The Stuck Residual: Cross-Shareholding Architecture in Japan Prime

The four-band architecture

The framework's B-01 indicator scores cross-shareholding intensity in four bands, anchored to share-of-net-assets ratio thresholds. The publicly-known ISS vote-against line at 20% of net assets serves as the production gate: firms above that line enter the Kill Switch population.^[2] Below it, the framework distinguishes elevated (10–20%), moderate (5–10%), and clean (≤5%) bands.^[3]

Universe distribution at the FY2025 yuho window:^[4]

The coverage gap consists predominantly of firms where the disclosure parsing yields no usable cross-shareholding total — typically issuers with non-standard yuho formats, recent IPO firms, or preferred-class duplicates. These firms still receive grade and archetype based on the other 17 indicators; they simply cannot be located on the B-01 band map.

What the distribution shows is a compressed-but-not-retired residual. The clean cohort (43%) is by far the largest, but it does not dominate the way it would in a market where cross-shareholding had been substantively eliminated. Roughly a quarter of Prime — 369 firms — sits at near-residual or above. This is the residual after a decade of CG Code revisions,^[5] the 2022 Prime Market reform,^[6] the 2023 TSE capital-cost initiative and FSA Action Program,^[7] and successive enhancements to cross-shareholding (政策保有株式) yuho disclosure rules across 2010, 2019, and 2025.^[1] The headline narrative says cross-shareholding is being reduced; the residual distribution says it has been compressed but not retired.

The B-01 × archetype mapping

The archetype distribution within each B-01 band is the centerpiece of this Note:^[4]

Three observations.

First, the stuck band IS the Kill Switch. Every firm above the ISS line fires KS-1 by construction — that is how the framework operationalizes the threshold. The 96 firms in the stuck band and the 96 firms in KS-1 are the same 96 firms. There is no partial overlap, no near-miss cohort. The threshold is binary.

Second, the near-residual band is overwhelmingly Chameleon. 82% of the 273 firms in the 10–20% band fall into Chameleon archetype. These are the structural Chameleons — the firms whose B-axis or R-axis weakness traces directly to elevated cross-shareholding even though the absolute level remains below the KS threshold. The 26 Celestials in this band are large-cap firms whose other axes (T-axis disclosure, B-axis board structure, R-axis other components) clear the relevant adequacy thresholds despite the cross-shareholding penalty.

Third, the clean band is the Celestial cohort. 55% of firms with cross-shareholding below 5% of net assets land in Celestial. The 268 Chameleons in this band are firms whose archetype lag traces to something other than cross-shareholding — independent-director structure, capital-efficiency profile, RPT pattern, or other B-axis or R-axis components.

The combined reading: cross-shareholding band is the single indicator with the cleanest archetype-predictive power in the Japan production stack. It is not the only force shaping the distribution — independent-director ratios, capital efficiency, RPT all contribute — but among individual indicators, none explains as much of the bipolar shape as B-01.

This is the mechanism behind Note 1. The bipolar Celestial / Chameleon split is, to a substantial degree, the cross-shareholding band split projected onto the archetype taxonomy. Firms with clean cross-shareholding tend to land in Celestial; firms with elevated cross-shareholding tend to land in Chameleon; firms above the threshold land in KS by construction.

The stuck residual: where 6.1% sits

The 96-firm stuck residual is not randomly distributed across sectors:^[4]

Banking and insurance combined account for 22% of the stuck residual — the legacy bank-network cross-shareholding cluster. Construction adds another 10%. Manufacturing-heavy sectors (chemicals, electric machinery, textiles) and trading-related sectors (wholesale, warehouse) account for the next layer. IT/Communications appearing at 10 is somewhat unexpected for a sector with limited keiretsu legacy — these are typically large-cap firms with bilateral cross-shareholdings tied to system-integration partnerships or carrier-vendor relationships.

What is not in the stuck residual: services, retail, pharmaceuticals (each at 3 firms or fewer). Sectors without keiretsu legacy or bank-network embedding rarely accumulate cross-shareholding above the ISS line.

Banking as the residual cluster

Banking deserves the closest attention because the sectoral pattern is the steepest in Prime.

Of 69 banks in the Prime universe, the full B-01 distribution is:^[4]

62% of banks sit at near-residual or above — the highest sector-level concentration in Prime. The grade distribution that follows: 28 firms at C, 19 at KS, 18 at B, 3 at A. The C + KS share — 47/69 = 68% of banks — lands in the watchlist or worse tier. This is, by some distance, the steepest sector-level grade gap in the Japan universe.

But the framing matters. The 18 stuck-residual banks are not the megabanks. The three Megabank holding companies — Mitsubishi UFJ, Sumitomo Mitsui, Mizuho — are not in the stuck-residual 18. They have publicly committed to and reported on cross-shareholding reduction over the past decade, and their CGSO disclosures show progressively retired positions.^[8] The stuck-residual 18 are predominantly regional and second-tier banks where the legacy mochiai (持合い) relationships with industrial-firm clients have not been retired at the same pace.

The distinction matters for the analyst. A blanket statement that "Japanese banks fail the framework" is wrong: 18 of 69 fail, the failures concentrate in regional banks, and the megabanks largely clear the threshold. The framework's residual reading is sector-specific in a way that headline aggregation obscures.

The near-residual: who is unwinding but not yet clean

The 273-firm near-residual cohort tells a different story. Here the sector concentration sits in trading houses and manufacturing:^[4]

The wholesale + manufacturing block (chemicals + machinery + electric machinery + transportation equipment) accounts for 41% of the near-residual cohort. These are the firms publicly engaged in unwinding — most have explicit reduction commitments in their CGSO disclosure — but the absolute level remains in the elevated band. The cohort is in motion, but the trajectory has not yet crossed the moderate threshold.

The contrast with the stuck residual is structurally meaningful. Stuck-residual firms have not initiated meaningful unwinding; near-residual firms are unwinding but the trajectory has not crossed below 10% of net assets. The two cohorts represent different governance postures: structural holdouts versus firms in active reform.

Keiretsu cluster differentiation

The six horizontal keiretsu (Mitsubishi, Sumitomo, Mitsui, Mizuho, Fuyo, DKB-Sanwa) and the two principal vertical keiretsu (Toyota Group, Honda Group) provide useful anchor structure for the cross-shareholding architecture.^[9] The framework's production data does not carry keiretsu-affiliation tags; cluster identification requires external mapping (Toyo Keizai 系列マップ, Daiwa 経済リサーチ keiretsu directories).

Anchor-firm-level B-01 distribution (citation-safe public-knowledge keiretsu hubs only) suggests qualitative differentiation across clusters. Among the six horizontal keiretsu hub firms surveyed, the Mizuho cluster carries the heaviest stuck-or-near residual concentration. The Mitsui cluster sits at the other end — most cleaned-up among the horizontal hubs. The Toyota Group hub firms sit at mid-cluster, with partial unwinding visible at the hub level and bilateral cross-shareholdings still material across major Group firms.

These are anchor-firm observations (n = 3 to 9 per cluster), not statistical claims about full keiretsu populations. The point is directional: keiretsu identity is not destiny — clusters differ meaningfully in their unwinding pace, and the framework's cross-sectional snapshot captures that differentiation.

A separate observation matters here. The largest single keiretsu unwind currently in motion is the Toyota Group restructure announced in April 2025, with Toyota Industries' tender offer transition driving a multi-year cross-shareholding reduction across the Toyota Motor / Toyota Industries / wider Toyota Group architecture.^[10] Note 3 reads this as the parent-subsidiary unwinding wave; in B-01 terms it is the most significant near-residual-to-clean migration on the Prime calendar.

The seven flavors that fold into one ratio

A methodological note matters here. Cross-shareholding in Japan is not a single phenomenon. The legacy literature identifies at least seven distinct flavors, each with different governance implications:^[11]

• transaction-partner (取引先): supplier-customer mutual holdings sustaining commercial relationships
• main-bank (銀行) — legacy main-bank: bank-network holdings tied to historical lending relationships
• insurance group (保険): insurer-industrial-firm mutual holdings
• keiretsu (系列): horizontal-keiretsu lateral cross-shareholdings within the six clusters
• founder family (創業家): founder-family holdings sustaining family control
• parent company (親会社): controlling-shareholder positions in listed subsidiaries
• anti-takeover (防衛策): cross-shareholding deployed as poison-pill alternative

Each flavor carries different governance weight. Founder-family holdings sustain control with intent. Bank-network holdings persist through inertia. Anti-takeover holdings are deployed defensively. Parent-subsidiary holdings entrench control at the cost of minority protection.

The framework's production B-01 indicator does not distinguish these flavors. All seven fold into a single share-of-net-assets ratio computed from the firm's cross-shareholding disclosure totals. The methodological compromise is honest: per-flavor classification would require natural-language processing on each disclosed issue's holding purpose (保有目的) free-text field, and the framework defers this to v1.2 NLP scope.

The trade-off is straightforward. Single-ratio measurement loses the flavor distinction but gains universe-wide comparability and parsing reliability. For the cross-sectional question — how much cross-shareholding does this firm carry, in aggregate? — the ratio is sufficient. For the deeper question — what kind of cross-shareholding, and with what governance implication? — the analyst needs to read the underlying CGSO disclosure.

Readers who want flavor-level reading should expect partial answers from the framework alone. The 96-firm stuck residual contains a heterogeneous mix: regional banks (legacy main-bank flavor), construction firms (transaction-partner flavor), the occasional founder-controlled large-cap. The 273-firm near-residual is more homogeneous — predominantly transaction-partner and keiretsu flavor with banks and chemicals carrying the bulk.

Trajectory: where the residual sits in historical context

The cross-shareholding architecture has been compressed substantially over the past two decades. Industry-aggregate trajectories show steady decline: by one widely-cited measure, the simple-average ratio of policy-held (政策保有) holdings to net assets across the TOPIX 500 has fallen from approximately 13.5% at fiscal year-end March 2015 to 8.4% at fiscal year-end March 2023.^[12] Most of the reduction has come from the megabanks, the major non-life insurers, and the largest non-financial firms; the residual that remains is concentrated in regional banks, mid-cap manufacturers, and family-controlled firms.

The framework reads the cross-sectional slice. It does not have year-on-year reduction tracking in the v1.0 production scope — that requires multi-year panel construction and the v1.2 work has not yet been completed. What the cross-section tells the analyst is what remains after a decade of pressure, not how fast the pressure is working.

The 6.1% above-ISS-line cohort matters not because it is large in number but because it is structural. These firms have absorbed the disclosure pressure of successive policy-held yuho rule enhancements,^[1] the market pressure of TSE 2022 Prime expectations,^[6] the regulatory pressure of the 2023 FSA Action Program and TSE capital-cost initiative,^[7] and the institutional pressure of stewardship-code-signatory investors voting against directors at firms with elevated cross-shareholding.^[13] They have absorbed all of that and still sit above the line. The reform mechanism has reached the firms that respond to mechanism; the residual is the firms that do not.

What the cross-shareholding signal implies

Three readings.

First, B-01 carries near-binary archetype-predictive power. A firm above the ISS line is, by construction, a Kill Switch firm. A firm in the elevated band is overwhelmingly Chameleon. A firm in the clean band is more often Celestial than not. Cross-shareholding is not the only signal that determines archetype, but among individual indicators it is the cleanest.

Second, sectoral context disciplines the reading. "Japanese banks fail" is wrong; "regional Japanese banks carry the residual" is correct. "Trading houses fail" is wrong; "trading houses are unwinding but not yet clean" is correct. The aggregate Prime distribution conceals the sectoral structure that the analyst needs.

Third, the 6.1% structural residual is the analytical center of gravity for governance-distress sourcing in Japan. Where Korea concentrates governance failure in the chaebol cross-holding architecture and India concentrates it in the promoter pledge cascade, Japan concentrates it in regional bank cross-shareholding and the longer tail of mid-cap holdouts. These are different mechanisms, and each market's framework reading reflects its own pathology.

The cross-shareholding architecture is the foundation. Note 3 reads the parent-subsidiary unwinding wave — the largest single category of cross-shareholding currently in active motion. Note 4 reads the shadow-governance layer of sodanyaku and komon (相談役 and 顧問) — the post-CEO advisor positions that operate alongside the formal board structure. Each is a vertical cut into the residual described here.

The bipolar archetype split and the Banking sector outlier flow primarily from this single architecture. Cross-shareholding is the keystone of Japan's archetype distribution.

Apex G-Score™ Japan Foundation Series. Production cohort: TSE Prime, 1,576 issuers (1,556 rated; 20 NR). The "FY2025" label denotes a rolling-window cohort — each firm's most recent yuho filed within 2025-03-11 to 2026-04-15. Fiscal year coverage by 決算月 (fiscal year-end month): 3月 (March) ~76.5% (reporting period 2024-04 to 2025-03); 12月 (December) ~13.7% (calendar year 2025); 9月 (September) ~5.1% (2024-10 to 2025-09); 6月 (June) ~4.6% (2024-07 to 2025-06). The December-end sub-cohort shares the calendar-year reporting window of the Korea cohort. Framework: Apex G-Score v2 (T 0.30 / B 0.30 / R 0.40); archetype classifier: unified-v1. Sample firm-level scores in this article are limited to the IP-guardrail-v2.0 disclosed set (Toyota Motor; Samsung Electronics and Reliance Industries appear in cross-market comparisons only). All other firm-level scores remain unpublished. Multi-year indicators where cited are noted explicitly with source vintages. This article is research output, not investment advice; readers should consult their own advisors before making investment decisions.