Radar Perene / Archive / science

Conflicts of interest in bank research: who pays the analyst who signs the report

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

Science

The report arrives well designed, dozens of pages long, with a detailed model and a conclusion about a company. The reader evaluates the argument, the numbers, the analyst's track record. Almost never does he evaluate the question prior to all of those: where the signer's salary comes from. Not because the answer is secret — it sits, by regulatory requirement, in the fine print of the document itself — but because the question feels impolite. In the reading of evidence, impoliteness is method.

A conflict of interest, in research, is any situation in which whoever produces the analysis has incentives tied to its conclusion — so that the reader cannot distinguish, from the text alone, finding from interest. The conflict does not prove the analysis wrong; it proves the analysis cannot be evaluated in isolation from the incentive.

The distinction in that second sentence matters. Naming a conflict is not alleging corruption — analysts at financial institutions include rigorous professionals, and bank reports carry real technical work. The point is structural, not moral: incentive is a force that acts even on the honest, through the selection of what gets covered, what gets emphasized and what goes unsaid.

The structure of the sell side

The research that circulates in the greatest volume in markets is the so-called sell side: produced by brokerages and investment banks and distributed to clients. The name itself describes the incentive — it is the research of the side that sells. The institution publishing it earns brokerage when clients trade, and mandates when companies hire it for issuances, offerings and advisory. The analyst is not paid by the report's reader; he is paid by a firm whose revenues depend, in part, on the companies he covers and on the activity his opinions generate.

From that arrangement come the tensions the international literature has documented for decades: the historical scarcity of negative recommendations relative to positive ones; the discomfort of downgrading a company that is, or may become, the bank's client; coverage that concentrates where there is business and thins where there is none. Brazilian regulation attacks the problem through transparency — CVM rules require conflict disclosures in the reports and formal separation between research and the institution's other areas — and transparency helps, but it transfers the work to the reader: the disclosure is there to be read, and most readers never read it.

The question that applies to everyone — including this house

The honest filter is not "bank research is suspect, the rest is not". It is: every producer of analysis has an incentive model, and adult reading begins by identifying it. Academic research has its own — prestige, publication, the temptation of eye-catching results, the subject of other articles in this section. Independent subscription-funded shops have the incentive to retain subscribers, which can reward alarm or comfort, depending on the audience. Free brokerage material has the incentive of turnover. There is no position without incentive; there is declared incentive and omitted incentive.

Applied to whoever writes this: Radar Perene sells no financial product, receives nothing from issuers and is not paid per reader transaction — no brokerage, no rebate, no mandate. The house's declared incentive is reputational: the public output with DOIs — the RPWP working papers, collected at /pesquisa — exists to be checked, and the house gains exactly to the extent that it survives the checking. That model carries its own risk, recorded here with the same candor: whoever lives on reputation has an incentive to appear rigorous. The defense the house chose is to make rigor verifiable — public data, declared method, an archive that records its own errors — rather than to ask for trust.

It is also out of coherence with this page that the house's readings contain no recommendations: describing the environment and stopping, with no "buy" and no "sell", removes from the table the genre's most common conflict — the analysis that concludes whatever generates activity.

How to read a report knowing this

From the house's editorial experience with sell-side material — a legitimate source of data and modeling, read every day — three habits have settled. Read the conflict disclosures before the argument, because they change the argument's weight. Separate the report's facts (numbers, series, comparisons, usually its most valuable part) from its conclusion, which is where the incentive lives. And pay special attention to what the report does not cover — agenda selection is the layer of conflict no mandatory disclosure captures.

Frequently asked questions

Is bank research useless, then?

The opposite — it usually contains the best data work available on the covered companies. Informed reading harvests the fact factory and treats the conclusion as an opinion produced under a known incentive.

What does Brazilian regulation require of analysts?

In broad terms: certification, conflict disclosure within the report itself and formal independence from the institution's commercial areas. The details are in the CVM's public rule on securities analysts.

Buy side versus sell side — what is the incentive difference?

The sell side publishes outward and lives off the activity its publications generate; the buy side (funds, asset managers) researches inward, for its own decisions, and rarely publishes. The second's incentive is to be right in silence; the first's, to be read.

Does independence eliminate bias?

No — it trades one set of incentives for another. The useful question is never "who is neutral?", but "is the writer's incentive declared, and can the work be checked without trusting it?".

---

This trail closes here: abstracts, pseudoscience, aggregation of evidence and incentive. The next one changes century — what happens to research integrity when part of the text may have been written by a machine: AI in scientific research: what journals already regulate.

House reading: the model described above produces, every day, the reading in the Diário.

Evaluating a specific report in light of these questions is the kind of second opinion the house gives on request.