Discounted cash flow (DCF) stocks: why I distrust it

2026-08-14 · By Lubin Danilo, founder of Lubin Investment

Discounted cash flow (DCF) projects a company's future cash flows over ten to twenty years, then discounts them back to today's value. The problem: a small change in the growth rate or the discount rate can swing the result by two times. My method only projects five years ahead, with a multiple anchored in the stock's own trading history.

What exactly is discounted cash flow (DCF)?

DCF starts from a simple idea: a company's value today is the sum of all the cash it will hand back to its owners in the future, brought back to today's value. In practice, an analyst projects free cash flow (the cash left over once bills are paid and reinvestment is funded) year by year, often ten to twenty years out. They then add a terminal value, an estimate of everything the company will be worth beyond that horizon, usually assuming it keeps growing at a modest rate forever.

Every future cash flow is then discounted: brought back to today's value using a discount rate, which reflects both the price of time (a dollar today is worth more than a dollar in ten years) and the risk taken. Add up all these discounted flows, plus the discounted terminal value, and you get a single number: the model's intrinsic value for the company.

Why did this become the standard method in finance?

The logic behind DCF is hard to argue with: a business is ultimately only worth the cash it can eventually return to its owner. Whether you are buying a rental building, a bakery, or a listed stock, the logic is the same: today's price should reflect the future rent, profits, or cash the asset will produce. This logic is taught in every finance program and used daily by investment banks to value an acquisition or an IPO.

Aswath Damodaran, a finance professor at NYU Stern and arguably the most cited academic authority on the subject, has built an entire career refining this method (his valuation courses and models are freely available). DCF has so many serious defenders for a reason: on paper, it is the most rigorous way to answer the question of what a company is really worth.

Where does it actually break down?

The problem is not the logic of DCF, it is what the logic requires you to guess. Projecting a company's cash over ten or twenty years means knowing its growth, margins, and reinvestment needs for years that have not happened yet. Nobody knew, fifteen years ago, that a virus would shut down stores for two months, or that a language model would upend entire corners of the software industry. A fifteen-year DCF built in 2020 for a movie theater chain or a software vendor would have missed the two events that mattered most for each of them.

Worse: in most DCF models, the terminal value, that estimate of everything that comes after the projection horizon, often makes up 60 to 80% of the final result. In other words, most of the price the model hands you today rests on an assumption about a period that only starts ten or twenty years from now. The model looks precise because it spits out a number with two decimal places, but that precision is an illusion built on quicksand.

Why can one tiny assumption swing the result by two times?

The best way to see it is to run the numbers. Take Fair Isaac (FICO), the company that owns the credit score used by nearly every US lender, whose business model I have already analyzed: as of writing, its free cash flow per share sits at roughly $30.90. Imagine I project that figure five years out under three different annual growth assumptions, only the growth rate changes, nothing else in the calculation moves.

Annual growth assumption	Projected FCF per share (5 years)	Fair buy price today
10%/year	≈ $49.76	≈ $990
20%/year (real recent growth, capped)	≈ $76.88	≈ $1,528.89
25%/year	≈ $94.30	≈ $1,875

A single assumption, the expected annual growth rate, moves the fair buy price from $990 to $1,875, almost double, without touching a single other input. And this example only projects five years out: a classic DCF, which projects ten or twenty, lets that same assumption compound over two to four times as many years. The gap between scenarios does not grow linearly with the horizon, it explodes exponentially.

What do I do instead?

Rather than trying to guess what happens in ten or twenty years, my method deliberately limits the exercise to five years, with only three ingredients I detail in full in my complete methodology. First, a free cash flow per share growth rate, based on what the company has actually delivered over the last two fiscal years, from which I systematically subtract two percentage points as a safety margin, capped between minus 20% and plus 20% a year: never an extrapolation of one great quarter projected forever.

Second, an exit multiple, meaning how many times that future cash I consider it reasonable to sell the stock for in five years. Finally, a required rate of return, fixed, used to bring that future exit price back to a fair buy price today. The next two sections detail these last two ingredients, exactly the two points where I diverge the most from a classic DCF.

Why only five years, not ten or twenty like a classic DCF?

Because a shorter horizon does not remove uncertainty, but it limits how many years a small growth error can compound over. As the Fair Isaac example above shows, the gap between my three scenarios already explodes over five years: over fifteen years, the same 10-to-25% annual growth divergence would produce results more than ten times apart, not less than double. Five years is long enough to avoid judging a company on a single quarter, but short enough to stay honest about what I actually know.

This choice has a cost I accept: I never say anything about what happens after those five years. A classic DCF claims to know, through its terminal value. I would rather use a model that openly admits its limits than one that gives the illusion of predicting everything.

How do I pick the exit multiple, instead of a theoretical perpetual growth rate?

A classic DCF calculates its terminal value by assuming the company keeps growing, forever, at a rate close to the global economy's, often 2 to 3% a year. The problem is that the math turning that perpetual growth rate into today's value is extremely sensitive: moving the perpetual growth assumption from 2% to 3% can, on its own, shift the final result by several tens of percent, for a difference in assumption that looks trivial on paper.

My method avoids that trap differently: instead of imagining a theoretical perpetual growth rate, I look at how many times its own free cash flow the market has actually been willing to pay for this exact stock, in median terms, over its own history, capped between 8 and 40 times depending on the case. For Fair Isaac, that cap kicks in fully: its historical median tops 40 times its cash, so I use 40 times, the highest ceiling I allow myself, rather than the higher raw figure its own past would suggest. It is the same principle I apply to other valuation multiples: I am just as wary of an EV/EBITDA multiple looked at alone, without comparing it to the stock's own history. The price the market has actually paid in the past is a verifiable fact; a theoretical perpetual growth rate is only a guess.

Where does my 15% required return come from?

In a classic DCF, the discount rate is supposed to be calculated precisely, usually through the weighted average cost of capital (WACC), a formula combining the cost of debt, the cost of equity, and a risk coefficient called beta. The problem is that each of these building blocks itself rests on choices: which market risk premium to use, over what period to measure beta, what future debt structure to assume. Two serious analysts, given the same starting data, can legitimately compute two different WACCs, and therefore two different valuations for the same company.

Rather than recomputing a different rate for each of the 5,000 stocks I track, I apply a single, fixed required return of 15% a year to all of them, in the spirit of the threshold Warren Buffett has long argued for when judging whether an opportunity deserves his capital (Berkshire Hathaway's shareholder letters, freely available, lay out this philosophy year after year). It is not scientifically more true than a custom-built WACC, but it is transparent, identical across every stock on the site, and impossible to quietly tweak to make the model say whatever I want it to say.

What does this actually look like on a real stock?

Back to Fair Isaac. Its recent free cash flow per share, once capped at my 20% annual ceiling, projects to roughly $76.88 in five years. Multiplied by my 40-times cash ceiling, that gives a theoretical exit value of roughly $3,075. Brought back to today using my 15% required annual return over five years, that gives a fair buy price of roughly $1,528.89, against a current price of roughly $1,110.70: my model shows a discount of roughly 37.7%.

One detail matters here: Fair Isaac's current price-to-free-cash-flow multiple, roughly 35.9 times its cash, looks high in absolute terms. But placed within its own five-year history, that level only ranks at the 39th percentile, meaning it is cheaper than the average price the market has paid for this exact stock most of the time. A multiple that looks expensive in absolute terms can therefore be, for this specific stock, fairly cheap relative to its own history. You can check these numbers live on Fair Isaac's analysis page, which runs on the same data quoted here, drawn from its latest regulatory filings.

Does my method have limits too?

Yes, and I would rather say so clearly than let it seem more precise than it is. Five years is still a bet on the future, not a certainty: nothing guarantees a company's recent growth holds up, even once capped for safety. A stock's own P/FCF history is not a perfect guarantee for the future either: a company can structurally shift how the market rates it, for better or worse, making its own past less relevant for judging its future multiple.

And my fixed 15% required return treats a century-old, highly predictable insurer the same way as a biotech whose key patent expires in two years, even though their real risks are nothing alike. No valuation model, mine included, removes the need to judge case by case. What I can guarantee, though, is that every assumption in my calculation is visible, explained, and applied the same way across the 5,000 stocks my site analyzes, rather than hidden in a twelve-tab spreadsheet only its author can really read.

FAQ

About the author

Written by Lubin Danilo, founder of Lubin Investment. A self-taught individual investor, I find fundamental analysis fascinating, and it has delivered excellent results. For three years now, my performance has beaten the S&P 500. But analyzing every stock took too much time: sites with incomplete data, calculation methods and criteria never aligned with mine. And spotting the best stocks was just as time-consuming, even with my own well-defined checklist. So I put my software development background to work to build this software, base my investment strategy on its results, and share it with people who share the same passion as me. It judges a company's quality and its price separately, using criteria drawn from the financial literature (Warren Buffett, Michael Mauboussin, Aswath Damodaran).