About Vector Research

An independent research practice operating at the frontier of human-led, computationally-augmented analysis.

The Old Order

For four decades, equity research was a guild. A small priesthood of sector analysts at universal banks set consensus on a few hundred names; the rest of the market traded the deviations. The model was depth-bound — six to fifteen names per analyst, beyond which the rest of the universe got a sentence in a daily wrap. It was also conflict-bound — research was subsidized by trading commissions and banking relationships, with disclosures expanding faster than candor. The model worked because, for most of its history, it was the only model on offer.

It broke under its own economics. Research-payment unbundling decoupled the cost of analysis from the cost of execution. Commission compression hollowed out the publishing budget. The sectors that had ten sell-side analysts in 2010 had three by 2020. By any reasonable measure, the supply of deep, single-name research has been in secular decline for fifteen years — even as the demand from sophisticated capital allocators for independent, conviction-coded analysis has, if anything, compounded.

The Algorithmic Interregnum

The vacuum left by the contracting sell-side did not stay empty. Quantitative methods — factor models, multi-factor scoring, statistical arbitrage, and eventually machine-learning ensembles — moved into the breadth gap. They produce signal at a scale no human team can match. They are also, in a structural sense, a different kind of object than fundamental research.

Statistical methods identify patterns without understanding them. They detect shifts in the data without knowing whether the shifts are real changes in the world or artifacts of the dataset. They generalize from history without a view on whether the future will resemble the past. The richest single-name theses — distress and recovery, structural inflection, regulatory windfall, compounding flywheel, governance turnaround — are precisely the ones quant approaches systematically miss, because the evidence for them is qualitative, scattered, and legible only to a mind that has built a working model of how the underlying world operates. Statistical methods are not in the business of building world models. They are in the business of finding regularities.

The Synthesis

What has changed in the last several years is the infrastructure available to a working analyst. Large language models, retrieval systems, and structured-data pipelines — deployed correctly — have, for the first time in the history of equity research, closed the breadth bottleneck. A single analyst can now ingest, cross-reference, and structurally compare primary-source material — regulatory filings, earnings transcripts, court records, foreign-language disclosures, sector- specific datasets — at a scale that no six-person sell-side team could have matched in 2015. The economics of what one analyst can see have been rewritten.

The depth bottleneck has not moved. The judgment that decides what evidence is load-bearing, which counter-cases deserve serious treatment, when conviction has been earned and when it is merely borrowed — none of this has been automated, and there is no near-term path to its automation. The work of building a thesis, defending it under adversarial review, and putting one’s name on a verdict remains the work of a human mind. What the new infrastructure does is collapse the cost of the supporting work — cross-referencing, screening, structured comparison, synthesis of pattern-matched analogs — and free the analyst to spend the marginal hour on the thing that still matters: judgment.

Vector Research is built on that asymmetry. The breadth that used to take a six-person team now takes an afternoon. The depth still takes a week. The new practice consists in spending the right week on the right name, and shipping a verdict only when one has been earned.

The Frontier

The economics of a research practice operating this way are different from the economics of either the legacy sell-side or a quantitative fund. They are also, at present, only thinly populated. A handful of independent shops are working out what the new register looks like; most of the field has not yet absorbed that the constraints have moved. Vector Research operates at that frontier.

Coverage is opt-in, not survey-based. We initiate on a name when there is something worth saying — when consensus has missed a structural change, when primary-source evidence is materially under-utilized in the public record, when a coming catalyst is inadequately mapped. We do not publish on every constituent of any benchmark; we do not initiate at the request of issuers, promoters, or any party with an economic stake in coverage.

Reports tend to be written when a thesis is worth defending, not on a schedule. They generally aim to carry a stated thesis in plain language, an indication of conviction, the evidence behind them, and an analyst attribution. Where a conviction level is attached, it reflects the analyst’s posterior probability that the thesis is correct — not a target return, not a position size, and not a recommendation suitable for any particular reader.

The Architecture

Operationally, Vector Research is a human-led, computationally-augmented practice. The architecture is simple to describe and demanding to execute:

• The analyst sets the agenda — selects coverage, frames the thesis, decides what counts as evidence, and ultimately puts a name on the conclusion.
• The computational layer handles breadth — ingests primary sources, structures them, surfaces candidate signal, runs comparisons, and lowers the cost of adversarial framing.
• The analyst closes the loop — reviews, weighs, judges, and decides what is published.

The analyst sits at both ends of the loop: originator of the question, and author of the conclusion. The machine works in between. This is not a return to the old model with new tools. It is a different practice altogether — one that is only possible at the intersection of two capabilities that, until very recently, did not exist together.

Editorial Standards

Our editorial practice generally aims for the conventions of institutional research: confidence-labeled statements, source discipline, conviction coding, pre-publication adversarial review, anti-scalping, dated corrections. These are described in greater detail in our Editorial Standards and our Methodology. The short version: we aim to publish verdicts rather than outputs — named, conviction-labeled, and evidenced.

What We Publish

Deep-dive, single-thesis analysis. The work spans macroeconomic and industry research, company fundamental and technical analysis, and startup and pre-IPO evaluation — whatever register the thesis demands. One question at a time, one verdict at a time. The full operational scope is set out in our Coverage page; the analytical architecture is in our Methodology.

Who We Read

Vector Research is written for sophisticated audiences — portfolio managers seeking independent coverage outside the sell-side consensus, family offices and private investors building conviction on specific names, operators and founders mapping competitive landscapes. Reports are not directed at any individual reader, are not tailored to any reader’s specific circumstances, and are not investment advice in any individualized sense.

Independence

We do not engage in investment banking. We do not market-make. We do not accept payment, equity, gifts, or consideration of any kind from covered companies, issuers, or third parties with an economic stake in our coverage. We publish what we conclude.

This is by design — and, in our view, it is the only structure under which independent research endures.