Data and methods

The evidence base behind everything we publish.

One sample, one published definition, and two constructed measures. Every paper, note, and diagnostic the Institute releases is built on what is documented here, and this page is versioned so a citation stays valid.

2,142U.S. venture capital funds
2000–2024Vintage years, more than three market cycles
$125MMedian fund size
1,149 / 993Emerging against established under the definition

Colibrí Institute, 2026 • Moncada, 2026
2,142 U.S.-based venture capital funds, 2000–2024.

Definition: what makes a manager emerging

Emerging managers are defined by systematic exclusion from institutional search processes. The definition builds on NexTier's Why Not the Best? framework and extends it empirically for venture capital. A fund is emerging if it meets any one of the following.

Scale
The firm oversees less than $100 million in total assets under management across all funds.
Sequence
The fund is Fund I, II, or III, regardless of the prior investing experience of the general partners.
Organizational age
The management firm has a vintage of less than five years, reflecting organizational youth rather than fund age alone.
Access
The founding team includes women or otherwise underrepresented general partners, where fundraising outcomes diverge from performance because of documented structural barriers.

The same definition governs both papers, which is what allows the second to extend the first rather than restate it.

Source

Fund-level performance data sourced from PitchBook and analyzed as part of doctoral research (Moncada, 2026).

Version

Evidence base v2, September 2026. The previous version covered 2,471 funds and supports White Paper I.

Reuse

Definitions, measures, and exhibits are free to reuse with attribution.

Glossary

Every term we use, defined once.

Open the glossary

The measures

Two measures, both built from fields commercial databases already carry.

An allocator can run this framework on an existing pipeline without proprietary data. Every measure is set against the fund's own vintage year rather than a fixed benchmark, which removes the J-curve objection structurally rather than by argument.

Return per unit of risk

A fund's IRR divided by how widely IRRs varied across every fund raised in the same year. It asks how much return a manager generated relative to the return actually available in that market that year.

Across the full sample the average fund scores 0.65.

Upside capture

A fund's TVPI divided by the median TVPI of every fund raised in the same year. It asks how far past the typical fund of its cohort a manager reached.

Across the full sample the typical fund scores 1.00, which is what typical means.

Worked example · the story of two funds

MeasureFund AFund B
Net IRR13.9%13.6%
TVPI3.25x2.00x
Return per unit of risk0.640.62
Upside capture2.481.53

Fund A lost fourteen of its twenty companies outright and returned one at sixty times. Fund B lost nothing and returned every company at roughly twice cost. On net IRR they are the same fund. A variance-based measure calls Fund A the riskier one and prices a discount.

Limitations

Stated here and repeated in the body of each paper beside the finding they qualify.

Association, not cause
The design compares funds at a point in time. Every result is an association between manager characteristics and performance.
Coverage is not neutral
Funds that appear in commercial databases are larger, more recent, and higher-performing than those that do not. Visibility correlates with the same access dynamics the research examines, so the documented premium is more likely understated than overstated.
Fund level, not deal level
The cash flow detail required for a true risk-adjusted return calculation is not available in commercial databases.
Uneven generalization
The advantage is clear among general venture and early-stage funds and absent among later-stage funds, where the sample is small.
U.S. only
The sample covers U.S.-based funds. We do not extend the findings to other markets.

Corrections policy

We publish our errata. When a figure, a citation, or a unit is wrong, we correct it, date the correction, and leave the record of what changed. Corrections that affect a headline finding are announced to anyone on the research list.

Corrections to date

None recorded against published work.

Questions about method

hello@colibri.institute