Discounted Cash Flow: Build One, Then Watch It Break
A discounted cash flow in six lines on an invented company, how much of the answer one terminal assumption carries, and the three inputs only you can supply.
See a sample reportThis is a method, not a recommendation. Nothing here, or anywhere else on Taufolio, is investment advice. Treat every example as a starting point for your own research.
A discounted cash flow is one sentence with arithmetic attached. This business will produce a stream of cash, money arriving in ten years is worth less than money arriving tomorrow, so the whole stream can be added up into one number for today. That is the idea, and it is a good one.
The trouble starts one line below it. The discount rate, the number that converts later into now, is not observable anywhere. You choose it. Move it by a single percentage point in the worked example below and the answer moves by roughly 15 percent, with nothing about the business changed. The model therefore does not replace the question of where today's price sits against the company's own history. It stands next to it.
Most writing about this model then picks a side. One camp treats the output as a value and sorts companies by it. The other says the whole exercise is guesswork with a spreadsheet. We think both are wrong in the same way, because both treat the model as a machine for producing an answer. It is an instrument for finding out which assumption your conclusion is standing on, and that is a more useful thing to own than a number. The same is true of the multiples most people reach for instead, which make the identical assumptions without asking you to write any of them down.
What a discounted cash flow actually is
The discounting part is the only piece that needs defending, and it defends itself. A thousand dollars promised in five years is not a thousand dollars. It might not arrive, and even if it does, you gave up five years of doing something else with the money. Dividing by one plus your required return, five times over, is the arithmetic of that sentence.
At a 9 percent required return, a thousand dollars five years out is worth about 650 today. At 12 percent it is worth about 567. The gap between those two numbers is not an opinion about the business. It is an opinion about you.
Everything else in the model is bookkeeping around that one idea.
The six lines you need to build one
Here is a full model. The company is invented, I will call it Halvard Instruments, and every number below is made up so that you can reproduce the arithmetic rather than trust it.
free cash flow, last year 200
growth, years 1-5 8% a year
growth after year 5 2% a year, forever
required return 9%
year 1 200 x 1.08 = 216.00 ÷ 1.09^1 = 198
year 2 233.28 ÷ 1.09^2 = 196
year 3 251.94 ÷ 1.09^3 = 195
year 4 272.10 ÷ 1.09^4 = 193
year 5 293.87 ÷ 1.09^5 = 191
sum of five years = 973
terminal value 293.87 x 1.02 ÷ (0.09 - 0.02) = 4 282
4 282 ÷ 1.09^5 = 2 783
value of the business 973 + 2 783 = 3 756
Free cash flow, the starting line, is operating cash minus the capital spending the business needs to keep running. It is the money genuinely available to everyone who financed the company, which is why following one dollar through the business is worth more preparation than the model itself.
Six lines, one output, about 3.76 billion for the whole enterprise. To get from there to what the shares are worth you would subtract debt and add cash, which is a step of its own and not the interesting part.
The interesting part is where those 3.76 billion came from.
The starting line is already a judgement call
Before any discounting happens, somebody decided what free cash flow means, and that decision is less settled than it looks.
Capital spending is the obvious argument. A company that spends 400 on plant may need only 150 of it to keep the current business running, with the rest buying growth that is not in the forecast yet. Subtract all 400 and you punish the company for expanding. Subtract only the maintenance share and you have made up a number, because almost nobody discloses the split. Which of the two is correct? Both are defensible, and they produce different valuations of the same business.
Share-based pay is the second argument and it is sharper. It costs the company nothing in cash, so it never leaves the operating cash line, yet it hands away a slice of the company every year. Treat it as free and your model quietly values a business whose owners are being diluted. Subtract it and your free cash flow drops by whatever the grant was worth.
So the first line of a six-line model is already two opinions deep. Write down which ones you took, because your future self will not remember.
Three quarters of the answer is one assumption
Look again at the two blocks. The five years you forecast in detail, argued over, checked against segment disclosures and management guidance, contribute 973. The single line that stands for every year after that contributes 2 783.
That is 74 percent of the result, produced by one formula about a period nobody can forecast.
The share is not a bug. A business is expected to run for far longer than five years, so most of its value genuinely lives beyond the forecast window. So where should the argument go? Not to year three. What follows from that 74 percent is a question about where your attention is worth spending. Move terminal growth from 2 percent to 3 percent, leaving everything else untouched, and the total rises from 3 756 to about 4 252. One point on a number describing the 2060s moves the answer by 13 percent, while a careful argument about next year's margin moves it by a fraction of that.
There is a second trap inside the same formula. Terminal growth has to stay below the discount rate, or the denominator turns negative and the model produces nonsense with a straight face. Set growth at 9 percent against a 9 percent return and the spreadsheet divides by zero. Set it at 8.9 and the denominator shrinks to a thousandth: the same sheet returns about 209 billion, fifty-six times the answer it gave a moment ago.
One point on the discount rate
Now change only the required return and leave the business alone.
| required return | value of the business | against the 9 percent run |
|---|---|---|
| 8 percent | about 4 400 | 17 percent higher |
| 9 percent | about 3 756 | the base case |
| 10 percent | about 3 273 | 13 percent lower |
Three runs, one input moved by a percentage point each way, and the top of the range sits about a third above the bottom. Nobody visited the company between those runs. Nobody read a filing.
So which of the three is right? None of them, and that question is the mistake. What the table actually says is that your conclusion about Halvard Instruments depends more on your own required return than on anything you learned about the business, and that is worth knowing before you spend an evening on the segment notes.
This is where the usual verdict gets applied, and it goes: the model is too sensitive, so drop it. The logic does not hold. A thermometer that moves when the temperature moves is not broken. Sensitivity is the output, and a model that reported the same value under every assumption would be useless rather than reliable.
The three inputs that have to be yours
Build the model twice and you notice that only three numbers really matter, and that none of them is in a filing.
The first is the required return. It bundles together what a safe asset pays today and how much extra you want for taking equity risk. Both parts move, both are opinions about the future, and there is no authority publishing the right answer for your situation.
The second is the growth path across the forecast years. Filings tell you what happened. The model asks what happens next, which is a different kind of claim and belongs to you.
The third is the terminal assumption, which as we just measured carries most of the weight. It asks what fraction of today's business still exists after your forecast window, and honest answers are rarely round.
Three subjective inputs, one output with three decimal places. That combination is exactly how a model built to test assumptions turns into a number people quote.
There is a tidy-looking way out of the first input, and it deserves a fair hearing. Textbook practice builds the discount rate from a weighted cost of capital. The cost of debt comes from what the company actually pays on its borrowings. The cost of equity comes from a rate on safe government paper plus a market premium, scaled by how much the share moves with the market. Every component is documented, and the result feels earned rather than chosen.
It is still a choice. The market premium has no observable value and estimates of it differ by several percentage points depending on the window and the method. The sensitivity of the share to the market changes with the period you measure. Two careful analysts using the same textbook will arrive at rates a point or two apart, which the table above already priced for you.
Why we do not run this model for you
Taufolio does not publish a discounted cash flow valuation of a company, and the reason is specific rather than philosophical.
All three inputs above belong to the reader, not to us. Any default we shipped would be a number invented at the moment it was needed, and then carried into a report where it would look like a finding. A model that outputs 3.76 billion when we quietly assume 9 percent, and 4.40 billion when we quietly assume 8, is not research. It is our preference wearing the costume of arithmetic.
There is a second reason and it is a rule rather than a judgement. We do not put a price on a real company anywhere on this site, in any form: not a number, not a range, not a percentage away from today's quote. What we publish instead is what a business earned, how that changed, and a link to the document behind every figure, so the reader can build whatever model they trust.
Does that make the model useless for a reader? The opposite. The model is useful precisely because it is yours, and a version produced by somebody else, with their required return baked in, is the one version that teaches nothing.
Analysts who do publish those numbers usually run a version of this model, among others. The useful part of their work is the assumption sheet, not the headline, which is one reason the last page of a research report is worth more than its first.
A multiple makes the same assumptions and hides them
The most common alternative is a valuation multiple read against a company's own history. It feels safer because nothing had to be forecast. That feeling is wrong.
Reading a multiple against a five-year median assumes the business is the same business it was five years ago, that margins hold, that growth continues near the old pace, and that the market keeps paying what it used to pay for those characteristics. Those are forecasts. They are simply not written anywhere, so nobody argues about them.
Both methods rest on the same beliefs. One prints them in cells you can attack, the other leaves them in your head where they cannot be attacked. If a colleague challenges your discount rate, you can answer. If a colleague challenges your median window, most people discover they never chose it on purpose.
The practical conclusion is not that one method wins. It is that the model you can argue with is the one that will teach you something.
Which is why we would rather a reader build a rough model and defend it than copy a precise one and trust it. A number carried across from somebody else's spreadsheet has no assumptions attached, and an assumption you cannot name is an assumption you cannot revisit when the company reports.
Turn the model around instead
There is a version of this arithmetic that never produces a value, and it is the version we find most useful.
Instead of choosing growth and solving for value, take the company's current market value as given and solve for the growth rate that would justify it. Same equation, different unknown. The output is not a price and cannot be turned into one. It is a sentence of the form: at today's market value, with a 9 percent required return, this business is expected to grow free cash flow at roughly this pace for a decade.
That sentence has a property no valuation has. It can be checked. You can hold the implied growth rate against what the company has actually delivered over the past five years, against the capacity it has disclosed, against what the industry grew at. If the number implied by today's market value is three times anything the business has ever done, you have learned something concrete, and you learned it without an opinion about what the shares are worth.
Two cautions travel with it. The implied rate depends on the required return you assumed, so it is not a fact about the market, only a statement conditional on your own number. And it must never be converted back into a price, a range, or a percentage from today's quote, because that turns a description of what is priced in into a forecast of where the quote is going.
Write the implied number down with the date, and revisit it after the next set of results.
What the model is actually good for
Build it. Break it on purpose. Then keep the wreckage.
The practical routine is short. Build the model once with your honest inputs. Run it again with the required return a point higher and a point lower. Run it a third time with terminal growth a point lower. If your conclusion survives all three, it rests on the business; if it flips, it rests on the discount rate, and you now know which evening's reading was wasted.
Then write the three surviving assumptions into whatever you use to record why you hold something, with a date and a condition that would break each one.
We take the other half of that work seriously: the Full report assembles what a company earns, how the numbers moved and where each one came from, and how those reports are produced is documented separately. The discount rate stays with you, because the risk does.
Which is the honest division of labour.
Nobody can pick your required return for you.
Frequently asked questions
What is a discounted cash flow model?
How do you calculate a DCF step by step?
Why is a DCF so sensitive to the discount rate?
What is terminal value and why does it dominate the result?
Is a DCF better than valuation multiples?
Does Taufolio calculate a DCF for a company?
- fundamentals
- research
- methodology
Posts are produced with AI tools and go through editorial review by the Taufolio team before publishing. How we write and correct posts