Peer group analysis

This is the company against others playing the same game, in the same period. Cross-section catches the drift a time series can't see. If margins fell 2pp but every peer's fell 4pp, the company is winning. If its margins rose while peers' rose faster, it's losing while feeling good. The whole method stands or falls on one question: what counts as a peer?

When you pick peers, business model beats sector label. Stock screeners, index committees and financial media group companies by labels like "retail," "autos" and "financials." Labels are filing conventions. A true peer shares the economic engine, whatever shelf tag it carries. Apply five tests, in order:

1. Same customer and need. Who pays, and what problem are they paying to solve?
2. Same economic engine. Where does the profit actually come from? Look at the unit economics, the take-rate, the spread. This is the test that does the most work.
3. Same capital intensity and balance-sheet shape. An asset-light franchisor and an asset-heavy owner-operator can share customers and still be incomparable on margins and returns.
4. Same regulatory and geographic regime, or a stated plan to adjust for differences in tax regimes, rate regulation, inflation or labor law.
5. Comparable scale and stage. A 25%-growth challenger and a mature incumbent can be compared, but you have to name the growth-stage difference, or the table will "discover" that the small one grows faster.

Why do sector labels fail as peer groups?

Here's how the label fails, in both markets:

• "Retail," India. Avenue Supermarts (DMart) runs owned stores and everyday-low-price grocery on thin margins and high turns. Trent runs lease-heavy fashion with high gross margins and brand economics. Their gross margins differ by ~30pp by design. Ranking them against each other tells you which model each one chose. It says nothing about who executes better.
• "Food & beverage," US. Costco vs Kroger. Costco's engine is the membership fee, which the common-size build exposes, so its merchandise margins are deliberately ~10pp thinner. Call Costco "worse at retailing" on gross margin and you've read the strategy as a failure.
• "Financials," both markets. A deposit-funded bank and an NBFC share a label, but funding, regulation and the meaning of leverage all differ. Banks get their own toolkit in Phase 5, and even ROIC doesn't apply to them.
• "Automobiles," India. Consolidated Tata Motors vs Maruti compares JLR, mostly, with an Indian small-car maker. That's a consolidation-basis failure as much as a peer failure, and the standalone-vs-consolidated rule in the hygiene layer deals with it.

How many peers do you need, and where do you find them?

Build the peer set from the company's own mouth. A 10-K names competitors in Item 1 and Item 1A, and Indian MD&As and investor decks describe the "competitive landscape." Cross-check with a screener's peer tab: screener.in for India, stockanalysis.com or TIKR for the US. Then audit the machine's list with the five tests, because it was built from labels. Aim for 3 to 5 peers. Fewer than three true peers is common, and DMart, for one, has no clean listed Indian twin. When that happens, the right answer is a smaller set plus clearly labelled references. A reference can be a foreign analogue like Costco or Walmart, compared with explicit geography caveats. Or it can be a partial peer matched on one dimension, such as Vishal Mega Mart on value-retail economics or a regional grocer on category. Never pad the table to five with false peers. Every fake row degrades the median you're about to trust.

How do you read a peer table?

Reading a peer table. Use the median as the group's center and leave the mean alone, because one outlier, a loss-maker or a hyper-grower, drags a mean of four numbers anywhere it wants. For each ratio, report the company's rank and gap against the median. Then apply the gap decomposition you installed at the start. Is the gap accounting, definition or economics? Only the third kind is a finding, and it should point somewhere physical: scale, mix, utilization, pricing, geography. The Infosys study does this line by line.