What a DCF actually assumes, and where it breaks
A discounted cash flow model is a machine for converting assumptions into a number that looks like a fact. Knowing which assumptions the number is actually made of is the entire skill.
Build a five-year DCF on almost any quality business and something uncomfortable emerges: seventy to eighty per cent of the value sits in the terminal period. The five years of forecasting you agonised over are the small part. The number is mostly a claim about what the business is worth for ever, expressed through two inputs you picked in about a minute.
The terminal value problem
A perpetuity growth terminal value at a nine per cent discount rate and two and a half per cent terminal growth implies roughly a fifteen times multiple on the final year's free cash flow. The market routinely pays far more than that for anything with durable economics. So a single-scenario DCF built on conventional assumptions will tell you that most good businesses are overvalued.
That is not a finding about companies. It is a finding about the terminal assumption, and it applies roughly equally to everything you value the same way. Which points at the useful correction: a bias that hits every name equally cancels in a cross-sectional ranking. The absolute output is close to meaningless. The relative ordering, computed the same way across a universe, carries real information.
The discount rate is doing more than you think
Move the discount rate by one percentage point and a typical valuation moves fifteen to twenty-five per cent. Very little else in the model has that leverage. Yet the discount rate is often the input people spend least time on, because it feels like a convention rather than a judgement.
It is a judgement. A capital asset pricing model cost of equity is a risk-free rate plus a beta multiplied by an equity risk premium, and every one of those three is a choice. Beta measured on daily returns is biased toward zero by bid-ask bounce and non-synchronous trading, which is why published betas are weekly or monthly. Measured betas also mean-revert toward one, which is why a raw regression slope is a biased forecast over a multi-year horizon and why practitioners adjust it.
The inputs that produce a confident wrong answer
The failure mode that matters is not a model that refuses to run. It is a model that runs, returns a plausible-looking number, and is built on an input that was never checked.
| Degenerate input | What the model returns | Why it goes unnoticed |
|---|---|---|
| Share count read as zero | An implied value per share of zero or infinity | A missing tag coalesced to zero looks like a real figure |
| Debt tag not reported | Net cash where there is net debt | The company files under a caption the screen does not read |
| Discount rate below terminal growth | A negative or infinite terminal value | A comparison against a non-finite value is silently false |
| Negative enterprise value | A healthy-looking terminal value share | Numerator and denominator are both negative, so the ratio looks fine |
The general rule: a missing balance sheet line is not a zero. Treating it as one asserts a fact about the company. The honest options are to widen the tags you read, to look back to the last period actually reported, or to abstain and say so. In that order.
How to use one properly
- 01Run it as a range, not a point. A single number invites false precision and gets quoted without its assumptions.
- 02Report what share of value sits in the terminal period. Above about seventy-five per cent, the model is a terminal value calculator wearing a forecast.
- 03Sensitise the discount rate and terminal growth first. They dominate everything else.
- 04Check the inputs the model did not compute: share count, net debt, and whether either was actually reported.
- 05Use the ranking, not the level, when comparing across names.
Run this on a company you know
Every idea above is a thing the terminal does for you. Bring a ticker you have an opinion about and see whether it agrees.
Free plan. No card required.