The conventional model of herding is straightforward enough to state in a sentence. Somebody observes others buying, infers that they hold information he himself lacks, and follows them into the position. The model assumes that decisions are visible to outsiders and that imitation therefore happens at the level of the decision. In an environment where information is scarce, that description works well enough to be useful. In an environment where information is overwhelming, it no longer explains what actually takes place.
When the available reading vastly exceeds what any person could process, the scarce resource stops being information itself. What becomes scarce instead is the capacity to decide which information deserves attention. Filtering begins to cost more than analysis does, and the cost is paid before any analysis starts. Under those conditions, outsourcing the filter is more attractive than outsourcing the decision, and considerably harder to notice as outsourcing at all. Few investors copy another person's positions outright, yet a great many accept, without examination, somebody else's list of things worth thinking about.
The result is an inversion of the usual sequence. Capital used to concentrate first, with attention following it afterwards, and now the order runs the other way. Well before any money moves, large numbers of people are already researching the same names and reading the same arguments. They are turning over the same set of questions, weeks or months ahead of any transaction. None of this activity appears anywhere in volume data. It has nonetheless already done most of the work of determining where the money will eventually go.
The mechanics of this reinforce themselves without anyone intending it. Attention can be measured in a way that conviction cannot, and anything measurable eventually becomes a target. A name already under discussion generates further discussion, because discussion is what the surrounding infrastructure exists to reward. Analysts, publishers and platforms face identical incentives to cover what is already being covered, since the audience for it demonstrably exists. Very little of this involves anyone forming a view about the underlying asset. By the time judgement is applied, the field of candidates has already been narrowed by other means.
This form of herding is unusually difficult to detect in oneself, because subjectively it is indistinguishable from independent work. The research really was carried out, and the data really was read. Contrary views really were considered, and the conclusion really was reached alone. The only part of the process that did not belong to the investor is the unrecorded first step. Why this particular name, out of the several thousand that were never examined at all? That question is rarely asked, and its answer is almost never written down anywhere.
Bringing that step back into view is a matter of asking a different question. More analysis will not reach it. Asking whether the analysis is correct assumes the subject has already been settled and merely checks the answer. Asking why this subject is being researched at all examines the premise instead. The origin of a premise is usually the weakest link in an entire chain of reasoning. Most quality problems in investment research are failures of selection. The inference is often sound, while the selection preceding it was never examined and never recorded.
None of this means that widely discussed assets are unworthy of study. Wide discussion is neutral in itself, and it sometimes reflects genuine and durable importance. The difference lies only in whether a person knows how their own attention came to be allocated as it was. Once that is known, the conclusion may not change in any respect. From that point onward, however, the conclusion belongs to the person who reached it. That difference is invisible in a portfolio and decisive over a long enough period.