Skip to main content

Sources & evidence

The full citation pack.

Every numeric claim made by the SWG Architect portfolio-sizing tool traces back to one of the published sources below. Titles, dates, findings and URLs are reproduced verbatim from the SWG Architect Portfolio-Sizing Methodology & Citation Pack (verified 3 June 2026).

Past performance is not a reliable indicator of future results. Capital is at risk; you may get back less than you invest. This is not investment advice.

List A — Sources supporting the sizing thesis

Tier 1–3 sources. Relevance is graded honestly. No source states the literal "30, ideally 60–80" band; that range is an interpolation off curves and thresholds these sources do report. "Direct" means the source directly addresses how portfolio size affects returns; it does not mean it endorses the exact numbers.

A1

How Portfolio Size Affects Early-Stage Venture Returns

Nigel Koh & Abraham Othman, PhD — AngelList Data Science

Date:
April 2020
Tier:
Tier 2 (institutional data-science research; not a fund marketing page)

Across 10,665 real LP portfolios, median IRR p.a. was 11.9% for investors with >50 investments vs 2.9% at ≤50; 11.1% (>25) vs 2.0%; 11.5% (>100) vs 4.1%. ~90% of investors with ≥90 investments were 'in the money' vs under 50% with ≤3. Each extra company added ~9.0 bps median / ~6.9 bps mean p.a. Strongest empirical anchor for 'more positions → higher and more reliable returns.'

Open sourcePDF loads in full.

A2

Thiel vs. AngelList. Who is Right?

David Coats — Correlation Ventures ('VC by the Numbers')

Date:
29 April 2022
Tier:
Tier 2 (named VC firm, ~$475m AUM, publishing its proprietary database analysis)

Monte Carlo on tens of thousands of portfolios (US, vintages since 2000, sizes 1–100): at 8 positions the chance of a fund-level loss (<1×) was 28% vs 6% at 100; chance of a 5×+ fund was 10% at 8 vs 2% at 100; median converges to mean by ~100. Firm's framing: expected (mean) return is roughly equal across sizes — only the risk shape changes.

Open sourceHosted on the firm's own channel; included under the institutional-self-publishing carve-out.

A3

US PE/VC Benchmark Commentary: Calendar Year 2024 (with the firm's investment-level dispersion analysis)

Cambridge Associates LLC

Date:
Benchmark commentary published Nov 2025 (CY2024 data)
Tier:
Tier 2 (the most-referenced institutional VC benchmark)

The US Venture Capital Index returned 6.2% in 2024 (rebounding after negative 2022–23) and has outperformed public markets over long horizons; the index is built from ~2,537 US VC funds (1981–2024). Cambridge's dispersion work shows median fund IRRs cluster around ~10% (net) while top-quartile upside and loss risk vary widely by strategy — VC carrying the widest top-to-bottom spread of any private strategy.

Open source

A4

UK Venture Capital Financial Returns 2025

British Business Bank (analysis of PitchBook, Preqin, survey of 50 UK GPs, and Bank data)

Date:
2 December 2025
Tier:
Tier 3 (government-affiliated; the BVCA CEO is quoted in the release)

UK VC funds (2002–2020 vintages) returned a pooled TVPI of 1.84 (US 1.95; rest of Europe 1.85); 2020–2023 UK vintages (1.22) edged the US (1.14). Crucially for sizing: VC showed the highest dispersion of any private asset class — an upper-quartile TVPI of 2.15 — and only 8% of UK funds reached TVPI ≥3 versus 13% (US) / 14% (Europe). The Bank's own conclusion: 'dispersion remains high in VC, so manager selection and vintage timing are important.'

Open source

A5

Siding with the Angels: Business Angel Investing — Promising Outcomes and Effective Strategies

Robert E. Wiltbank, PhD (Willamette) for NESTA & the British Business Angels Association

Date:
May 2009
Tier:
Tier 3 (government-affiliated, academic authorship)

158 UK angels, £134m, 1,080 investments, 406 exits: 56% of exits returned <capital while the 9% returning >10× produced ~80% of all positive cash flows; overall 2.2× / ~22% gross IRR over ~4 years; ≥20 hours of diligence reduced failures. Honesty flag: establishes the concentration (9% → ~80%) but does NOT prescribe a position count.

Open sourcePDF loads in full.

A6

Returns to Angel Investors in Groups

Robert Wiltbank & Warren Boeker — Ewing Marion Kauffman Foundation & Angel Capital Education Foundation

Date:
November 2007
Tier:
Tier 3 (foundation research, academic authorship)

~1,137 angels, >1,100 exits: average 2.6× in 3.5 years (~27% IRR), heavily skewed (many losses, occasional very large wins). Re-analysis (DeGennaro & Dwyer, European Financial Management, 2014) estimates expected returns ~70% over the risk-free rate, with large variance.

Open sourceAbstract / citation live; full PDF via SSRN download. DOI 10.2139/ssrn.1028592.

A7

Startup Growth and Venture Returns

Abraham Othman, PhD — AngelList Data Science

Date:
December 2019
Tier:
Tier 2

Fitting AngelList data, winning seed investments after ~5 years draw from an α < 2 power law (unbounded mean), so an investor raises expected return by indexing into 'every credible deal' at seed — a result that does NOT hold at later stages. Lays out the three regimes that govern the whole debate.

Open sourcePDF loads in full.

List B — Theoretical & foundational sources

B1

The Power Law: Venture Capital and the Making of the New Future

Sebastian Mallaby — Penguin Press / Penguin Random House, 496 pp.

Date:
February 2022
Tier:
Tier 4 (book, recognised publisher)

ISBN 9780525559993 (US hc); 9780525560005 (ebook); 9780141988948 (UK paperback). The organising thesis — most venture bets fail while a tiny number succeed at a scale that more than compensates, and this asymmetry drives the whole VC model — is set out in the Introduction and recurs through the Sequoia, Kleiner Perkins and Accel chapters. Page numbers differ between US hc and UK pb; cite by chapter. Mallaby argues the disposition the power law demands (back outliers, let winners run); he does not prescribe a numeric position count.

Open source

B2

The Risk and Return of Venture Capital

John H. Cochrane — Journal of Financial Economics 75(1), 3–52 (NBER WP 8066)

Date:
January 2005
Tier:
Tier 1 (peer-reviewed)

Selection-bias-corrected MLE gives an arithmetic mean return ~59%/yr but SD ~100% and a heavily right-skewed (lognormal) distribution; Cochrane describes VC investments as option-like — small chance of a huge payoff, high chance of total loss — and notes only sufficiently large VCs can 'effectively diversify.'

Open sourceOpen NBER PDF.

B3

Power-Law Distributions in Empirical Data

Aaron Clauset, Cosma Rohilla Shalizi, M.E.J. Newman — SIAM Review 51(4), 661–703

Date:
2009
Tier:
Tier 1 (peer-reviewed)

The principled method — maximum-likelihood estimation of the scaling parameter plus a Kolmogorov–Smirnov goodness-of-fit test — for detecting and validating power laws; warns that common least-squares fitting is biased. This is the method the AngelList papers rely on.

Open sourceSIAM. DOI 10.1137/070710111.

B4

Private Equity Performance: What Do We Know?

Robert S. Harris (Virginia), Tim Jenkinson (Oxford), Steven N. Kaplan (Chicago) — Journal of Finance 69(5), 1851–1882 (NBER WP 17874)

Date:
October 2014
Tier:
Tier 1 (peer-reviewed)

Using a research-quality Burgiss dataset of ~1,400 US buyout and VC funds, buyout has consistently beaten public markets (outperformance vs the S&P 500 averaging 20–27% over a fund's life); VC beat public equities in the 1990s but lagged in the 2000s — cross-validated against Cambridge Associates, Preqin and Venture Economics.

Open sourceOpen NBER PDF.

List C — Counter-evidence & qualifiers

The thesis is robust as a downside-management argument for a broadly diversified strategy. It is weaker (a) as an expected-return claim and (b) for an active value-add manager.

C1

The expected-return caveat lives inside the pro-diversification camp

Correlation Ventures (see A2)

Date:
29 April 2022
Tier:
Counter-evidence / qualifier

Correlation Ventures — the firm demonstrating diversification's benefit — also states that expected (mean) returns are the same across portfolio sizes; larger portfolios cut loss probability and cut home-run probability, and average deal quality tends to fall as size grows. Implication: a 'higher multiple at 60–80' claim is the exposed one; 'lower loss probability + higher median' is defensible.

Open source

C2

Diversification in Private Equity Funds: On Knowledge Sharing, Risk Aversion, and Limited Attention

Mark Humphery-Jenner — Journal of Financial and Quantitative Analysis (JFQA)

Date:
2013
Tier:
Tier 1 (peer-reviewed). Full text paywalled; Cambridge abstract live.

Beyond a point, finite partner attention/monitoring dilutes the value-add that drives returns — over-diversification can erode fund-level returns ('limited attention').

Open sourceCambridge abstract live; full text paywalled.

C3

Diversification, risk, and returns in venture capital

Buchner, Mohamed & Schwienbacher — Journal of Business Venturing 32(5), 519–535

Date:
2017
Tier:
Tier 1 (peer-reviewed). DOI 10.1016/j.jbusvent.2017.05.005. Full text paywalled.

Diversification across industries can RAISE VC returns — but via enabling riskier, higher-upside bets, not via simple variance reduction; a specialisation literature (Gompers, Kovner & Lerner) cuts the other way. Net: the academic picture is mixed; avoid a clean 'more = higher multiple' claim.

Open sourceScienceDirect / DOI live; full text paywalled.

C4

The α<2 assumption is contested, and it changes the claim

AngelList (A7) vs Correlation Ventures (A2)

Date:
Tier:
Counter-evidence / qualifier

AngelList (A7) estimates α<2 (more positions raises the mean); Correlation's flat-mean result (A2) implies 2<α≤3 (more positions raises the median, not the mean). If UK EIS sits at α>2 — which the flat-mean evidence makes plausible — the honest benefit of 30→60–80 is narrower dispersion and a higher median / lower loss probability, not a higher expected multiple. Frame it that way.

Open source

C5

'Credible deals' and access are load-bearing

AngelList (A1, A7); SyndicateRoom (deal-sourcing side)

Date:
Tier:
Counter-evidence / qualifier

The indexing result (A1/A7) holds only if the investor can sample the true market distribution; diversifying within a weak or adversely-selected funnel will not replicate market returns. For the tool: position count is necessary but not sufficient — deal-source quality/breadth must be scored alongside it, or the diversification claim is overstated.

Open source

C6

Even the originator now de-emphasises the multiples

SyndicateRoom

Date:
Current canonical research framing
Tier:
Counter-evidence / qualifier

SyndicateRoom has retired the pages stating 3.7×/4.7× and moved to a variance/probability framing ('50/year to minimise variation'). This corroborates C1/C4: the durable claim is risk-shape, not headline return.

Open sourceCurrent canonical research home (open).