Intrinsic value is an estimate, and the margin of safety admits it

Intrinsic value cannot be observed

Intrinsic value is what a business is worth on the cash it will produce over its life, brought back to what that cash is worth today. It is not a figure you can look up, because the cash has not been produced yet. Every intrinsic value anyone quotes is the output of a model.

That sentence is the whole article, and it is worth slowing down on, because the way intrinsic value is usually introduced quietly implies the opposite. “The market price is what you pay, intrinsic value is what you get” sets the two side by side as though they were the same kind of object — two numbers, one visible and one merely harder to find, like a price hidden behind a curtain.

They are not the same kind of object at all. A price is a fact: somebody handed over that amount this morning, and it is remade every second the market is open. An intrinsic value is a conclusion drawn from assumptions about a future that has not happened. One is measured, the other is constructed.

This is the same asymmetry that runs through the price ratios, only pushed further. A price ratio divides a fact by an accounting figure — an opinion about the past. A discounted value divides nothing; it builds a number entirely out of opinions about the future, and then reports it to the rupee.

None of which makes the exercise pointless. Working out what a business is worth on its own cash flows is the only valuation question that does not depend on what somebody else will pay you. It is simply an exercise whose output carries an error bar, and the honest version of it starts by saying how wide.

Three inputs, and all three are assumptions

The machinery is short enough to state in full. Take the cash the business is expected to produce for its owners in each future year, convert each year's figure into what it is worth today, and add them up. That conversion is the only piece of finance involved, and it is the subject of the time value of money: a rupee arriving in five years is worth less than a rupee arriving now, so a rate is applied to shrink it.

The cash being forecast is usually what the business generated from operating, less what it had to spend on assets to keep going — free cash flow. What that figure includes, and how many ways it can be moved without anything real changing, is set out in free cash flow. It is already an estimate before anyone starts forecasting it.

So a discounted valuation needs three things, and here is what each one actually is:

• The cash flows — how much the business produces each year for some stated number of years. This is a forecast of a business you do not run, in an economy nobody can describe in advance.
• The discount rate — the rate at which future rupees are shrunk. It is meant to capture what the money could earn elsewhere and how uncertain these particular cash flows are. There is no observable market quote for it; it is chosen.
• The terminal assumption — what happens after the last forecast year. Since a business does not stop, the model has to assume something about the rest of time, usually a modest growth rate carried on forever.

Notice what is missing from that list: anything measured. The share count is a fact and the current cash balance is close to one, but the three inputs that decide the answer are all judgements. The model is arithmetic performed on opinions, and arithmetic does not improve its inputs.

The same business, valued nine ways

Arguing about this in the abstract gets nowhere. So here is a business — an illustration, not a real company — that produces ₹100 crore of owner cash flow next year. Assume it grows at some rate for ten years, then settles into a steady growth rate forever, and discount the lot at some rate. Those three rates are the only things that change below.

Every rate in this table was chosen to be legible. None is typical, none is recommended, and there is no such thing as a typical value for any of them — saying otherwise would itself be a claim. The point is not which row is right. The point is the distance between the rows.

Read the first row and the last two together. Nothing about the business changed across those three lines — not a rupee of cash flow, not a customer, not a factory. Growth moved by two percentage points, the discount rate by two, the terminal rate by one. The answer ran from ₹1,115 crore to ₹2,782 crore, which is two and a half times from one end to the other.

Individually each input is well behaved, and that is what makes the combination deceptive. Two percentage points off the growth rate costs 12%; two percentage points off the discount rate adds 33%. Neither is alarming on its own. They compound when they move together, and one person fills in both from one view of the business — optimism about growth and comfort about risk come from the same place. That is a judgement about how models get filled in rather than anything measured, and it is the reason to read the corners of a grid rather than its middle.

Divide by an illustrative 10 crore shares and the base row is ₹164.37 a share, with the corners at ₹111.45 and ₹278.23. Both decimals in ₹164.37 are real arithmetic and both are entirely spurious. The model can only be as precise as the assumption it is least sure about, and that assumption is a ten-year growth rate for a business nobody has met.

Most of the number lives past the forecast

There is a second, quieter problem, and the base row above shows it plainly. Split that ₹1,644 crore by where it comes from.

• Years 1 to 5: 26% of the total value.
• Years 6 to 10: 23%.
• Everything after year 10 — the terminal assumption: 51%.

Just over half the answer rests on a single line stating what a business will do after the tenth year, forever. It is also the input with the least available to discipline it — no part of it can be checked against a filing, a competitor or a quarter — so it gets settled by picking a small growth rate that sounds unobjectionable. That one rate carries more weight than the entire decade of forecasting above it.

The arithmetic behind that concentration is worth seeing, because it explains why the terminal figure is so touchy. A perpetual value divides next year's cash by the gap between the discount rate and the growth rate. With an 11% discount rate and 2% growth the divisor is 9 percentage points; move growth to 3% and the divisor becomes 8. The value of everything after year 10 rises by more than a tenth because the denominator shrank, not because anything was learned.

And this produces a specific, recognisable mistake. Effort in a valuation flows to the near years — the next four quarters get modelled line by line, margins debated, one order win argued over — because those are the years anyone can say something concrete about. Those years are 26% of the answer. The half of the answer sitting in the terminal line gets a round number and no discussion at all. Attention is inversely proportional to weight, and the model does not warn you.

The trade-off has to be stated rather than glossed. A longer explicit forecast moves value out of the terminal line and into years you have at least thought about — but the extra years are the ones you know least about, so what you have bought is not accuracy. It is a redistribution of the same uncertainty into cells that look more considered.

What the margin of safety is actually for

It absorbs error in your own estimate. It is not a discount that turns into extra profit when the price catches up — that is the usual framing, and it is the wrong way round.

It is often described as buying below value so you make more when the price catches up — a discount as extra profit. That reading assumes the estimate is right and the price is wrong. The grid above says you cannot assume that. The discount is there because your own number may be wrong, and the gap is the amount of wrongness you can absorb without paying more than the business is worth. The phrase is Benjamin Graham's, and the logic in it is structural rather than empirical: it does not claim to improve returns, it claims to make being wrong survivable.

The arithmetic is simple and worth carrying around. Buying at 30% below an estimate means paying 70 paise for each rupee you think is there. Your estimate can therefore be overstated by 1/0.70 − 1 — about 43% — before you have paid more than the business is actually worth. The discount is stated on price; the error it tolerates is larger, because it is stated on the smaller number.

Now put that back against the grid, which is the part most explanations skip. Take 30% off the base row and you get ₹1,151 crore. The pessimistic corner of the same grid — growth two percentage points lower, discount rate two higher, terminal rate one lower — was ₹1,115 crore. A 30% discount buys back roughly, and not quite, one corner of the grid. It is not a cushion on top of a reliable answer. It is barely enough to cover the range the answer already had.

Which is exactly why this article will not tell you what discount to use. A number stated as a rule — a third off, half off — is a recommendation dressed as arithmetic, and worse, it is stated without reference to the one thing that should set it: how wide your own range came out. A business whose corners sit 15% apart and one whose corners sit 150% apart cannot sensibly take the same discount, and only the second calculation tells you which you are holding.

The cost is real and gets left out of most descriptions. A gap between price and your estimate exists only where the market disagrees with you, so the wider the gap you insist on, the more the rule requires that disagreement to be large — and the more of the time it will find nothing that qualifies. Insisting on a large discount means owning nothing for long stretches, and waiting has a price, paid in the returns of whatever you held instead. A margin of safety is not free — it buys error tolerance with time out of the market, and pretending otherwise is the free-lunch version of the idea.

One more limit, because it is the failure people actually run into. The discount protects the estimate, not the business. If the cash flows deteriorate for reasons the model never contained — the product stops selling, the regulation changes — the value moves rather than the error, and the gap is spent immediately. Error room absorbs an assumption that was off. It does nothing about an assumption that was answering the wrong question.

Run the model backwards instead

There is a use of the same machinery that avoids most of the trouble above, and it is the most useful thing in this article.

Instead of estimating value and comparing it with the price, take the price as given and solve for what it implies. Feed the market's own valuation into the model and ask which growth rate makes the arithmetic balance. The output is no longer a verdict on what the business is worth; it is a plain statement of what the price already assumes. Running the same machinery in that direction is a reverse discounted cash flow.

With the illustrative business above — ₹100 crore of cash flow next year, an 11% discount rate and 2% growth after year 10 — a market value of ₹1,644 crore implies 8% growth for ten years. At ₹2,000 crore the implied rate is 10.9%. At ₹2,500 crore it is 14.2%. Those are the same three chosen inputs run in reverse, and every one of them is re-derivable from the model.

The reason this is better is a matter of what kind of question you end up answering. “What is this business worth?” requires you to produce a number out of nothing and has no check on it. “Can this business compound cash at 14.2% for ten years?” is a question about a specific business that its own history, its industry structure and its capital requirements bear on directly. You will still often answer it wrongly. But you can tell that you are wrong later, which is not true of a number you invented.

Note carefully what this does not do. It does not say a high implied growth rate means a share is expensive or a low one means it is cheap — the implied rate is only as meaningful as the two inputs you fixed to compute it, and moving the discount rate moves it substantially. It converts an unanswerable question into a checkable one. Whether the answer to the checkable one is favourable is a judgement, and it stays yours.

Five ways this goes wrong in practice

Each of these is common enough to be worth naming, and each one is produced by the model rather than by carelessness.

Precision theatre. The output arrives as ₹164.37 a share and is discussed as though the paise were information. Nothing about a ten-year forecast supports two decimals. The defence is to carry the range rather than the point — report the corners, not the middle — and to notice that a model reporting one number has hidden its own error bar as a formatting decision.

Solving backwards to a conclusion you already hold. The inputs are adjustable and the answer is sensitive, so whatever value you set out to justify is reachable, entirely honestly, one defensible tweak at a time. This is anchoring and confirmation bias with a spreadsheet attached, and the tell is that the inputs were revised after the first answer rather than before.

Applying error room to the wrong quantity. A 30% discount tolerates an estimate 43% too high. If the honest range on your own inputs is wider than that — and the grid above ran 2.5 times from corner to corner — then the discount is not covering the error, it is decorating it. The size of the gap should follow from the width of your range, which means the range has to be computed first.

Mixing real and nominal. If the cash flows are grown at a rate that already includes rising prices but the discount rate is chosen as though it did not — or the other way round — the answer is wrong by a factor nobody will spot, because both numbers look reasonable in isolation. Keep both in the same terms; what that means is in inflation.

Treating the model as the analysis. The forecast has to come from somewhere, and the somewhere is the accounts: how the profit was earned in the profit and loss account, what capital it took to earn it in return on capital, and what obligations sit against it in the balance sheet. A discounting formula applied to numbers nobody interrogated is a very tidy way of being confidently wrong.

Where the inputs come from, and what a tool can do

Everything above needs source data, and the sources are ordinary. Cash flow statements, profit and loss accounts and balance sheets sit in the company's annual report and in its exchange filings. Read the consolidated statements rather than the standalone ones where a group has subsidiaries. Share counts, including the diluted count, are in the same filings.

What software genuinely helps with is the part people skip: computing the range rather than the point. Running the same model across a grid of growth and discount assumptions takes seconds and turns a single figure into a spread, which is the form the answer was always in. Running it backwards from the current market value — market capitalisation plus debt, less cash — produces the implied rate discussed above.

FNOTrader's Market Pulse scanner covers price, volume and market breadth across NSE, and the discipline that applies to any filter — knowing exactly what each field is computed from before ranking anything on it — is set out in our screening guide. A screen narrows a list. It does not value a business, and no field on it is an intrinsic value.

Past performance does not indicate future results, and nothing here is a recommendation to buy, sell or hold any security. FNOTrader is not a SEBI-registered investment adviser or research analyst; the figures in this article are illustrations chosen to demonstrate arithmetic, not statements about any company or any market.

Common questions

Continue reading

More in Stocks & Screening · App: Stocks · Definitions: glossary · Free tools: calculators · All: every article