Understanding DCF Sensitivity: Why Small Assumptions Create Large Valuation Swings
Ask almost anyone in finance how you should value a company and they will eventually steer you toward a discounted cash flow model. The logic is hard to argue with. A business is worth the cash it will throw off over its life, discounted back to today. Estimate the cash flows, pick a discount rate, and you get a number.
The trouble is how much that number moves when you nudge the inputs. Shift the discount rate a little and the valuation jumps. Shift the terminal growth rate a little and it jumps again. Stack a few of those small changes together and the range of defensible answers gets so wide that the model can justify nearly any price you walked in wanting. I have watched two capable analysts build a DCF for the same company, both using assumptions they could defend, and land nowhere near each other.
None of that means you should throw out DCF. It means you should know exactly where the model is fragile, so you can lean on it where it is honest and ignore it where it is fantasy.
Where the sensitivity actually lives
A DCF has two pieces. There is the explicit forecast, usually five to ten years of projected cash flows you build out year by year. Then there is the terminal value, which stands in for everything after that. In most models the terminal value is the majority of the total valuation, often the large majority. So the assumptions that drive the terminal value carry far more weight than the ones you spent all your time on in the forecast.
Terminal value usually comes from the Gordon Growth Model: Final Year FCF x (1 + g) / (WACC - g), where g is the perpetual growth rate and WACC is your weighted average cost of capital. Look hard at the denominator. It is WACC minus g. When those two numbers sit close together, the terminal value blows up.
Say WACC is 10% and g is 3%, so the denominator is 7%. Bump g to 4% and the denominator drops to 6%, which lifts the terminal value by roughly a sixth. Push g to 5% and the denominator falls to 5%, and now the terminal value is up something like 40% from where you started. Nothing about the business changed. You moved one assumption by two points. The formula is unstable whenever the growth rate creeps toward the discount rate, which happens to be exactly the zone where most growth-company valuations live.
Why the discount rate hits so hard
The discount rate pulls in the opposite direction, and the effect compounds. A higher rate shrinks the present value of every future cash flow, and it punishes the far-out years much more than the near ones.
Take a hypothetical company generating $100 million in free cash flow, growing 8% for ten years and then 3% forever. Run it at a 9% WACC and you might land near $2.5 billion. At 10% it slides to roughly $2.1 billion. At 11% it is closer to $1.8 billion. That is a $700 million spread, most of the way to a third of the mid-point value, from moving the discount rate two points. Same cash flows. Same story about the business. Only the rate changed.
The reason this is so slippery is that the discount rate is itself a stack of estimates. It comes from the cost of equity, usually via CAPM, plus the cost of debt, weighted by the capital structure. The risk-free rate moves every day. The equity risk premium gets estimated differently by nearly everyone who touches it, and reasonable published figures span a few points on their own. Beta is backward-looking and jumps around depending on the window you measure. Two careful analysts can build the same rate from the same textbook and still end up a couple of points apart, and a couple of points is a large swing in the final number.
Growth assumptions in the forecast period
The growth you pencil in for the explicit years matters too, and it matters twice. A company you assume grows revenue 15% a year for five years finishes with a much bigger terminal-year cash flow than one you assume grows 10%. That fatter final-year number then gets fed straight into the terminal value calculation, so the gap you created in the forecast gets amplified in the part of the model that dominates the answer.
The catch is that multi-year growth is genuinely hard to forecast. Next year is a stretch. A five-year revenue path is an educated guess built from recent trends, management guidance, and your read on the competition. That is why two analysts can use defensible inputs and still land far apart. Neither has to be wrong. They are just making different, plausible bets about a future nobody can see.
Margins and capex, the quieter levers
Revenue does not turn into free cash flow directly. It runs through operating margins, capital expenditure, and working capital, and each of those adds its own sensitivity.
Margins bite hardest for businesses with high operating leverage. When most of the cost base is fixed, margins fan out fast as revenue climbs and collapse fast when it falls. Move your assumed long-term operating margin by a couple of points and you have moved every year of the forecast, plus the terminal value that trails behind it, so the valuation shifts by more than the margin change alone would suggest.
Capex is the one people gloss over, especially for capital-intensive companies. The line between maintenance capex, what it takes to keep the doors open, and growth capex, what you spend to expand, feeds straight into free cash flow. Overstate maintenance capex and you starve the cash flow and depress the valuation. Understate it and you inflate the valuation with cash the company is actually going to have to plow back in just to stand still.
Using sensitivity analysis the right way
You are not going to find the correct assumptions. Nobody does. The useful move is to map the range of plausible outcomes and decide from there.
Start with a two-variable sensitivity table. Put your two heaviest assumptions on the axes, which for most companies are the terminal growth rate and the discount rate. Run WACC across the axis from your low estimate to your high one, maybe a two to three point band, and run terminal growth from something like 1% to 4% on the other. Each cell shows the valuation per share those two inputs produce.
Now compare against the current price. If the stock sits below the model output across most of the grid, you have a reasonably robust case for undervaluation. If it sits above the output in most cells, the market is already paying for the optimistic corner of your table. If the grid is a mix of upside and downside, the name is roughly fairly valued and your thesis really rests on which assumptions you believe. The question quietly changes from "what is this worth?" to "under what assumptions does this stock make sense?", which is a far more honest thing to ask.
Scenarios, not just a grid
A sensitivity table flexes two numbers. Scenarios let you tell whole stories. Build three to five of them. A bull case where the company takes share, widens margins, and outgrows consensus. A base case near consensus with modest tweaks. A bear case with competitive pressure, margin compression, and softer growth.
Put rough probabilities on each one. If the probability-weighted average clears the current price with room to spare, the setup has positive expected value. The real benefit is that scenarios force you to build the version of the model you would rather not look at. Most people build the DCF that flatters the thesis they already hold. Making yourself write down the bear case, and then weighing it honestly against the rest, is most of the discipline right there.
A few habits that reduce errors
Use shorter explicit forecasts for businesses you cannot really predict. A five-year forecast for a mature utility is worth more than a ten-year forecast for a fast-moving tech company. Yes, a shorter window pushes more weight onto the terminal value, but at least you are not pretending to know what a volatile business looks like in year eight.
Cross-check the output against another method. If your DCF says $150 a share while comparable companies trade around 15x earnings and imply something closer to $75, you owe yourself an explanation. Either your assumptions run hot, or the comps are cheap, or they are not really comparable. Do not just let the gap sit there.
Watch the terminal growth rate closely. Over the long run no company can compound faster than the whole economy, or it eventually becomes the economy, so the perpetual rate should stay at or below long-term nominal GDP growth. Plenty of models sneak past this without meaning to by running aggressive final-year growth that quietly feeds an unrealistic terminal number.
And sanity-check the whole thing against reality. Build your model for a company that recently changed hands at a known price. If you cannot get anywhere near that price using reasonable inputs, your structure or your assumptions need work before you trust the model on a name where you do not have an answer key.
The honest version of the pitch
DCF is the most theoretically sound way to value a business and also the easiest to bend to whatever you want it to say. Both things are true at once. The model is only as good as the inputs, and the inputs are forecasts of an uncertain future.
Handled well, it gives you a structured way to think about value and a map of where you are exposed. Handled badly, it hands you false precision you can wrap around any thesis. The whole difference is whether you treat the output as a single answer or as one point inside a range. So build the sensitivity tables, run the scenarios, cross-check against other methods, and never quote a DCF value to the second decimal place. That kind of precision just tells the room you do not quite trust the tool you are holding.