Method

How the holdings are chosen

A rules-based quantitative process chooses every holding. It looks for patterns in market data and a small number of reported financial figures that have repeated in the past. We developed its rules by testing a very large number of combinations on about 70 years of past market history, with the help of AI tools.

Each list has up to ten companies, which may be held in equal or different amounts. The process runs on a fixed schedule, and we follow it without discretionary overrides. We do not publish its rules, inputs or settings.

What testing can and cannot show

The method assumes that some patterns that repeated in the past will go on repeating. Everything depends on that assumption: if those patterns stop repeating, the lists could do much worse than the market and lose money. Testing on this scale finds patterns that were luck as well as patterns that last, and no one can fully tell them apart in advance. The rules may do much worse than they did in testing, or stop working altogether.

That is why we publish no backtested, simulated or hypothetical results. The only results on this site are real ones: what our holdings did from the first list onward, at the prices we actually paid. See the record.

Risks

Each list holds only a few companies, so it can rise and fall much more than the market as a whole, and it can fall a long way. Holdings can change every month. Share prices go down as well as up, and you can lose money. Past results do not predict future results.

Changes to the method

None since the first list. If we change the rules, we will say so here, with the date, before the first list that uses them.

How AI is used

AI tools helped us research and test the rules, and they help draft our notes and company pages. A person reviews everything before it is published. The holdings are chosen by fixed rules, not by an AI making judgements.

Fundamentals

Selection uses market data and a small number of reported financial figures. The fuller analysis on this site is context and is not used in selection.

When each list is published

We trade first, then publish. Each list is sealed the evening before it takes effect. Our orders fill in the opening auction, at the official opening price, and we publish the list as soon as they have filled, on the same day.

We do this so that nobody can trade ahead of our orders. It means we buy and sell before readers see a list, so we may benefit if prices move after publication. The seal shows that the list was fixed before we traded, and we do not trade the listed holdings again until the next list.

The seal

Before a list is published, we seal it. We publish a fingerprint of the list (a SHA-256 hash of the list plus a random value), with an independent timestamp. When the list is published, we publish the exact sealed file as well, so anyone can check that the list was fixed before we traded. The random value stops anyone working out the list from its fingerprint by trying possible lists. See the archive for every seal and how to check one.

The record

The record uses the prices at which our orders actually filled, and month-end values from our broker statements. It is updated monthly. Corrections are added and dated; nothing is changed silently. See Corrections.