GeminIQ
Subscribe

How to Build a Defensible Peer Comparison Set

Chad Hartman

By

Published · Last updated

A peer comparison is only as good as the peers in it, and an industry-label peer group can fail before a single ratio gets calculated — not because the math is wrong, but because the companies being compared were never actually comparable in the first place. Two companies can share the same industry classification and describe meaningfully different businesses: different scale, different growth trajectory, different margin structure, different capital intensity. Averaging them together doesn't produce a benchmark; it produces a number that looks precise and describes nothing real.

This guide covers how to build a peer comparison set that actually holds up: why industry classification codes alone aren't sufficient, a real, underused source of professionally-vetted peer groups sitting in plain sight in every company's own proxy statement, and the specific filters that separate an actual comparable from a company that merely shares a sector label.

Start your 7-day free trial →

Table of Contents

Why "Same Industry" Isn't the Same as "Comparable"

Industry membership is a necessary filter and a wildly insufficient one on its own. A regional community bank and a global systemically important bank share a sector classification and almost nothing else that matters for comparison — different risk profiles, different regulatory capital requirements, different growth ceilings, different cost of capital. A diversified industrial conglomerate and a pure-play manufacturer in one of that conglomerate's segments might share the same primary industry code while running businesses with almost no operational overlap.

The gap between "same industry" and "actually comparable" is exactly where peer analysis goes wrong, and it goes wrong quietly. A comparison built on a flawed peer set doesn't produce an obviously broken number — it produces a plausible-looking multiple or margin benchmark that's simply measuring the wrong thing.

The SIC Code Problem

Every SEC filer discloses a primary SIC (Standard Industrial Classification) code on the cover page of its registration statements, and that code is the most immediately available starting point for finding companies in a similar business. It's also a badly outdated system. The SIC classification structure was last substantively revised in 1987, and the federal government stopped maintaining or updating it for new industry categories in 1997, when it was formally replaced by NAICS for economic statistics purposes. The SEC never made the same switch — SIC codes are still what's on the filing's cover page today, frozen in a taxonomy built for an economy that looked meaningfully different than the one companies operate in now.

The practical consequence is that SIC codes systematically undercount services-heavy, technology-heavy business models that didn't exist in their current form when the taxonomy was last updated, and they group companies by product category rather than by the underlying economics of how those companies actually compete and get valued.

GICS: A Better Starting Point, Still Not the Answer

GICS (Global Industry Classification Standard), developed jointly by MSCI and S&P starting in 1999, groups companies by economic activity and end markets rather than by the product-category logic SIC was built around — a meaningfully better proxy for real economic comparability, before any company-specific filter gets applied. GICS organizes companies hierarchically, from 11 broad sectors down to over a hundred specific sub-industries, which allows for a much finer-grained starting filter than a single SIC code provides.

GICS still assigns each company to exactly one classification based on its primary business activity, which remains a real limitation for any company operating meaningfully across more than one business line. A diversified company gets filed under whichever segment generates the largest share of its revenue, even when a second segment is large enough to matter for real comparability — the classification system, however refined, is still a single label applied to what's sometimes two or more distinct businesses running under one ticker.

The Underused Source: A Company's Own Disclosed Peer Group

There's a source of peer group data that skips the classification-code problem entirely: the compensation peer group disclosed in nearly every company's own DEF 14A proxy statement. Public companies routinely disclose the specific, named list of companies their compensation committee — often working with an outside consultant — selected as comparable for executive pay benchmarking purposes, chosen explicitly on criteria like revenue range, market capitalization, business model, and operational complexity, not on an industry code alone.

Carvana's 2025 proxy statement, filed March 25, 2025, makes the underlying logic explicit — and shows how far a board-selected peer set can diverge from a classification code. Carvana's committee said it chose peers on "size, industry focus, growth rates, customer base, and market for talent," and further weighted "our online business model, revenue size, and complexity of operations." The list it produced:

Peer What a code-based screen would have said
AutoNation, CarMax, Lithia Motors, Penske Automotive, Genuine Parts The auto-retail peers a SIC or GICS lookup finds on its own
Chewy, Wayfair, eBay, Expedia E-commerce and online marketplaces — a different industry code entirely
DoorDash, Uber, Lyft Consumer platforms, matched on business model rather than product
Zillow, Opendoor Online transaction platforms for a large, infrequent, high-ticket purchase — the closest structural analogues, in an unrelated sector

Only five of those names come from Carvana's own industry classification. The other nine were selected because the committee judged the business model comparable, and the last pair is the most instructive: Zillow and Opendoor sell houses, not cars, but they share the thing that actually determines how Carvana's economics behave — moving a large, infrequent, high-consideration purchase online. That is a board-level judgment about comparability, made by people with every incentive to get it right, sitting in a public filing most equity research never reads for this purpose.

The caveat is real: a compensation peer group is selected for pay benchmarking, not equity valuation, so company size and role complexity get more weight than a valuation-focused peer set might use. Genuine Parts is a distribution business with little in common with Carvana operationally. But as a starting universe of board-vetted comparables, it is a stronger foundation than a raw industry code lookup — and it is free. For what the underlying business actually did over this period, our analysis of Carvana's crash and recovery covers the filings behind it.

Four Filters That Separate a Real Peer From a Same-Sector Name

Once a starting universe exists — from GICS, from a disclosed compensation peer group, or both combined — four filters, checked against actual filed data rather than assumed from the industry label, separate the actual comparables from the companies that merely share a sector.

Filter What you compare Range the filter allows What it distorts when ignored
Revenue scale Peer revenue against the subject company's revenue Roughly a half to two times the subject company's revenue — close enough to share similar operating leverage and market position Well outside that range, scale effects start dominating the comparison more than the underlying business quality does
Growth profile Revenue growth rate Within roughly 10 percentage points of the subject company's TTM revenue growth rate A double-digit grower measured against low-single-digit growers produces a multiple comparison that's really measuring the growth premium, not relative valuation
Margin structure Gross margin and operating margin Gross margin within roughly 10 points of the subject company's A broad industry classification spans very different margin profiles, so mismatched margins signal a different underlying business model
Capital structure Leverage Net Debt-to-EBITDA within roughly 1.5x of the subject company's, or both companies on the same side of the net-cash line Wildly different leverage produces distorted equity multiples, unless the comparison is built on an enterprise-value basis that neutralizes the difference in the first place

A peer that clears the industry filter and fails two or more of these four checks isn't a peer; it's a same-sector company that happens to also be public.

Running the filters: Carvana's own peer group

The four filters are easy to agree with in the abstract. Applied to a real board-selected peer group, they are less forgiving than they look.

Carvana's TTM figures at the quarter ended June 30, 2026: revenue of $25.06 Billion, revenue growth of +54%, a 19% gross margin, and Net Debt-to-EBITDA of 1.11x. That sets the four bands: revenue between $12.5B and $50.1B, growth between 44% and 64%, gross margin between 9% and 29%, and Net Debt-to-EBITDA between -0.39x and 2.61x.

Peer Revenue scale Growth profile Margin structure Capital structure Filters passed
AutoNation Pass Fail Pass Pass 3 of 4
CarMax Pass Fail Pass Pass 3 of 4
Lithia Motors Pass Fail Pass Pass 3 of 4
Penske Automotive Pass Fail Pass Pass 3 of 4
Genuine Parts Pass Fail Fail Fail 1 of 4
Chewy Pass Fail Fail Fail 1 of 4
Wayfair Pass Fail Fail Fail 1 of 4
Expedia Group Pass Fail Fail Fail 1 of 4
DoorDash Pass Fail Fail Fail 1 of 4
eBay Fail Fail Fail Pass 1 of 4
Uber Fail Fail Fail Pass 1 of 4
Opendoor Fail Fail Fail Pass 1 of 4
Lyft Fail Fail Fail Fail 0 of 4
Zillow Group Fail Fail Fail Fail 0 of 4

None of the fourteen clears all four. Every one of them fails the growth filter, for a single reason: Carvana grew revenue 54% over the trailing twelve months and not one company on its own disclosed peer list grew faster than 34%. The four auto retailers — AutoNation, CarMax, Lithia and Penske — pass scale, margin and leverage cleanly and fail only there.

That result is not an argument against using the proxy peer group. It is the argument for screening it. The compensation committee was solving for market for talent and operational complexity, and by that standard DoorDash and Uber belong on the list. For a valuation comparison they do not, and the filters say so in one pass. What the disclosed list gives you is a defensible starting universe of fourteen names, assembled by people with real accountability for getting comparability right; what the filters give you is the subset that survives a different question.

The practical read for Carvana specifically: there is no clean peer set. A grower at 54% in a sector where the nearest comparable business models grow in the low single digits does not have peers on a multiples basis, and any comparison built anyway is measuring the growth premium rather than relative valuation. Knowing that before running the comparison is more useful than a peer-average multiple that quietly encodes it.

How Many Peers Is Enough

There's a real tension between a peer set narrow enough to be actually comparable and one large enough to average out company-specific noise. A set of two or three peers is vulnerable to one company's idiosyncratic situation — a recent acquisition, a one-time charge, a temporary demand spike — distorting the whole comparison. A set of twenty or more, built by relaxing the four filters above until enough companies qualify, usually means the comparison has drifted back toward "same industry" rather than "actually comparable." A peer set in the range of five to ten companies, each of which clears all four filters rather than most of them, is generally the more defensible middle ground — small enough that each company's inclusion can be individually justified, large enough that no single company's noise dominates the average.

Building the Comparison Correctly

Once the peer set itself is defensible, the comparison still has to control for scale by using ratios and margins rather than absolute dollar figures, and it has to account for fiscal year misalignment. Comparing a company's most recently reported trailing twelve months against a peer's stale, six-month-old trailing figure understates the comparison's precision even when every other filter was applied correctly. Checking the actual period each peer's most recent figures cover, not just pulling the latest number each platform happens to display, is the final step that turns a defensible peer set into a defensible comparison. Platforms differ in how they handle that period alignment, and a platform that recalculates or normalizes older figures can quietly break exactly this kind of comparison in a way that as-filed data does not.

Building the Peer Set on GeminIQ

Every filter covered here requires comparing filed data across multiple companies side by side: revenue, growth rates, margins, and capital structure for a candidate list, checked against the subject company's own figures before any company gets included. GeminIQ's Custom Tables builder assembles exactly this kind of multi-company comparison from as-filed data, and Calculated Metrics like Revenue Growth and Gross Profit Margin are pre-built from the same tags across every company in the database, which means the four filters above can be checked directly rather than reconstructed by hand for each candidate peer.

For the compensation peer group specifically, the actual named list lives in the subject company's own DEF 14A on EDGAR — a five-minute read that often does most of the initial screening work a raw industry code search would otherwise leave undone.

Frequently Asked Questions

Why isn't SIC code enough to build a peer group?

SIC codes were last substantively revised in 1987 and haven't been updated by the federal government since 1997, when NAICS replaced them for general economic classification. The SEC still requires SIC codes on filing cover pages, but the taxonomy reflects an economy that looked meaningfully different than the one companies compete in today.

What is a compensation peer group, and can I use it for valuation?

It's the specific list of companies a company's own compensation committee — often with an outside consultant — selects as comparable for executive pay benchmarking, disclosed in the DEF 14A proxy statement. It's a useful starting universe of professionally-selected comparables in its own right, but it's built for pay benchmarking, not equity valuation, so company size and role complexity carry more weight in that selection than a pure valuation comparison might use.

How many companies should be in a peer comparison set?

Somewhere around five to ten is generally the more defensible range. Fewer than that leaves the comparison vulnerable to any one company's idiosyncratic noise; more than that usually means the selection criteria were relaxed enough to blur back toward "same industry" rather than "actually comparable."

What's the difference between SIC and GICS codes?

SIC is the older classification system, still required on SEC filing cover pages but not updated since 1987. GICS, developed by MSCI and S&P starting in 1999, organizes companies by economic activity and end markets rather than product category, with finer-grained sub-industry categories that group companies whose stock returns move together more closely than a SIC-based grouping does.

A peer comparison built on an industry code alone answers "who's nearby." A peer comparison built on revenue scale, growth, margin, and capital structure — checked against actual filed data — answers the harder and more useful question: who's actually comparable.

Start your 7-day free trial →

Sources: Carvana Co.'s compensation peer group and the committee's stated selection criteria are drawn from its definitive proxy statement (DEF 14A) filed March 25, 2025, publicly available on SEC EDGAR. GICS sector and sub-industry counts reflect the classification structure maintained by MSCI and S&P Dow Jones Indices.

Wall Street's data. Main Street's price.

Institutional terminals charge thousands a year for as-filed accuracy. GeminIQ gives you the same thing for a fraction of the cost: financials built directly from raw SEC EDGAR filings, not third-party APIs, with full XBRL traceability back to the original 10-K or 10-Q. No normalized guesswork, just calculated metrics, charts, screeners, and watchlists built on numbers exactly as the company reported them. Start researching now at GeminIQ.com.

Disclaimer: The content in this blog is for educational and entertainment purposes only and does not constitute financial, legal, or tax advice. Investing involves risk, including the loss of principal. The views expressed are my own and not intended as financial advice or a guarantee of future performance.