Shareholder Rerating

Shareholder Rerating

_{<a href="README.ko.md">한국어</a>}

In one paragraph

This manager looks for Korean companies that are giving money back to their shareholders — through dividends, share buybacks and the retirement of the shares they buy — and that can keep doing it out of the profit and capital they actually have. Its bet is not that the shares are cheap. It is that a company doing this repeatedly, in public, eventually gets priced closer to what it is worth, and that the gap between today's price and that value is what a patient holder is paid for. Positions are usually held six to twelve months, and every judgement it produces says three things apart: what closing the discount is worth, what the dividends are worth to you, and what the company's buyback is worth to the company. It proposes; you approve every position, and it never places an order.

The methodology

The mechanism has to be visible. Plenty of companies are cheap. This one only wants the ones where there is a reason the discount should close, and the reason it will accept is a shareholder return programme that is being carried out. So a high dividend yield on its own never selects a company; nor does a low price-to-book; nor does a big fall in the share price.

Announced and executed are different facts. A company can announce a buyback and buy almost nothing. This manager measures how much of a programme has actually been done against how much of the programme's own time window has passed, and a programme running at less than half its own schedule is treated as evidence for the announcement and nothing else.

Banks, insurers, brokers and factories are read differently — and only two of them are read here. For a bank or a bank-led financial holding company the question is capital: how much sits above the ratio the company itself has promised to hold, which is what a dividend or a buyback is paid out of. For an operating company the question is cash: what is left after the investment the business cannot skip. Asking a bank ratio of a shipbuilder produces a number about nothing, and the package refuses to do it rather than produce one.

An insurer, a securities firm and a mixed group are told apart from a bank, and then left alone. An insurer's solvency is measured by K-ICS and a broker's by the NCR — different ratios, answering different questions, out of which a different amount is distributable. This manager implements the bank and the operating-company arithmetic and no others, so it says which of the five kinds a company is, states the disclosure it read that off, and then reports the other three as not evaluated rather than pushing an unfamiliar ratio through a bank's formula. That is a wait with a reason, not a rejection of the company — and being financial is never by itself a reason to be read as a bank.

Two things are called a sector and they belong to different people. One is the fund's risk-management sector: your Mandate's consistent classification of the whole account, which is what a "no more than 25% in one sector" limit is measured over. That one is the host's, and this manager only reads it. The other is the kind of business a company is in, which decides which capital arithmetic is even meaningful; that one this manager decides from filings, for one company at a time. If your Mandate states a sector ceiling and this manager cannot work out the account's total in that sector — because the candidate is unclassified, or because a position or pending proposal somewhere in the account is — it will not increase a position. It waits and says which row it could not classify. Holding, analysing and reducing carry on as normal: what an unverifiable limit withholds is the thing the limit constrains.

Three yields are not one yield. The expected return has exactly two parts — what re-rating is worth, and the cash dividend after tax. A company's buyback is in neither: it pays you nothing directly, and what it is worth to you is already inside the first part. Adding it to a dividend yield counts the same money twice, and the arithmetic in this package refuses the addition rather than quietly accepting it.

What it deliberately does not do. It does not require a share to be "oversold" or to be above a moving average — a good company executing a return programme into a rising price is exactly what it is looking for. It does not buy more simply because the price fell; every stage of a position re-checks the reasoning, the remaining discount and the remaining risk budget. It does not run a basket or output a screen: one completed judgement per run, or a stated reason there is none.

<details> <summary>A short glossary</summary>

• Re-rating — the price moving closer to what the business is worth, without the business changing.
• Buyback / cancellation — the company buying its own shares; retiring them permanently is cancellation. Bought-and-kept shares can come back onto the market, retired ones cannot.
• Payout ratio — the dividend divided by earnings. Above 1 means paying out more than was earned.
• CET1 ratio — the core measure of how much loss-absorbing capital a bank holds. Every bank states a target it runs to, usually above what the regulator demands.
• Loss to invalidation — how much of a position is lost if the price reaches the level at which the reasoning is no longer true. It is what the position's size is calculated from.
• Mandate — the limits you set: how much of your account may go into one name, one sector, and how much may be put at risk.

</details>

How a run works

flowchart TD
  A[Invocation, asOf, mandate] --> B[Read the book: holdings, cash, open proposals]
  B --> C[Read own folder: staged ledger, unfinished research]
  C --> D[Review existing theses first]
  D --> E{Case?}
  E -->|re-rated / policy retreat| F[Stage down, or re-argue]
  E -->|intact| G1[Resume unfinished research and retry failed ranges]
  G1 --> G2[Sweep new return disclosures from the last receipt read]
  G2 --> G3{Did this run look at anything?}
  G3 -->|no universe, no lane, budget gone| G4[Report discovery_not_run — not 'no candidates']
  G3 -->|yes| G[Score on all three axes: discount, earnings quality, execution rate]
  G --> G5[Short of three axes: keep as watching with the missing axis named]
  G --> H[Complete one thesis: cause, evidence, difference, catalysts, scenarios, refutation]
  H --> I[Argue against it]
  I --> J[Two-leg total return: re-rating + net dividend, buyback reported apart]
  J --> K[Issuer kind: capital for a bank, cash for an operating company, unevaluated otherwise]
  K --> L{Enough evidence?}
  L -->|no data| M[WAIT — data_missing, arm a re-check]
  L -->|unfinished| N[WAIT — research_incomplete, name what is left]
  L -->|refuted| O[Record thesis_refuted with the evidence]
  L -->|yes| P[Size: risk budget / loss to invalidation, under every cap]
  P --> Q{Account headroom?}
  Q -->|full| R[WAIT — risk_limit_exceeded, thesis untouched]
  Q -->|room| S[One proposal: BUY, RESIZE or the next stage]
  F --> S
  S --> T[Arm the next review and the disclosure watches]
  T --> U[Write the roster and the receipt cursor back, so the next run resumes]
  G4 --> T
  G5 --> U

Where a run starts looking. Not from a list of names it already knows: from the disclosures companies filed since the last run finished — dividend resolutions, decisions to buy treasury stock, the reports saying how much was actually bought, decisions to retire it, and the value-up plans that claim all of it is policy. It records where it stopped, so a run that failed halfway is resumed rather than restarted, and a range it could not read is retried before anything new is collected.

Cadence. A price and risk review after the Korean close on weekdays, a deeper re-argument of each held thesis about monthly, and a watch on results, return-policy and cancellation disclosures for anything held or being researched. Every judgement — including a wait — arms what wakes the manager next, so a failed run has a defined way back.

What it needs

• Market. Korea-listed single equities (XKRX), priced in won.
• Data. Corporate filings through OpenDART — dividend and buyback resolutions, treasury acquisition and cancellation receipts, quarterly and annual statements. Prices, corporate actions and traded value through your own broker connection. A company's stated return policy usually lives in its investor-relations material rather than in a filing, so it is filed as a dated, cited reading rather than treated as a fact from a vendor.
• How it reads them. Filings come out of Aumos's own stored copy of OpenDART rather than being re-fetched every run — it reads that store, refreshes it when it is behind, and never keeps a second copy of its own. The issuer's IR statement is read on the web and filed through the one route that gives a web reading a receipt Aumos can show you; what it keeps in its own folder is a pointer and a question, never the documents.
• What a key costs. OpenDART is free and requires registration. The broker connection is one you already have; Aumos relays the vendor's answer and never exposes the credential to this package.
• Settings. Nine, all optional: how often a held thesis is fully re-argued, how many theses this instance carries at once, the smallest position worth opening — as an amount, in a currency you name, or as a share of the book — your dividend withholding rate, the share of the book risked on one idea, how many filings one run reads before it stops collecting and starts researching, and how many candidates it has to finish rather than leave half-read. Nothing in them can loosen your Mandate, and the risk budget may only be lowered.
• If your account is not in won. The smallest position worth opening is an amount of money, and the 500,000 KRW this package publishes is a fact about a Korean venue. On a book denominated in anything else that floor simply does not apply: the run says so and nothing else changes. Give it your own — an amount plus its currency, or a share of the book — if you want one.
• Where you approve. Everywhere. It proposes one decision per run; you approve it, Aumos sizes and routes the order. There is no capability in this package that can place one.
• What your Mandate must state. A single-name cap — your concentration limit. This package carries no default for it, so an account that states none gets a wait and an explanation rather than a size somebody guessed. It does carry its own risk budget, 1% of the book on one idea, because no Mandate has a field for that and a package that refused without one would never trade; you can lower it in the settings and not raise it. A sector ceiling is optional — state none and the sector axis simply does not apply. State one and it is checked; if the account's sector classification is incomplete, purchases wait until it is not.

What it is bad at

• A market that never closes discounts. The whole thesis is that a return programme eventually changes the price. In a market or a sector where governance discounts persist for years despite the payouts, this manager is right about the company and still loses money, slowly.
• Cyclical peaks. A cyclical company at the top of its cycle shows enormous cash, a small payout ratio and a low multiple. The package pushes hard against this — recurring earnings, mid-cycle coverage, ratio-versus-amount policies — and it is still the shape most likely to fool it.
• Fast markets. A six-to-twelve-month holding period with a monthly deep review learns about a sharp deterioration late. It is not a trading system and it will not behave like one.
• A run that could not read the disclosure window. If the filing route is down, or no universe of companies was declared, or reviewing what you already hold used up the run's budget, it reports that — in as many words — rather than "no candidates found". The two look identical on the screen and they are opposite facts, and this is the failure it spends the most machinery refusing. A "nothing qualified" from this manager means it swept a declared list, read every lane it needs, and nothing passed.
• Announcements, which it will not accept as results. A company that announces a buyback and buys nothing is a company this manager keeps on a watch list, not a position. That means it is slow by construction on a programme that is genuinely about to start, and it will say so rather than anticipate.
• Companies with little disclosure. Its evidence is filings. A company that discloses little gives it little, and the honest answer it produces is "I could not tell", repeatedly.
• Insurers, securities firms and mixed groups. It can tell them apart from a bank and it cannot size their return capacity. Install it expecting an explicit "not evaluated" on those, not a verdict.
• Who should not install it. Anyone wanting broad diversification from one manager — it holds a small number of researched positions; anyone who wants an answer every day; anyone who wants a quantitative screen. And anyone expecting the historical example below to be a promise.

Notes

• Provenance. This is a port of a personal research methodology from morethanmin/trading-harness, at the commit named in aumos.json, published with its copyright holder's request. NOTICE.md carries the licence notice. None of the source repository's order code, credentials, database or personal position ledger came with it.
• ⚠️ No track record, and one withdrawn claim. The methodology was used historically on a Korean financial holding company, and the investor reported a positive outcome. That result was never re-audited, it is not a validated edge, and it is not this package's performance. Nothing in this package quotes a return or a backtest, no threshold in it was chosen to make that case come out well, and Aumos measures a package's behaviour from actual runs rather than from anybody's recollection.
• Known limits. It reads one market. It judges one company at a time and has no view on how its positions correlate with each other beyond the sector limit. It cannot see private credit quality or a management intention that has not been written down anywhere.
• Installed beside other managers. Exposure is measured across the whole account — real holdings and unapproved proposals from every manager — and the binding limit is always the smallest that applies. Two managers' limits never add up into a bigger one. If several managers will run on one book, read ARCHITECTURE.md (ARCHITECTURE.md) first: it states which state this package owns, which the host owns, and what the host has to support for the arrangement to be safe.
• For maintainers. ARCHITECTURE.md (ARCHITECTURE.md) carries the state-ownership table, the deterministic modules and what each is for, the fixture inventory, and every fixed threshold with its formula and its rationale.