Discounted Cash Flow: Run It Backwards
A discounted cash flow model estimates what a business is worth by forecasting its future cash and reducing each year's figure to present value. Because it needs three estimated inputs and the terminal value usually dominates the result, small changes in assumptions move the answer a very long way.
How it works
A discounted cash flow model values future money in today’s terms. Forecast the cash a business will produce, reduce each future year by a rate, and add the results.
The discounting is not arbitrary. Money available now can be invested; money promised later carries the risk of not arriving. The rate is the price of both facts combined.
The input is free cash flow, not profit. Cash from operations minus the capital spending needed to keep the business running — which is what an owner could actually take out.
Three estimates, one answer
Three inputs, none of them observable. The cash flows are forecasts, the discount rate is a judgement about risk, and the terminal value stands in for everything beyond the forecast period.
The terminal value usually dominates. In a typical five-year model, the value attributed to everything after year five is the majority of the total — which means most of the answer comes from the input you know least about.
A single percentage point on the rate changes the answer materially. At 8% the illustrative model gives 2,020; at 9% it gives 1,690; at 7% it gives 2,480 — a spread of nearly half the base figure, from an input nobody can pin down.
The growth assumption behaves the same way. Two per cent gives 2,020 and three per cent gives 2,420, which is a twenty per cent swing from a number chosen by feel.
Which is the real criticism of the method. With three flexible inputs, any conclusion between 1,400 and 3,100 can be reached with entirely defensible assumptions. A model built after forming a view will produce that view, and the arithmetic gives it an authority it has not earned.
In practice: run it backwards
The reverse version is the one worth doing. Take the market price as given and solve for the growth rate that justifies it, then compare that figure with what the company has actually achieved.
That question is answerable and the forward one is not. A price implying 9% growth from a business that has delivered 4% for a decade is a specific, checkable claim. It also cannot be tuned toward a conclusion, because the price is a fact rather than an input.
The output is the whole business, not the shares. Subtract debt and add cash to get equity value, then divide by the share count — a step that is easy to skip and changes the answer substantially in a leveraged company.
It works where cash flows are predictable. A regulated utility is a reasonable candidate; a company whose product may not exist in five years is not, and the model will produce a confident number for both.
The output is a range. Running the model at several combinations of rate and growth and reporting the spread is more honest than a single figure, and it makes the sensitivity visible rather than hidden.
One structural choice makes the model far more honest and almost nobody makes it: shorten the forecast period. A ten-year forecast is not more informative than a five-year one — it is the same guess extended, and it moves more of the answer into a period nobody has any grounds for.
The same applies to precision in the inputs. A cash flow forecast quoted to the nearest pound implies a confidence the method does not have, and rounding it to two significant figures changes nothing about the answer while changing a great deal about how it reads. Match the precision of the output to the precision of the worst input, which in every discounted cash flow model is the terminal value.
What a discounted cash flow model is not
It is not precise. Three estimates cannot produce a precise answer.
It is not objective. Every input is a judgement.
It is not a price target. It is a range, and a wide one.
And it is not applicable everywhere. It needs forecastable cash.
When it fails
It fails hardest where it is used most enthusiastically. A fast-growing company with negative cash flow has no forecastable input, so the entire value sits in a terminal figure derived from assumptions about a business that does not yet exist.
The second failure is a discount rate borrowed from a textbook. The rate should reflect the risk of these specific cash flows, and a single number applied across an entire portfolio does not.
A third is forgetting the debt. Enterprise value is not equity value, and the gap is the whole capital structure.
A fourth is a terminal growth rate above the economy’s. A business growing faster than the world forever eventually becomes the world.
And a fifth is building the model after deciding. The inputs will accommodate the conclusion, and the spreadsheet will make it look like analysis.
The original data
Of the 31,760 trading and investing videos in this site’s corpus, 1 has “discounted cash flow” in the
title at 143,078 views and 2 have “DCF” at a median of 77,502. “Valuation” returns 9 at a median of 17,868
across 7 channels, and “intrinsic value” returns 1 at 1,019,621 views. The counts are in
research/corpus-coverage.json, produced by site/measure_corpus.py.
The figures in the diagrams on this page are illustrative and chosen to make the sensitivity legible. The 2,020 base case, the 1,690 at a nine per cent rate and the 2,480 at seven are the same model run three times, which is the only part of the method worth memorising.
The habit worth taking is the sensitivity table. Build the model once, then vary the discount rate and the terminal growth rate by a point in each direction and record the four extra answers. If the range spans the current price, the model has not told you anything — and knowing that is considerably more useful than a single number that appears to.
Related
Intrinsic value is what the model is estimating. Valuation covers the alternatives and when each applies. And cash flow statement is where the input figures come from.
I stopped building these to produce a number and started building them to produce a question. The useful version asks what the market is already assuming, because that assumption is testable against what the company has actually delivered. The forward version just tells me what I already thought.
— Michael Whitman
This page is educational, not financial advice. Test every idea on your own charts before risking money.