Peer Group Construction: Why Your Benchmark Might Be Comparing Apples to Oranges

Data sourced from Dakota Private Markets, the private fund performance platform powered by Dakota. Learn More | Request Access

The Shift

A fund can be top-quartile in one benchmark and second-quartile in another using the same cash flows and the same strategy label. The difference isn't performance. It's how the peer group was built. Dakota's own benchmark data shows how much a single classification choice can move the numbers: top-quartile net IRR for US private equity funds has ranged from 11.5% for the 2005 vintage to 27.3% for the 2023 vintage (Dakota Marketplace Performance Benchmarks, data as of June 30, 2026). Vintage year alone produces that spread. Add inconsistent strategy filters, sector mixing, and peer groups too thin to be reliable, and "top quartile" starts to mean whatever the peer group was built to show. This post breaks down where peer group construction goes wrong and what to check before citing a benchmark, whether you're the one being ranked or the one doing the ranking.

The Data

Vintage year is the variable every peer group anchors to, and the swings it produces are the starting point for everything else. But vintage alone isn't a finished peer group, it's a first filter. A properly constructed peer group layers several more:

Classification Layer	What It Controls For	What Happens If Skipped
Vintage year	Market timing and entry conditions	Funds deployed in different cycles get compared as if conditions were equal
Strategy (buyout, growth, VC)	Return profile and risk structure	Growth-stage multiples get benchmarked against buyout leverage-driven returns
Sector of underlying portfolio companies	Exit multiples and holding-period dynamics	A software-focused fund and an industrials fund from the same vintage get pooled despite facing different exit cycles
Fund size	Deal sourcing and portfolio construction	A $200M fund and a $2B fund are compared as if they access the same deal flow
Geography	Regulatory, currency, and market maturity effects	Developed-market and emerging-market PE get blended into one number

Source: Dakota Private Markets, How to Benchmark Private Equity Fund Performance, 2026.

Common Construction Errors

1. Blending Vintages to Hit a Sample Size

When a peer group is too thin, filtered by strategy, vintage, and geography, the fastest fix is to widen the vintage window until enough funds show up. A 2019 buyout fund gets benchmarked against a "2018-2020" cohort instead of its true vintage peers.

Combining vintages erases the market-timing effects that made each fund's environment different in the first place (Dakota Marketplace, 2026). A tough capital-deployment year and a strong one get averaged into a single, misleading midpoint. For fund managers, this means a "beat the vintage benchmark" claim is only as good as whether the vintage was defined narrowly. Ask what window was used before citing the number.

2. Using a Single Metric Alone

A benchmark built on net IRR alone misses whether returns have actually been realized (DPI) or are still sitting on paper (RVPI). Two funds can show identical IRR while one has returned real cash to allocators and the other hasn't returned a dollar. Citing IRR in isolation is the fastest way to make an unrealized fund look identical to a realized one.

3. Ignoring Peer Group Size

A quartile ranking built from five comparable funds is far less reliable than one built from fifty, even when the filter criteria are identical. A single outlier fund can swing the top-quartile threshold by several points of IRR when the sample is that thin. As the TVPI data above shows, this isn't hypothetical: benchmark providers routinely publish quartile rankings on samples an order of magnitude smaller than the PE median.

The instinct to widen the vintage window (Error 1) is often a direct response to this problem: the peer group was too small, so the fix was to make it bigger by making it less precise. Neither fixes the underlying issue. The right fix is a larger underlying dataset, not a looser filter.

4. Skipping Sector-Level Filtering

Stopping at "middle market buyout" without accounting for what those portfolio companies actually do produces a peer group that includes fundamentally different return profiles under one label. A software-focused middle market buyout fund and an industrials-focused middle market buyout fund from the same vintage face different multiple-expansion environments and different exit cycles, even though they'd be pooled together under most default peer group definitions.

What This Means When You're the One Being Benchmarked

Allocators running diligence on a "top-quartile" claim increasingly ask about construction before accepting the ranking.

What makes a benchmark claim credible:

The vintage definition is stated explicitly and matches how the fund itself defines its vintage.
The peer group is filtered to sector and industry of underlying portfolio companies, not just headline strategy.
The sample size behind the ranking is disclosed.
More than one metric is shown, since IRR alone hides whether a return is realized.

What makes allocators discount a claim:

A benchmark citation with no stated sample size or peer group definition.
A single metric presented as the whole story.
A peer group that was visibly widened to hit a sample size threshold rather than held to a strategy and sector filter.

The practical result: a GP who can name the exact filter criteria behind a benchmark claim has a stronger position in diligence than one who cites a headline quartile number without explaining how the peer group was built.

How to Build (or Request) a Peer Group You Can Trust

State the vintage definition before comparing anything. Confirm the peer group and the fund being measured use the same definition.
Filter by sector and industry of underlying portfolio companies, not just strategy label. "Middle market buyout" is a starting filter, not a finished peer group.
Check the sample size before trusting the quartile. A peer group in the single digits should be treated as directional, not definitive.
Ask which metrics were used. A benchmark built on net IRR alone is incomplete. Request TVPI and DPI alongside it.
Confirm the vintage window wasn't widened to hit a sample size. If a "2019 vintage" comparison spans 2018 to 2020, that's a blended benchmark, not a vintage benchmark.
Match fund size and geography, not just strategy. A $200M fund and a $2B fund in the same vintage and sector still face different deal access.