Reading a Comps Set: What Makes Two Companies Actually Comparable
Anyone can pull ten companies from the same SIC code and take a median. The work is in defending the set — and in knowing which of those ten quietly doesn't belong.
Comps are the easiest valuation method to build and the easiest one to build badly.
The math is arithmetic. Pull a peer group, calculate a multiple for each, take the median, apply it to your company. Nothing in that sequence is hard. What's hard is the first step, and it's the step nobody grades themselves on: deciding which companies belong in the group at all.
Get that wrong and every number downstream is precise and useless.
This post is about reading a comps set critically — yours or someone else's. For the mechanics of building one from scratch, see the trading comps guide.
What “comparable” actually means
Not “same industry.” Same economics.
A multiple is shorthand for what the market will pay for a dollar of something — revenue, EBITDA, earnings. Two companies deserve the same multiple when a dollar of that metric means roughly the same thing at both: it grows at a similar rate, it converts to cash at a similar rate, and it carries similar risk of not showing up next year.
That's the whole test. Industry labels are a proxy for it, and often a bad one. Two companies can share a sector, a customer base, and a conference circuit while running completely different businesses underneath.
The screens that actually filter
Business model and revenue mix
Start here, not with the sector code. A software company selling annual subscriptions and a software company selling perpetual licenses with a services attach have different revenue durability, different margin profiles, and different working capital behavior. The market knows that and prices them differently.
Read the revenue disaggregation in the filings rather than the one-line company description. Segment tables are where two “peers” usually stop looking alike.
Size
Larger companies tend to trade at higher multiples than much smaller ones in the same business — more diversified customers, better access to capital, more analyst coverage, more liquidity in the stock. A peer set spanning an order of magnitude in revenue isn't a peer set. It's two peer sets stapled together.
Growth and margins
These are the two inputs a multiple is most sensitive to, and the two most often waved past. Faster growth and higher margins support a higher multiple, all else equal.
So when a peer trades far above the rest of the set, the first question is whether it grows faster or earns better margins. If it does, it isn't an outlier — it's correctly priced and your company probably shouldn't get that multiple.
Geography and end market
Same product, different demand drivers, different regulatory exposure, different currency. A domestic operator and a global one with half its revenue overseas are exposed to different things, and accounting rules can differ across regions in ways that reach the multiple itself.
Capital structure — sort of
This one gets handled by the multiple you choose rather than by the screen. EV/EBITDA is capital-structure-neutral, because enterprise value and EBITDA both sit above the effect of financing. P/E isn't: interest expense hits net income, so a levered company and an unlevered one aren't comparable on P/E even if they're identical operators.
Leverage still matters for risk, and extreme leverage will show up in the equity multiple no matter what. But it's the wrong reason to throw a company out of an EV/EBITDA set.
Matching the numerator to the denominator
The single most common technical error in a comps set, and the easiest one to check.
Enterprise value belongs on top of metrics available to all capital providers — revenue, EBITDA, EBIT, unlevered free cash flow. Equity value belongs on top of metrics that are already net of interest — net income, EPS, levered free cash flow, book value of equity.
EV/net income is wrong. P/EBITDA is wrong. They're wrong for the same reason: one side of the ratio has been paid to debtholders and the other hasn't. If you want the full bridge between the two values, it's in the enterprise value guide.
Which multiple to lead with depends on the set:
- EV/EBITDA — the default for most mature, profitable companies. Neutral to capital structure and to differences in tax rate.
- EV/EBIT — better when capital intensity differs a lot across the set, since EBIT is after depreciation and EBITDA isn't. A company that has to keep replacing its asset base looks artificially cheap on EBITDA.
- EV/Revenue — for companies with negative or barely positive earnings, where the earnings-based multiples stop meaning anything. Weakest of the three, because it assumes the margin structure across the set is similar.
- P/E — standard for financial institutions and useful where the set is genuinely similar in leverage and tax. Distorted by both otherwise.
- Sector-specific metrics — EV per subscriber, per bed, per square foot, and so on. Worth showing alongside the standard multiples, never instead of them.
The adjustments that make the set honest
A comps set is only as consistent as the definitions behind it. Every company reports slightly differently, and the point of these adjustments is to make ten different filings answer the same question.
- Calendarize the periods. A company with a June fiscal year end and one with a December year end aren't describing the same twelve months. Put every peer on a common calendar basis before you compare anything.
- Use the same period convention throughout. Last twelve months for everyone, or forward estimates for everyone. Mixing an LTM multiple into a set of forward multiples makes a fast-growing company look expensive purely by construction.
- Strip out non-recurring items. Restructuring charges, litigation settlements, impairments, gains on asset sales. If it isn't going to repeat, it shouldn't sit in the denominator of a multiple meant to price the ongoing business.
- Use diluted share count, not basic. In-the-money options and other dilutive securities are usually brought in through the treasury stock method: assume they're exercised, and assume the proceeds are used to buy back shares at the current price. The net addition is real dilution and it belongs in equity value.
- Build enterprise value the same way for every peer. Equity value plus debt, minus cash, plus preferred, plus minority interest. Minority interest gets added because consolidated EBITDA includes 100% of a partly-owned subsidiary while equity value only captures the share actually owned — leave it out and the multiple is understated.
- Watch lease accounting across regions. US GAAP and IFRS treat operating leases differently in how they hit the income statement, which means EBITDA for an American company and a European one isn't always built the same way. Mixed geography sets need a look here before the EV/EBITDA column means anything.
Reading someone else's comps set
You'll see far more comps sets than you build, and most of them arrive as a finished page with a median already calculated. Four things to check before you use it.
Look at the range, not the median
A set trading between 8x and 10x is telling you something. A set trading between 4x and 22x with a median of 9x is telling you the median is an accident. Wide dispersion usually means the group isn't really one group.
Prefer the median to the mean
One company at 40x drags a mean somewhere no individual company in the set actually trades. The median is more resistant to that, which is why banker pages lead with it.
Ask who's missing
The most informative question about any peer set is which obvious comparable didn't make it in. Sometimes there's a clean reason — it just IPO'd, it's under a pending acquisition and its price reflects deal terms rather than its own prospects, its financials are being restated. Sometimes the reason is that it trades somewhere inconvenient for the conclusion.
Both answers are useful. Only one is defensible.
Remember what comps can and can't tell you
Comps price a company against how the market is valuing similar companies right now. If the whole sector is expensive, a comps analysis inherits that and reports it as fair value. It's a relative measure by construction, which is exactly why it gets shown next to a DCF rather than instead of one.
How this gets asked
The multiples are the output. The peer set is the argument.
Interviewers know that, which is why “how did you pick these companies” is a far more common follow-up than anything about the arithmetic. Have an answer that starts with economics and not with a screen.