How we value stocks
Every stock on ACCE is run through the same set of valuation models, mechanically, with no analyst discretion. This page explains what each model asks, how it is calculated, and — the part that matters more — where each one stops working and why we switch it off there.
The short version. A fair value is a confidence-weighted blend of the models that are valid for that particular business, after outliers are trimmed. Models that cannot meaningfully value a company are suppressed rather than fudged, so a bank is not valued on free cash flow and a software company is not valued on book equity. Where the models disagree sharply, the page shows the range instead of implying a precision that is not there.
The models
Six produce a fair value. A seventh, the reverse DCF, produces no valuation at all — it reads the market’s own assumption back out of the price.
Two-stage DCF
What is the cash this business will produce, in today’s money?
Five years of projected free cash flow discounted at the cost of equity, then a Gordon terminal value at 2.5%. Growth glides from the company’s own filed revenue CAGR down to the terminal rate rather than dropping off a cliff in year six. Critically, growth is charged for the capital it consumes: the identity g = ROIC × reinvestment means a business growing 5% on a 40% return retains 12.5% of profit, while one growing 5% on a 6% return retains 83%. Same growth, very different cash left over. Growth is also capped at the return on capital, because a company cannot sustainably grow faster than it earns without raising outside money.
Where it stops working. It cannot value a business the market prices far above what a five-year window plus a 2.5% perpetuity can produce. At a 10.5% cost of equity the model pays about 12.5× free cash flow at 2% growth and 21× at 15%. A company trading at 45× is not necessarily mispriced — the market may be underwriting durability the frame cannot express. It is also switched off entirely for banks and property, below.
Reverse DCF
What is the market already assuming?
Runs the same machinery backwards: instead of producing a fair value, it solves for the growth rate that would justify the current price, then compares it with what the company has actually delivered. This is the only model that never votes in the consensus — it is an anchor, not an opinion.
Where it stops working. The Gordon terminal breaks as growth approaches the discount rate, so a five-year solve cannot represent prices that imply high growth for longer. Where that happens we report the implied PERPETUAL growth instead, which is always solvable, and label it as such rather than quietly reporting the solver’s ceiling as if it were the answer.
P/E comparables
What do similar companies trade at?
The market’s own pricing of comparable businesses, applied to this company’s forward earnings. It needs no forecast of its own, which makes it a useful counterweight to the DCF.
Where it stops working. It assumes the company deserves the sector median. For a business whose growth or margins diverge sharply from its peers, that assumption is doing all the work. It also inherits whatever the sector is currently mispricing.
EV/EBITDA comparables
What would an acquirer pay for the whole business?
Capital-structure neutral, so it compares a debt-heavy company with a debt-free one on the same basis, then bridges back to equity by subtracting net debt.
Where it stops working. EBITDA ignores capital intensity, so it flatters businesses that must constantly reinvest. Switched off for lenders, where enterprise value is not a meaningful construct.
Graham number
What would a strict value investor pay?
Benjamin Graham’s rule of thumb — no more than 15× earnings and 1.5× book, multiplied together. Deliberately old-fashioned and deliberately conservative.
Where it stops working. It was designed for asset-heavy industrials in an era before intangibles dominated corporate balance sheets. On modern software and pharmaceutical companies it reads almost everything as expensive, which is why it is often excluded from the blend as a structural lowball rather than treated as a fair value.
Residual income
Does this business earn more than its capital costs?
Book value plus the present value of returns earned ABOVE the cost of equity. A company earning exactly its cost of equity is worth its book value and nothing more, however fast it grows. This is the standard frame for banks and insurers, whose assets are largely financial instruments carried near fair value.
Where it stops working. It runs on lenders only. For an asset-light business, buybacks and unrecognised intangibles leave book equity far below economic capital — Apple carries roughly $5 of book value per share — so the model would read almost anything as expensive. Property fails the other way: a REIT’s book value is depreciated historical cost, well under market.
Shareholder-yield discount model
What is the cash actually returned to owners worth?
A Gordon model on total shareholder yield — dividends PLUS net buybacks, measured from the change in diluted share count. For most large caps repurchases return more cash than the dividend, and an owner is indifferent to the form it takes. Apple returns about 0.35% as dividend and 2.6% as buybacks; counting only the first would call it a non-payer.
Where it stops working. A single-stage perpetuity is only meaningful for a mature business returning most of its profit. It is excluded from the consensus below a 2.5% total yield, where the cash return is not what drives the value.
How the models are combined
The headline fair value is a confidence-weighted average of the models that survive three filters. Anything implying more than ±200% against the current price is dropped as a probable data fault rather than a signal. Anything more than 3× above the median of the remaining models is trimmed as an outlier, and anything below 0.4× the median is trimmed as a structural lowball — a model that is simply the wrong frame for that company. At least two models must survive, or no fair value is published at all.
The filters are asymmetric on purpose. The failure we see in practice is a model reading far too low because its assumptions do not fit the business, not one reading absurdly high.
Currencies
A company can report its accounts in one currency and trade in another — HSBC files in US dollars and trades in Hong Kong dollars and in pence. Some venues also quote in a sub-unit: the London main market quotes ordinary shares in pence while the accounts are in pounds, a factor of 100.
Both are handled before any model runs. The price and the market capitalisation are converted into the currency the financial statements are denominated in, so every model works in one unit, and per-share results are converted back afterwards so the fair value you see is in the same currency and unit as the price beside it. Where the pairing cannot be established — the reporting currency is not yet on file, or no exchange rate exists for the pair — the affected models are suppressed rather than guessed at.
What this is not
These are not price targets or forecasts. A mechanical fair value is a statement about what a set of standard assumptions implies, and the gap between it and the market price is often a statement about how hard a business is to model rather than a view on the business itself. A company can trade well above every model’s output for years and be right to.
Nothing here is investment advice, and no model output should be read as a recommendation. See the disclaimer.
See it applied
The market valuation snapshot aggregates this engine across the whole covered universe — by sector, country and exchange — and every company page shows each model’s output, its assumptions, and whether it was included in the blend.