Decoding RaymondsIndex: Why We Validated on Separation, Not Coverage

On 7 July the Korea Exchange published its monthly update on corporate value-up disclosure. The number that travelled was 85.5% — the share of total market capitalisation represented by companies that have filed a value-up plan since the programme began in May 2024. Underneath it: 741 companies cumulatively, 347 on KOSPI and 394 on KOSDAQ, with 12 new filers in June.

Two other figures in the same release did not travel. First, the coverage number splits sharply by market — 89.4% of KOSPI capitalisation, 32.5% of KOSDAQ. Second, the number of companies that have submitted a periodic filing including an assessment of how the earlier plan actually went: 116.

Tokyo's version of this looks much the same from a distance. The Tokyo Stock Exchange asked Prime and Standard Market companies in March 2023 to disclose actions reflecting cost of capital and share price. Prime Market disclosure passed 90% by March 2025, Standard sat near 50% as of April 2026, and more than 60% of respondents had updated their disclosures by July 2025. Over three years from the initiative's July 2022 starting point, average PBR moved from 1.1 to 1.4 and ROE from 8.4% to 9.0%.

Coverage is a different variable from behaviour

None of this is a criticism of either programme. It is an observation about what kind of number a disclosure rate is.

A coverage statistic answers one question: how much of the market has spoken. It is a denominator problem — you count filers over listed companies, or filer market cap over total market cap. It is easy to compute, easy to compare across years, and it moves monotonically upward by construction, because a plan once filed never un-files.

A behaviour statistic answers a different question: among firms that said something, how many did the thing. That requires a second observation at a later date, matched to the first. The 741-to-116 gap is exactly the distance between those two designs. It does not tell us that 625 companies failed to act — it tells us that for 625 companies, no matched second observation exists in this dataset. Which is precisely the condition under which a metric cannot support the inference people want to draw from it.

The KOSPI-KOSDAQ split makes the same point from another direction. A 56.9-point spread inside a single headline figure means the headline is carried by the largest names. If the thing you are trying to detect concentrates in smaller, thinner, less-covered companies — which relational risk generally does — then a market-cap-weighted coverage number is structurally blind to your target.

What we required instead

This is the design constraint behind how RaymondsIndex was validated. We did not ask how many companies the deterioration signal could be computed for. We asked how far apart the signal placed two groups that later turned out to be different.

Across 3,109 Korean listed companies, the deterioration signal cleared an effect size of d = 0.8. Cohen (1988) set that threshold as a "large" effect, and its practical reading is the useful part: at d = 0.8, the average member of one group sits above roughly 79% of the other group. The distributions still touch, but they no longer sit on top of each other. Separately, ordering agreement with realised outcomes — the concordance statistic formalised for prognostic models by Harrell, Lee and Mark (1996) — came in at 85.9%.

Those two numbers do different work, and reporting only one is how measurement systems drift. Concordance tells you the ranking is right. Effect size tells you the ranking is far enough apart to act on. A model can pass the first and fail the second, and a model that fails the second is an interesting research result rather than a usable warning.

For an individual investor, the practical translation is short. When any market statistic is offered as evidence that something improved, ask what the denominator is and whether a second observation exists. Coverage without follow-up is participation data wearing the clothes of outcome data.

General structural observation based on publicly released exchange data. Not investment advice.

#RaymondsRisk #RelationalRisk #CorporateGovernance #MeasurementDesign #ValueUp #EffectSize