The Table Where Japan Keeps Its Hostages

In the early summer of 2019, ninety-three large Japanese companies filed annual securities reports under a rule that had just been rewritten. The rule asked them to state, holding by holding, the quantitative effect of owning shares in another listed company, connected to their own management strategy, business content and segment information. A survey of those filings — Russell/Nomura Large Cap constituents whose reports were out by 24 June 2019 — counted the companies that supplied a quantitative effect. The count was zero. All ninety-three treated the effect as difficult to state, and not one produced a figure. Where reasons were offered they included commercial confidentiality and information belonging to the counterparty; some companies offered no reason at all.

The survey's other finding is the one that ages better. Accounts of why a holding was reasonable were mostly abstract, and concrete descriptions of the method used to verify that reasonableness were, by the surveyor's assessment, relatively rare.

The alternative route is not a loophole discovered by clever counsel. It is in the ordinance. Form No. 2, note (58)d(f) of the Cabinet Office Ordinance on Disclosure of Corporate Affairs requires the quantitative effect of the holding, and then, in the same sentence, permits a statement that such disclosure is difficult together with the method used to verify reasonableness. A duty with a stated substitute is an effort-based duty. The 2019 result should be read with its own limits attached: ninety-three is a large-cap first-year sample, not a census of the market, the survey counted only what appeared in the shareholdings section of the report, and that source cannot tell you whether the practice changed in later years.

The table itself dates to a Cabinet Office ordinance amendment promulgated and effective on 31 March 2010. Individual names had to be disclosed where the balance-sheet carrying amount exceeded one per cent of stated capital, with a floor of thirty names. The tidy version of this — Japan has disclosed thirty policy holdings since 2010 — is wrong in both directions. Full application began with the fiscal year ended March 2011; the March 2010 year ran on a reduced transitional test with a floor of ten. Banks and insurers were phased in separately and landed somewhere else entirely: one per cent of stated capital, capped at fifty names, with a floor of thirty.

At the same time the regulator set out its view that stating the purpose as policy investment was not enough, and that what kind of policy purpose was at work should be described specifically. That view was about words. Nine more years passed before anyone was asked for a figure.

The figure arrived with the amendment promulgated on 31 January 2019, which raised the individually disclosed names from thirty to sixty. The policy-shareholding items were not staggered: the expanded name count, the holding policy, the method for verifying the rationality of holdings, the content of the board's verification of whether individual holdings should be retained, and the quantitative effect itself all applied from fiscal years ending on or after 31 March 2019, alongside the executive-remuneration items. What was deferred to fiscal years ending on or after 31 March 2020, with early adoption permitted, was a different block entirely — management policy, risk factors, management's discussion and analysis.

The regulator's own announcement of the shareholding change describes it as requiring disclosure of the method of verifying reasonableness. The words purpose and effect live one level down, in the notes to the form.

The mechanics of that form repay a slow reading. Specified investment shares and deemed holdings are disclosed individually where the carrying amount per name exceeds one hundredth of stated capital, and where the total of such names falls short of sixty, the sixty largest by carrying amount are disclosed instead. If eleven or more deemed holdings would qualify, the sixty splits into the ten largest deemed holdings and the fifty largest specified investment shares. A holding company reports through its largest holding subsidiary, and for the others sixty is read as ten. The revision consulted here is a snapshot that still refers to quarterly reports and predates the 2025 change described below, so the note numbering may have moved since.

The point survives the numbering: the population of the table is a function of the filer's stated capital, which is a number set by history rather than by exposure.

Alongside the quantitative-effect field, the same note requires the purpose of the holding, an outline of the business transactions or alliance where that is the stated purpose, the reason for any increase in shares held, and, at the end, whether the issuer of those shares holds shares in the filer. That last item is a yes-or-no box. Inside the policy-shareholding table it is the one field that records the reciprocal leg as a stated fact rather than as something the reader assembles from two documents; elsewhere in the annual report, the voting-rights section carries its own line for mutually held shares. It asks whether the other side is also holding. It does not ask what happens if you stop.

The Aggregates and Their Compilers

The aggregate picture is genuinely one of decline. Nomura's estimate of the cross-shareholding ratio — the market value of listed shares held by listed banks and listed corporates, over total market capitalisation — was 6.2 per cent for fiscal 2024, down 1.1 points. The broader measure including life and non-life insurers was 9.7 per cent, down 1.6 points and below ten per cent for the first time. Both were sixth consecutive record lows, and the decline was the first exceeding a full point since fiscal 2009. By holder: listed corporates 4.1 per cent, life insurers 2.8, listed banks 2.1, non-life 0.7.

These are private estimates built from large-shareholder data and filings, not official statistics, and they are measured at market value, which introduces a problem the compiler flags rather than hides. A different series in the same report — policy holdings measured against net assets rather than against total market capitalisation — moved the other way in the middle year: 11.1 per cent in fiscal 2022, 12.3 in fiscal 2023, then 9.5 in fiscal 2024 for all sectors. Non-financials went 7.5, 8.2, 6.4; financials 26.9, 30.8, 23.9. The share of companies above the twenty-per-cent-of-net-assets threshold that institutional voting policies tend to use moved 8.1, 9.4, 7.7. The middle year was TOPIX up thirty-eight per cent; the last was TOPIX down four.

The report ranks insurance highest at 50.3 per cent for fiscal 2024 and banking next at 22.4, and separately gives the highest non-financial reading as warehousing and transport-related at 23.3 — a number larger than the one ranked above it. The compiler also states that these figures do not correspond to the carrying amount of non-pure-investment shares in annual reports, and approximate but do not match the definitions institutional investors use.

A separate series from the same house puts the policy-holding ratio — the share of listed companies' equity holdings held for policy rather than pure investment purposes — at 29.4 per cent for fiscal 2024, down 1.4 points and below thirty for the first time since the series begins in 1990. Only the summary of that report was available here, so whether the denominator is measured by amount or by number of names is unconfirmed, and it is a third measure again, different from the 6.2 and 9.7 above rather than a refinement of them.

The exchanges publish the one figure in this area that is an official statistic, and it is not a policy-holding statistic. The four Japanese exchanges' fiscal 2024 shareholding distribution survey, as of 31 March 2025, records business corporations at 18.7 per cent of market value, down 0.6 points and the fifth consecutive record low, holding 177.3 trillion yen ($1.08 trillion), down 16.85 trillion yen ($103 billion) or 8.7 per cent. City and regional banks held 1.8 per cent, or 17.5 trillion yen ($107 billion), down 18.7 per cent. Life insurers held 2.7 per cent, or 26.0 trillion yen ($159 billion), down 13.5 per cent. Non-life insurers held 0.6 per cent, or 6.03 trillion yen ($37 billion), down 34.5 per cent.

Total surveyed market capitalisation was 948.0 trillion yen ($5.79 trillion), down six per cent. The exchanges themselves write that reduction of policy holdings by financial institutions appears to be behind the movement. The same document contains a trap: business corporations were net buyers of 10.2 trillion yen ($62 billion) for the twenty-first straight year, because treasury shares are booked to that category. Foreign investors reached a record 32.4 per cent.

Below the official statistic, the aggregates diverge by construction. A think-tank tally of TOPIX 500 constituents, based on reports filed from October to September, puts corporate holders at roughly 10,500 names and 29 trillion yen (about $177 billion) in 2023, falling to roughly 9,400 names and 24 trillion yen (about $147 billion) in 2024 — down eleven per cent by name count and eighteen per cent by value. Financial institutions went from roughly 8,200 names and 28 trillion yen (about $171 billion) to roughly 7,000 names and 21 trillion yen (about $128 billion). Combined, roughly 57 trillion yen (about $348 billion) to roughly 45 trillion yen (about $275 billion).

A newspaper tally of Nikkei 225 constituents puts fiscal 2024 policy holdings at 54.596 trillion yen ($333 billion), down about twenty per cent, the largest fall in five years. Different universes, one at market value and one apparently at carrying amount, both current, and about 9.6 trillion yen (roughly $59 billion) apart, which is a fifth of the larger one. The most eloquent table in Japan has no total row.

Disposals are the number that moves fastest and settles slowest. Newspaper tallies of annual reports put sales of policy holdings at 9.2 trillion yen ($56 billion) for the year ended March 2025, up about fifty-three per cent, across more than two thousand companies; and 9.77 trillion yen ($60 billion) across roughly four thousand reports filed between January and November 2025, a second consecutive record. The two are not the same measure and cannot be joined into a series. They are also provisional by nature: the equivalent figure for the year ended March 2024 was first reported at 3.6 trillion yen (about $22 billion) and later ran to about 6.1 trillion yen (about $37 billion) as filings accumulated. Records set by rolling collection tend to be broken by arithmetic.

The one series immune to share prices is acquisition cost, disclosed for other securities. Over the year to March 2025, for five major banking groups it fell from 4.09 trillion yen to 3.43 trillion yen ($25.0 billion to $20.9 billion), a reduction of 0.66 trillion yen ($4.0 billion). For five life and non-life insurers, from 3.09 trillion yen to 2.71 trillion yen ($18.9 billion to $16.5 billion). For regional banks, from 2.87 trillion yen to 2.90 trillion yen ($17.5 billion to $17.7 billion) — flat, and slightly the wrong way, though the compiler does not say how many banks stand behind that line.

Acquisition cost has the property that reclassifying a holding from policy to pure investment does not reduce it, which is why it is used to test whether reductions are real. It also falls on impairment, so not all of the movement is sale, and the bank and insurer baskets are five names each rather than industries.

Reclassification was worth a rule of its own. The amendment promulgated and effective on 31 January 2025 requires disclosure, for shares moved from policy to pure investment purposes within the most recent five fiscal years including the current one and still held at year end, of the name, share count, carrying amount, the fiscal year of the change, the reason for it, and the policy for holding or selling afterwards. It applies from fiscal years ending on or after 31 March 2025. The regulator's stated reason was that its review of annual reports identified cases where the purpose had changed and the substance had not. As of July 2026 this remains the most recent amendment touching policy holdings; the ordinance changes promulgated on 20 February 2026 concern sustainability and human capital disclosure.

The Code on a Parallel Track

The governance code moved on a parallel track and in the same direction. The 2015 original had three elements under a heading that called them so-called policy shareholdings: disclose a policy, verify mid- to long-term economic rationality annually at board level for major holdings, and establish and disclose criteria for voting. The words reduction, individual issue, cost of capital and specific criteria did not appear, and there were no supplementary principles. The 2018 revision, effective 1 June, deleted so-called, required disclosure of a policy on reduction, moved verification from major holdings to individual holdings, set the test as whether the purpose is appropriate and whether the benefits and risks cover the cost of capital, required disclosure of what the board found, upgraded voting criteria to specific criteria applied in practice, and added two supplementary principles: do not obstruct a holder who wishes to sell, and do not continue transactions that lack verified economic rationality.

The 2021 revision changed the principle by zero words; a machine comparison of the 2018 and 2021 texts returns an exact match, despite substantial comment-letter pressure. The 2026 revision, applying from 21 July 2026 with revised governance reports due by end-July 2027, restructures the code by abolishing supplementary principles, and promotes those two into the body of Principle 1-4 as items (3) and (4). The substance is unchanged, and no interpretive guidance is attached.

Then there is the compliance data, which is where the exercise becomes interesting. As of 12 July 2024, across 3,244 Prime and Standard listed companies, Principle 1-4 was complied with by 3,046 and explained by 198, a rate of 93.9 per cent — low, by the standards of this code. The supplementary principle on not obstructing sales ran at 99.8 per cent, and the one on transactions lacking economic rationality at 100.0 per cent.

In April 2025 the regulator published a column, based on its review of annual reports, on pressure not to sell: cases in which an issuer seeking stable shareholders hinted at reducing existing business with a holder that wanted to exit. Its hearings with the companies doing the holding put the proportion of names subject to such pressure at roughly five to forty per cent by name count. The range is a hearing-based estimate with undisclosed sampling, which is the honest way to describe it. The regulator's own formulation is that a company reporting compliance while organising such pressure would undermine the spirit of the code and could amount to a serious governance problem — could, not does. The principle promoted into the code body in 2026 is the one with the 99.8 per cent compliance rate.

What the exchange has consistently declined to do is set a target. Its published interpretation, repeated across public consultations in 2018, 2021 and 2026, is that policy shareholdings include one-way holdings and deemed holdings, not merely mutual ones, and that the principle does not require uniform reduction — only that holdings found inappropriate on board verification are expected to be reduced. In the 2026 consultation, which drew 147 respondents overall, the responses numbered 270 to 278 addressed policy shareholdings: disclosure of reduction targets, of sale deadlines, and a norm of eventual zero. Each was received as a comment. None became a rule.

The Decade That Assembled the Structure

The structure being disclosed was assembled in a decade of policy that ran in the opposite direction. Inward direct investment was liberalised from a cabinet decision of June 1967 in five rounds, the first in July 1967 covering seventeen industries at a hundred per cent and thirty-three at fifty, the fifth in May 1973 making conditional full liberalisation the rule with agriculture, forestry and fisheries, mining, petroleum, leather goods and retail left outside. The standard account, offered as historical interpretation rather than estimated causation, is that the prospect of foreign acquisition accelerated large mergers and stable-shareholder arrangements — a paper industry three-way merger contract in March 1968, the Yawata-Fuji combination that formed Nippon Steel in March 1970. Corporate-dominant ownership is generally dated as complete around 1970.

Where it peaked is harder to say than the round numbers suggest: a narrow cross-holding estimate reached 21.6 per cent of listed market value in fiscal 1990 on a series that only begins in fiscal 1987, a broad measure stood at 50.5 per cent at the end of fiscal 1990 on a series that begins there, and the exchanges' own long ownership series peaks at 46.0 per cent for financial institutions in fiscal 1989 and 27.5 per cent for business corporations in fiscal 1973 under definitions that changed several times in between. Banks and life insurers went from above thirty per cent to twelve by 2004.

Sixteen years of amendments have made the table longer, the names more numerous, the purposes more specific, the verification more procedural and the reclassifications traceable for five years back. The last of the required items still asks whether the issuer on the other side holds shares in the filer, and records the answer as yes or no. It does not ask what the answer costs. That was the field for the quantitative effect, and in the first year it was required — among the ninety-three large-cap filers with March 2019 year-ends whose reports the Nomura Institute of Capital Markets Research had in hand on 24 June 2019 — it was answered ninety-three times with the observation that answering it was difficult. What later years did with it, that survey does not say.

Yen amounts are converted throughout at 163.8 yen to the dollar, the rate on 24 July 2026, including figures reported for fiscal 2024 and 2025. The dollar figures therefore carry no information about currency movements and should not be compared across time as if they did. The yen is the original unit here; the dollar is a convenience. The 1990 and 2010 material above is ratios and rules, which need no conversion at all.