Margin of safety: how big a discount is big enough?

Margin of safety is Benjamin Graham's phrase and, by some distance, the most casually used idea in value investing. It gets deployed to mean “this looks cheap,” which is not what it means at all. A margin of safety is an allowance for being wrong about a number you calculated yourself. Its size should be governed by how wrong you could plausibly be — not by how much of a discount happens to be on offer.

Which means the honest question is never “is 30% enough?” It is “enough for what?”

What the margin is actually protecting you from

Three separate sources of error, and they do not behave the same way:

1. 01The inputs. Reported figures get restated. A ten-year average of earnings can be dominated by one exceptional year, or flattered by an acquisition that will not repeat. Share counts move. Book value contains goodwill from deals that did not work.
2. 02The model. Every valuation carries an assumption doing most of the work — a discount rate, a growth rate, a multiple. Change it a little and the answer moves a lot. This is usually the largest of the three errors, and the one people account for least.
3. 03The world. Interest rates, competition, regulation, technology and demand can all move against a company that was correctly valued at the time. No discount removes this; a wide enough one buys you time to notice.

A margin of safety is not insurance against a bad outcome. It is room for the estimate to be wrong and the outcome still to be acceptable.

Sizing it: scale with the fragility of the estimate

The useful principle is that the discount should be inversely proportional to your confidence in the value figure. Businesses whose earnings power is legible and stable justify a smaller allowance than businesses where the multiple is doing all the work.

That table is a framing, not a rule, and deliberately gives no percentages. Any specific number would be a fake precision — and the ordering is the part that actually transfers.

A worked example, on hypothetical numbers

The arithmetic below uses round, invented figures for a company that does not exist. It is here so you can follow the mechanics on paper, not so you can copy the output.

Two independent methods, on the same hypothetical company, disagreeing by about 19%. The conservative move is not to average them. It is to work from the lower one and treat the disagreement as a statement about your confidence.

Now the entry question. An entry band running from a third of the floor to half of it — $26.93 to $40.40 in this example — is demanding to the point of being uncomfortable. Against a floor of $80.80 that is a discount of 50% to 67%. Most companies never trade there, and a band like that will keep you out of the market for long stretches.

That is the function, not a flaw in it. A band that demanding says: at this price, the arithmetic would have to be badly wrong before the business, rather than the stock, cost you money. Its job is to be rarely satisfied.

A margin of safety you can find every week is not a margin of safety. It is a preference.

Three ways the margin gets quietly spent

• Paying up because the story improved. The story improving is usually already in the price. If the discount you demanded was 40% and you bought at 20% because the narrative got better, you did not reduce your risk — you moved your threshold.
• Averaging down into a broken thesis. Buying more as the price falls is only rational if the value estimate is unchanged. If the reason for the fall is a reason to lower the estimate, adding is not discipline; it is anchoring.
• Letting the price re-anchor the estimate. When a stock runs, the temptation is to revisit the assumption that now looks conservative. A model whose inputs move with sentiment has no margin of safety in it at all, whatever the arithmetic says.

What a margin of safety cannot do

It does not shorten the wait. A price can sit below a well-reasoned estimate of value for years, and may never converge on it — because the market is not obliged to agree with your model, and because the estimate can be wrong in exactly the way you were trying to protect against.

And it offers no protection at all against a business whose earnings power is genuinely in decline. A discount to a value figure that is itself falling is not a discount; it is a moving target you are measuring from the wrong side. This is why a margin of safety has to be paired with a reason the earnings power persists. The discount tells you the price is interesting. Only the business tells you whether the value is real.

How Hety handles this

Hety values every S&P 500 and Dow 30 company two independent ways from its own filings — a decade of reported earnings priced at a fixed sector multiple, and its balance sheet priced at a fixed sector price-to-book — then works from the lower of the two answers as its floor. Under that floor it prints an entry band running from a third of the floor to a half of it, which is deliberately, almost insultingly conservative. The sector multiples are fixed and published, so you can see exactly which assumption produced any figure and discount it yourself. And because a cheap price is a reason to open the filing rather than a reason to act, the same page carries the seven Graham criteria against the company's own numbers, the disclosed risks from its 10-K, and a read on how candidly management argues.