1. From the idea to the hypothesis that can be refuted

Level 12: A process that survives

The five previous levels describe a territory: where an edge can exist, who erases it, what makes it last a while and why it ends up being used up. This level adds no territory: it adds a method for walking through it. And it starts with the step almost everybody skips, because it looks administrative and it is the one that decides the outcome: turning a hunch into a statement the world can contradict.

A hunch is not a hypothesis

A hunch sounds like this: small quality companies are cheap. It is an agreeable sentence and practically useless, because it does not say what counts as small, what quality is, cheap relative to what, over how long, compared with what alternative, nor how much you expect to make. Above all, it does not say what observation would show it up. A sentence that can accommodate any result is not an investment idea: it is a mood with a financial vocabulary.

A usable hypothesis, by contrast, has parts that can be argued over separately. A concrete universe. A rule another person could apply without asking you anything. A horizon. An explicit comparison, because a return with no reference means nothing. An expected magnitude, because if you do not write it down you will end up accepting any positive figure as good. And two elements almost nobody writes down and which separate serious work from entertainment: the reason and the counterparty.

Without a reason there is no hypothesis, only a pattern

A pattern is a regularity observed in some data. A hypothesis is a statement about why that regularity exists and why it should still exist tomorrow. The difference matters because financial data are short, noisy and full of coincidences, and because, as became clear in level 2, a price is not a neutral number: it is the conclusion of somebody who has looked at the same thing you have and has decided to sell. If your explanation for why the pattern exists is that the market has not noticed, you have to be able to say who that distracted market is and why it pays them to stay distracted.

Levels 3 and 4 raised the bar higher still. If your idea is a version of a known factor, you are not discovering anything: you are collecting a premium that already has a name, that was published, that got crowded and whose capacity is finite. That may still be reasonable, but it is a different hypothesis and it has to be written down as such.

Who is on the other side and what mistake they are making

This is the most useful filter in the whole course, and it can be stated in one line: if you do not know what mistake is being made by whoever is selling to you cheaply, the one who is probably wrong is you.

There are legitimate reasons why somebody sells at a price that does not reflect value, and none of them is the seller's stupidity:

• They sell because they have no choice. A mandate forbids them from holding the asset after a downgrade, redemptions from unitholders oblige them to liquidate, or a margin call forces them to close. The price is irrelevant to them: their constraint is not value, it is the date.
• They sell because their objective is not to make money on that trade. Level 5 showed the cleanest case: an index fund does not buy or sell because something looks expensive or cheap to it, but so as not to separate from its benchmark.
• They sell because the professional cost of holding on is greater than the expected benefit. Explaining three years of results worse than the index costs you the job; being right in your analysis does not save you.
• They sell because the asset is uncomfortable to hold: wide spreads, little volume, scant information, or a sector with a bad reputation among their clients.

None of these sellers is a fool: all of them are acting rationally within their constraints. That is what level 2 called limits to arbitrage, seen from the other side of the table. If your answer to the counterparty question is a shrug, the hypothesis is not finished.

The bar: if nothing could refute it, it is not a hypothesis

Campbell Harvey has spent years maintaining an idea the industry finds uncomfortable: the usual standard of evidence is too low. When hundreds of factors have been published on essentially the same series of prices, accepting a result because it has less than one chance in twenty of being due to chance is a guarantee that a good part of what is published is well-dressed noise. His proposal is to raise the bar explicitly: the more that has been searched for, the more extraordinary the finding has to be for you to believe it.

Translated into your notebook, the bar becomes a single question, and it has to be answered before you look at anything: what specific observation would make me abandon this idea. If there is none, you do not have a hypothesis, you have a faith. And a faith cannot be tested, only defended.

A finished hypothesis fits into seven lines:

• Statement: what I buy, what I sell or avoid, and by what exact rule.
• Reason: why the price difference exists.
• Counterparty: who is selling cheaply and what constraint is making them do it.
• Expected magnitude and horizon: how much and over how long.
• Capacity: up to what size it makes sense, according to level 4.
• Refutation condition: the observation that kills it.

Real case: the fallen angels of 2020

In the spring of 2020, several large companies lost their investment grade rating and their debt came to be considered high yield. Ford was the most talked-about example, with tens of billions of dollars of bonds affected, and the total volume downgraded that year ran to the order of hundreds of billions. Many institutional holders were forbidden by mandate from holding debt below that threshold: they had to sell, and within a few days. The counterparty was perfectly identified and its mistake was not analytical, it was contractual.

The hypothesis wrote itself, refutation condition included: this stops working if a buyer appears who is insensitive to price and has unlimited capacity. Weeks later, the Federal Reserve announced that its credit purchase programs would include some of those downgraded bonds. The condition written in advance was met, and whoever had written it did not have to improvise an explanation.