Lesson 4 of 9
Setting the level
The level decides whether a guardrail is a test or an ornament. The market's own cross-section is the wrong place to set one from.
By Javier Sanz
Founder of ValueScreener. Built Ninety Nine, a securities services provider, and sold it to a tier-one global bank; ran operations and internationalization at Alpaca, the brokerage API.
Who writes thisEditorial processLinkedIn (opens in a new tab)X (opens in a new tab)
In one minute
- The level decides whether a guardrail is a test or an ornament, before anything happens.
- The right reference is the company's own worst filed year, not where it sits against the rest of the market.
- If the reason survived the worst year, the level belongs below it. If it did not, that is the level.
The level decides whether a guardrail says anything. Placed at the middle of the market's cross-section, it splits companies in half for no reason tied to your thesis. Placed far into a tail, it clears almost everything or almost nothing.
One company's eight filed years, with the level under its worst one
- Gross margin
- Guardrail level
Show the numbers
| Gross margin | Guardrail level | |
|---|---|---|
| Yr1 | 42% | 34% |
| Yr2 | 41% | 34% |
| Yr3 | 39% | 34% |
| Yr4 | 42% | 34% |
| Yr5 | 38% | 34% |
| Yr6 | 41% | 34% |
| Yr7 | 37% | 34% |
| Yr8 | 36% | 34% |
The market's cross-section, and what it is worth
Across the 3,673 US companies with a computed current ratio on 2026-09-07, banks, equity REITs, insurers and mortgage REITs excluded, the spread looked like this. Graham's floor of 2.0, from chapter 14 of The Intelligent Investor, is marked on it.
Where Graham's current ratio of 2 sits
Graham's floor, 2.0: between the median (1.86x) and the 75th percentile (3.59x).As a floor it admits between a quarter and a half of the cohort.
- 10th percentile: 0.69x
- 25th percentile: 1.10x
- median: 1.86x
- 75th percentile: 3.59x
- 90th percentile: 8.06x
Across the 4,029 companies with a computed return on equity the same date, exact zeros excluded, the tenth percentile was minus 100.50 percent. Any intuition about an ordinary return was formed on large profitable companies, not on the market that actually files.
Why the market is the wrong reference
A guardrail is a condition on this company's reason, not a statement about where it ranks. The market cohort answers one question only: is the level you arrived at ordinary or extreme. It does not tell you where to put it.
The procedure runs backwards from how people do it. Take the reason. Pull the company's own filed years, as far back as EDGAR goes. Find the worst year. Ask whether the reason survived it.
If it did, the level belongs below that worst year: the guardrail exists to catch the reason failing, not the figure dipping, and the series has already shown the figure can reach that point with the reason intact. If it did not, the level is found, and found from evidence rather than preference.
Tight, loose, and the cohort a level came from
A tight level fires often; a reader told four times that nothing important happened stops reading the fifth message. A loose level fires after the reason has already been failing for several filings. Neither is a mistake. The trade is between false alarms and late ones.
Published levels carry the population they were fitted on. Graham's floor sits between the median and the 75th percentile in the figure above, admitting roughly a quarter to a half of the cohort. Altman's Z-Score zones of below 1.81 and above 2.99 were fitted on public manufacturers, and the score itself is built from market value of equity; the median across 2,773 US companies with a computed Z-Score on 2026-09-07, from prices last stored on 2026-08-26, with banks, capital markets firms, REITs, insurers and specialty finance excluded, was 2.37: it transfers badly to banks, insurers and asset-light service companies. A level borrowed from either is a level fitted to a different population for a different question.
Check yourself
What is the right reference for setting a guardrail's level?
- The market's median for that figure
- The company's own worst filed year
- The figure's value on the day you buy
Show the answer
Answer: B.
The market cohort tells you whether a level is ordinary or extreme. It does not tell you where the level should sit for this company's reason.If a reason survived the company's worst filed year, where should the level go?
- At the worst year's exact value
- Below the worst year, since the reason has already shown it can survive that point
- At the market median instead
Show the answer
Answer: B.
A level at the worst year itself would fire on a repeat of a year the reason already survived.What does it mean that Altman's Z-Score zones were fitted on public manufacturers?
- The zones apply equally to every industry
- The zones carry that population, and transfer badly to banks, insurers and asset-light service companies
- The score cannot be computed for any other industry
Show the answer
Answer: B.
A published level is fitted to a specific population for a specific question. Borrowing it elsewhere borrows an assumption that may not hold.Sources
- Graham, The Intelligent Investor, revised edition 1973, chapter 14, on the defensive investor's fixed criteria.
- Altman, Journal of Finance, 1968, on the Z-Score and the manufacturer population it was fitted on.
What this leaves out
- Every cohort figure on this page was computed on 2026-09-07 and is rebuilt on a cadence; a later reader sees different breakpoints.
- Each cohort's exclusions are stated because they change what the number means. A company absent from one is absent because the metric does not apply to it.
- The worked series and level are round invented numbers, chosen to show the procedure rather than any company's history.
- Nothing here is advice about any security.