The peer set (or "comp set") is the group of publicly traded companies whose trading multiples you average to value a target firm in comparable company analysis. It is the single most consequential — and most subjective — judgment in the entire method: the output multiple is only as good as the comparability of the firms feeding it. The core tension is between finding enough comparables to form a stable average and keeping them genuinely similar in the things that drive valuation. A textbook starting point is "same industry," but the deeper standard, articulated by Aswath Damodaran, is that a comparable firm is one with similar cash flows, growth, and risk — which may or may not coincide with industry membership.

How the peer set is built

Practitioners (and the standard investment-banking workflow, e.g. Street of Walls, CFI) build the set in a funnel:

1. Start with industry classification. Screen by GICS, NAICS, or SIC code, then refine using the "Competitors" / risk-factors sections of the target's 10-K, sell-side research coverage groups, supply-chain databases, and merger-proxy "selected companies" lists. 2. Filter on business model. Keep firms with similar products/services, customer base, distribution channel, and revenue model (e.g. SaaS subscription vs. perpetual license). Two firms in the same SIC code can be economically unalike. 3. Filter on financial profile. Narrow to firms within a comparable band of: - Size — revenue and market cap (a common rule of thumb is roughly within a 0.5x–2x range of the target; this band is convention, not a derived threshold). - Growth — historical and forecast revenue/earnings growth. - Profitability — operating, EBITDA, and net margins; ROIC/ROE. - Capital structure & risk — leverage, cyclicality. - Geography — region of operations and listing, which affects accounting, tax, and country risk. 4. Land on a working set. Most guides suggest roughly 5–10 comparables — enough to dampen the noise of any single outlier, few enough to stay genuinely similar. Banker desk practice often shows a broad set then a tighter "core comps" subset for the headline conclusion.

Damodaran's more rigorous alternative replaces subjective banding with a regression: regress the target multiple across a wide universe on the fundamentals that theory says drive it (for P/E: expected growth, payout, risk; for P/B: add ROE; for P/S: add net margin). The fitted equation predicts a multiple for the target, controlling for differences rather than hoping they wash out in an average.

How it's used in practice

The peer set is the engine that turns the method's logic — "similar assets should trade at similar prices" — into a number. Once chosen, the analyst computes each peer's chosen multiple, takes the median (preferred over the mean because it resists outliers), and applies it to the target's corresponding metric to imply a value. The peer set also frames the relative read: a target trading at a discount to its comp-set median is flagged as potentially cheap, a premium as potentially rich — but only if the peers are truly comparable, otherwise the gap reflects mismatch, not mispricing.

Two craft points dominate good practice. First, the right peers depend on the multiple — a sibling concern handled under Choosing the Right Multiple and Normalizing for Comparability. Capital-structure differences, for instance, make P/E comparisons across differently levered peers unreliable, pushing analysts toward EV-based multiples. Second, transparency about the set is the deliverable — a comps table that hides which firms were excluded, and why, is not analysis. Analysts typically show the full screen, the dispersion (high/low/median), and any firms dropped as non-comparable.

Adoption, debate & evidence

Comparable company analysis is the most widely used valuation approach in practice — dominant in equity research, IPO pricing, fairness opinions, and M&A — precisely because it is market-anchored and quick. But peer selection is its acknowledged soft underbelly. Street-of-Walls and CFI both concede the process is "somewhat subjective," and Damodaran is blunter: subjective comp choices "all too often confirm the analyst's biases about companies," because a banker pitching a high valuation can lean toward richly priced peers, and vice versa.

The strongest empirical evidence on selection comes from Bhojraj & Lee (2002), "Who Is My Peer?" (Journal of Accounting Research 40(2): 407–439). They estimated cross-sectional regressions of EV/Sales and P/B on proxies for growth, profitability, and risk to derive a "warranted multiple," then selected as peers the firms with the closest warranted multiple. Tested on out-of-sample prediction of one- to three-year-ahead multiples, their valuation-based peers delivered "sharp improvements" over peers chosen by industry classification (e.g. SIC code) alone. The takeaway is well-supported: fundamentals-matched peers beat industry-matched peers for valuation accuracy. This validates Damodaran's stance and is the main reason quantitative shops favor regression/clustering over pure industry buckets. The counterpoint — adoption inertia — is that subjective same-industry selection remains the default in most deal and research settings because it is fast, defensible to clients, and requires no modeling.

Strengths & limitations

Works best when the industry is populated with many economically similar, liquid public firms (e.g. large-cap retailers, integrated oils, money-center banks) so a tight, fundamentally-matched set is achievable. Fails when the target is a near-unique business, an early-stage/loss-making firm, a conglomerate, or in a thin industry — there the analyst is forced to either widen the net (importing noncomparable firms) or accept a tiny set (one outlier swings the median). A peer set drawn from a market or sector that is itself overvalued imports that error wholesale: comps measure relative, not intrinsic, value, so a set of expensive peers will make an expensive target look fair.

The single most common misuse is treating "same SIC/GICS code" as sufficient and ignoring differences in growth, margin, and risk — the exact error Bhojraj & Lee show degrades accuracy. The second is selection bias: cherry-picking peers (or quietly excluding inconvenient ones) to land a predetermined valuation. The discipline that guards against both is showing the full screen and the dispersion, not just the median.

Sources

• Aswath Damodaran, "Controlling for differences in relative valuation" (Little Book of Valuation companion) — pages.stern.nyu.edu/~adamodar/New_Home_Page/littlebook/controldifferences.htm — "a comparable firm is not one in the same business but one with the same growth, risk and cash flow characteristics"; regression to control for differences.
• Damodaran, "Relative Valuation" lecture notes (Stern) — pages.stern.nyu.edu/~adamodar/pdfiles/execval/relval.pdf — fundamentals driving each multiple.
• Bhojraj, S. & Lee, C.M.C. (2002), "Who Is My Peer? A Valuation-Based Approach to the Selection of Comparable Firms," Journal of Accounting Research 40(2): 407–439 — warranted-multiple selection outperforms industry-classification peers.
• Street of Walls, "Comparable Company Analysis" training module — selection criteria (sector, products, customers, channel, geography), "almost never a perfect comparable."
• Corporate Finance Institute (CFI), Comparable Company Analysis / Pitchbook Comps templates — screening workflow, ~5–10 comps convention, median over mean.
• Valutico / Phoenix Strategy Group, "How to choose comparable companies" — size/growth-stage banding conventions (rules of thumb, not derived thresholds).

Dispute flagged: the "right" basis for selection is genuinely contested — academic/quant evidence (Bhojraj & Lee) favors fundamentals-matched peers, while mainstream banking practice still defaults to industry-classification peers for speed and defensibility. Numeric bands (size 0.5x–2x; 5–10 comps) are industry convention, not statistically derived.