The shape of the process is public, because you should know what you are buying before you buy it. The thresholds, the scoring anchors and the sizing arithmetic are for members, because that is the work.
Deep value, rebuilt from the balance sheet up.
Deep value is easy to describe and hard to do. Buy a business for less than it is worth, wait, get paid. The part that kills people is the waiting, because a stock can be cheap for a very good reason and stay cheap for a decade while the reason quietly eats the business.
So the method spends most of its effort on two questions that are not "is it cheap". First: are these accounts real, and will any of this value reach a minority shareholder. Second: what specific event, on what date, forces the gap to close.
One more thing sits alongside the fundamentals. Every name gets read through the IVT terminal, which measures whether large volume is actually accumulating the stock yet. That read never changes what a business is worth. It only changes how fast a position gets built. Fundamentals decide what to buy. The timing overlay decides how quickly to build it.
Overlooked companies anywhere in the world. US small and micro caps plus special situations are permanently in the pond, with one non-US region rotating alongside: Japan, Europe ex-UK, the UK, Korea, then Canada, Australia and the Nordics. Roughly half US, half international over a rolling year.
Outside the US, the mispricing usually comes from language and neglect. In the US it comes from structure: coverage deserts, spin-offs, post-bankruptcy equities, self-tenders, going-private filings, index deletions, tax-loss selling, holding-company stubs. A US name that three good writers have already covered is not overlooked, and it fails the mispricing pillar on the spot.
The most common error in value research is treating a screen result as a verdict. It is not. The funnel has four stages and each answers a different question. They are kept apart on purpose.
| Stage | Question | Output |
|---|---|---|
| 1. Screen eligibility | Is this name allowed into the funnel at all? | Pass or fail, per gate |
| 2. Forensic survivability | Are the accounts real, and will value reach a minority holder? | Clean, Noted, Raised or Fatal |
| 3. Underwriting | What is it worth, what is the range, how confident am I? | Score out of 25, targets, expected return, tier |
| 4. Entry and sizing | How much, how fast, and where do I admit I am wrong? | Gate, size, invalidation level |
Stage 1 carries no opinion. Passing eight gates is not a rating. "Passes seven of eight screen gates" sounds like seven eighths of a buy and it means nothing of the sort. A screen pass is an invitation to read a filing, and that is all it is.
Stage 4 never rescues Stage 3. A green timing gate on a name that scored badly is still a name that scored badly.
Run the same way every month so months are comparable. Thresholds live in one place in the code and are never loosened for a name that nearly made it. Nine gates, in this order.
| Gate | What it tests | Threshold |
|---|---|---|
| Investability | Can a normal reader actually buy and sell it | Members |
| Size | Small enough to be neglected, large enough to be real | Members |
| Neglect | How many analysts already cover it | Members |
| Profitability | Has it earned money across recent years, not just one | Members |
| Quality | Return on invested capital on a normalised year | Members |
| Safety | Debt against cash, and against equity | Members |
| Cheapness | Five tests, one of which must pass | Members |
| EV rebuild | Enterprise value recomputed, never taken from a data vendor | Members |
| Value-trap test | Margin now against margin at the same revenue in the past | Members |
Two of those gates exist because of specific errors, not theory. One name's vendor enterprise value was wrong by a quarter, because lease liabilities were being carried as debt and a pile of investments had been left out of cash. That single error moved the valuation multiple by more than two turns and produced a wrong rating. Another name failed every classic cheapness test while holding two thirds of its market cap in net cash against a profitable business, which is how the fifth test came to exist.
A screen is a commodity, so it is one of six sourcing channels. The others are forced and indifferent sellers, insider and controller behaviour, governance and activist catalysts, structural arbitrage such as holding-company discounts, and reverse engineering from the filings of investors worth reading. Names from those channels still go through everything below. The channel gets a name looked at. It does not get it published.
Deep value on small companies fails for two reasons more than any others. The accounting was not real, or the cash was never going to reach you. Two checks catch both, and they run before any valuation work, because valuing a company whose auditor has just resigned is a waste of an afternoon.
Binary. Any one of them ends the work, whatever the multiple says.
Each one gets logged with the date and the evidence. That log is what stops a rejected name being re-litigated next month when the screen surfaces it again three per cent cheaper. Some of the most useful things published here are the notes explaining why a cheap stock was passed on.
Every flag is marked Clean, Noted or Raised, and the register is published inside the issue with a note against each one.
| # | Flag | The question |
|---|---|---|
| 1 | Cash conversion | Does the profit turn into cash, over five years |
| 2 | Receivables drift | Is revenue being booked from customers who are not paying |
| 3 | Inventory drift | Is stock building faster than sales |
| 4 | Accounting policy changes | How often have the rules been rewritten |
| 5 | Margin outliers | Is the margin credible against peers, or too good |
| 6 | Auditor | Who signs it, for how long, with what changes |
| 7 | Is the cash real and free | Restricted, pledged, trapped, or unable to come home |
| 8 | Quasi-debt | Pensions, leases, guarantees, deposits, warranties |
| 9 | Related parties | Where does money flow to people close to the controller |
| 10 | Share count history | Ten years of dilution, options and converts |
| 11 | Capital allocation | What management actually did with cash for a decade |
| 12 | Control and alignment | Votes, classes, cross-holdings, insider behaviour |
Twelve rows of statuses is an audit trail, not something anyone absorbs. So every issue also carries four lines, each rated Strong, Medium or Weak, on earnings quality, cash and balance sheet quality, governance quality, and capital allocation quality. Those four lines tell you where the weak joint is.
What a Noted flag costs, what a Raised flag costs, and how many of each caps or kills a rating, is set out in the members section below.
Headline net cash is nearly always overstated for a deep-value purpose. Screeners take cash minus debt and stop. That figure includes money the company cannot spend, money that belongs to customers, money the pension scheme has a claim on, and money the business needs just to open the doors on Monday.
So every cash-based thesis carries the bridge, in full, in the issue.
Both numbers get published in every issue, headline and adjusted, each as a percentage of market cap. The adjusted figure is the one that drives the valuation, the score and the downside floor. It reads like this: the company screens with a large pile of net cash, and after the bridge, a materially smaller one.
That gap is the whole point. It is also the most common reason a cheap balance sheet turns out not to be cheap.
Not all cheapness is the same thing, and the differences decide what counts as downside protection. Every issue names one primary kind, and at most one secondary.
| Label | The stock is cheap because |
|---|---|
| Cash-backed | Net cash is a large share of the market cap |
| Asset-backed | Tangible book, net current assets or property support the price |
| Earnings-backed | The multiple on normalised earnings or free cash flow is low |
| Sum-of-the-parts | Segments or holdings are worth more apart than together |
| Event-backed | A defined event should release value |
| Cyclical mean reversion | Earnings sit below mid-cycle |
The label is not decoration. It fixes which downside floor is allowed to be used and which checkpoint has to be monitored afterwards, and it constrains how earnings may be normalised. Those rules are in the members section.
The bridge says how much cash is really there. It says nothing about whether a minority shareholder ever sees a yen of it. A hoard controlled by a family that has never returned capital deserves a permanent discount, and counting it at face value is how a thesis ends up right about the balance sheet and wrong about the outcome.
Eight factors get scored: insider ownership and alignment, buyback history and whether shares were actually cancelled, dividend history, acquisition discipline, governance and minority protections, whether the business genuinely needs the money, how the country treats minority holders, and where the cash physically sits.
The result is a High, Medium or Low rating, and it is not decorative. A weak rating cuts how much of the cash counts toward the downside floor and caps the cheapness score outright, which can take a Conviction Buy off the table on a name that is genuinely stuffed with cash. The haircuts and caps are in the members section.
One principle worth stating in public: the haircut applies to the floor, never to the bridge. The bridge is a statement of fact about a balance sheet. This is a judgement about people. Mixing them would corrupt a number that has to stay clean.
Bear, base and bull, each with a probability, a target, the arithmetic behind the target, and a time horizon. Probabilities are set before the targets are calculated, they sum to 100, and they are not reverse engineered until the expected value looks good.
Every target shows the multiple, the earnings figure, the share count, and where each came from. No target rests on a multiple above the name's own five-year median unless the reason is written out.
The horizon is what makes it honest. A 60% gain over five years is a 10% annual return and is not interesting for an illiquid foreign microcap. The same 60% over eighteen months is a different investment entirely. So the output is an expected annual return, and it has to clear a hurdle set by the rating tier. A good score that misses its hurdle gets downgraded, which is the mechanism that stops a high score smuggling in a mediocre return.
Earnings underneath those multiples are normalised, never trough and never peak. The normalisation method is ranked, stated in the issue with its rank, and tested against the company's own history before it may be used. Publishing a normalised figure well above anything the business has ever earned requires clearing a specific burden of proof.
Alongside the three worlds sits a comparison table of three to six real peers on the multiple, price to book, return on capital and operating margin. Cheaper than all of them, and the issue has to say specifically why the gap is undeserved. Not cheaper than them, and the cheapness is against the market rather than the industry. That gets said out loud.
Five pillars, zero to five each, twenty five in total. Each pillar is scored in isolation, in order, against written anchors, and before the verdict is written. Writing the bull case first biases every pillar upward, so it is written last.
| Pillar | The question it answers |
|---|---|
| 1. Cheapness | Is it cheap on adjusted cash and normalised earnings, rather than on the headline? |
| 2. Business quality | Is it merely cheap, or does the business have some durability? |
| 3. Catalyst and clock | What specific event, on what date, forces recognition? |
| 4. Balance sheet and governance | Can it survive, and will minority holders actually benefit? |
| 5. Mispricing conviction | What specifically is the market getting wrong, and why is that non-fundamental? |
The median pillar score across every name scored, including the ones that never get published, should be 3 and not 4. A scale where the usual answer is 4 carries no information. It is checked quarterly across the whole population, because checking only the published names is circular: they were selected for quality in the first place.
Two more rules keep the scale honest. Pillar 5 may not restate Pillar 1 or Pillar 3, because "it is very cheap" is not a reason the market is wrong. And "management says they will" is a weak catalyst, not a strong one, unless they have already done it once.
Verification is a gate, not a grade. Thesis-critical figures are checked against the primary filing and published in a log with the document, the page and the date. Nothing is ever docked a point for being unverified. It simply does not publish.
The written anchors for each pillar, the deduction schedule, the tier boundaries, the floors that stop one strong factor carrying a weak idea, and the return hurdles are all in the members section.
This is the part that belongs to the other half of the business. The IVT terminal reads volume structure, anchored VWAP and regime across timeframes, and it answers one narrow question: is institutional money accumulating this name yet, or is it still being distributed.
The timing overlay does not determine what a business is worth. It has no view on intrinsic value and it never overrides the score. All it decides is whether a rated name is bought now, bought in pieces, or watched.
| Gate | Meaning |
|---|---|
| Accumulate | Buy now, up to the computed size. |
| Scale-In | Start with a fraction. Add on the published trigger, not on a feeling. |
| Wait | No new buying, whatever the fundamental tier says. |
Two things are published with every gate. A single invalidation level, and a flip condition: the specific observable event that moves the name to the next gate. The flip condition is what turns a Wait rating from a dead end into a promise, and when it fires, a timing note goes out within 24 hours.
Where volume is genuinely thin, the issue says so. Volume-based signals are more reliable on a liquid US name than on a foreign microcap that trades a few tens of thousands of dollars a day, and pretending otherwise would be dishonest about the tool.
Position size is computed, not asserted. "I own some" is not research, so every rating produces a weight against a stated reference portfolio, adjusted for liquidity, downside, currency, outstanding flags and the timing gate, and constrained by portfolio limits. The arithmetic is in the members section.
Most write-ups are beautiful on day one and silent afterwards. That is the real failure mode of this format and it is worse than being wrong. Every issue carries four to six thesis checkpoints, each with a current value, a date, what would confirm the thesis and what would break it. Every date goes in the calendar. A checkpoint that is not diarised does not exist.
| Obligation | Deadline |
|---|---|
| Earnings note on a covered name | Within 72 hours of the result |
| Checkpoint update | On the date, even when nothing happened |
| Timing alert when a gate flips | Within 24 hours |
| Catalyst note | Within 72 hours |
| Exit note when a thesis breaks | Within 72 hours, published whether or not it lost money |
| Portfolio review | Within two weeks of each quarter end |
Declared in advance, and never changed after the fact. This is the part that decides whether any of the above is worth anything in three years.
Exact position and size band disclosed on every issue. No trading in a covered name within three trading days either side of publication. No paid coverage, ever. Corrections are published at the top of the issue and never edited in silently.
The record is also honest about one structural limit. Any performance statistic computed over the same history the parameters were chosen on is in-sample, and it is labelled as such. Forward results are kept separately.
The live scorecard is here: the Deep Value Portfolio.
Every issue opens with a conclusion box: eight questions, and the current action. Free readers see the questions, which tells you exactly what the work covered. Members see the answers, because the answers are the report.
| Question | Answer |
|---|---|
| Is it cheap on adjusted, normalised figures? | Members |
| Is the balance sheet safe enough? | Members |
| Do the accounts survive the forensic pass? | Members |
| Is there a catalyst with a clock? | Members |
| Is management aligned, and will value reach minorities? | Members |
| Is it liquid enough to enter and to leave? | Members |
| Does it clear the return hurdle for its tier? | Members |
| Do I hold it? | Members |
| Current action | Members |
Free: the thesis, the business, why it is overlooked, and the bear case in full. The bear case is free deliberately. Publishing the strongest argument against your own idea, for nothing, is the most credible thing an independent research letter can do.
Paid: every number and every answer. The conclusion box, the scorecard detail, the verification log, the net cash bridge, the three worlds and the comp set, the catalyst timeline, the flag register, the entry levels and the invalidation, the computed position size, the checkpoints, and the tear sheet.
The tear sheet is the page people keep. One page, every number needed to make and monitor the decision, and a trail back to where each one came from. That is the test the whole method has to pass before an issue ships.
Enter your Cantillon Research subscriber email to unlock the working half of the method: every screen threshold, the flag levels and what each one costs, the bridge policy and the cash realization haircuts, the normalisation hierarchy with its burden of proof, the written scoring anchors, the tiers, floors and return hurdles, and the position sizing arithmetic.
One flagship deep dive a month on an overlooked company, plus watchlist notes, timing alerts and the published record. Every issue built from this method, in this order.
Subscribe