.7 billion in 2022, and stayed there for two more. Fundraising fell by more than two-thirds from its peak. Deal value halved.

PitchBook-NVCA Venture Monitor, Q4 2025 edition, published January 2026. Exit value is aggregate reported transaction value and is dominated in any given year by a small number of very large events.

Three things in that picture deserve attention. Fundraising in 2025 fell to $66.1 billion, its lowest level since before 2019 and less than a third of the 2022 peak — limited partners are still retreating even as deal activity recovers. Exit value in 2025 nearly doubled to $297.6 billion, the first genuine sign of a reopening. And the gap between what funds deployed and what they raised has now been negative for three consecutive years, which is a polite way of saying the industry has been spending down a war chest it is no longer replenishing.

The Ledger: Claim by Claim

Eleven propositions from 2020, scored against the data available in September 2026.

Two of the eleven survive intact and unqualified. Four require material restatement. Three are broken. Two can no longer be tested at all, because the data providers stopped publishing the cut. That last category deserves its own comment: a meaningful share of the 2020 paper's evidence has not been contradicted so much as withdrawn. Cambridge Associates no longer splits its venture index by stage. The loss-ratio table has not been refreshed. PitchBook's benchmark reports moved behind a client wall in 2025. An argument built on public benchmark data in 2020 cannot be rebuilt the same way in 2026, and any investor relying on one should know the ground under it has thinned.

The Reversal, and What It Actually Means

The centrepiece exhibit of Nothing Ventured was a horizon-return table showing the Cambridge Associates US Venture Capital Index beating the S&P 500, the Russell 2000, the Russell 3000 and the Nasdaq over one, three, five, ten, fifteen, twenty and twenty-five years. Every cell was positive. The paper's comment was triumphant and, at the time, fair: "the longest bull market in history did not outperform the venture index on any of the time horizons shown."

Here is the same index, six years later.

Cambridge Associates US Venture Capital Index. 2019 figures as cited in Nothing Ventured from the 2019 Q3 US VC Benchmark Book. 2025 figures from the Q4 2025 US VC Benchmark Book, 2,816 funds formed 1981–2024, $764 billion of value. Pooled returns are net of fees, expenses and carried interest.

The one-year, ten-year, fifteen-year and twenty-year numbers are as good or better than they were. The three- and five-year numbers are materially worse. And the twenty-five-year figure fell from 34.4% to 8.1%.

That last collapse is the most instructive number in this entire addendum, and it is not a story about venture performance. In 2019 a twenty-five-year window reached back to 1994 and therefore contained the entire dot-com run-up. By 2025 that window began in 2000 and contained the crash instead. Nothing about the asset class changed. The window moved.

I make this point at Adam's expense and my own, because we both signed off on that table. The discipline it implies is one I now apply to every benchmark I am handed: before asking what a horizon return says about an asset, ask which years it is made of, and what happens when the earliest of them rolls off.

The Comparison That Actually Matters

Strip out the window artifact and the substantive finding remains, and it is not comfortable. The proper comparison is a public market equivalent — the return an investor would have earned putting the same cash flows into an index on the same dates. In 2020 that table was positive against every index at every horizon.

2019 column from the mPME value-add table reproduced in Nothing Ventured. 2025 column computed as the CA US Venture Capital Index pooled return less the S&P 500 mPME return over the same horizon, both from Cambridge Associates' US PE/VC Benchmark Commentary for calendar year 2025.

Venture beat the public market equivalent over one and twenty years, roughly matched it at ten and fifteen, and lost — badly, in one case — over three, five and twenty-five years. Against the Nasdaq it lost at every horizon published. Against small-capitalization equity it won at every horizon. That is the honest restatement of the 2020 claim: venture reliably beats small-cap equity, roughly matches broad large-cap equity over long periods, and has lost decisively to concentrated large-cap technology over the last five years. A much narrower claim than the one we made — and, unlike the original, one likely to survive the next window roll.

The Thing the 2020 Paper Could Not See

Reread Nothing Ventured with a pen and you will notice something: with the single exception of one reference to funds "returning capital," the entire argument is conducted in internal rates of return and total-value multiples. Both are computed from valuations. Neither requires a dollar to have moved.

In 2020 that was a defensible simplification, because for the preceding two decades marks converted into cash on a fairly predictable schedule. A fund carrying a 2.3x total value multiple at year ten had, historically, distributed most of it. That relationship has broken, and it has broken more completely than any other statistic in this paper.

PitchBook Benchmarks, Q4 2024 edition with preliminary Q1 2025 data, North America venture capital, marks as of 31 December 2024. Vintages after 2020 remain young, and a low DPI is expected at that age — the point is the rate of convergence.

The gap between TVPI and DPI is unrealized value: money a limited partner has been told they have but cannot spend. In 2008 it was a sliver. It is now the whole picture.

The obvious objection is age: of course a 2021 fund has distributed little. But the comparison that controls for age is worse, not better. At the three-year mark, 25% of 2017-vintage funds had begun distributing; only 9% of 2021-vintage funds had. At five years, 59% of 2017 funds had begun; only 39% of 2019 funds had. Recent vintages are not merely young. They are converting more slowly than their predecessors did at the same age.

Meanwhile the marks have recovered. The Cambridge index returned 21.1% in calendar 2025, and the 2023 vintage returned 50.5% in that single year, driven overwhelmingly by artificial-intelligence markups concentrated in a small number of names. Median total value multiples have risen for essentially every vintage from 2017 to 2024 over the past six quarters. Cash distributions have not moved with them. The gap is currently widening from both ends at once.

Why This Undoes the Original's Central Argument

Nothing Ventured built its case for risk on a Kahneman and Tversky coin flip: if losses feel twice as heavy as gains, an investor needs a two-times multiple to justify a fifty-percent chance of loss. The paper then produced a 2.4x gross multiple against a 20% loss ratio and declared the trade rationally compelling.

The arithmetic was sound. The omission was time. A coin flip that pays two-to-one resolves when it lands. A venture fund that carries a 1.5x mark for twelve years and then distributes it has offered a materially worse trade than the multiple implies, and one that no loss-aversion framework in the paper captured.

This is the single largest correction I would make to the 2020 paper. It does not argue against venture. It argues that the underwriting question was posed incorrectly. The question is not "what multiple will this fund mark." It is "what has to be true for this fund to convert its marks into cash, and what does that require of the market that I cannot control?"

The Floor Gave Way

"Since 2007," the 2020 paper wrote, "the bottom quartile has produced, on average, positive IRR in every year. This finding should shift the investor notion of risk in venture — the worst performers produced small returns, not zeros."

It was the paper's most quoted line in conversations with prospective investors, and for good reason. It reframed venture from a lottery into an asset with a floor. I should note at the outset that I could not locate the original Cambridge source for that specific claim in any currently public document; whatever it rested on is no longer where the footnote points. But the substance can be tested against current benchmark data, and it fails.

From 2009 through 2020 the floor held, and mostly held comfortably — bottom-quartile managers in the 2012 through 2017 vintages returned between nine and twelve percent annually, which is a respectable result for the worst quarter of any asset class. Then it gave way: −3.8% for 2021, −5.6% for 2022, −13.0% for 2023 (see the vintage table above).

Two honest caveats. These are young vintages carrying J-curve effects, and some marks will recover; and a bottom-quartile IRR is not the same as capital loss. But the direction is unambiguous, and the dispersion around it is wider than at any point in the series.

Which produces an uncomfortable interaction with the original's other big claim. Nothing Ventured argued that manager selection was overrated because dispersion reflected structural features of early-stage investing rather than skill. That argument was made when the downside of picking badly was a small positive return. It reads differently when the downside is minus thirteen percent. The case against worrying about manager selection was always partly a case about a floor that no longer exists.

Access: Right on the Merits, Wrong on the Market

The boldest section of Nothing Ventured attacked the idea that venture returns require access to brand-name managers. It argued three things: that persistence in venture reflects deal access rather than skill; that new and developing funds dominate the top of each vintage; and that an ordinary investor can therefore reach top performance without an invitation to Sand Hill Road.

All three still hold. PitchBook's 2023 study of persistence found that a top-quartile fund is followed by another top-quartile fund only 34.8% of the time against a 25% random baseline, and that predecessor performance explains under two percentage points of successor IRR. Its 2024 emerging-manager study found emerging managers outperforming established ones vintage by vintage from 1997 onward, with the gap reaching 4.5 points in the 2010–2014 cohort.

And then the capital left anyway.

PitchBook-NVCA Venture Monitor, Q4 2024 edition. The NVCA's 2026 Yearbook, using a stricter first-time-fund definition, reports 101 first-time funds launched in 2025 against 457 in 2021 — a decline of 77.9% and the lowest count since 2011. The two series are not directly comparable and are not combined here.

Emerging-manager capital fell from $24.2 billion across 441 funds in 2021 to $5.2 billion across 103 funds in 2024. In the first half of 2025, forty-four new managers raised $1.8 billion between them — less than half of what a single established firm raised in the same period. The count of unique active venture investors in the United States stood at 11,425 as of September 2024, which PitchBook noted was 45.5% of the 2021 figure.

The capital that remained went to the top. Of all US venture capital raised in 2024, nine firms took approximately 50%, the next twenty-one took 25%, and every other firm in the market shared the remaining quarter. A single firm took more than eleven percent.

So the 2020 claim inverts in an unusual way. It is not that the paper was wrong that access is unnecessary. It is that the paper assumed the market would eventually notice. It did not. Limited partners responded to a difficult period by concentrating into the largest brands — the precise behavior the evidence says is unrewarded — and in doing so they made the emerging-manager segment thinner, cheaper and less competitive.

For an investor willing to act against that flow, the position is better than it was in 2020, not worse: the same body of evidence, less competition for allocations, and managers who have had to survive the hardest fundraising market since 2009 to be in business at all. That is a real edge, and it is the one part of the original thesis I would state more strongly today than we did then.

What Aged Well: Small. Focused.

Size

The original relied on a Preqin table showing pooled internal rates of return of 20.0% for funds under $100 million against 2.4% for funds over $1 billion. That specific table has no current successor — I could not find any provider publishing the same four buckets on current data, and I would rather say so than substitute a lookalike. But the finding replicates on every independent cut available.

Carta's analysis of 687 funds between $1 million and $10 million against 210 funds over $100 million found the small funds ahead on most measures, with 2017-vintage median net IRR of 13.8% against 9.8%. A separate study found that 25% of funds under $350 million reached a 2.5x total value multiple against only 17% of funds over $750 million.

The mechanism is not mysterious and has, if anything, strengthened. A fund must return itself before it returns anything. A $50 million fund needs a plausible number of ordinary exits. A $2 billion fund needs outcomes that occur a handful of times a year across the entire market. When the exit environment narrows — as it did from 2022 — large funds are structurally dependent on precisely the events that stop happening.

Focus

The 2020 claim was that sector-focused funds beat multi-industry funds roughly 70% of the time. PitchBook's 2023 study across 1,306 venture funds from the 2000–2020 vintages refines this usefully: across the whole population the difference is not statistically significant, but for funds under $250 million specialists are clear winners on both internal rate of return and total value multiple. Above $250 million, in the most recent 2015–2020 cohort, generalists led by roughly five points.

That is a more interesting result than the original's, and it points the same direction as the size finding. Focus is an edge when it means knowing an industry better than anyone else pricing it. At sufficient scale, focus stops being expertise and becomes concentration — the same bet, larger, with fewer places to put it.

Stage — a Claim We Can No Longer Test

The original cited Cambridge Associates early-stage pooled returns of 19.7% against 11.6% for late stage, and noted that the early-stage index beat every public benchmark on every horizon. Cambridge no longer publishes separate early- and late-stage venture indices; the last edition I could locate carrying that split was the benchmark book as of June 2018. PitchBook's public benchmark reports segment by geography and fund size, not by stage.

So the stage claim is not broken. It is untestable on public data, which for a document of this kind amounts to the same practical conclusion: I would not repeat it as evidence. And in any case, the live question in early-stage venture has moved from when you invest to at what price.

The Risk the 2020 Paper Never Priced

Nowhere in Nothing Ventured does the word "valuation" appear as a risk. This is not an oversight so much as a reflection of when it was written: in 2020, entry prices were the least of a venture investor's problems. That is no longer true, and it is now the variable I would put first.

PitchBook-NVCA Venture Monitor, Q4 2024 edition. These are not peak-of-bubble figures — they are 2024 medians, after the correction.

2026 figure as of mid-year. Other published software indices use different weightings and universes and show somewhat higher absolute levels; all show compression of roughly 60–75% from the 2021 peak.

Put the two together. An investor buying seed equity in 2024 paid four and a half times what a 2019 investor paid, into an exit market that values the resulting company at roughly a third of the 2021 multiple. Compression is arriving from both ends of the same trade, and no amount of company-level execution closes a gap of that size.

Two further pressures compound it. Graduation from seed to Series A collapsed: cohorts formed between the second half of 2021 and the end of 2023 reached Series A at roughly half to sixty percent of the 2018–2019 base rate, recovering only partially by 2024. And the down-round rate has fallen to a three-year low of under fourteen percent — which sounds like good news and is better read as evidence that markdowns are being deferred rather than taken, since it coincides with the slowest distribution environment on record.

I hold this view strongly enough that it now organizes how we invest at Golden Section, and I would rather state the bias than pretend to neutrality: we hold ourselves to a defined entry multiple and decline good companies that clear every other test at a price that requires the exit environment to cooperate. Reasonable investors disagree, and the counter-argument — that in a power-law asset class, price discipline systematically excludes the outcomes that produce the returns — is a serious one that the distribution data does not settle either way. What I would say is that the 2020 paper offered no view on this at all, and that a framework which cannot express a price is not a framework for the market we are now in.

Liquidity and Correlation, Restated

Two arguments that were technically correct and practically misleading.

The Illiquidity Argument Survived the Arithmetic and Failed the Behavior

The original leaned on a 2019 BlackRock study concluding that an investor with annual spending needs below eight percent of assets could carry private allocations as high as sixty percent without breaching liquidity constraints, even under crisis conditions. Nothing has come along to contradict that arithmetic; I found no successor study overturning it.

What the arithmetic did not describe is what allocators would actually do. Between 2022 and 2025 private equity and venture positions rose as a share of portfolios not because anyone bought more but because public markets fell and then private marks stayed put — the denominator effect. Institutions found themselves over-allocated to assets they could not sell at par, and a number of them sold anyway. Yale explored and executed a secondary sale of private equity and venture fund stakes reported at $2.5 to $3 billion in 2025. MIT researched sales and cut a private equity commitment by two-thirds. The University of Illinois considered the same.

The secondary market that absorbed this grew from $85 billion in 2019 to roughly $204 billion in 2025. It did not absorb venture at par.

Jefferies 2025 Global Secondary Market Review. Pricing for limited-partner portfolio sales, averaged across transactions. Funds under five years old priced at 95% of net asset value; funds over ten years old at 73%.

Venture stakes cleared at 78 cents on the stated dollar while buyout stakes cleared at 92. That eighteen-cent difference is the market's own estimate of how much venture marks overstate venture value — and it is a more honest number than anything in a quarterly valuation report.

The restatement, then: an investor who can bear illiquidity and an investor who will are different people, and the 2020 paper modeled the first and met the second. Liquidity planning for venture should assume not only that distributions may be slow but that the escape hatch, if used, costs about a fifth of stated value.

The Correlation Number Was Measuring the Reporting Calendar

The original reproduced a Cambridge table showing venture's correlation with large-cap equity at −0.06 and drew the natural conclusion: venture is a diversifier of unusual power. The table was real. The interpretation was not safe.

Private assets are marked quarterly, by appraisal, with a lag. A measurement process that updates slowly produces a return series that is smooth by construction, and a smooth series shows low correlation and low volatility regardless of the underlying economics. Cliff Asness named this "volatility laundering," and the academic work since has put numbers on it: correcting for appraisal smoothing raises the measured public-market beta of private real estate funds from 0.07 to 0.34 — nearly a fivefold understatement — while the most rigorous venture-specific estimates put startup betas between 1.8 and 3.5 once selection effects in observed valuations are corrected.

None of this means venture offers no diversification. It means the diversification is smaller than an appraisal-based correlation implies, that a meaningful part of what looks like low correlation is simply reporting lag, and that an investor building a portfolio on a −0.06 input is optimizing against a number that partly describes an accounting convention. The 2022 experience — when public equities fell hard and venture marks followed a year later — is what that lag looks like in practice.

The Doctrine, Revised

Four instructions become five. Three of the originals stand. The two additions are the ones the last six years actually taught.

1. Underwrite to distributions, not to marks. The original measured everything in multiples and internal rates of return, both computed from valuations. Ask instead what specific event converts each position to cash, who the buyer is, at what size, and on what timetable — and treat a fund that cannot answer as unpriced rather than as merely early. A mark is a manager's opinion about a future transaction. Distributions to paid-in capital is the only figure in a venture report that cannot be argued with.
2. Buy at a price that survives multiple compression. Entry prices rose roughly fourfold across stages between 2019 and 2024 while public software multiples fell by two-thirds. Every other input to a venture return is a forecast; the entry price is a decision. Underwrite to the exit multiple that prevails now, not the one that prevailed in 2021, and be willing to pass on good companies at prices that require the market to help.
3. Stay small — the arithmetic is unchanged and the environment is worse. Small funds still lead on every current cut of the data. The mechanism has strengthened rather than weakened: a large fund is structurally dependent on the billion-dollar outcomes that became scarce after 2021, while a small fund can be made whole by exits that happen every quarter. Fund size remains the most reliable publicly observable predictor of return dispersion in venture.
4. Back focused managers, and keep the focus honest. Specialization still wins below roughly $250 million of fund size and stops winning above it. But establish what the focus is made of. With artificial intelligence taking two-thirds of all venture dollars, a great deal of what is marketed as sector expertise is a position in a single crowded trade. Durable focus looks like knowing an industry's buyers, budgets and switching costs better than the people pricing the round.
5. Back emerging managers precisely because capital has left them. The performance evidence for emerging managers is stronger than it was in 2020 and the competition for their allocations is a fraction of what it was. First-time fund formation is at a fourteen-year low and nine firms took half of 2024's committed capital. This is the clearest instance in the whole paper of the market moving away from what its own data supports — which is ordinarily where returns come from.

What Could Not Be Verified

The original printed a table that cut against its own argument. This is the equivalent gesture: everything in this addendum that I would not want a reader to rely on without checking.

• The bottom-quartile claim's original source. I could not locate the Cambridge Associates document supporting the 2020 statement that the bottom quartile produced positive IRR in every year since 2007. The claim is tested against current PitchBook data in Section V, but its provenance in the original paper is unverified.
• Capital loss and impairment ratios. No successor to the 2002–2015 Cambridge table appears in public materials. The 20.0% figure cannot be updated and should not be cited as current.
• Early- versus late-stage returns. Cambridge Associates has discontinued the stage-split venture indices. The last public edition I found carrying the split was as of 30 June 2018. The 19.7%/11.6% comparison cannot be refreshed.
• Performance by fund size on the original buckets. No provider currently publishes the four Preqin buckets the 2020 paper used. The small-fund finding rests on differently constructed Carta and third-party analyses, not on a refresh of that table.
• Pooled returns for funds above $250 million. That section of the PitchBook benchmark document could not be read reliably across repeated attempts, and is excluded rather than reported.
• Median and average US venture fund size. The series the 2020 paper charted appears to have undergone a definitional change: the same chart title now reports medians three to four times lower, almost certainly because far more micro funds are captured. The series is not extended here because the two ends are not comparable.
• Micro-venture firm counts and global deal volume. No current equivalent to the 2020 paper's "over 900 micro VC firms" could be sourced, nor could the origin of that figure. The original's global deal-volume series could not be matched to any publisher's current numbers either, so this addendum uses a single consistent United States series rather than splicing a continuation onto it.
• Endowment venture allocations. The NACUBO-Commonfund study no longer breaks out venture capital as a standalone line. The 2020 paper's Yale comparison — 21.5% to venture alone — has no like-for-like successor.
• Benchmark reports behind a client wall. Every PitchBook Benchmarks edition after Q4 2024 is now gated. The vintage-year figures here are as of 31 December 2024, and more recent marks — which, given 2025's markups, are likely higher on TVPI and broadly unchanged on DPI — were not available to verify.

Principal Sources

• Cambridge Associates, US PE/VC Benchmark Commentary: Calendar Year 2025 (July 2026); US Venture Capital Benchmark Book, Q3 and Q4 2025.
• PitchBook, Q4 2024 PitchBook Benchmarks with preliminary Q1 2025 data, North America venture capital; Allocator Solutions studies on persistence (Aug 2023), emerging managers (Apr 2024) and specialists versus generalists (Sep 2023, updated Jun 2026).
• PitchBook-NVCA Venture Monitor, Q4 2024, Q4 2025 and Q1 2026 editions; National Venture Capital Association 2025 and 2026 Yearbooks.
• Carta, VC Fund Performance Q4 2025 and Q1 2026; State of Private Markets Q4 2025 and Q1 2026; VC fund size and performance, April 2025.
• SaaS Capital Index, median enterprise value to revenue for public B2B software companies, 2021–2026.
• Jefferies, 2025 Global Secondary Market Review; Setter Capital Volume Report; Evercore H1 2026 Secondary Market Review.
• AQR / C. Asness, Volatility Laundering (2023); Couts, Gonçalves and Rossi, Unsmoothing Returns of Illiquid Funds, RFS 37(7) 2024; Korteweg and Sorensen, Risk-Adjusting the Returns to Venture Capital, NBER 19347.
• BlackRock Investment Institute, The core role of private markets in modern portfolios (March 2019) — the study cited in the original paper.
• A. Day, Nothing Ventured, GSV Investors Management, August 2020 — the paper this document updates.