Note · Opinion, not rated
Most of a discounted cash flow is a number you made up.
The ten years of forecasts everybody argues about are the minority of the answer. The majority sits in a single line called terminal value, built from two assumptions nobody can check. This is a criticism of our own primary method, which is why it is worth making.
Where the answer actually comes from
A discounted cash flow has two parts. An explicit forecast, usually ten years, where you project cash flows line by line. And a terminal value, one number standing in for everything after that, because the business does not politely stop existing in year eleven.
Analysts spend almost all their time on the first part. Here is what it is worth.
Worked example
$100M of free cash flow growing 8% for 10 years, 9% cost of capital, 2.5% perpetual growth after that
- Present value of the explicit 10 years
- $951M
- Present value of the terminal value
- $1,438M
- Total value
- $2,389M
- Share sitting in terminal value
- 60.2%
A decade of careful forecasting produces 39.8% of the answer. The rest is one formula fed by two numbers you chose.
Forecasting further out does not fix it
The obvious response is to extend the explicit period. It helps, and it helps much less than you would hope, because cash flows twenty years out are discounted so heavily that they barely register.
| Explicit forecast | Explicit period | Terminal value | Share in terminal |
|---|---|---|---|
| 5 years | $486M | $1,506M | 75.6% |
| 10 years | $951M | $1,438M | 60.2% |
| 15 years | $1,394M | $1,373M | 49.6% |
| 20 years | $1,818M | $1,311M | 41.9% |
Doubling the forecast from 10 years to 20 moves the terminal share from 60.2% to 41.9%. You have doubled the work, added ten more years of assumptions that are themselves guesses, and the assumption at the end still carries more than 42% of the answer.
At a five year horizon, which is what a lot of published models use, terminal value is 75.6% of the valuation. The forecast is decoration on the assumption.
What those two assumptions are worth
Terminal value in the standard form needs a perpetual growth rate and a cost of capital. Neither is observable. Both are chosen. Here is the entire valuation across a range where every cell is defensible and no cell is checkable.
| Perpetual growth | 8% cost of capital | 9% cost of capital | 10% cost of capital |
|---|---|---|---|
| 1.5% | $2,562M | $2,185M | $1,899M |
| 2.0% | $2,700M | $2,280M | $1,967M |
| 2.5% | $2,864M | $2,389M | $2,043M |
| 3.0% | $3,060M | $2,516M | $2,130M |
| 3.5% | $3,300M | $2,667M | $2,231M |
From $1,899M to $3,300M. A spread of 74% on the same company, the same cash flows, and the same ten years of forecasting, driven entirely by two numbers reasonable people disagree about.
Half a point on the perpetual growth rate moves the valuation 5.3%. Half a point off the cost of capital moves it 9.1%. Both are rounding choices.
The honest complication
We would be overstating the case if we stopped there, so here is the part that cuts against the argument.
Double the explicit growth rate, from 8% to 16%, and the valuation moves 83%. That is far more than either terminal lever. So the explicit forecast is not irrelevant.
But look at why. The explicit period matters mostly because it sets the base that terminal value is computed from. Grow faster for ten years and you arrive at year ten with more cash flow, and the terminal formula multiplies that larger number. The forecast is not doing its work by predicting years three through seven. It is doing it by determining where you end up.
Which is a different claim from the one implied by a model with quarterly revenue built out to 2036. The detail is not where the value is. The destination is.
What we do about it
Run it backwards. This is the real reason we publish reverse DCFs rather than forward ones. If the answer is mostly assumption, then solving for the assumption the market has already made is more honest than inventing your own and presenting the output as a valuation. It converts an unanswerable question into a checkable one.
Show the range, not the point. A single fair value implies a precision the method does not have. When our reports give one number, the scenarios and the assumptions behind it are on the page so a reader can move them. A fair value you cannot reverse-engineer is an assertion.
Sanity check the terminal value as a multiple. Divide it by the final year cash flow. If that implies the business trades at forty times in perpetuity, the perpetual growth rate is doing something you did not intend.
Prefer methods whose assumptions are visible. For a business whose economics turn on one ratio and whose reinvestment needs are modest, a normalized earnings multiple is often more honest than a ten-year projection, because it makes the judgment explicit instead of burying it in a spreadsheet. That is exactly the reasoning we used on UnitedHealth.
None of this makes the DCF useless. It makes it a framework for organizing assumptions rather than a machine for producing answers. The number at the bottom is not what the company is worth. It is what you believe, arithmetic.