Graham set the rule. Buffett set the condition. We keep the two answers apart.

And where we go further

Ranks, not thresholds

Graham wrote “P/E below 15” in 1949. A fixed threshold describes its author's market, not today's. Every stock is scored against the whole universe of the day, and within its own sector: an 8% margin does not mean the same thing in retail as in software.

Leaving a list, not just entering one

Neither Graham nor Buffett ever formalized when to stop looking at a company. Each of our rankings has two frontiers, not one: a stock enters by crossing, and leaves only by crossing back with a margin. Without that gap, a name would flicker in and out from one morning to the next on a rounding difference.

Rules written before they are applied

No rule is kept because it sounds plausible. It is written down first, replayed on real data using at each date only what was known that day, and kept only if it still holds when the year closes. Several convincing ideas were dropped at that stage.

The families of criteria are public, and taught in full in the Curriculum. They are the ones any serious value investor uses: our work is not having found them, it is having calibrated them — and having refused to merge them.

Listed property trusts are excluded and financial companies are handled separately: their accounts cannot be read with the same ratios, and mixing them would build a ranking that means nothing to anyone. The universe is bounded to U.S. market capitalizations between $200M and $5B — where the fundamental data is deep enough and clean enough to support a daily ranking.

The exact weights inside each rating, the percentile frontiers that put a stock into a ranking, and the margin it must cross back to leave one, all remain ours. This is not coyness: the calibration is precisely what separates a list of well-known criteria from a method that produces a result.

That calibration is not fixed, and it is where the work lives. At every closed year, every rule is re-examined against up-to-date data and kept only if it still holds. Several settings that had a good reason to exist in 2021 have since been removed, because the reason had stopped being true — and the result improved for it.

What you can check for yourself is left whole: the two rankings (Avoid, Shortlist) are published every morning, every rating and every position is dated. The portfolio, our third pillar, is open to members. The method works in plain sight, even if its calibration does not read.

A historical test is easy to make flattering: you only have to forget one detail. Here is the complete list of what ours handles before a rule is kept.

Strictly point-in-time

At every calculation date, selection uses only the data published by that date. No later information ever enters.

Real filing dates

A financial year becomes usable on its actual filing date, not its closing date. A company closing on 31 December reports in February or March: using it from 1 January would mean collecting the market's reaction to results nobody has seen. Where the filing date is missing, a 90-day lag applies. Filing dates that precede the close — the provider does produce them — are discarded.

Delisted companies included

The universe holds 2,242 companies that disappeared over the period: bankruptcies, takeovers, delistings. A backtest built only on survivors can literally never lose money on a bankruptcy.

Delisting treated as such

A liquidated company leaves the selection at its last quoted price. Without that rule it would sit there for ever, marked at a price that no longer exists, occupying a slot without ever costing or earning anything.

The market cap of the day

The size filter uses capitalization reconstructed at each date — the share count of the time multiplied by the price of the time — not today's. Keeping a 2021 company because it grew into a mid cap by 2026 is knowledge of the future.

Eligibility over the observable window

A company enters the universe if it occupied the size band during the period tested, not if it occupied it ten years earlier.

Ratios refreshed to today's price

Enterprise value and free cash-flow yield are recomputed at today's price from the price observed at the report date. A ratio frozen at publication ages by several months between two financial years.

Dividends reinvested

Distributions are collected and reinvested, on the selection and on the comparison indices alike. Gross: tax depends on each investor's wrapper and residence, and computing it would mean inventing a figure.

Transaction costs, and what they cover

Every buy and every sell bears a cost of 10 basis points. We do not model the bid-ask spread line by line — it depends on a given day's order book, which no historical database restores honestly. Sensitivity is therefore tested up to 100 basis points, ten times the working assumption: at that level the simulated cost comfortably absorbs the bid-ask spread of a liquid U.S. small cap. A method whose result rested on an optimistic cost assumption would not be one.

Whole shares

The backtest buys whole numbers of shares with finite capital, like any investor. Fractional positions would produce a result nobody can reproduce.

Past performance is no guarantee of future results. Investing in equities carries a risk of capital loss, up to and including the entire amount invested. Value Investing Club is an information and education service; it provides no personalized investment advice and takes no account of your situation, your objectives or your tolerance for risk.

One point deserves to be said plainly, because it decides the outcome more surely than the selection itself: this method demands that you hold. A discount takes months, sometimes years, to close, and the path is never a straight line. Nothing guarantees that it closes at all. An investor who sells at the trough, driven out by noise, does not collect a smaller return: they collect the loss. If price swings keep you awake, or if you may need this money in the short term, this approach is not for you.

Read the full risk warning