How OakBag computes fair value

Every stock on OakBag carries a single traffic-light colour — green, orange or red. That colour is shorthand for a fuller judgement, and the number doing most of the work behind it is the estimated fair value: what the business looks worth on its fundamentals, set against what the market is charging for it today. This guide opens up how that fair value is computed, method by method, and the guardrails that stop a fragile estimate from painting a stock green.

One thing to hold onto from the start: everything below is a model estimate produced from third-party data, not a price target and not advice. The point of the machinery is to be honestly conservative — when the evidence is thin or contradictory, the model would rather stay quiet than flag a false bargain.

An ensemble, not a single formula

There is no one correct way to value a company, so OakBag does not lean on a single formula. It runs several independent valuation methods, each answering the same question — what is a share worth? — from a different angle, and then combines them. The methods that feed the number are:

• Peer P/E — the company’s earnings multiplied by a price-to-earnings multiple drawn from the median of its profitable sector peers, so a cheap-looking multiple is judged against the right cohort.
• Peer EV/EBITDA — operating cash earnings valued on the median enterprise-value multiple of the sector, then bridged from enterprise value back to equity per share.
• Machine-learning valuation — a LightGBM model predicts a justified EV/EBITDA multiple from a company’s own fundamentals, which is then applied the same way as the peer multiple.
• Cash-flow DCF — a Gordon-growth discounted-cash-flow leg on free cash flow per share, with the growth rate capped well below the discount rate so the arithmetic can never run away.

A fifth figure, the Graham number, is shown on each stock page for reference but is deliberately kept out of the combined estimate. Balance-sheet businesses are handled separately, which is worth a section of its own below.

How quality tilts the multiple

A great business deserves a richer multiple than a mediocre one, so the peer multiples are nudged by a quality adjustment: a company whose profitability, balance sheet and earnings quality rank above its peers earns a modest premium, and a weaker one takes a discount. The tilt is gentle by design and bounded, so it shades the estimate rather than dominating it.

This creates a subtle trap the model guards against carefully. Quality already helps decide a stock’s overall rank, so if it also inflated the fair value, the same virtue would be counted twice — once in the rank and again as a discount. To prevent that double-count, the green decision is re-tested with the quality tilt switched off: a company that looks cheap only because its quality premium lifted the fair value does not clear the bar. The premium stays in the displayed number, but it is never allowed to be the sole reason a stock turns green.

Clipping the outliers, taking the median

Individual methods misfire. A single distorted input — a one-off earnings dip, a data gap, an unusual capital structure — can send one leg to an absurd value. Two habits keep that from poisoning the result.

First, every method’s raw output is clipped to a sensible band around the current price, from 0.45× at the floor to 2.2× at the ceiling. A leg that lands outside the band is capped to the nearest edge and flagged on the stock page with a small ▲ or ▼ marker, so a reader can see it was reined in rather than measured cleanly. Second, the fair value is the median of the surviving legs, not their average — the median shrugs off a lone extreme reading in a way an average never can. The upshot is that the fair value shown at the top of a stock page is exactly the middle of the clipped method legs listed just below it, which is why the two always reconcile. Reading that table leg by leg is its own short guide, covered in the per-method breakdown so you can see agreement and disagreement between methods at a glance.

From fair value to margin of safety

Once there is a fair value, the margin of safety follows directly: it is the gap between that estimate and the price, expressed as a fraction of fair value — (fair value − price) ÷ fair value. A positive number means the model sees the shares trading below what the business looks worth; a negative one means the opposite. This single figure is what the traffic light keys off, and it is worth understanding on its own terms, which is the subject of the companion guide on reading a margin of safety. A positive margin alone is never enough to earn a green light, for reasons the verification gates make clear.

The verification gates: why not green?

A screener that only ever asked “is it cheap?” would hand out green lights to value traps. OakBag adds several checks that can withhold a green even when the margin of safety looks generous, because it would rather show a cautious orange than a false green. When one of these fires, the stock page spells out the reason under a “Why not green?” heading. The main gates are:

• Ceiling-pin — if every valuation method came in far above the price and was capped to the same 2.2× ceiling, the apparent discount is a clipping artefact, not a measurement. Unless at least one method independently clears the bar below the ceiling, the discount is not trusted.
• Thin ensemble — a fair value resting on fewer than two surviving methods is too fragile to flag as a buy on its own.
• Quality double-count — the quality-neutral re-test described above; if the discount evaporates once the quality tilt is removed, no green.
• Operating loss — a company losing money at the operating line is never flagged as a buy, however cheap the multiples look.

These gates are the difference between a number that merely looks low and one the model is willing to stand behind.

Red flags and knock-outs

Cheapness is only half the story; the other half is not stepping on a landmine. Before valuation even enters the picture, each company runs a set of knock-out safety checks, and tripping any one of them forces a red light regardless of how attractive the price looks. Those checks screen for bankruptcy risk via the Altman Z-score, negative shareholder equity, excessive leverage, poor liquidity, penny-stock or micro-cap fragility, and stale fundamentals. A separate earnings-manipulation screen, the Beneish M-score, flags accounts that show the statistical fingerprints of manipulation.

Alongside the hard knock-outs, the Piotroski F-score — a nine-point checklist of financial health — feeds the quality assessment, so a company with weak fundamentals is marked down rather than rewarded. You can see every one of these checks, passed or failed, on any stock’s page: browse them from the full stock list or jump straight to the stocks rated green today. Two contrasting names make the point well — a cyclical industrial such as A. O. Smith is valued on the full earnings-and-cash-flow ensemble, while a highly levered insurer like Prudential Financial is valued on book value instead, for the reasons in the next section.

Why banks and insurers are valued differently

Cash-flow DCF and EV/EBITDA assume a business turns operations into free cash the way an industrial or a retailer does. Banks, lenders and insurers do not: their “EBITDA” and “free cash flow” are accounting artefacts of deposit and premium flows, not spendable earnings, and valuing them on those legs produces nonsense. So OakBag routes genuine balance-sheet financials — identified by industry code and a real leverage test, to exclude shells miscoded into the sector — onto a peer price-to-book method instead, with the peer median taken among structurally similar institutions. For those names, a green light additionally requires the book-value leg itself to show the discount, because a cheap-looking earnings multiple on volatile insurance or cyclical-bank profits is not a trustworthy basis for a buy on its own.

What the number is — and isn’t

Put together, the fair value is a deliberately conservative, multi-method estimate with the outliers trimmed, quality counted once, and several ways to say “not confident enough.” It refreshes with the data on a regular cycle, so a colour can move as prices and fundamentals change. What it is not is a forecast, a price target or a recommendation — no model can promise a future return, and OakBag does not try to. It is a research tool to help you see quickly what looks cheap, what looks dear, and where the risks sit, so your own judgement has a running start. If you would like the whole toolkit — every method leg, price history and portfolio alerts — the pricing and plans page lays out the free plan and optional Pro.