Radar Perene / Archive / science

Sector baskets in a concentrated market: when a few companies decide the whole sector

◦ Index methodology v2.2 (working papers with DOI). See the methodology.

Science

Ask an American analyst what "the financial sector" is and they will point to a list with dozens of banks, insurers, asset managers and payment processors. Ask the same question in Brazil and the honest answer is uncomfortable: a handful of institutions accounts for nearly all of the sector's market value. The sentence "the financial sector rose" describes, in practice, the day of three or four companies. The plural is grammatical, not statistical.

A sector basket is a grouping of assets meant to represent the behavior of an entire sector of the economy. In concentrated markets, that ambition runs into a structural problem: when a few companies dominate the group's weight, the basket stops measuring the sector and starts measuring those companies — and every "sectoral" conclusion inherits, without warning, the idiosyncratic risks of half a dozen names.

The literature that founded the instrument barely addresses this problem, because it was born where the problem does not exist. In the United States, with thousands of listed stocks, a sector basket dilutes any individual company; the group's behavior is genuinely collective. The imported formula assumes that dilution. The Brazilian market does not supply it.

What concentration does to the measurement

The effect is not subtle. In a diluted basket, one company's event — an accounting fraud, a change of control, an exceptional quarter — is noise the average absorbs. In a concentrated basket, the same event is the sector's reading for that month. Whoever consults the series later has no way to tell: was that jump the energy sector responding to some common condition, or a single company living through an episode of its own?

Public sector indices of the Brazilian exchange live with this arithmetic: some carry only a few dozen constituents, and the composition rules allow the largest weights to sit with a small core of companies. That is no fault of the calculator — it is the portrait of an exchange where entire sectors have two or three liquid representatives. The distortion is not in the index; it is in reading the index as if it were American.

There is a second, less visible effect. Concentration shrinks the effective sample: a basket of twenty stocks in which three carry most of the weight behaves, statistically, like a basket of three or four. Any test run on it — correlation with another sector, response to cycles, seasonality — has fewer independent observations than it appears to. It is the sectoral version of a problem Track 1 covered in sample size in financial series: the imported ruler counts assets; statistics counts assets that actually move on their own.

What the house bench found

The house maintains its own study of Brazilian sector baskets — part of the same line of work that examines what imported formulas assume and the local market does not deliver. The design and the results are bench work, but the central diagnosis can be stated: building a sector basket in Brazil demands decisions the source literature never had to make. How many names it takes for the group to exist as a group. What to do when a sector has one dominant representative and several illiquid satellites. How to keep the basket from changing identity when a single company enters or leaves.

None of those questions has an answer in the American manual, because the American manual never had to ask them. And each one, badly resolved, produces a series that looks sectoral on the label and is individual in its content.

The reading error this produces

The practical risk is not technical — it is interpretive. Sectoral analyses feed large conclusions: rotation between sectors, defense and offense, cycle reading. When the underlying basket is concentrated, all of those conclusions carry a stowaway: the particular history of the dominant companies. A "defensive sector that held up in the decline" may be a single company with a good quarter. A "rotation into consumer names" may be a corporate event. The series does not lie; it simply answers a different question from the one the reader asked.

The antidote begins before any statistics: look at the composition. How many names, at what weights, with what liquidity. It is a thirty-second habit that reclassifies a good share of the sectoral readings published out there — from collective evidence to corporate anecdote.

Frequently asked questions

Is sector concentration a problem exclusive to Brazil?

No — it is a common trait of small and emerging markets. Brazil is an acute case in some sectors, but the same question applies to any exchange where a sector has few liquid representatives. The dominant literature, written for deep markets, rarely poses it.

Is a concentrated basket useless?

No. It is useful for what it actually measures: the joint behavior of a small, identifiable group of companies. The problem is never the basket; it is presenting it as the portrait of an abstract sector it does not contain.

Does alternative weighting solve it?

It mitigates, it does not solve. Equal weights, per-company caps and other adjustments redistribute influence, but they do not create companies that do not exist. When the sector has three liquid names, no weighting scheme turns three into thirty.

How can a reader tell whether a sectoral reading suffers from this problem?

By checking the basket's composition before accepting the conclusion: number of constituents, weight of the largest, liquidity of the smallest. If the answer to "how many companies decide this number?" is small, the sectoral conclusion deserves to be reread as a conclusion about those companies.

---

Continue the trail: Three incidents, one lesson: what fails when a concentrated sector basket "breaks" →

House reading: today's reading is in the Diário; the sectoral episodes on file, in the Atlas.

Building or reviewing a specific sector basket, with the questions the imported literature does not ask, is an exercise the house carries out in conversation.

Characters: Method