Mitsubishi UFJ Financial Group (8306)
The Multiple Is the Whole Trade
Research Report — Mitsubishi UFJ Financial Group (8306)
The Multiple Is the Whole Trade
A complete research report, published in full.
Document type: Synthesised research report Sealed: 2026-10-01 Data cut-off: 2026-09-30 Coverage: 8306.T (TSE Prime). Fiscal year ends 31 March. Basis: Twelve completed, verified, sealed research passes in my own research archive Version policy: This document is sealed. All twelve underlying passes remain immutable and were not modified in producing it. Where this document and a prior pass differ, the prior pass governs.
What this document is not. It contains no price target, no rating, no buy or sell recommendation. It is a research artefact, not a recommendation. Section 2 explains what the market must currently believe in order to justify the price — not whether I think it should.
Editor’s note (added in the English edition, not part of the sealed report)
This report presents the research process in full: the derivation chain, the epistemic status assigned to each number, the record of corrections, and eighteen questions the author left unresolved.
Four conventions are used throughout, and they are worth internalising before reading:
- Every figure carries a label. FACT, DERIVED, ASSUMPTION, UNRESOLVED. This is not decoration. A reader should be able to tell, sentence by sentence, what MUFG disclosed from what the research inferred.
- Sensitivity tables are scenarios, not forecasts. Where the report gives a table of outcomes, the axis is an input the analyst chose. The report says so explicitly and repeatedly.
- Two inconsistencies found during the English translation are flagged in-line rather than silently corrected. They are marked as Editor’s note and are described in full. The sealed Chinese original is immutable and still contains the original wording.
- Internal study names are preserved verbatim (
8306-return-model,8306-cet1-decomposition, and so on) because they are how the adjudication record in Appendix A stays traceable. They are not securities or data products.
One unit convention you must know before reading Section 3. MUFG labels its fiscal year by the starting calendar year. Its “FY2025 Databook” covers the year ended 31 March 2026. Throughout this report:
- FY2025 = fiscal year ended 31 March 2026 (MUFG’s “2026年3月期”)
- FY2026 = fiscal year ended 31 March 2027 (MUFG’s “2027年3月期”)
MUFG’s “FY2026 target” and its “Q1 FY2027 results” therefore sit one year apart in Western calendar terms. Getting this wrong inverts the sequence of events in Sections 4 and 5.
How to read this report — four kinds of statement
Every number in this report is labelled. Please read by the label:
| Label | Meaning |
|---|---|
| FACT | A disclosed figure from company IR, official financial statements, or regulatory filings — or a market figure already public from a third party |
| DERIVED | Computed or inferred by this research from FACTs. Not a company forecast and not company guidance |
| ASSUMPTION | A scenario input the model requires in order to run. Not a prediction |
| UNRESOLVED | A question that has not been closed. This report does not complete it |
Specific warning on MUFG’s own “forecast.” For the year ending March 2027, MUFG discloses 業績予想に代えて目標値を記載 — “a target value disclosed in place of an earnings forecast.” That number is a company-set target, not a forecast. This report refers to it as the company target throughout, never as guidance or consensus.
1. Executive summary
The question
In September 2026 the market pushed MUFG’s (8306) PBR from roughly 1.6x to 1.88x. The research question was narrowed to one sentence:
Under the currently verified earnings and capital-constraint assumptions, what exactly is the market paying for — and what has to be true for this price to hold?
The core valuation frame
The causal chain established after the eleventh pass is not a conventional three-stage valuation. It is:
Business / interest rates
↓
EPS, ROE (already established as true)
↓
The market broadly accepts this earning power
↓
The genuine free variable = PER / PBR
↓
Future shareholder return
Key finding (FACT / DERIVED): between 2026-05-15 and 2026-09-07, MUFG’s forward ROE was essentially flat (12.51% → 12.57%) while its PBR rose 14.37%. The market did not re-price ROE improvement — it re-priced the multiple.
Final base case
With exit PER held at 14.9x, three-year annualised total return:
| Scenario | Annualised total return |
|---|---|
| Company FY2025 payout policy maintained (60.7% total payout ratio) | 8.86% |
| Company FY2026 actual authorisation (43.8% total payout ratio) | 9.70% |
Base case band: 8.86% – 9.70% annualised total return (DERIVED; ASSUMPTION: ROE 12.11% and PER 14.9x held constant).
The single most important outcome driver
Exit PER. On an identical earnings path:
| Exit PER | Annualised total return |
|---|---|
| 12.0x | 1.86% |
| 14.0x | 6.78% |
| 14.9x | 8.86% |
| 16.0x | 11.30% |
The multiple moves about 7 percentage points. Every payout-related variable combined is worth 0.84 percentage points.
The largest downside driver
Multiple compression to 12x. Even on the most investor-friendly low-payout scenario (g_EPS 7.07%), a 12x exit PER yields only about 2% annualised total return.
Additional FACT: MUFG’s current PBR of 1.873x is higher than every one of the seventeen year-end observations on record (previous high 1.32x at 2026-03-31), and PBR÷ROE is above the entire five-year year-end range by 22%.
Conclusions the evidence supports
- The earnings improvement is real. Over five years NII contributed 51.4% of profit growth and fees 42.5%, together 94% (DERIVED from FACT).
- But the market did not re-price it. Forward ROE was flat across the re-rating window (DERIVED).
- The capital constraint is real. CET1 was 9.2% at 2026-03-31, below the company’s own 9.5% floor (FACT, verbatim from company IR).
- That constraint directly drove the FY2026 buyback cut from the ~¥500bn scale down to ¥100bn (FACT, company states it is “considering CET1 ratio”).
- The market is not pricing MUFG’s dependence on Morgan Stanley (DERIVED from FACT).
Still unknown
- The market cost of equity for this name — never observed, and seven passes confirmed that no defensible PBR can be cited
- The path of the exit PER — this report does not forecast it
- Whether the 9.5% CET1 floor is a hard constraint or a soft one
- What proportion of future RWA growth “earns well”
- The 25% gap between MUFG’s disclosed quarterly equity-method income and its 23.95% stake
2. Investment thesis
This section is not a recommendation, not a rating, and contains no price target. It describes what the market must currently believe and where that belief is supported and where it is fragile.
What the market must currently believe
At a share price of ¥3,700, the market must simultaneously believe three things:
- ROE can be sustained around 12% — this is the company’s own medium-term target (FACT, IR verbatim: 「東証定義 ROE12%程度 かつ PER の維持・向上」, “ROE around 12% on a TSE-defined basis, and maintenance/improvement of PER”).
- The market will keep paying 14.9x forward PER — higher than any year-end PBR in MUFG’s seventeen-year record.
- The earnings contributed by MS will not reverse — 32%–39% of net income over five years (DERIVED from FACT).
Of the three, only the first is even partly within the company’s control.
How MUFG’s earnings, capital allocation and per-share growth connect
The causal chain this research established (FACT / DERIVED):
ROE 12.11% (rolled forward from a consensus base)
↓
Payout: dividends 40.1% + buybacks 3.7% ~ 20.6% = total payout ratio 43.8% ~ 60.7%
↓
Retained portion compounds at ROE → equity growth
Buyback portion → share count reduction → EPS accretion
↓
g_EPS
↓
At a fixed PER, the share price grows at the same rate as g_EPS
↓
Plus dividend cash
↓
Annualised total return
No step on this chain is a “re-rating.” Every step is the company earning the money back.
What a reasonable return looks like at a 14.9x exit PER
ASSUMPTION: ROE 12.11%, PER constant at 14.9x, dividend payout held at 40.1%.
Three-year total return, annualised: 8.86% – 9.70%, depending on whether the total payout ratio lands at 43.8% or 60.7%.
Composition of that number:
- roughly 6.2 – 7.1 percentage points from per-share EPS growth
- roughly 2.7 – 2.9 percentage points from dividend cash
The effect of multiple compression on the outcome
| Scenario | Annualised total return |
|---|---|
| exit 12.0x | 1.86% |
| exit 12.0x (low-payout scenario) | about 2.0% |
| exit 14.9x | 8.86% |
| exit 16.0x | 11.30% |
Moving from 14.9x to 12.0x removes about 7.0 percentage points of annualised return.
Why the payout assumption matters less than the exit multiple
Two numbers side by side settle it:
| Variable | Range it can move through | Impact on annualised return |
|---|---|---|
| Total payout ratio (43.8% – 60.7%) | 16.9 percentage points | about 0.84pp |
| Exit PER (12.0x – 16.0x) | 4.0 turns of multiple | about 7pp |
The payout ratio can move across 17 percentage points and deliver less than 1 percentage point of return. The multiple can move 4 turns and deliver 7 percentage points.
The reason is structural: at ROE 12.11%, retained earnings compound at 12.11%, while buybacks accrete EPS at 1/PER = 6.71%. Shifting payout from dividends to buybacks barely changes the per-share outcome; moving the multiple from 14.9 to 12 changes the terminal price enormously.
3. The business and earnings engine
ROE (FACT / DERIVED)
| Fiscal year | Consolidated ROE |
|---|---|
| FY2021 | 6.68% |
| FY2022 | 6.51% |
| FY2023 | 8.09% |
| FY2024 | 9.28% |
| FY2025 | 11.34% |
| forward (rolled forward from consensus base) | 12.11% (DERIVED) |
FACT — official ROE uses period-average shareholders’ equity. Verified: 2,427,229 ÷ 21,397,158 = 11.34%, an exact match.
FACT — DuPont decomposition (FY2024→FY2025): net margin +1.84pp, asset turnover +0.28pp, leverage −0.47pp. Leverage was a drag (DERIVED from FACT).
Three growth rates that must never be conflated
This is the most important structural element of the model in this report.
| Symbol | Name | Formula | Value (high payout / low payout) |
|---|---|---|---|
| g_B | Equity and net income growth | ROE × (1 − total payout ratio) | 4.76% / 6.81% |
| g_S | Share count change | −buyback payout ratio ÷ PER | −1.38% / −0.25% |
| g_EPS | EPS growth | (1 + g_B) ÷ (1 + g_S) − 1 | 6.23% / 7.07% |
The identities behind these:
- g_S = −b/p, because buyback spending of b×NI retires (b×NI)÷P shares, and P = p × EPS = p × NI/S
- g_NI equals g_B, because at a constant ROE net income grows at the same rate as equity
- In steady state BVPS grows at the same rate as EPS, because B and NI grow together and the share count is the common denominator
Why the three cannot be collapsed into one
This research made all three of these errors at some point:
| Error | Where | What was wrong |
|---|---|---|
| Definition | 8306-valuation |
g = ROE × (1 − 40.1% dividend payout) = 7.25%. Used the dividend payout ratio only |
| Definition | 8306-multiple-anchor |
g = ROE × (1 − 60.7% total payout) = 4.75%. Caught the direction of the problem, but this is net income growth, not per-share growth |
| Correct | 8306-return-model |
Split into three quantities; for a per-share investor the answer is g_EPS = 6.23% |
All three attempts (7.25% → 4.75% → 6.23%) were wrong, in different directions.
A second-layer error: the effect of buybacks on book value.
BPS_new = (B − Pq) / (S − q)
Δ(BPS)/BPS = (q/S)(1 − PBR) / (1 − q/S)
FACT — MUFG’s PBR is 1.873. At q/S = 1.38%, the calculation gives Δ(BPS)/BPS = −1.22%.
Editor’s note (English edition). The sealed Chinese original prints −0.94% at this point. The source pass’s own calculation sheet (
8306-return-model/.../data/return_model.md) prints −1.22%, and substituting the stated inputs into the stated identity gives (0.0138 × (1 − 1.873)) ÷ (1 − 0.0138) = −1.22%. The −0.94% in the sealed report appears to be a transcription error; it is reproduced here corrected to −1.22% with the discrepancy disclosed rather than silently fixed. The conclusion the number supports is unchanged and directionally identical: when PBR > 1, buybacks reduce book value per share.
When PBR is greater than 1, buybacks reduce book value per share. MUFG trades at 1.87x, so buybacks are not book-value accretive. Plugging the total payout ratio directly into a retention formula is not a valid identity.
Dividend and buyback history (FACT)
Original table was denominated in ¥100mn units; converted to ¥bn below. Conversions are exact.
| Fiscal year | Dividends (¥bn) | Buybacks (¥bn) | Total payout ratio |
|---|---|---|---|
| FY2020 | 321.8 | 0 | 41.3% |
| FY2021 | 357.1 | 150.0 | 44.8% |
| FY2022 | 390.0 | 450.0 | 75.2% |
| FY2023 | 488.0 | 400.0 | 59.6% |
| FY2024 | 742.0 | 400.0 | 61.3% |
| FY2025 | 976.0 | 500.0 | 60.8% |
FACT — the dividend payout ratio alone is only 40.3%, but including buybacks the total payout ratio is 60.8%. The 40% policy number understates half of the distribution.
4. The capital constraint and CET1
What changed from 2025-03 to 2026-03
Both endpoints are on the same basis (FACT): 10.5% (2025-09) and 9.2% (2026-03) carry exactly the same footnote — the finalised Basel III basis, effective March 2029, deducting AFS unrealised gains/losses.
This is the basis on which this report rules out a “basis artefact” explanation.
| Step | Effect | Ratio |
|---|---|---|
| Value as reported 2025-03 | 10.8% | |
| Less: FX effect of the subsidiary results change date | −30bp | 10.5% |
| Value as reported 2026-03 | 9.2% | |
| Substantive decline | −1.3pp |
Attribution of the −1.3pp (FACT — management’s own account)
| Attribution | Share | Approximate |
|---|---|---|
| The Shriram Finance investment | about half | about −0.65pp |
| RWA from loan growth | most of the remainder | about −0.65pp |
FACT — the estimated impact announced at the time of the Shriram Finance transaction was “approximately negative 60 basis points.” A single investment accounted for roughly half of the full year’s substantive deterioration, in a year MUFG earned a record ¥2.43tn.
From a 9.5–9.7% estimate to an actual 9.2% (FACT — management’s itemised walk)
| bp | Item | Did management treat it as a surprise? |
|---|---|---|
| +10 | Net income ahead of estimate (excluding incremental payout and FX) | No |
| −5 | Additional FX effect on Shriram Finance | Mechanical |
| −9 | Period-end loan balance ¥3tn above the usual trend | No — verbatim, “within our expectation” |
| −4 | Increase in retained earnings at equity-method investees (principally Morgan Stanley) | No |
| −20 | Operating RWA increase from high-margin business | Yes — the only item management called a surprise |
Total −28bp.
Independent verification from regulatory filings (FACT — Basel III KM1, current basis, ¥mn)
| Date | CET1 capital | RWA | Ratio |
|---|---|---|---|
| 2025-09-30 | 15,605,282 | 110,808,030 | 14.08% |
| 2025-12-31 | 16,452,727 | 117,496,950 | 14.00% |
| 2026-03-31 | 15,002,259 | 120,281,761 | 12.47% |
| 2026-06-30 | 16,202,176 | 125,020,709 | 12.95% |
DERIVED — RWA grew 8.5% over six months (¥110.8tn → ¥120.3tn) while CET1 capital fell 3.9% (¥15.61tn → ¥15.00tn). Both moved in the wrong direction, and the denominator moved more.
DERIVED — the ratio recovered 48bp in the 2026-06 quarter alone (12.47% → 12.95%).
The relationship to the FY2026 buyback cut from ~¥500bn to ¥100bn
FACT — MUFG’s statement on the FY2026 buyback: “Resolved share repurchase up to ¥100bn for FY26H1, considering CET1 ratio and organic growth”, with footnote 5, “Utilized a part of the capital release effect through the sale of MUFG Union Bank.”
FACT — FY2026 total payout ratio = (dividends 1,082,700 + buybacks 100,000) ÷ 2,700,000 = 43.8% (DERIVED, with the company target of ¥2,700,000 as the denominator).
What the 9.5–10.5% target range means
FACT — MUFG’s CET1 target range is 9.5% – 10.5% (the floor is the regulatory requirement plus buffer; the ceiling is operating policy).
FACT — management’s framing of the 9.2% reading: “we expect to return to the target range within this fiscal year.”
FACT — management’s framing of the basis: “we must manage the target range for the CET1 ratio, excluding unrealized gains under the finalized Basel III with greater sensitivity since the finalized and fully implemented basis will be in 3 years or so. We consider the gap between our CET1 ratio under the current Basel III basis and the finalized and fully implemented basis as a buffer.”
Adjudication of the causal chain
| Causal chain | Status |
|---|---|
| CET1 decline → constrained capital allocation → buyback cut | Supported (company’s own account + independent verification from regulatory filings) |
| CET1 decline ← Shriram Finance investment | Supported (company’s own account) |
| CET1 decline ← RWA from loan growth | Supported (company’s own account + verified by RWA data) |
| CET1 decline mainly caused by a change in basis | Ruled out (both endpoints on the same basis) |
Explicitly not claimed:
- That the constraint is permanent — management expects a return to the range in H1, and the filings already show a 48bp recovery in June
- That the FY2027 payout returns to 60%
- That 9.5% is a hard constraint — management’s wording reads soft, but cutting the buyback reads hard. This report does not adjudicate this.
5. Morgan Stanley
This section deliberately refuses to simplify MS into “a positive” or “a negative.” The same asset pushes in opposite directions on the earnings side and the capital side.
The stake (FACT)
| Item | Value |
|---|---|
| Voting rights (2026-03-31) | 23.9% |
| Beneficial ownership | 23.95% |
| Shares held | 380,307,520 |
| Board representation | 2 directors |
| Standstill expiry | 2028-10-13 |
Earnings contribution (FACT / DERIVED)
| Fiscal year | Equity-method investment income (¥mn) | Share of net income |
|---|---|---|
| FY2021 | 441,500 | 39.0% |
| FY2022 | 425,800 | 38.1% |
| FY2023 | 531,800 | 35.7% |
| FY2024 | 596,900 | 32.0% |
| FY2025 | 845,500 | 34.8% |
FACT — the five-year range is 32.0%–39.0%, with no upward trend.
DERIVED — this is not “a newly emerging risk.” It is a permanent structural feature. This corrected the framing in the original 8306-morgan-stanley pass.
DERIVED — 35.7% of FY2025 segment net operating profit (¥2,365,383mn).
MS’s own position in the cycle (FACT — SEC filings)
| Period | Net income attributable to MS | ROTCE |
|---|---|---|
| CY2024 | $13,390M | 18.8% |
| CY2025 | $16,861M | 21.6% |
| 2Q2026 | $5,581M | 26.6% |
DERIVED — MS’s 2Q2026 pre-tax margin was 34.4%. MUFG’s single largest earnings source is attached to a cyclical high in its own core business that has not yet come down.
DERIVED — in a flat scenario (2H2026 matching 2H2025), MS’s CY2026 comes to $20,155M, which corresponds to MUFG equity-method income of ¥757.9bn, 10.4% below the ¥845.5bn actually recorded in FY2025.
DERIVED — a 30% reversal in MS earnings reduces MUFG’s ROE by approximately 0.99 percentage points, making the company’s “around 12%” target unreachable even with zero deterioration in the domestic business.
Capital consumption (new finding in the eleventh pass, FACT — management verbatim)
“One unique aspect for us is that this portion [exceeding the 15% threshold for specified items] increases in proportion to our stake in Morgan Stanley as its retained earnings increase. While Morgan Stanley’s significant profit increase is positive in terms of our share of earnings, it has aspects that are not so welcome from a capital ratio perspective.”
That is the −4bp in the CET1 walk.
DERIVED — the capital drag from outperformance is roughly 4bp per quarter.
The effects in each direction (which must be kept separate)
| Dimension | Direction | Magnitude |
|---|---|---|
| Net income | Accretive | 32–39% of net income |
| Segment NOP | Accretive | 35.7% |
| EPS growth | Accretive | via g_EPS |
| CET1 ratio | A drag | about −4bp per quarter |
The reverse of the market-rule assumption also holds: the only FX channel here is a yen-depreciation channel — if BOJ tightening causes the yen to appreciate, this item is damaged.
Editor’s note (English edition). The sentence above reads 「MarketRepo 假設的反向效果也成立」 in the sealed Chinese original. “MarketRepo” appears to be an internal label — most likely a model or rule name that survived from a working note. It has been left untranslated rather than guessed at. The substantive claim (the FX channel runs only through yen depreciation) is self-contained and does not depend on the label.
Is the market pricing this?
DERIVED — the three megabanks’ PBR÷ROE converged into a 0.1618–0.1661 range (2026-09-28). MUFG at 0.1661 versus the JPX banks index at 0.1797 — MUFG is 7.8% cheaper than its own sector, and cheaper on PER as well (15.5 versus 17.98).
The market is not pricing the MS dependence — neither on earnings quality nor on capital cost.
UNRESOLVED
The 25% gap between FACT and DERIVED: MUFG’s disclosed quarterly equity-method income implies an MS quarterly net income of $6,956M, whereas MS actually reported $5,581M for 2Q2026 — a ratio of 1.246. Proportionality roughly holds at the annual level (MS is 75.0% of the total equity-method amount), but not at the quarterly level.
This report does not explain the gap. Candidate reasons, all unverified: the Economic Interest Percentage in the Investor Agreement may be higher than 23.95%; other equity-method affiliates may contribute above the full-year run-rate; the actual conversion rate may not be 157; book-basis differences may exist; the equity-method income line may contain items other than MS net income.
Until this is resolved, every absolute-value estimate of MS contribution carries systematic uncertainty.
6. Valuation
This report refuses to cite a “fair PBR”
FACT — the four methodological conclusions of the 8306-valuation pass:
- The common g = 2% assumption implies a payout ratio of 82.9%, which contradicts the official 40.1% → withdrawn
- Substituting a self-consistent g makes the Gordon model pathological
- The well-behaved substitute is to invert the problem: the expected return implied by the current price
- The cost of equity has never been observed, and the Gordon form is pathological for this name → no single fair PBR should be quoted
This position survived twelve passes without being overturned. This report maintains it.
Exit PER sensitivity (DERIVED; ASSUMPTION: ROE 12.11%, PER constant, 40.1% dividend payout)
| Exit PER | Annualised total return |
|---|---|
| 12.0x | 1.86% |
| 13.0x | 4.38% |
| 14.0x | 6.78% |
| 14.9x (current anchor) | 8.86% |
| 15.5x | 10.20% |
| 16.0x | 11.30% |
The effect of payout policy
| Scenario | Total payout ratio | Buyback share of payout | g_EPS | Annualised total return |
|---|---|---|---|---|
| FY2025 payout maintained | 60.7% | 20.6% | 6.23% | 8.86% |
| FY2026 actual authorisation | 43.8% | 3.7% | 7.07% | 9.70% |
Cutting the payout ratio from 60.7% to 43.8% raises EPS growth from 6.23% to 7.07%, and raises annualised total return from 8.86% to 9.70%.
⚠ Counter-intuitive but mechanical: a lower payout ratio produces better per-share growth. Retained earnings compound at ROE 12.11%; buybacks accrete EPS at 1/PER = 6.71%. At 14.9x, the compounding speed of retention is close to twice that of buybacks.
Historical anchors for the multiple (FACT)
| Fiscal year | Year end | PBR | PER | Bank sector median PBR | MUFG / sector |
|---|---|---|---|---|---|
| FY2014 | 2014-03-31 | 0.62 | 8.15 | 0.51 | 1.22x |
| FY2019 | 2019-03-29 | 0.44 | 8.22 | 0.31 | 1.42x |
| FY2023 | 2023-03-31 | 0.59 | 9.35 | 0.29 | 2.03x |
| FY2024 | 2024-03-29 | 0.93 | 12.49 | 0.39 | 2.38x |
| FY2025 | 2025-03-31 | 1.13 | 12.57 | 0.45 | 2.51x |
| FY2026 | 2026-03-31 | 1.32 | 12.20 | — | — |
| Now | 2026-09-30 | 1.873 | 14.90 (forward) | — | — |
⚠ Basis warning: the historical PERs are trailing earnings, whereas the current 14.9x is on expected earnings. Comparing them directly overstates how stretched the multiple is. PBR is the clean comparison (both sides use book value at the time).
FACT — the current price PBR is higher than every year-end observation in the seventeen-year record.
FACT — MUFG’s premium to the bank sector median PBR has risen from 1.22x in FY2014 to 2.51x in FY2025.
⚠ The nature of these numbers
Everything in the table above is an ASSUMPTION-driven model scenario, not a forecast.
- It assumes ROE stays at 12.11%
- It assumes PER at the end of the holding period equals the specified exit value
- It assumes the payout policy does not change
- It assumes no OCI movement
The real path will not look like this. This report does not forecast the exit multiple.
7. Return decomposition
The four components
| Component | Nature | Magnitude (high / low payout) | Source |
|---|---|---|---|
| EPS growth | per-share | 6.23% / 7.07% | g_EPS (model derivation) |
| Dividend cash | cash | 2.86% | 40.1% dividend payout ÷ PER |
| Share count reduction | per-share | already inside g_EPS | must not be added again |
| Multiple change | valuation | scenario axis | never added into a return |
Three boundaries that must not be crossed
① Dividends are a cash return. The shareholder receives cash directly.
② Buybacks affect EPS through share count, not through cash. The shareholder receives no money; the benefit appears as a larger claim on earnings.
③ If buybacks are already inside EPS growth, they cannot also be counted as a separate return component.
This error actually occurred.
The effect register
The checkable form of Rule 16 of my research protocol (added to this project after the tenth pass):
| output_id | effect_id | representation | occurrences |
|---|---|---|---|
| g_B | E_RETAINED_EARNINGS |
per-share | 1 |
| g_S | E_SHARE_COUNT |
per-share | 1 |
| g_EPS | E_EPS_GROWTH |
per-share | 1 |
| Dividends | E_DIVIDEND_CASH |
cash | 1 |
| Buybacks | E_SHARE_COUNT |
per-share | same effect as g_S |
| Multiple | E_MULTIPLE |
valuation | scenario axis, not an addend |
Hard rule: the same effect_id may appear only once in a given decomposition.
This check immediately caught an error whose arithmetic was entirely clean: one pass added “dividends 2.69% + buybacks 1.38% = shareholder yield 4.08%” into total return, even though E_SHARE_COUNT was already inside g_EPS. That figure has been formally invalidated.
The register forces a distinction that narrative tends to blur:
E_MS_CAPITAL_DRAG(capital side) andE_EPS_GROWTH(earnings side) are both driven by Morgan Stanley’s profits, but in opposite directions. A strong MS pushes EPS up and CET1 down.Collapsing them into “MS is good” or “MS is bad” is a definitional error.
The correct decomposition
Annualised total return = g_EPS (per-share) + dividend yield (cash)
= 6.23% + 2.86% → compounded 8.86% (high payout)
= 7.07% + 2.86% → compounded 9.70% (low payout)
E_SHARE_COUNT appears only inside g_EPS.
8. What the model actually says
If every currently verified assumption holds, what genuinely determines the outcome?
Risk decomposition
| Variable | Impact | Status |
|---|---|---|
| Earnings / retention assumptions | Limited | Fully modelled (g_B enters the formula) |
| Payout / buyback mix | Limited to moderate (0.84pp) | Fully modelled |
| CET1 / capital constraint | Moderate (sets the payout level) | Causality closed through to the FY2026 round |
| Exit PER | Dominant (about 7pp) | Not modelled as a forecast — this is an honest blank |
The core conclusion
Improvement in operating assumptions does not eliminate valuation multiple risk.
Multiple compression from 14.9x to 12x affects annualised return far more than any change in the current payout assumptions.
This conclusion has been stable across twelve passes, and each pass has strengthened it:
- First stated in the eighth pass (“the multiple is a level risk, not a growth requirement”)
- Ninth pass added the historical anchor (the current PBR is above all seventeen year-end observations)
- Tenth pass found it more valid once the true g_EPS of 6.23% replaced the assumed 10% per year
- Eleventh pass quantified it (payout 0.84pp versus multiple 7pp)
- Twelfth pass confirmed that closing the causality on the capital constraint does not change the ranking
This is neither “cheap” nor “expensive”
FACT — MUFG’s PBR÷ROE is 7.8% cheaper than its own sector, and it is cheaper on PER. FACT — MUFG’s PBR is above every year-end observation in seventeen years, and its PBR÷ROE is above the entire five-year year-end range.
Both are facts. That is precisely the substance of the valuation multiple question: there is no single “cheap/expensive” answer, because the answer depends on what multiple you are willing to hold a bank at.
9. Evidence quality
Evidence grading
| Tier | Weight | What it covers |
|---|---|---|
| Company IR / official disclosure | Primary | Earnings summaries (決算短信), results presentations (決算説明資料), IR presentations, Basel III related data, shareholder return policy |
| Regulatory filings | Primary | Basel III KM1 (current-basis CET1 / RWA time series) |
| Earnings call transcript | Moderate | FY2025 results call (third-party transcription; key figures cross-checked against official documents) |
| Third-party market data | Limited | Consensus EPS, share price, BPS, PER, sector multiples |
| Research derivation | All valuation conclusions | g_B / g_S / g_EPS, return decomposition, PBR÷ROE historical series |
| Scenario assumptions | All sensitivity tables | ROE constant at 12.11%, specified exit PER, 40.1% dividend payout, no OCI movement |
The classification list
FACT (company or regulatory disclosure)
- ROE series 6.68% / 6.51% / 8.09% / 9.28% / 11.34%
- Total payout ratio series 41.3% – 75.2%
- CET1 9.2% (2026-03-31), target range 9.5%–10.5%
- MS stake 23.95%, 380,307,520 shares
- MS net income and ROTCE (SEC filings)
- FY2026 target of ¥2,700,000mn (a company target, not a forecast)
- Official financial figures including net assets per share, liabilities, shareholders’ equity
DERIVED (research derivation)
- Forward ROE 12.11%
- All of g_B / g_S / g_EPS
- Annualised total return 8.86% / 9.70%
- PBR÷ROE historical series
- The exit PER sensitivity table
- The “earnings growth + dividends” decomposition
ASSUMPTION (scenario)
- 14.9x as the anchor (this is a current observation, not an assumption; assuming it holds is an assumption)
- exit 12.0x / 14.0x / 16.0x scenarios
- ROE 12.11% unchanged over three years
- No OCI movement
- Buybacks executed at the prevailing market price
UNRESOLVED
- The market cost of equity
- The path of the exit PER
- The hardness of the 9.5% floor
- The quality split of RWA growth
- The 25% quarterly equity-method gap
- The shareholder-equity bridge in the 8306 passes (verified against the official statement of changes in equity, but the dataset used at the time had an extraction residual)
⚠ Specifically flagged
The 12.95% figure requires a basis note. It is the ordinary-share-equivalent Tier 1 ratio on a current (non-pro-forma) basis: 12.47% at 2026-03-31 and 12.95% at 2026-06-30. It is not the same quantity as the pro-forma 9.2%. Whenever the two appear together, the basis and date must be stated.
10. Unresolved questions and research limits
This report does not answer any of the following.
Valuation
- The market’s cost of equity for MUFG — never obtained in twelve passes. This is the direct reason the Gordon model cannot be cited.
- The path and probability of the exit multiple — direction only, no probabilities. The historical multiple series from IRBANK was collected as annual snapshots (five observations), not as a full distribution.
- The 8% – 10% cost-of-equity range is an analyst-set plausible range, not an observation.
Capital
- Whether the 9.5% CET1 target is a hard constraint or a management goal — management’s wording (“best to return to the lower range as soon as possible”) reads soft, but cutting the buyback reads hard. This report does not adjudicate.
- How much of future RWA growth is in high-return assets — MUFG discloses the rotation (cutting low-yielding business, adding high-yielding business), but planned and achieved figures cannot be matched from the sources obtained. This determines whether ROE improvement and shareholder returns move in the same direction.
- Whether the capital cost of MS earnings growth has long-run significance — the specified-item deduction rises with MS retained earnings, so the stronger MS gets, the tighter CET1 becomes. How this scales over time is not modelled.
- Whether Shriram Finance is one-off or part of a broader India investment plan — if the latter, the capital constraint is structural. The investment plan was not obtained.
- Whether the FY2026 H2 buyback is executed — this directly determines whether the base case lands at the 8.86% end or the 9.70% end.
- RWA growth rate versus internal capital generation rate — determines the long-run sustainable payout level.
Business
- The 25% gap between MUFG’s quarterly equity-method income and its 23.95% stake — see Section 5.
- The cause of the PBR move from ~1.6 to 1.88 between 2026-03 and 2026-09 — once TOPIX was added in the seventh pass, the abnormal return against all four beta benchmarks was positive (+0.15% to +0.22%), which refutes the “already priced in” inference. But the cause of the jump itself remains unexplained.
- The regional bank cross-section — only partial data (79 rows, mean PBR 0.58 / ROE 5.3% / PER 9.9, all third-party). The JPX sector index is 68% the three megabanks, so “sector” here mostly means the three megabanks themselves.
- The 7x contradiction between Q1 JGB-related gains/losses and the full-year figure — Q1 FY2027 was −¥35,036mn, annualising to roughly −¥140,000mn, whereas full-year FY2024 was −¥991,400mn. Basis or time distribution has not been separated.
- MUFG’s adjusted ROE — Mizuho discloses an “adjusted ROE (実力値) of 10.5%”; MUFG and SMFG make no equivalent disclosure. ROE rankings therefore carry a caveat.
Methodological limits
- Every sensitivity table in this report assumes ROE, PER and the payout ratio are constant. The real path will not be.
- September was a rotation month; interest-rate transmission and sector rotation cannot be separated within a single window.
- The historical multiple series is an annual snapshot, not a distribution. “Above the five-year range” holds on five observations.
- Most of this report’s key derivations come from MUFG’s own account (earnings call). This is the company’s explanation of its own numbers, not third-party audit.
11. Falsification and update triggers
This report should be reopened if any of the following occurs.
Capital
| Trigger | Why it matters |
|---|---|
| CET1 still below 9.5% at FY2026 year end | Management committed to returning to the range in H1. A miss invalidates the “the constraint is temporary” reading, and the payout may be constrained further |
| H2 buyback not executed and the FY2027 authorisation cut again | The base case should be revised down to the low-payout end or below |
| The 9.5% floor is explicitly lowered, or management publicly describes it as a soft target | Directly changes the weight of the capital constraint |
| A large investment beyond Shriram Finance | The capital constraint becomes structural rather than one-off |
Valuation
| Trigger | Why it matters |
|---|---|
| Any market cost-of-equity observation obtained | The largest unresolved gap in twelve passes. Once obtained, the pathology of the Gordon model can be re-evaluated |
| PER breaks below 13x or moves above 16x | Enters directly at the two ends of this report’s sensitivity table |
| MUFG changes its policy language on PER (for example abandoning “maintain/improve”) | The company’s own medium-term target is one of this report’s supporting arguments |
Business
| Trigger | Why it matters |
|---|---|
| MS ROTCE falls materially from 26.6% | The flat scenario holds, and the inference that flat MS equals −10.4% is validated |
| The 25% quarterly equity-method gap is resolved | Affects every absolute-value estimate of MS contribution |
| ROE deviates from the 12% target | Forward ROE of 12.11% is the input to every model here |
| The structure of RWA growth changes (the high-yielding additions target is met or missed) | Determines whether ROE improvement and shareholder returns move in the same direction |
| Dividend / buyback policy clearly changes | The payout end of the base case moves |
Methodological
| Trigger | Why it matters |
|---|---|
| The effect register rules prove insufficient — for example a third representation of the same economic effect appears | The double-counting control would need redesign |
| The cause of the PBR move from 1.6 to 1.88 is explained | Affects the strength of the conclusion that the multiple is the dominant variable |
Appendix A: Adjudication record
Where passes conflicted, this report adopts the last completed pass carrying an explicit verdict. The full settlement history follows.
| Topic | Evolution | Adopted | Adjudicated by |
|---|---|---|---|
| Return at 14.9x | 11.7% → 8.86% → 8.86–9.70% band | 8.86% – 9.70% | 8306-return-model → 8306-capital-return |
| g (per-share growth) | 7.25% → 4.75% → 6.23% | g_EPS 6.23% (high payout) / 7.07% (low payout) | 8306-return-model → 8306-capital-return |
| shareholder yield 4.08% | asserted → invalidated | dividends only, 2.69–2.86% (buybacks already inside g_EPS) | 8306-capital-return effect register |
| “FY2025 showed zero underlying growth” | asserted → withdrawn | Withdrawn. Five-year NII 51.4% + fees 42.5% | 8306-bond-cycle |
| “MS dependence is a new risk” | framed → reframed | Permanent structural feature (32–39% over five years, no trend) | 8306-bond-cycle |
| “MUFG is expensive versus the sector” | hypothesis → partly refuted | MUFG PBR÷ROE 0.1657 vs index 0.1797, i.e. 7.8% cheaper than its own sector | 8306-redteam |
| “Fully priced” | UNRESOLVED → inference refuted | Not supported. Abnormal return positive on all four beta benchmarks | 8306-required-earnings |
| CET1 9.2% vs 12.95% | called a “conflict” → judged not a conflict | Different bases. 12.95% is the current-basis line and it has itself moved (12.47% in March) | 8306-capital-return → 8306-cet1-decomposition |
| Cause of the CET1 decline | causally undecided → closed | Mainly substantive capital consumption (Shriram about half, lending RWA most of the remainder) | 8306-cet1-decomposition |
| Attribution of the 11.7% figure | challenged → reconciled | It is the 14.9x total-return row, not the 16x row (16x is 14.2%) | 8306-return-model |
Adjudication principle: prior passes are not modified. A reversal is declared in a new pass and recorded in this appendix; this report cites only the final verdicts.
Appendix B: Sources
Company IR / official disclosure (PRIMARY)
- FY ended March 2026 earnings summary (決算短信, Japanese basis) — full consolidated income statement, statement of changes in shareholders’ equity, BPS
- Earnings summaries for the years ended March 2025 / 2024 / 2023 — the five-year income series
- Securities report data set 2026 (有価証券報告書データ集) — five-year consolidated financial series
- IR presentation
speech2603_en— the 9.2% CET1 figure verbatim, the target range, the buyback explanation - FY ended March 2026 results presentation
slides2603_ja— capital allocation and dividend policy - January 2026 investor presentation — medium-term target “ROE around 12% on a TSE-defined basis, and maintenance/improvement of PER”
- FY2025 interim results meeting materials — CET1 10.5% at 2025-09, RWA rotation targets
- Shareholder return policy — dividend policy verbatim
- Q1 of the year ending March 2027 earnings summary — Q1 financials
- FY2025 interim results data (data corner) — breakdown of revenue from funds employed
Regulatory filings (PRIMARY)
- Basel III related data, 2027 Q1 / 2026 Q2 / 2026 Q3 — the KM1 series of CET1 capital, RWA and ratios
Earnings call (PRIMARY content, third-party transcription)
- FY2025 results call transcript — management’s attribution of the −1.3pp CET1 decline, the itemised −28bp walk from a 9.5–9.7% estimate, and the MS specified-item language verbatim. Key figures were cross-checked against official documents; the attribution detail has not been second-confirmed by an official document.
Third-party market data (SECONDARY / THIRD_PARTY)
- Traders / 96ut — share price, BPS and PER series
- Minkabu — consensus EPS
- IRBANK — 17-year year-end multiple snapshots
- kabutore.biz — bank sector median PBR (not primary)
- KABU Lab (citing JPX) and the JPX banks index factsheet
Internal
- Twelve passes in the 8306 series (
research/8306*/), all immutable
Appendix C: What this report did not do
Listed explicitly to prevent misreading:
- No new research was performed. This report only synthesises twelve existing passes; no new data was searched for
- No new targets, valuation methods, scenarios, or unverified assumptions were added
- No existing pass was modified
- No searching was done to fill unresolved questions
- No intermediate assumption was written up as a fact
- No model derivation was written up as company guidance
- None of the questions in Section 10 were answered here
- No price target, buy/sell recommendation, or rating is provided
- No economic effect was double-counted
Document control
| Item | Value |
|---|---|
| English edition created | 2026-10-01 |
| Source document | Sealed Chinese report, run 20261001-215625 (immutable) |
| Research passes synthesised | 12 |
| Unresolved questions preserved | 18 |
| Falsification triggers listed | 14 |
| Contested numbers adjudicated | 10 |
| Typos found during the original audit | 2, both fixed in the English edition’s source audit record |
| Editorial interventions in this English edition | 2 figures flagged in-line (see Section 3 note and Section 5 note); unit conversion of the payout history table to ¥bn; a fiscal-year labelling note added to the front matter |
This document contains no price target, no rating, and no recommendation. It is a record of research, published for its method as much as for its conclusions.