1. What a factor is and where the idea came from

Level 9: Factors

Level 1 gave you the map of where an edge can exist and level 2 explained why the market resists giving it away. This level goes into the family of edges that has withstood the most scrutiny and that is still within reach of someone with no special resources: factors. And it starts with what almost nobody bothers to clarify before selling you one, which is what exactly a factor is.

A shared characteristic, not a company

A factor is a measurable characteristic shared by many shares that explains part of the return of that group. No more and no less than that. If you separate all the shares in a market into piles according to something objective (how expensive they are relative to their profits, how big they are, how much debt they carry, how much they have risen over the past year) and the piles behave systematically differently over decades and across many countries, that characteristic is a factor.

Three things are worth pinning down from the outset, because almost every later misunderstanding is born of skipping them:

• A factor is a difference, not a level. It is measured by comparing one group with another. The value factor can have an excellent year while the whole market collapses, because what it measures is how much better the cheap ones did than the expensive ones, not whether you made money.
• A factor describes a group, not a company. That a particular share is cheap says nothing reliable about that share. The claim only means something applied to hundreds of them at once and over a long period.
• A factor is a measurement, not a cause. The name you give it explains nothing. The explanation is a separate job, and it is the one covered in the last module of this level.

From a single factor to several

The first model that brought order to all this held that only one thing mattered: how much a share moved with the market as a whole, its beta. According to that model, the expected return of any share depended solely on that sensitivity, and the rest of its swings was noise that disappeared on diversifying and that nobody had any reason to pay you for. Elegant, coherent and testable. That last part was its problem.

When it was tested against data in the 1970s and 1980s, whole groups of shares appeared that returned systematically differently from what their beta predicted. Small companies returned more than they should have. Those trading cheaply relative to their book value did too. At first they were called anomalies, which is the word used when something does not fit the model but the data keep insisting on it.

Eugene Fama and Kenneth French formalized the way out in the early 1990s: if one characteristic is not enough, use several. Their model added two more to the market, size and value, and explained far better why some diversified portfolios returned differently from others. Shortly afterwards momentum was incorporated, and years later Fama and French themselves extended theirs with the company's profitability and investment policy. What matters is not the names or the dates, but the movement: the field went from one characteristic to several because the data did not fit into one.

What it was really for: allocating the credit

Here is the original use of all this, and it was not to make anyone rich. Before multi-factor models, a manager who beat his index for ten years was, by definition, a talent. Afterwards, the question changed shape: did he beat it because he picked well, or because his portfolio was permanently tilted toward small, cheap companies, which on average did better over that period?

That question can be answered with numbers, and the answer turned out to be uncomfortable for the industry. A three-characteristic model explained more than ninety percent of the differences in return between diversified portfolios, considerably more than the market managed on its own. A good part of what was being sold as talent was exposure. And exposure, unlike talent, can be bought cheaply and does not depend on one particular person continuing to get out of bed motivated.

Explaining the past and predicting the future are different operations

This is the point that separates someone who understands factors from someone who repeats them. A well-built factor explains the past almost by definition: it was picked, from among many possible characteristics, precisely because it sorted returns that had already happened. That it manages to do so is not a discovery, it is the procedure.

The question that really matters is a different one: is it going to keep paying? And that one is not answered with the same data. Measuring that tall players score more baskets describes games already played; it guarantees nothing about the next tall player you sign, still less if every team has spent a decade signing tall players.

The error always slips in at the same point. A characteristic is found that sorts the past well, a serious-sounding name is given to it, an economic story that sounds reasonable is then built around it, and the package is presented as a source of future edge. Each step is defensible on its own and the conclusion does not follow from them. A factor is not a money machine: it is a description. It only becomes a reasonable expectation if you can answer why the difference existed and why that reason is going to still be there tomorrow.

Real case: four points almost nobody actually collected

In the US market, the average annual difference between the group of cheap shares and the group of expensive ones has been, in the most widely used long series, on the order of four percentage points since the late 1920s. It is a real figure, measured over almost a century, and it is the sales argument of an entire industry.

It is also a figure that gets misread daily. Four points on average over ninety years does not mean four points in almost every year: it means whole decades in the red offset by short, spectacular runs, and an investing experience almost nobody endures in full. The average existed. The investor who collected all of it, practically not at all. That is what the next module is about.