How our valuation model works: from company filings to a range of outcomes
In short
Our model turns a company's filings and a set of written-down assumptions into a range of possible values for the end of 2030. It simulates the business one quarter at a time, 50,000 times, and reports the middle outcome, the spread, and the chance of beating a 9%-a-year hurdle. It is ordinary statistics with no AI inside, and every result is checked before it is published.
Every Our View page ends with a simulation of what the company could be worth at the end of 2030. This post explains how it works: what goes in, what comes out, and how we check it.
What the model is for
The model answers one question: given what a company has reported and a set of clearly stated assumptions, what range of values could its stock reach by the end of 2030? We made five design choices along the way.
A range, not a target
Nobody knows what a company will earn in 2030: it depends on its products, its competitors, rates and markets, and on events that may or may not happen. A single price target hides all of that; a range shows both the middle outcome and how wide the uncertainty is.
One quarter at a time
Companies report quarterly, and many things happen in a particular quarter: a trial result, a product launch, a share sale. So the model moves forward quarter by quarter instead of jumping straight to 2030. That lets cash, debt and the share count evolve together, quarter by quarter, under our assumptions.
A fixed date, with the method labelled
Most companies are valued directly at the end of 2030, at a multiple of their sales or earnings over the following four quarters. For a company whose main product is only starting to sell, an end-2030 valuation may not capture the mature business we assume it becomes. For such companies we project the business until its sales mature, value it there, and discount that value back to the end of 2030. Each Our View says in one line which method was used.
A hurdle
A value means little without a yardstick. We compare each outcome with a hurdle: the end-2030 price that would earn 9% a year from today’s price. The share of simulations above it tells you how often our assumptions produce that return.
No AI inside
The model is ordinary statistics and accounting. It contains no AI model, does not read the news, and does not change its own assumptions. Its random draws start from a fixed seed, so the same inputs and the same version of the model reproduce every simulated path and every summary result exactly, which lets us check the calculation (more on that below). We use AI tools around the model, to help research, draft and check pages, but people set the inputs and approve what is published.
What goes in: the inputs
Each company in our model has between 55 and 213 inputs, with a median of 144; most have 130 to 150. Most inputs are ranges, not single numbers: for example “growth between 5% and 15% a year, most likely about 10%”. Each one is written down with its source.
The ranges are researched, not invented. Company-specific inputs come from the company’s filings and its own guidance. Many ranges come from the history of groups of comparable companies:
- revenue growth and margin changes from peers at a similar stage;
- the odds of trial results and approvals from industry success rates by phase of drug development;
- exit multiples from what comparable companies have traded at;
- how valuations react to company news, from a study of past moves.
Where our view of a company differs from analysts’, we blend the two, with weights set from how accurate each kind of forecast has been in the past. The model’s inputs fall into seven groups:
- Starting financials from filings: the latest reported revenue, costs, cash, debt and share count, taken from the company’s 10-K and 10-Q.
- Revenue paths and growth: how sales develop; for a drug company, patients treated and net price per patient; for a chip or software company, market growth and its share.
- Margins and costs: gross margin, research and other operating costs, and how they scale with sales.
- Cash and dilution: how profit turns into cash, stock-based pay, buybacks, convertible notes, and when a company would have to raise money.
- Company events: trial results, approvals, launches, patent expiry, competitors’ launches, lawsuits or policy changes, each with its odds, its timing and its effect. For some companies a takeover is also modelled as an event: a chance in each period and a price premium if it happens.
- Exit multiples: what investors might pay for the business at the end of 2030, based on the company’s own valuation history and on comparable companies. We compare with an average of recent quarters, not one day’s price: for a company valued on profit, when the average of the last four quarters is below its longer history, the multiple leans mainly on that average. These weights are our judgment. Published Our Views take this at their next update.
- Economy and discount rate: a shared simulated economy (recessions, interest rates, the stock market’s overall valuation and sector cycles) that all companies experience together, plus the discount rate used for companies valued at maturity.
Four or five inputs usually matter most:
- The growth path sets the size of the business in 2030.
- The exit multiple turns that business into a value; a small change moves the result a lot.
- The odds and timing of the big events, such as a trial result or an approval, decide which branch a simulation follows.
- The margin path decides how much of the revenue becomes profit and cash.
- The cost of funding: whether the company needs to sell shares, and at what price, decides how much of the value belongs to today’s shareholders.
What comes out: the results
Each run produces dozens of results; an Our View page shows the main ones:
- Value per share at the end of 2030: the middle outcome (P50) and the spread from P10 to P90. One simulation in ten ends below P10 and one in ten above P90.
- The chance of beating the hurdle: the share of simulations above the 9%-a-year hurdle.
- The chance of a loss: the share of simulations that end below today’s price.
- The implied end-2030 multiple: the sales (and, where earnings are positive, earnings) multiple that our end-2030 value implies, shown next to the company’s multiple today and, where we use it, its average over the last four quarters, so you can compare it with what investors pay now.
- Cash and dilution: how often a company runs short of cash and raises money, and how many shares it ends up with.
- What drives the range: which assumptions move the result most, shown as a short table of “what would move the result”.
We also run a few alternative versions, called sensitivities, with one assumption changed, and show how the result moves.
How one quarter is simulated
Inside one simulation, each quarter follows the same steps:
- Draw the events that are due: if a trial result, an approval or a competitor’s launch is scheduled for this quarter, the model draws whether it happens, using its odds.
- Update revenue, costs and profit: sales move along the growth path, adjusted for any event that just happened, and costs and margins follow.
- Update cash, financing and shares: profit becomes cash; if cash falls too low, the company raises money at the simulated share price, and stock-based pay adds shares.
- Record the quarter, including the value per share the simulated market would put on it at that point.
The model then moves to the next quarter and repeats, up to the end of 2030. At the end, the company is valued: directly at a multiple, or at maturity and discounted back. That is one simulation. The whole process is repeated 50,000 times, each time with different draws from the same ranges, using a fixed seed. The 50,000 end values form the range we publish.
Why 50,000 simulations and a fixed seed
With only a few hundred simulations, the middle outcome and the tails would shift noticeably from one run to the next, simply by chance. Using 50,000 simulations reduces that random noise.
The fixed seed is about reproducibility, not certainty. With the same inputs and the same version of the model, a re-run draws exactly the same random numbers, so every one of the 50,000 simulated paths repeats exactly, and so do the summary results: P10, P50 and P90, the chance of beating the hurdle and the chance of a loss. That makes the calculation checkable. If a number on a page changes, it is because an input or the model itself changed, not because the dice fell differently, and we can trace which one it was. The seed does not make the future any more certain.
With a different seed, the individual paths change, but in our tests the summary results moved only a little. In an October 2026 test on three companies, each run with 50,000 simulations and two other seeds, the middle outcome moved by less than 1%. The chance of beating the hurdle moved by at most half a percentage point, and the chance of a loss by about a fifth of a point. The bad case (P10) moved by up to about 4% for one company whose P10 is a very low price; the good case (P90) by less than 1%.
A simple, made-up example
Suppose a stock trades at $50. The hurdle for the end of 2030, a little over four years away, is about $72. If the middle outcome is $80, P10 is $30, P90 is $160 and 60% of runs end above $72, then on our assumptions a typical outcome clears the hurdle, but one run in ten ends at $30 or less and four in ten fall short of 9% a year. These numbers are invented and describe no company.
How the model was built and checked
The work runs in a fixed order, and each step is checked:
- Data: we start from the company’s filings with the US Securities and Exchange Commission and from market prices.
- Inputs: each input is set from those facts, from comparable companies’ history or from a stated judgment, and written down with its source.
- Simulation and valuation: the engine runs the 50,000 simulations and values each one.
- Checks:
- Automated tests: about 280 automated tests check the engine.
- Stored reference results: each company has a stored reference result. If a change to the engine moves any company’s numbers without a change to its inputs, the run stops until we find out why.
- Near-term sanity check: the model’s next quarters and fiscal year are compared with the company’s own guidance, its recent results and analysts’ consensus. Differences beyond a set tolerance are either explained in writing or fixed by revising the inputs.
- Independent fact-check: a second AI model checks every page against the filings, and its findings are fixed before publication.
- The Our View page: only then is the result written up and published.
We update the numbers when a company reports a new quarter or a major event happens, so a page always says which filings and which market close it reflects.
What the model cannot know
A simulation is only as good as its inputs, and many are still judgments: how a launch will go, how competitors respond, where rates will be. History is short for some industries, past relationships can change, and rare events are hard to measure. That is why we publish the assumptions and the sensitivities alongside every result, and why we treat the range as a description of our assumptions, not as a forecast.
How to read an Our View page
Each Our View starts with a short summary, the middle outcome, the range and the hurdle, explains the business, its risks and its finances, and ends with the simulation: the method, the main assumptions and what would move the result. Our guide to reading scenarios explains P10, P50, P90 and the hurdle in more detail.
An Our View is not a buy or sell recommendation. It shows what our inputs imply, so you can judge for yourself whether those inputs look right.
Quoting this? You may quote short excerpts under our reuse policy. Please credit WhyMyStockMoved.com and link to this page.