Banco Bradesco S.A. (NYSE: BBD) — An Insurer Stapled to a Bank, Priced as One Mediocre Bank

Report date: 2026-08-01 · Independent fundamental research · Initiation of coverage

Security: NYSE: BBD (ADR, 1 ADR = 1 BBDC4 preferred share) · B3: BBDC4 (preferred), BBDC3 (common) · NYSE: BBDO (ADR over the common) Price: US$3.61 per ADR (close 2026-07-31) · R$18.43 per BBDC4 · BRL/USD ≈ 5.07–5.10 Market capitalization: ≈ US$37.9bn · Sector: Financials — Diversified Banks & Insurance (Brazil) SEC CIK: 0001160330 · Foreign private issuer: files 20-F and 6-K, not 10-K/10-Q

Timing warning, stated once and carried throughout. Bradesco reports Q2 2026 on 2026-08-05 — four days after this report date. The latest reported quarter is Q1 2026 (released 2026-05-07). Every credit, margin and capital figure in this memo is pre-print and has a four-day shelf life on precisely the lines that matter most. Separately, on 29 July 2026 — two days before this report date — Bradesco announced a capital increase of up to R$10bn. Both facts are load-bearing.

The analysis in Sections 1–15 below is written position-free: it carries no investment recommendation and no price target. The single, deliberately fenced-off exception is the author’s own opinion block immediately following.

⚡ Claude’s Take

The author’s own subjective opinion, deliberately fenced off from the position-free analysis that follows. General information only — this is not investment advice, and it is not a recommendation to buy or sell any security. Sections 1–15 below carry no recommendation and no price target of any kind.

AVOID at $3.61. Not a short. Accumulate only below ~$2.60 (~0.75× book).

Tag: The turnaround is real. It’s also fully paid for.

The call, and the arithmetic behind it. Bradesco is a competent, genuine, nine-quarter operational repair that the market has already re-rated 46% off its December-2024 low — and, having done so, now prices the company at almost exactly what management has promised to deliver. Correct the share count that every vendor screen gets wrong (10.57bn shares, not 9.25bn — provable from Bradesco’s own 5.721967934% preemptive ratio) and the stock trades at ~1.02× book, not the widely-repeated 0.95×. It is not below book. At 1.02× the Gordon identity collapses to ROE = cost of equity independent of growth, which means the market is making no forecast at all: it is pricing indifference. At a 15% cost of equity, today’s price implies ~15.1–15.7% ROE. Management guides FY2026 to 15.5% and delivered 15.8% last quarter. There is no unpaid-for upside from the plan working exactly as advertised — and the skew around it is bad: a stall back to the 2020–22 average of 12% is −40%; JP Morgan’s explicitly-framed ceiling of ~17.5% is +19%. Two identifiable risks are entirely absent from the price, and both are bearish: the tax shield is worth 31% of recurring net income and moving to the effective rate Itaú has already guided to is −26% with no operating deterioration; and 36% of insurance operating income is float carry at a Selic that Bradesco’s own house view says falls to 9.5%.

What the market is pricing correctly, and the framing. This is not a mispricing — it is a value trap risk dressed as a value opportunity, and the framing is deep-cyclical recovery, twenty months in, with the easy re-rating already banked. The tape agrees and is worth reading: momentum loading is +0.04 (nil), short interest is 0.4%, the five-year Sharpe is 0.006 and the ten-year is −0.056 — this is neither a crowded long nor a falling knife, it is a Brazil country instrument (Country:Brazil factor beta 1.53, higher than Itaú’s 1.43, with 21% idiosyncratic vol you must underwrite). The bull anchor — “cheap versus 1.8× book in 2019” — is wrong twice: the level is wrong, and 2019 is the wrong regime. Against the post-2021 PIX/Open-Finance world, today is the 76th percentile. Underneath it all sits the real business: a ~22%-ROE insurer on 28% of the capital producing 41% of the profit, stapled to a ~13%-ROE bank that has lost 40% of its demand-deposit share in three years and has a smaller client base than in 2022. Mark the insurance at the price the market already pays for it — including Bradsaúde’s own R$46.6bn listing — and the banking stub implies a 12.0–12.8% perpetual ROE against the ~12.4% it earns. Both halves are priced about right. There is no hidden free bank.

Conviction: medium. Bradesco is genuinely better than it was, the Bradsaúde reorganisation was a good decision, and at an 8% yield with the credit book demonstrably cleaner this is emphatically not a short — the bear case only works if the multiple breaks, and holding the multiple flat the bear scenario still returns +16%. What would flip me bullish: two consecutive quarters of recurring ROAE above 17% with the 12-month efficiency ratio through 47% and coverage stable — i.e. evidence that the bank, not the insurer, is driving the improvement. What would flip me bearish: the 5 August print showing coverage down another 5 points below 161% while MSME and rural NPLs keep rising — or management guiding a 2026 effective tax rate above 27%.

One timing note, stated plainly: Q2 2026 reports on 5 August, four days after this memo. Everything here is pre-print.

📈 Stock Price Action — Five-Year Event Map

Factual price history — not a recommendation and not a price target.

The arc. Over the trailing five years the BBD ADR fell ~58% into a $1.58 low (18-Dec-2024) — a 14-year low — then more than doubled to a five-year high of ~$4.12 (17-Apr-2026) before drifting back to ~$3.61, roughly 12% off that high and mid-range in a 52-week band of $2.79–$4.30. Zoom out and the round trip is far less flattering: the ADR traded near ~$10.39 in July 2019 on the same split-adjusted basis, so today’s price is ~65% below the pre-COVID level, five-year total return is +10.2% (~+2.0%/yr) and ten-year total return is −0.9% — an entire decade of nothing, with an 8.15% trailing dividend yield doing the heavy lifting.

Cycle narrative. (1) 2021’s decline was rate-and-politics, not company-specific — BBD and ITUB fell together — and the early-2022 bounce was the global value/commodity rotation into cheap EM banks. (2) The April–November 2022 slide is where BBD’s path separated: the 9-Nov-2022 Q3 print produced an 18.6% single-day loss against ITUB’s 6.1%, a 12.4pp idiosyncratic gap, as unsecured consumer credit written in 2021 defaulted and management raised provision guidance. (3) The Americanas accounting fraud then forced a R$4.9bn full-cover provision in the Q4-22 report, collapsing quarterly ROE to ~4% and, on 10-Feb-2023, taking BBD down 7.5% on a day ITUB rose 1.0%. (4) 2023’s recovery was borrowed from the macro — the Selic easing cycle and the global November rate rally — not from Bradesco’s own P&L, which is why it faded. (5) 2024 was the capitulation: the market marked down a five-year turnaround plan by 15.5% in a day (a 14.1pp idiosyncratic hit), then Brazil’s fiscal-package disappointment and a record-low real did the rest. (6) 8-May-2025 is the cleanest inflection in the dataset — a +18.9% day with 17pp of idiosyncratic outperformance versus ITUB, on the first quarter where the transformation showed in the numbers. (7) The Q4-25 print delivered the plan’s stated milestone, ROAE above cost of capital, into a record Ibovespa and the start of Selic easing. (8) The move since April is not a Bradesco story: BBD’s −6.6% on 3-Mar-2026 came with ITUB −5.2%, and the Brazil country factor is itself down ~6% over 126 days (z −1.02) — the pullback is the country factor, not the bank.

1. Executive Summary

No recommendation and no price target appear in this section or anywhere in Sections 1–15.

Banco Bradesco is two businesses stapled together, and the consolidated statements obscure it. Insurance produces 16.2% of revenue but 41% of net income, earning 21.6% on 28.5% of group equity. The residual — the universal bank, its 1,938 branches, R$1.09trn loan book and 82,095 employees — earns roughly 13.1%, at or below any defensible Brazilian cost of equity with the Selic at 14.25–14.75%. Management’s declaration at the Q4-2025 print that group ROAE had exceeded its cost of capital “for the first time” is a group-level statement that does not survive disaggregation. The correct framing is a ~22%-ROE insurer with a ~13%-ROE bank attached.

The turnaround is real and deserves to be stated without hedging. Recurring net income has grown for nine consecutive quarters; ROAE has climbed from ~10.0% (FY2023) to 15.8% (1Q26); 28% of the branch estate has gone since 2023; and the credit book is measurably cleaner — secured mix +3.8pp to 60.8%, restructured portfolio −20% y/y, problem assets −28%, Stage 1+2 at 92.9%. The Bradsaúde reorganisation (February–May 2026) crystallised the group’s best assets into a separately-listed R$46.6bn platform at no cash cost while adding ~250bp to pro-forma CET1. That is competent work.

But it is a repair, not a re-founding, and four things undercut it. First, the profit is in the insurer, not the bank. Second, the margin expansion is bought with credit risk and leverage: gross NIM is +50bp y/y while net NIM is −10bp; cost of risk jumped to 3.5% (retail 5.5%); coverage has fallen 22 points to 161.0%; the fastest-growing lines are vehicle (+25.4%), rural (+17.5–24.0%) and MSME working capital (+16.3%) — exactly where management flags deterioration; and assets/equity has risen from 11.0× to 13.5×, accounting for roughly 83% of the five-year ROE improvement. Third, the capital is thin and low-quality: CET1 fell 100bp in one quarter to 10.2%, prudential adjustments are 34.5% of stated equity, and net deferred tax assets of R$109.3bn equal 61% of book equity. Fourth, the tax line is fragile: a ~20% recurring effective rate against 45% statutory means roughly a third of recurring net income is a tax-structuring artefact.

On competitive position the verdict is direct: no durable advantage in banking; a narrow but genuine one in insurance. The dispositive evidence is Bradesco’s own — 58% of branches removed since 2015 with no loss of loan market share. An asset you can more than halve without losing the share it defended was never a barrier to entry. Greenwald’s market-share-stability test fails comprehensively in banking (demand deposits 11.0% → 6.6% in three years; private payroll 12.0% → 6.6% in one; INSS benefit payments −4.4pp) and passes cleanly in insurance (22.7 / 22.9 / 22.8% across 2023–25). The cost position is inverted: a 49.2% efficiency ratio against Itaú’s 36.2%, with derived cost-to-serve of ~US$9–10 per client per month against Nubank’s ~US$0.80. The strategic tension is that 93% of pension sales — the group’s most captive book — are branch-originated, so the bank must run a bank-shaped cost base to protect an insurance-shaped profit stream.

Growth is low-quality with one high-quality exception that management guides to decelerate. The client base is smaller today (74.3m) than in 2022 (77.1m); checking-account holders peaked in March 2025; real loan growth is ~2.9% p.a. over five years. Every headline growth metric is a re-segmentation of a shrinking base. The one engine genuinely earning its cost of capital is health insurance, which supplied 100% of Bradesco Seguros’ FY2025 profit growth while life and pension shrank 14.8% — and it is guided from +20.4% down to +6–8%.

Capital allocation has been poor, and the July 2026 equity raise is the bill arriving. Over FY2021–25 Bradesco distributed R$56.5bn — ~57% of net income — sustained straight through the 2023 trough, while CET1 fell to 10.2% and leverage rose; then recalled R$10bn. Buybacks totalled R$1.68bn, 3% of distributions. The decade’s M&A record is R$16bn on HSBC Brasil for scale in a business with no cost advantage (most of it since closed), R$2.15bn more of Cielo in 2024 — an acquirer whose profit pool PIX is expropriating, returning 1.55% in cash — and nothing sold in five years. Variable compensation keys to absolute accumulated net income with no ROE, efficiency or per-share metric anywhere in the scheme, under a board where 4 of 11 directors are independent.

On valuation, one correction changes the headline. Every vendor screen carries ~9.25bn shares; the true count is 10.57bn, provable from Bradesco’s own 5.721967934% preemptive ratio and confirmed by the 20-F. Book value per share is R$17.01, not R$19.43, and the stock trades at ~1.02× book — at book, not below it. At that level the Gordon identity collapses to ROE = cost of equity independent of growth, so the market is making no forecast: it prices indifference. At a 15% cost of equity the price implies ~15.1–15.7% ROE against guidance of ~15.5% and 1Q26 delivery of 15.8%. The market is already paying, precisely, for management to hit its own plan. The scenario skew is unattractive: a stall to 12% is −40%; the guide is −2%; JP Morgan’s stated ceiling of 17.5% is +19%. Two large risks sit outside the price, both bearish — normalising the tax rate to Itaú’s guided band is −26%, and normalising insurance float income to a 9.5% Selic removes ~80bp of group ROAE.

The single question, on which everything turns: is ~15.5% ROE this franchise’s ceiling or its floor? At ~1.02× book the market says ceiling, with no credit for success and no penalty for failure. Q2 2026 reports on 5 August — four days after this memo — and coverage, MSME/rural NPLs, and any effective-tax-rate guidance are the three lines that begin to answer it.

2. Business Overview

2.1 What Bradesco actually is

Banco Bradesco S.A. is Brazil’s fourth-largest financial institution by assets (10.3% system share, March 2026) and the country’s largest insurance group (22.8% of insurance premiums plus pension contributions plus capitalization income, SUSEP/ANS, September 2025). It carries R$2,435bn of total assets (Mar-2026), R$180.3bn of equity, 82,095 employees, 1,938 branches, and — on third-party BACEN-derived rankings — roughly 110m customers, of which the company itself discloses 37.9m checking-account holders and 28m fully digital clients. [FY2025 Form 20-F, filed 2026-03-25, Section 4.B; 1Q26 press release (6-K), filed 2026-05-07]

Structurally, it reports two operating segments plus a de-minimis “Other Activities” line:

2.2 The revenue split does not resemble the profit split — and that is the whole company

This is the single most important structural fact about Bradesco, and it is visible only when you put the two disclosures side by side.

FY2025 managerial revenue [4Q25 press release, 6-K filed 2026-02-06]:

Audited segment net income [FY2025 Form 20-F, Note 38 “Operating Segments”]:

FACT: insurance produces 16.2% of revenue and 41% of net income. In FY2023 — the trough of the retail-credit and Americanas crisis — it produced 58%. INTERPRETATION: for three consecutive years the insurance franchise has been the reason Bradesco was profitable at all. Strip it out and the FY2023 “bank” earned R$6.0bn on roughly R$110bn of attributable equity — a ~5.5% ROE.

Push the arithmetic to the current quarter. Bradesco Seguros carried R$49,521m of consolidated equity at Mar-2026 against group equity of R$173,549m — 28.5% of the group’s capital. It earned R$2,760m in 1Q26 (~22.3% annualized; 21.6% on reported average equity). The residual is therefore R$4,051m of banking profit on R$124,028m of equity — a ~13.1% ROE.

INTERPRETATION, and it is the spine of this memo: Bradesco is best understood as a ~22%-ROE insurance and health business stapled to a ~13%-ROE bank. With the Selic at 14.75% at the end of 1Q26 and a Brazilian bank cost of equity plausibly 14–17% nominal, the universal-banking operation — the branches, the loan book, the cards, the 82,095 employees — is at or below its cost of capital and has been for years. Only the insurer creates economic value. Management’s declaration at the 4Q25 print that group ROAE had exceeded cost of capital “for the first time” is true at the group level and does not survive disaggregation. (The derivation uses Bradesco Seguros’ consolidated equity as a proxy for capital allocated to insurance; the parent does not disclose segment capital. Label: INTERPRETATION built on primary disclosure.)

2.3 How the bank makes money — the mechanics

Net interest income (1Q26, R$m) [1Q26 press release]:

Two observations that matter more than the headline. First, market NII is now ~3% of total NII (R$553m of R$20,051m). It went deeply negative in 2023 (−R$312m in 1Q23) and management declared it “a closed chapter” by 3Q23. The recovery from that hole is fully banked — there is no remaining “treasury normalizes” tailwind to underwrite. Client NII is 97% of the margin and its +16.3% y/y is a volume-and-spread story (average balance +10.7%, gross NIM +50bp).

Second, and less comfortably: gross NIM is up 50bp y/y while net NIM — after provisions — is DOWN 10bp y/y and 30bp q/q. Client NII net of provisions fell 4.5% q/q. Bradesco is buying its margin expansion with credit risk. Provisions are running +26.5% y/y against 8.4% portfolio growth.

Fee and commission income (1Q26, R$m):

Cards are 43% of fees (turnover R$95bn in 1Q26, +9% y/y; high-income clients ~51% of card revenue, +24% y/y). Asset management runs R$1,489bn of AuM (+16.4% y/y, 16.2% market share) with R$2.9tn under custody. The two lines in outright decline — checking account (−6.9%) and collections/payments (−7.0%) — are precisely the two most exposed to PIX. That is not a coincidence; it is the fee line reporting the regulatory event described in Section 3.

Insurance (1Q26, R$m):

FY2025: insurance net income R$10.1bn (+11.2%), ROAE 21.9%, faturamento R$118.5bn.

One caveat must travel with every insurance number in this memo. R$2,301m of the R$6,384m (36%) of 1Q26 insurance operating income was financial results — investment income on R$482.8bn of assets at a 14.75% Selic. Bradesco’s own house forecast is Selic 12.50% end-2026 and 9.50% end-2027. A meaningful slice of the 21–22% insurance ROE is a rate carry, not an underwriting moat, and underwriting-only ROE is not separately disclosed anywhere.

2.4 The loan book

Expanded loan portfolio, March 2026 [1Q26 press release]:

Two readings, both required. The quarter’s book was flat (+0.1% q/q) against FY2026 guidance of +8.5% to +10.5% — MSME fell 2.3% q/q, large corporates 0.2%, and only individuals grew. Either Q2–Q4 accelerates sharply or the credit-growth guide misses. And the year-on-year mix is the uncomfortable part: the fastest-growing lines are vehicle (+25.4%), rural (+17.5% / +24.0%) and MSME working capital (+16.3%) — exactly the mass-retail and small-business exposures that produced the 2022–23 blow-up, and exactly where management now flags deterioration (see Section 6).

2.5 Recurring vs. non-recurring — three different “net income” numbers exist

The gap between recurring and book in 1Q26 is −R$1,781m, from adherence to Brazil’s Programa de Transação Integral (PTI) tax-settlement programme plus other tax provisions. Management’s headline — “+16.1% y/y, ninth consecutive quarter of growth” — is the recurring number. On the book number, 1Q26 net income fell 13.3% y/y and 22.3% q/q. That is a 26% difference in a single quarter’s earnings, and every headline in the financial press used the flattering one.

2.6 Verdict — Business Overview

Bradesco is a genuine universal financial group with real national scale in every product it sells, and it is the market leader in the one product line in Brazilian financial services that the state has not attacked. But its economic architecture is lopsided in a way the consolidated statements deliberately obscure: 28.5% of the capital produces 41% of the profit, and the other 71.5% of the capital earns roughly 13%. The correct analytical posture for the rest of this memo is to treat Bradesco Seguros as the asset and Banco Bradesco as the increasingly commoditized, share-losing distribution and credit business attached to it — and then to ask what price the market is charging for each.

3. Industry Dynamics

3.1 The pool, sized

FACT. Brazilian SFN credit stock reached R$7.356 trillion in June 2026, +9.7% y/y, equal to 55.8% of GDP. Households +10.8% y/y, corporates +7.9%. The Brazilian banking system earned a record R$255bn of net income in 2025 on a system ROE of 16.76% — the highest since 2021. The five bancões together earned ~R$107bn. Over the trailing 16 years Brazilian banks have averaged a 17.4% ROE against 11.02% for US financial institutions. [BCB Nota de Política Monetária e Operações de Crédito, June 2026; SpaceMoney/Economatica 2026; Folhapress via Diário do Comércio, 2026-05-04]

(The R$255bn “system” figure and the R$107bn “big five” figure use different perimeters — the former is the full SFN including BNDES, cooperatives, non-bank credit institutions and insurance-linked results. Do not compute a big-five profit share from the pair.)

By any static measure Brazil is one of the most attractive banking markets in the world. Credit penetration at 55.8% of GDP leaves genuine runway; barriers to entry are real (BACEN authorization, capital rules, deposit-insurance and settlement access); and concentration remains high — the five largest institutions held 63.4% of deposits in September 2025 [FY2025 20-F Section 4.B.60.01].

3.2 Concentration is falling — slowly, and to the wrong people

Individual 2024 credit shares: Caixa 19.7%, Banco do Brasil 16.5%, Itaú 11.2%, Bradesco 10.4%. Five-bank asset shares (March 2026): Itaú 15.0%, BB 14.2%, Caixa 12.2%, Bradesco 10.3%, Santander Brasil 7.2% — top five 59.3%, top ten 72.2%. [Poder360 reporting BACEN REF 2H2024; RankingsLatAm 2026]

INTERPRETATION. The oligopoly is not collapsing; it is being taxed. RC4 has lost roughly 1.5–3 points over five years and the leakage has gone to credit cooperatives (+0.9pp) and non-bank institutions (+0.9pp) at least as much as to the headline neobanks. But aggregate concentration masks a violent redistribution within the incumbent group and within product lines — which is precisely where the Greenwald test bites (Section 4).

3.3 PIX — the largest single act of profit-pool expropriation in modern banking

FACT. In the first five months of 2026 PIX moved R$16 trillion across 36.3 billion transactions, +26% in value and +19.3% in count y/y — nearly 3,000 operations per second. June 2026 alone: 7.69bn operations, ~R$3.59tn. In 2025 PIX took 42% of online purchase value, edging past credit cards at 41%; EBANX projects 45% in 2026 and 50% by 2028. [BCB PIX statistics; PaymentsCMI; EBANX via PYMNTS]

INTERPRETATION. PIX is not killing the payments pie — it is socializing it. The rail is central-bank-owned, free to individuals, and instant. What it destroys is precisely the set of rents that funded incumbent branch networks: transfer fees, float on demand deposits, and debit interchange. The direct evidence sits in Bradesco’s own filings: demand-deposit market share 11.0% → 6.6% in three years, and the two most PIX-exposed fee lines shrinking ~7% y/y (Section 2.3).

And the agenda is not finished:

PIX Garantido is the one that matters, and it is the highest-impact unquantified regulatory risk in this thesis. Brazilian consumer credit is dominated by parcelamento on credit cards. A central-bank-run, free-rail installment product would attack the fattest remaining consumer pool for a bank like Bradesco. OPEN QUESTION: BCB’s committed launch window and pricing model, and whether it permits issuer-set credit spreads.

3.4 Open Finance — switching costs dismantled by decree

FACT. Brazil reached 128 million active consents by January 2026 — the world’s largest Open Finance ecosystem, ahead of 78 other regulating nations. The trajectory: 43m (Jan-2024) → 62m (Jan-2025) → 128m (Jan-2026), a doubling in the most recent year. Credit portability — one-click migration of a loan between institutions — launched February 2026. [TI Inside, 2026-01-22; FEBRABAN; BCB]

INTERPRETATION, honestly hedged. Consent counts are a weak proxy for mobility: a consent is granted whenever a challenger app asks to read an incumbent’s data, so the metric measures challenger acquisition activity, not consumer switching. The defensible read is that Open Finance has collapsed the information asymmetry — a challenger can now underwrite using the incumbent’s own transaction history — while consumer inertia still protects the primary relationship. Credit portability is the genuine step-change, because it converts an informational right into an executable transfer, and it lands on the incumbent’s stickiest, highest-duration assets. OPEN QUESTION: what portability volumes have actually executed since February 2026? No reliable public series located.

In Greenwald’s taxonomy, customer captivity is one of only three genuine advantage types. In Brazil, the regulator has made destroying it an explicit policy objective. Any thesis resting on Bradesco’s customers being locked in rests on an asset the state is actively expropriating.

3.5 Crédito do Trabalhador — the cleanest natural experiment on Brazilian banking moats in a decade

FACT. The federal programme opened private-sector payroll-deducted lending to a national, portable, digitally-originated market via the eSocial/CTPS infrastructure, replacing a fragmented employer-by-employer model. Concessions rose 183.6% in 2025 to R$54.5bn across 17 million loans; the total private-payroll portfolio passed R$100bn in March 2026, +142% y/y. [Ministério do Trabalho e Emprego, Jan-2026; Viva/Seu Dinheiro 2026]

A new, large, fast-growing, structurally good-credit pool opened up on a portable digital rail. If incumbency conferred a real advantage — distribution, employer relationships, data, brand — Bradesco should have won share. Instead its share halved in twelve months, 12.0% → 6.6%, while Itaú’s book went from ~R$12bn to R$19.5bn. Bradesco grew its own book 43% in a market that grew 142%. Growing 43% while losing 45% of your share is the textbook signature of no barrier to entry.

And the pool is already showing stress: private-sector payroll delinquency has reached 7.9% — extraordinary for a payroll-deducted product, and evidence the land-grab was underwritten loosely across the industry. This is a Marathon capital-cycle signature in miniature: capital rushed into a newly-opened “safe” pool, competed away the return, and is now discovering the loss content — inside eighteen months.

3.6 Tax — two opposite-signed changes the market conflates

FACT — JCP. Juros sobre capital próprio is booked as a financial expense that reduces the IRPJ and CSLL base — for decades the mechanism behind Brazilian banks’ below-statutory effective tax rates. From 1 January 2026: withholding tax rose from 15% to 17.5% (Lei Complementar 224/25), and the calculation base was restricted so current-year profits can no longer support current-year JCP. The base had already been narrowed by Lei 14.789/2023 effective 2024. The corporate-level deductibility survives — this is primarily a shareholder-level tax increase plus a base narrowing, not the removal of the corporate shield.

FACT — CSLL, restructured. Under MP 1.303/2025 and LC 224 of 26-Dec-2025, effective April 2026: banks 20% (down from 22%); insurers, brokers, card administrators and credit unions 15%; payment institutions, exchanges and settlement entities 12% (Apr-2026–Dec-2027), rising to 15% from January 2028; SCFI and capitalization companies 17.5% until end-2027, then 20%.

INTERPRETATION — these run in opposite directions and the second is badly under-appreciated. The JCP restriction is a system-wide, permanent hit to Brazilian bank net profitability and to the tax-advantaged character of distributions; Itaú has already guided its 2026 effective tax rate to 29.5–32.5%, far above JCP-shielded historical levels. But the CSLL restructuring is explicitly pro-bank: banks’ rate falls while the fintech attacker cohort — payment institutions, credit fintechs — is taxed up toward bank rates by 2028. The tax arbitrage that let payment institutions undercut banks on an after-tax basis is being narrowed by statute. This runs directly against the “the regulator is uniformly anti-incumbent” narrative. Net: the state is taking rents from banks on the distribution side while partially levelling the operating playing field — but JCP is a hard, immediate, quantifiable EPS hit while the CSLL levelling is a slow, second-order relative benefit.

3.7 CMN Resolution 4,966 — expected-loss provisioning, effective 1 January 2025

FACT. Resolução CMN 4.966/21 replaced Resolução 2.682/1999’s incurred-loss model — in force since 1999 — with an IFRS-9-equivalent expected-loss model. It affects an estimated ~80% of a bank’s balance sheet: higher provisions, more instruments at fair value, Stage 1/2/3 classification, and revenue-recognition changes on impaired assets. Banco do Brasil’s provisions jumped 104% to R$15.9bn in Q2-2025, explicitly attributed in part to 4,966. [KPMG Brasil; Deloitte; MarketScreener]

Three consequences for reading any Brazilian bank’s numbers: (1) comparability is broken across 1-Jan-2025 — any five-year provision/NPL/coverage trend line through that date is contaminated; (2) provisions are front-loaded and pro-cyclical, so arriving alongside a record household-delinquency cycle, 4,966 amplifies the reported shock; (3) it removes managerial discretion to smooth — a quality-of-earnings improvement and a reported-earnings volatility increase simultaneously. For a bank mid-turnaround, it means the recovery is harder to flatter.

3.8 The credit cycle — worse than the mid-2026 consensus framing

FACT — the worst household-credit configuration in the available series:

[BCB; CNC Peic; Senado Notícias May-2026; DIEESE Síntese Especial nº23; FGV/IBRE Boletim Macro nº176]

INTERPRETATION — and this is a direct correction to a framing that has become common in mid-2026 commentary. Analysis focused on Itaú routinely describes Brazilian credit quality as “pristine” and “benign.” That is true of Itaú’s book (1.9% 90-day NPL) and false of the system: NPLs are at all-time highs in every available series, unsecured personal credit defaults at 14.2%, and even payroll-deducted private lending — structurally the safest consumer product in Brazil — runs at 7.9%. The correct reading is a severe bifurcation: elite incumbent books (affluent, secured, corporate) are holding, while the mass-market, unsecured, low-income pool is in the worst shape on record. That pool is where the neobanks sit — and where Bradesco’s legacy mass-retail base sits.

The configuration is also unusual in a way that matters for the forward path: record-low unemployment (5.4–5.8%) coexisting with record household delinquency. That says the distress is not a labour-market failure but a debt-service failure at a 14–15% policy rate. The fix is a rate cut, not a jobs recovery — which makes the easing cycle that began in March 2026 the single most important variable for 2027 credit quality, and makes the October election, which determines whether Copom is permitted to keep cutting, the single most important political one.

And it cuts favourably for Bradesco in one specific respect. This is the first cycle that tests underwriting rather than cost structure — the one dimension on which a 90-year-old incumbent with a full credit-bureau history and a secured/affluent mix out-competes an eight-year-old app. Nubank’s 90+ NPL is 6.5%, above the system’s 4.7%, and its shares are down roughly a third in 2026 on credit-cycle downgrades.

3.9 Spreads at record widths — which is not the good news it looks like

FACT. The Brazilian banking spread exceeded 15 percentage points in 2026 — 15.17pp in March, the highest since the ICC series began in 2013; an alternative BCB calculation put it at 21.84pp in April, the highest since March 2017. Delinquency accounted for 33.45% of the average spread decomposition over the three years to 2024. [Folhapress via Diário do Comércio, 2026-05-04; FEBRABAN spread study, Oct-2025]

INTERPRETATION. Record spreads and record NPLs are the same fact viewed twice. Roughly a third of the Brazilian spread is expected-loss compensation; when loss expectations spike, so does the spread. The spread is not evidence of pricing power — it is evidence of risk. The corollary is decisive for valuation: the system’s 16.76% ROE is being earned on a spread that embeds peak loss assumptions. If losses normalize, spreads compress with them, and “record profitability” does not simply persist.

3.10 The capital cycle (Marathon)

Capital is being withdrawn from physical banking. 7,252 Brazilian bank branches closed over ten years; the count peaked above 21,000 in 2017 and fell to 17,215 by 2022. Bradesco alone closed 342 branches and cut 2,564 staff in the most recent measured period; Itaú closed 227. 82% of banking transactions now occur via mobile or internet. This is textbook favourable capital-cycle behaviour on the incumbent side — supply of physical distribution capacity is contracting and the industry is not adding into a downturn.

And capital is no longer flooding into digital credit. Applications to open new fintechs at BACEN have fallen from ~15 per month to ~2 per month following stricter requirements; Resolution BCB 494/2025 pulled the regularization deadline for unauthorized payment institutions forward from December 2029 to May 2026; BACEN reports “accelerated movements of mergers, capitalizations and voluntary exits.” An 87% collapse in licensing applications is a supply-side capitulation signal, and Marathon’s framework says returns recover where supply exits.

The counter-argument is important and I do not think it is weaker than the argument. The survivors of a consolidation are stronger, not weaker. Nubank is deploying R$45bn in Brazil in 2026 and calls it an “inflection year”; C6 earns a 43% ROE with JPMorgan owning 46%; Mercado Pago’s Brazilian credit book grew 90%; PicPay raised $434m on Nasdaq and clears 11% of all PIX transactions. Capital is not leaving Brazilian digital financial services — it is concentrating into five or six well-capitalized winners with permanent balance sheets. For Bradesco that is arguably a worse competitive environment than a hundred underfunded startups, because these firms no longer need to raise capital to compete.

3.11 Where the profit pool has actually moved

3.12 The insurance sub-industry — where the profit actually is

FACT. The CNseg full-market pool (insurance + pension + capitalization + net health reimbursements) was R$764.5bn in 2025, projected at ~5.8% of GDP in 2026. Supplementary health alone was R$349.4bn (+10.8%). Excluding VGBL — a savings product wearing an insurance wrapper whose flows swing with the Selic — the underwriting market (damages, liability, life, health) is growing 6–11% nominal.

Why Brazilian insurance ROEs are structurally high, in order of importance:

1. Float invested at 14–15% risk-free. The dominant and most under-appreciated factor. A Brazilian insurer’s technical reserves earn the Selic; at 14.25% the investment return on float alone can generate a mid-teens ROE before a single point of underwriting margin. This is a high-nominal-rate EM artefact, not a competitive advantage — and it mean-reverts as Copom eases.
2. Bancassurance distribution at near-zero marginal cost. Selling insurance, pension and capitalization through an existing branch/app/relationship base has a customer-acquisition cost approaching zero versus broker distribution. This is a genuine Greenwald economies-of-scale-plus-captivity advantage — and it is the one place where Bradesco’s much-maligned branch network is an asset.
3. Licensing and capital barriers. SUSEP/ANS authorization, solvency capital, actuarial reserving and ANS network-adequacy rules are real. Critically: there is no PIX for insurance, no Open Finance-equivalent portability regime with teeth, and no Drex analogue. The regulator has not decided to compress insurance rents the way it decided to compress banking rents.
4. Health: contractual inflation pass-through. The VCMH medical-cost index triggered a 15.1% repricing for 2026 on corporate health contracts. A business that contractually reprices at 15% in a 4–5% CPI economy has genuine pricing power — provided claims inflation is passed through faster than incurred.
5. Concentration. Top-5 groups hold 50–55% of risk products; the top-10 health operators hold 42% of beneficiaries and 50% of revenue.

The risk to that thesis is health-cost inflation. Brazil has 53.08m supplementary-health beneficiaries across 668 active operators; operators collected ~R$340bn in 2025 and paid ~R$275bn to providers — an 82% loss ratio against a ~70% break-even. The Brazilian health-insurance industry is, in aggregate, underwriting at a loss and surviving on float income — the same Selic dependence as point (1), doubled. Bradesco’s creation of Bradsaúde (vertically integrating Saúde + Odontoprev + Atlântica Hospitais, >13m beneficiaries) is the rational structural response: owning provider capacity is the only durable answer to VCMH at 15% while ANS caps individual repricing.

3.13 The October 2026 election

First round 4 October 2026; runoff 25 October. June polling: Lula 41–43%, Flávio Bolsonaro 28–34%; a BTG/Nexus simulated runoff in late March had them tied 46–46. The fiscal deficit breached 7% of GDP in 2025.

INTERPRETATION. For a Brazilian bank ADR the election is the dominant near-term driver of the multiple, transmitting through three channels: the BRL (already appreciated ~10% from the 2025 average, which looks like partial pricing of a market-friendly or continuity-with-discipline outcome); the Selic path (fiscal loosening forces Copom to keep rates high, which is exactly what generates the delinquency cycle); and direct sector policy risk — banking spreads at 20-year highs alongside record household distress in an election year is a standing invitation to rate caps, mandatory-renegotiation mandates, or a Desenrola 3.0. The asymmetry is unattractive: a market-friendly result is largely already in the currency, while a fiscally-loose result reopens both the rate and the intervention risk.

3.14 Verdict — is Brazilian banking a structurally good or bad industry?

It is a structurally good industry that is being administratively degraded — and it is now a materially better industry for insurers, for the lowest-cost digital operator, and for the scale wholesale/affluent franchise than it is for a legacy mass-retail branch bank. Bradesco is the last of those four.

The case for is strong and should not be dismissed: a 17.4% sixteen-year average ROE, a record R$255bn system profit pool, spreads at their widest since 2013, credit at only 55.8% of GDP, a 63.4% top-five deposit concentration, and a regulator that has just made entry harder (licensing applications down 87%) while taxing the challenger cohort up toward bank rates by 2028. On a Greenwald reading, the industry passes the barriers-to-entry test.

The case against is that the state has decided incumbent rents were too wide and is dismantling the specific mechanisms that produced them: PIX destroyed transfer fees, float and debit interchange and is now moving on recurring billing and installment credit; Open Finance reached 128m consents and added executable credit portability in February 2026; Crédito do Trabalhador proved incumbency confers no advantage on a portable digital rail; and JCP removes a permanent structural tax shield from every Brazilian bank’s EPS.

Two swing variables point in opposite directions. The credit cycle is at its worst point on record, arriving simultaneously with Resolution 4,966’s expected-loss regime which front-loads recognition and removes the ability to smooth — a genuine, quantified, present danger, worse than the mid-2026 consensus framing. But it also runs against the challengers, testing underwriting rather than cost. And the Selic easing cycle that began in March 2026 is the mechanism that fixes household debt service at record-low unemployment — while simultaneously deflating the float income that produces both the record bank spread and the 21.9% insurance ROAE. Record profitability at a 14.25% Selic is not a run-rate; it is a rate-cycle artefact.

Bottom line: a good industry, getting slowly worse for this specific participant, currently earning peak-cycle spreads on peak-cycle losses, with the durable franchise concentrated in the insurance subsidiary rather than the bank.

4. Competitive Position

Greenwald recognises exactly three genuine advantages: supply-side (cost), demand-side (customer captivity), and economies of scale reinforced by captivity. Everything else is a regulatory artefact or a cost dressed as a moat. Each candidate below is tested against a financial metric that would deteriorate if the advantage were absent.

4.1 The physical network — the historic “moat.” Its own management has proved it was a cost.

The killer test. Between FY2023 and FY2025 Bradesco removed 26% of its branches and 36% of its own ATMs. Over the same window its loan market share went 10.2% → 10.1% → 10.4%, and its share among private institutions went 17.8% → 17.6% → 17.7% [FY2025 20-F Section 4.B.60, source BACEN].

INTERPRETATION: you can remove more than half of an asset over a decade with no measurable loss of the market share it supposedly defended. By Greenwald’s definition that asset was never a barrier to entry — it was a fixed cost the competitive equilibrium no longer supports. The correct reading of the branch closures is not “management is executing”; it is “management is liquidating a stranded asset, ten years late.”

The counter-evidence is real and it is the strategic tension at the heart of the group.

FACT: 93% of pension-plan sales — the product line sitting on R$385bn of the group’s R$455bn of technical provisions — are still originated in branches, and that dependence has increased every year. The branch network is a value-destroying cost in banking and the sole distribution moat in the group’s best business. Cutting it too fast damages the insurance franchise; not cutting it keeps the bank’s ROE below its cost of capital. Management has not, in any public disclosure, quantified this trade-off. It is the most important undisclosed number in the company.

(On Bradesco Expresso: the 39,335 correspondent points have been flat for five years and are not the post-office franchise they once were — Bradesco held the Banco Postal concession from 2002 and lost it at auction in May 2011 to Banco do Brasil, effective 1-Jan-2012. What remains is supermarket/pharmacy correspondents: a rented, non-exclusive, replicable channel available to every competitor on identical terms.)

Verdict on the network: not a moat. A legacy cost base in managed decline, with one genuine residual function — insurance distribution.

4.2 Customer captivity — structurally destroyed by the regulator

FACT: the total client base is smaller today (74.3m) than in 2022 (77.1m), and checking-account holders peaked at 38.2m in March 2025 and have since declined. Over the identical window Nubank added roughly 50 million Brazilian customers, overtook Bradesco in the BACEN customer ranking in January 2026 (112.0m vs 110.5m at Q4-2025; 114.7m vs a declining Bradesco at Q1-2026), and now processes 276.5m monthly transactions against Bradesco’s 103.9m — Bradesco runs ~38% of Nubank’s throughput on a near-identical headline customer count.

INTERPRETATION: that gap is a dormancy problem masked by a large headline number. Bradesco discloses no average client age, no principality/primary-bank rate, and no monthly-active-client figure. Nubank discloses an 83% monthly activity rate. The absence of a principality disclosure from Bradesco is itself informative.

Who is actually captive?

Verdict: switching costs are theoretical, not real, for the mass-retail client. Real friction remains only where a contract binds — payroll/INSS arrangements, mortgages, corporate cash management, and above all insurance/pension policies with surrender penalties and tax lock-ins. That is a much smaller book than the 74m-client headline implies, and its centre of gravity is inside the insurance company.

4.3 Cost — Bradesco has a cost disadvantage, and it is quantified

This is the cleanest refutation of a scale moat. If scale produced a cost advantage, the #2 private bank in Brazil would not be 11–14 points worse on efficiency than the #1 at comparable scale.

Derived cost-to-serve. FY2025 personnel (R$27,991m) + administrative (R$23,199m) = R$51,190m of opex across a 74.3m client base = R$689 per client per year (~US$10.4/month). Excluding the insurance segment’s R$5,139m, the bank spends ~R$620/client/year (~US$9.4/month). Nubank’s disclosed cost-to-serve is ~US$0.80 per active customer per month — roughly a 12× gap. Bradesco’s revenue per client (~R$1,848/year ≈ US$28/month) is only ~1.8× Nubank’s ARPAC of ~US$16, so the revenue premium does not come close to covering the cost gap.

And the cost base is rigid. Branches fell 26% across 2023–2025 while headcount fell only 4.8%; 1Q26 personnel expense still rose 4.7% y/y against a 5.68% collective-bargaining increase. Structural payroll was actually flat (−0.1% y/y) — the entire personnel increase is profit-sharing, up 35.6%. Technology spend (data processing and communications) is +29.8% y/y and is the explicit strategy, so branch-closure savings are being redeployed, not banked. Most damningly: the risk-adjusted cost-to-income ratio ROSE to 76.9% from 76.1% a year ago — once you charge the P&L for credit cost, the entire efficiency gain disappears.

On the 40%-by-2028 target: the 12-month ratio has moved 52.2% → 50.0% in two years. Reaching 40% by 2028 requires roughly 5× the annual rate of improvement achieved so far, against a unionised cost base indexed at 5.68% and rising technology spend. INTERPRETATION: the target is not credible on the current run-rate and should be treated as an aspiration, not a plan. That said, the size of the prize is real: on ~R$118bn of FY2025 revenue, closing half the gap to Itaú would be worth roughly R$6bn pre-tax — larger than the entire 2023-to-2025 recurring earnings recovery. That is simultaneously the bear case (they have not closed it in a decade) and the bull case (the option is enormous and, on today’s multiple, unpriced).

Verdict: the opposite of a cost advantage. This alone disqualifies the scale-economies moat under Greenwald.

4.4 The market-share-stability test — failed in banking, passed in insurance

Greenwald’s central diagnostic: a genuine barrier to entry produces stable market shares over time. Bradesco publishes BACEN/SUSEP-sourced shares in its own 20-F, which makes the test unusually clean.

[FY2025 and FY2024 Forms 20-F Section 4.B.60; sources BACEN, SUSEP/ANS/Fenaprevi, ANBIMA, INSS/Dataprev. 2025 base dates Sep/Nov-2025.]

Reading it honestly, in four parts:

1. Where Bradesco buys share with its balance sheet, share is stable. Total loans and private-bank loans have not moved in three years. That is real and should not be dismissed — but note what it is: a bank can hold lending share simply by pricing and accepting risk. Bradesco’s NPL-90 is 4.2%; Itaú’s is 1.9%. Share held at the cost of credit quality is not evidence of a moat; it is evidence of a price-taker.
2. Where the relationship lives, share is falling fast. Demand deposits −40% relative in three years. INSS benefit-payment share −4.4pp. INSS payroll book −2.6% y/y in absolute terms. Auto-consortia −5.4pp; agri-consortia −7.9pp. These are the metrics that measure primacy — being the client’s main bank — and every one is deteriorating. Greenwald would call this an unambiguous failure of the stability test.
3. The one-year private-payroll collapse (12.0% → 6.6%) is a regime change, not purely a Bradesco failure — it reflects the Crédito do Trabalhador platform re-plumbing the market (Section 3.5). But that is the point: when the market was re-plumbed onto a portable rail, Bradesco’s incumbency bought it nothing.
4. Insurance passes cleanly. 22.7 → 22.9 → 22.8% across 2023–2025, and 25.5% back in 2015 — materially the flattest series in the table. Stable share + a 21.9% ROE = a real barrier, under Greenwald’s own test.

4.5 Head-to-head — Bradesco is the weakest large private bank in Brazil

The read. Against Itaú — same country, same regulator, same cycle, a larger balance sheet — Bradesco earns 9 points less ROE, runs 11–14 points worse efficiency, and carries 2.2× the NPL rate. That is not a moat differential; it is an execution and cost-structure differential, and it has persisted for a decade.

And it is a comparatively recent divergence, which matters. In Deutsche Bank’s September-2011 Global Banking Industry Primer (third-party sell-side research, 2011 vintage — used here for structural framing, not as current data), Bradesco screened as the most capital-resilient of Brazil’s big three under a two-year severe-recession stress test (−60% of tangible equity versus Banco do Brasil’s −82%) and was named a DB preferred global name alongside Itaú — “lower beta stocks with strong capital, above average pre-provision profitability and superior asset quality metrics.” Fifteen years ago Bradesco was regarded as Itaú’s peer, not its laggard. The ~9-point ROE gap and ~11-point efficiency gap are the product of the last decade, not of structural inferiority — which is precisely why the bull case that they are fixable deserves a hearing.

4.6 The insurance franchise — the only candidate that passes

Grupo Bradesco Seguros is the largest insurer in Brazil at 22.8% of premiums + pension contributions + capitalization income, with R$118.5bn of FY2025 revenue, R$455.2bn of technical provisions and R$482.8bn of financial assets. Its five largest competitors combined (SulAmérica, Porto, BB Seguridade, HDI, Tokio) held ~35.1% as of September 2025.

The advantage is a genuine economies-of-scale + captivity combination, and unusually it is multi-sourced:

1. Regulatory capital as a barrier. SUSEP/ANS solvency requirements mean the market cannot be entered cheaply. Bradesco’s regulated insurance entities carry Adjusted Shareholders’ Equity of R$16.0bn against Minimum Capital Required of R$13.7bn.
2. Captive, hard-to-replicate distribution. 93% of pension sales and 35.9% of insurance sales through the bank’s own branches. A digital entrant cannot rent that channel.
3. Health-network scale and vertical integration. Bradesco Saúde is the flagship (R$41bn revenue, 3.9m beneficiaries), and the group has verticalised into providers — Atlântica Hospitais (3,600 beds), primary-care and oncology clinics, and a 25% stake in Grupo Fleury. Where medical-cost inflation is the binding constraint, owning capacity is a real supply-side cost advantage.
4. Actuarial and claims data at national scale, plus the largest dental book in Brazil (Odontoprev, >9.4m beneficiaries).

Financial proof it converts: insurance ROAE 22.4% FY2025 / 21.6% 1Q26 on ~28% of group equity delivering ~41% of group profit, combined ratio 82.9%, technical provisions +9.9% y/y. This is the metric that would deteriorate without the moat, and it hasn’t.

Two hard caveats, both material. First, the float dependency: 36% of 1Q26 insurance operating income was financial results at a 14.75% Selic, against a house forecast of 9.50% by end-2027. A meaningful slice of the 21–22% ROE is rate carry. Second, Bradesco Seguros is not uniquely good — it is competitively matched. Porto Seguro (PSSA3) delivered a 22.7% ROE in FY2025, its highest ever, with an insurance-vertical annualized ROE of 34% in 1Q26 — at a fraction of Bradesco’s scale. In Jan–Apr 2026 SUSEP premium rankings reportedly showed Porto (R$7.9bn) narrowly ahead of Bradesco Seguros (R$7.78bn) in property/liability lines. Bradesco is bigger, not better: this is a scale-and-distribution advantage that produces size, not superior unit returns. It is a moat (it protects the 22% ROE against entrants), not a superiority (it does not produce a 30% ROE).

4.7 Verdict — Competitive Position

No durable competitive advantage in banking. A narrow but genuine advantage in insurance and health. Stated in Greenwald’s terms:

• Supply/cost advantage: absent, and inverted. A 49–50% efficiency ratio against Itaú’s 36–39% and Nubank’s high-teens, at comparable or larger scale, is a cost disadvantage of 11–14 points. Scale that does not lower unit costs is not economies of scale; it is a fixed-cost trap. Derived cost-to-serve of ~US$9–10/client/month against Nubank’s ~US$0.80 makes the point unarguable.
• Demand/customer captivity: destroyed by the regulator, not by competitors. PIX removed the payments float and habit; Open Finance (128m consents, executable credit portability since Feb-2026) removed the friction. Bradesco’s own disclosures show demand-deposit share down 40% relative in three years, checking-account holders declining, total clients below 2022, INSS benefit-payment share down 4.4pp, and the most PIX-exposed fee lines shrinking 7% y/y.
• Economies of scale + captivity — the only durable combination: fails in banking, passes in insurance. In banking both legs are broken. In insurance both hold: regulatory capital as a genuine entry barrier, 22.8% national share stable across three years, captive branch distribution, owned health capacity, and a 21–22% ROE as financial proof.
• The branch network — the historical “moat” — is now definitively a liability with one residual function. The dispositive evidence is Bradesco’s own: 58% of branches removed since 2015 with no loss of loan market share. Its remaining economic function is distributing insurance and pensions to a cohort that is aging — which means the group is being forced to run a bank-shaped cost base to protect an insurance-shaped profit stream.

What would change this verdict. Only two things. (1) The 12-month efficiency ratio falling below ~44% by end-2027 while NPL-90 converges toward 3% — which would evidence a genuine emerging cost advantage rather than a repair; the current 2.2pp-in-two-years run-rate makes this unlikely. (2) Evidence that the affluent rebuild (Principal/Prime, 4.7m clients targeted for 2026) is taking share from XP/BTG/Itaú Personnalité rather than internally re-labelling existing clients. Bradesco has disclosed 3.1m “upgrades” and no net-new-affluent-client figure — a conspicuous omission.

5. Growth History and Forward Opportunities

5.1 The customer base is not growing — and that reframes everything management calls growth

FACT: the client base peaked in 2022 at 77.1m and is 74.3m today — 3.6% lower. Checking-account holders peaked at 38.2m in March 2025 and were 37.9m at March 2026.

INTERPRETATION — this is the single hardest fact to reconcile with the “transformation is working” narrative. Every growth metric management leads with — 3.1m affluent “upgrades,” 19m → 28m fully-digital clients, MSME share, NPS from 56 to 74 — is a re-segmentation or re-channelling of a customer base that is not growing. The metrics that genuinely grow are the monetisation metrics: AuM +16.4%, custody +15.4%, card turnover +9%, high-income card revenue +24%. That is a share-of-wallet story on a shrinking customer footprint, in a country where Nubank alone has 113m Brazilian customers. It is not worthless — deepening wallet share across 74m clients is a legitimate strategy — but it is a fundamentally different, and lower-ceiling, proposition than customer growth, and management does not frame it that way.

(Consortia deserve a specific flag: fee income grew +19.5% y/y and management calls it a lever, but outstanding quotas fell 2.4% q/q in 1Q26 — auto quotas declined, only real-estate quotas grew. The fee growth is ticket size and mix, not units.)

5.2 Loan growth — the fastest-growing lines are the deteriorating lines

The test for this section is simple: is Bradesco growing where returns are high, or where they are being competed away? Rank the book by growth rate and the answer falls out of Bradesco’s own disclosure without any third-party data.

INTERPRETATION. On the very call at which those growth rates were presented, the CEO said agribusiness is “still not back to normal,” attributed the entire MSME Stage-2 build to the government-guaranteed lines growing +81% y/y — “because the grace period is over, and so they are not paying” — and took a discretionary provision against a specific large-corporate case. Cost of risk rose to 3.5% (retail 5.5%) against a guided full-year 3.3%. The growth and the credit deterioration are in the same lines, and management has said so on the record.

Three inorganic distortions must be stripped before the growth optics are believed:

Bradesco discloses no organic/inorganic bridge. OPEN QUESTION: what is rural growth ex-Banco John Deere? Without it, the +24.0% figure is uninterpretable.

5.3 The growth is levered, not earned

ROE decomposes as ROA × leverage. From FY2020 to FY2025, ROA moved 0.99% → 1.02% (+3%) while assets/equity moved 11.4× → 13.4× (+18%). Roughly 83% of the five-year ROE improvement is balance-sheet expansion, not asset productivity.

From the FY2023 trough the split is more flattering — roughly 73% asset-return recovery, 27% leverage — and that recovery is real and should be credited. But asked directly by a sell-side analyst whether the sequential ROAE gain was mostly leverage, the CEO did not dispute it: “leverage is part of our business. We remain comfortable with our level of leverage” [Noronha, 1Q26 call, 2026-05-07]. Eleven weeks later the bank raised equity at book value.

Five-year IFRS loan growth works out to roughly 2.9% per annum in real terms, with an outright contraction in 2023. That is not a growth franchise. It is a large balance sheet being re-mixed.

5.4 Forward opportunities, assessed sceptically

5.5 Verdict — Growth

Low-quality growth, with one high-quality exception that management is guiding to decelerate.

The fastest-growing lines are the deteriorating lines, and management has said so on the record. Roughly 83% of the five-year ROE improvement is leverage rather than asset productivity. The customer base is smaller than it was in 2022, so every headline growth metric is a re-segmentation of a shrinking footprint — legitimate as a wallet-share strategy, but structurally capped. Real loan growth is ~2.9% per annum over five years with an outright contraction in 2023. And the one engine that genuinely earns its cost of capital — health insurance, which supplied 100% of Bradesco Seguros’ FY2025 profit growth while life and pension shrank — is 16% of group revenue and is guided from +20.4% down to +6–8%.

What would change this verdict, in order of weight: (i) a disclosed net-new affluent client count showing Principal/Prime is taking share rather than re-badging; (ii) MSME growth sustaining without the FGI/FGO guarantee crutch and with the NPL flat rather than rising; (iii) evidence in the 5-August print that the fast-growing 2025–26 vintages — vehicle, rural, FGI/FGO — are not rolling into Stage 2/3 on the schedule the Resolution 4,966 cohort tables imply. Absent those, this is a re-mix, not a growth story.

6. Financial Quality

6.1 The multi-year record — the trough was worse than the recurring numbers admit

Recurring net income and ROAE, management basis (BR GAAP):

(FY2020–FY2022 recurring figures are secondary-sourced from the Brazilian financial press and carry lower confidence than FY2023 onward, which are primary or near-primary. Flagged in Open Questions.)

The audited IFRS basis tells a harsher story:

*Annualized on IFRS book net income; not comparable to the 15.8% recurring ROAE headline.

INTERPRETATION. On the audited basis, ROE bottomed at 8.7% in FY2023 — not the 10.0% management reported — and FY2024 was only 10.3%. That is four consecutive years (2022–2025) in which the bank earned at or below any plausible Brazilian cost of equity. Peak-to-trough, IFRS net income fell 39%. This was not a soft patch; it was a capital-destroying period.

And the recovery is real. Recurring net income by quarter: 1Q25 R$5,864m → 2Q25 R$5,972m → 3Q25 R$6,204m → 4Q25 R$6,516m → 1Q26 R$6,811m. Quarterly ROAE 14.4% → 14.6% → 14.7% → 15.2% → 15.8%. A genuinely monotonic nine-quarter sequence, driven by revenue rather than provision release. That deserves to be said plainly before the criticisms that follow.

6.2 Credit quality — the deterioration is in SMEs, rural and wholesale, and coverage is falling

Peak 90-day NPL in this cycle was 5.9% in 2Q23, from 3.5% in 2Q22.

FACT — coverage has fallen 22 percentage points in twelve months (183.1% → 161.0%) while the provision balance was essentially flat (R$57.8bn → R$58.2bn) against a portfolio that grew 8.4%.

INTERPRETATION — this is the single biggest quality-of-earnings issue in the 1Q26 print, and it genuinely cuts both ways. The bull reading is mix: Stage 3 fell 8.0% → 7.1%, the restructured book is down 20% y/y, problem assets are down R$7.6bn (−28%), and the secured share is up 3.8pp — a lower-loss-severity book genuinely needs less coverage, and new-Stage-3 coverage actually rose to 118% from 103% in 4Q25, which argues against a smoothing release. The bear reading is that Bradesco is releasing balance-sheet cushion into the P&L at precisely the moment the forward indicators turn: MSME NPLs +60bp y/y, cost of risk +50bp q/q to 3.5%, retail cost of risk 5.5%, and management itself flagging “deterioration… particularly in agribusiness” and the “legacy rural credit portfolio” — into which rural lending grew +17.5% (individuals) and +24.0% (companies) y/y. My read: roughly 60% mix, 40% cushion release. Another 5pp coverage drop in the 5-August print with MSME NPLs still rising would be a red flag.

Where Bradesco sits in the cost-of-risk band matters more than almost any other number here. The peer triangulation: Itaú at 2.6%, Santander Brasil at 4.14%, against a well-informed sell-side definition of a healthy Brazilian retail bank as “~20% ROE with cost of risk below 4%.” Bradesco at 3.5% and rising sits in the middle of that band, closer to Santander than to Itaú.

6.3 Costs and the footprint reduction

The cost story is real but flattered, and the honest version is the 12-month 49.2% rather than the seasonally-assisted quarterly 46.9% (4Q always carries year-end admin and marketing — advertising fell 32% q/q and outsourced services 25%). Structural payroll is flat; the entire personnel increase is profit-sharing. Technology spend is up 29.8% and is being redeployed, not banked. And the risk-adjusted ratio moved the wrong way.

6.4 Tax — the least trustworthy line in the P&L, and it just flipped hard

Statutory rate for Brazilian banks: IRPJ 25% + CSLL 20% = 45% (CSLL cut from 22% effective April 2026).

INTERPRETATION. On the IFRS basis Bradesco booked net tax benefits in FY2023, FY2024 and FY2025 — a −13.8% effective rate on R$21.0bn of FY2025 pre-tax income. That is the combined effect of the JCP deduction, deferred-tax-asset activation, and the hedge of overseas investments, which moves the tax line inversely to the BRL. In 1Q26 the rate flipped to +33.2% — a swing worth roughly R$3bn annualized against the FY2025 run-rate. Part is the PTI charge; the recurring rate held at ~20%. But the direction is unambiguous: the FX hedge that generated benefits while the BRL was weak reverses when it strengthens (USD fell 5.14% in 1Q26), the JCP base is narrower, and withholding is higher.

The practical consequence is large and under-discussed. A ~20% recurring effective rate against a 45% statutory rate means roughly R$2.2bn per quarter of Bradesco’s recurring pre-tax income never reaches the tax authority — about 32% of recurring net income. Any further narrowing of JCP deductibility is a first-order threat to reported ROE, and it would hit the reported number without touching a single operating line.

6.5 Capital — thin, deteriorating, and just plugged with an equity raise

*Pro-forma for the Bradsaúde consolidation benefit (+250bp CET1).

FACT — CET1 fell 100bp in a single quarter, 11.2% → 10.2%. The bridge: net income +2.0pp, shareholder remuneration −0.6pp, prudential adjustments/other −0.4pp, RWA growth −0.8pp, regulatory changes −0.1pp with operational-risk RWA up R$22.4bn (+19.5% q/q). Prudential adjustments jumped R$7.5bn in the quarter to R$59.9bn — 34.5% of stated shareholders’ equity is deducted from CET1.

And the deferred-tax-asset stock is the most under-discussed item on this balance sheet:

Gross DTAs of R$116.7bn and net DTAs of R$109.3bn at Dec-2025, against IFRS shareholders’ equity of R$178.4bn. Net DTAs are 61% of book equity. Strip goodwill and intangibles (R$25.7bn) and net DTAs, and the “hard,” cash-generating equity is roughly R$43bn — about 24% of stated book value. The DTA is a real asset — Brazilian regulation permits it, it converts to cash as future taxable profits are earned, and it is precisely why the effective tax rate is 20% rather than 45%. But it earns nothing itself, R$59.9bn of it is already deducted from CET1, and it means the reported 13.6% IFRS ROE is generated on a capital base that is substantially an accounting claim on future tax payments. This is exactly why the July equity raise “increases tangible book and permits expanded utilisation of DTAs” — the raise is, in part, about buying the tangible-capital headroom needed to consume the tax credits.

(This is not a Bradesco-specific accounting abuse; Deutsche Bank flagged in 2011 that Brazilian system capital was already “inflated by a large volume of deferred tax credits.” But Bradesco carried an unusually large DTA through the 2020s provision cycle, and the magnitude relative to equity is not comparable to Itaú’s.)

6.6 Litigation and contingencies — a permanent R$8.5bn-a-year tax on the franchise

Brazilian retail banks carry enormous labour, consumer and tax litigation books. Bradesco’s is large enough to change how you read the earnings.

Provisioned (probable), R$ thousands [FY2025 20-F, Item 8.A.20 and Note 22]:

Possible but NOT provisioned (off balance sheet), R$ millions:

FACT. Against parent equity of R$178,415m (Dec-2025), the unprovisioned tax contingency alone equals 24.2% of book equity; all three possible categories together equal 31.2%; provisioned plus possible equals R$73.7bn — 41.3% of equity. The largest single possible case is an IRPJ/CSLL 2012–2015 CDI-interest disallowance of R$11.1bn, followed by a COFINS credit-offset case of R$10.5bn and a goodwill-amortisation disallowance of R$7.7bn.

FACT, and this is the item that matters most: Bradesco paid out R$8,519m in litigation settlements in FY2025 (FY2024: R$8,334m) — roughly 34% of FY2025 recurring net income — and that cost sits inside the recurring number, not as an add-back. This is not an unusual event awaiting resolution; it is a permanent, structural operating cost of running a Brazilian mass-retail bank, and it is a cost the digital-native competitors, with no branch estate, no unionised workforce and no thirty-year consumer-litigation tail, largely do not carry. It belongs in any honest comparison of Bradesco’s cost structure with Nubank’s.

Two directional reads, opposite in sign. The genuinely good news: on 24 May 2025 the STF declared the Bresser, Verão, Collor I and Collor II economic plans constitutional and reaffirmed the 2017 collective settlement, setting a 24-month adhesion window to roughly June 2027. Bradesco’s FY2025 20-F rewrote that risk factor from “are currently deciding… which may increase our costs” to “we remain subject to residual risks.” A three-decade tail risk on the entire Brazilian banking system has materially narrowed — and the possible-tax stock fell R$3.8bn in 2025, the first decline in the series, consistent with the PTI settlements. The less good news: the labour provision nearly doubled (R$2.6bn → R$4.4bn) on a “refined model of the measurement parameters” — a methodology change worth roughly R$2.7bn of gross provisioning, and model-driven provision swings deserve scepticism in both directions. And the JCP-deductibility assessment quintupled from R$197m to R$933m (+374% y/y) — small in absolute terms, but it means the tax authority is now directly attacking the instrument that underpins Bradesco’s entire distribution policy and roughly a third of its reported earnings.

And the tax reform is quantified and dated. The FY2025 20-F’s most materially expanded risk factor replaced FY2024’s generic language with concrete law: under EC 132/23, LC 214/2025 and LC 227/26, financial-services rates rise progressively from 10.85% in 2027 to 12.50% in 2033; LC 224/25 raised CSLL on the financial sector from 1 April 2026 (banks 22% → 20%, but up for capitalisation and payment businesses inside the group) and raised JCP withholding from 15% to 17.5%; and from 2027 insurance operations are taxed as “financial services” under IBS/CBS on top of the 2026 IOF-Insurance schedule. Management’s own risk disclosure concedes “there is no definition of the IBS/CBS tax rate applicable to the general regime, which covers much of our services.”

INTERPRETATION — the risk-factor drift is unusually informative. Across FY2022 → FY2025 Bradesco removed or softened risk factors on LIBOR transition, market-risk losses, insurance claims, negative publicity, international interest rates and inflation, and materially de-escalated the economic-plans litigation. It expanded exactly one: tax. Management is measurably less worried about macro and legacy litigation and measurably more worried about the tax line — which is precisely what Section 6.4 shows in the numbers.

6.7 Balance sheet, funding and leverage

*Every mainstream data vendor carries a Bradesco share count near 9.25bn, producing a book value per share of R$19.43 and the widely-repeated claim that the stock trades below book. That count is wrong by roughly 14%. Bradesco’s own FY2025 20-F reports 10,577,012,028 shares, and the July capital-increase filing confirms it arithmetically: 604,852,753 new shares at a 5.721967934% preemptive ratio implies 10,570,712,034 existing shares (~5.293bn ON + ~5.277bn PN). The corrected book value per share is R$17.01, and the prior-year figures above are restated on the same basis. See Section 10.1 — this single correction removes the “trades below book” headline and materially changes the valuation conclusion.

Funding (R$m, Mar-2026):

INTERPRETATION — funding is ample but the mix is getting more expensive. Loans-to-funding fell to 85.0%: comfortable liquidity, no funding stress. But cheap sticky savings deposits fell 5.2% y/y while expensive time deposits (+19.4%), securities issuance (+23.1%) and repos (+25.5%) carried the growth. At a 14.75% Selic that mix shift is a direct headwind to client NII that the +50bp gross NIM had to overcome. If the Selic falls to 12.5% by end-2026 as Bradesco forecasts, this reverses and becomes a tailwind — genuinely the strongest bull argument for FY2027 margins.

And leverage is doing real work in the ROE recovery. Assets/equity moved from 11.0× (FY2020) to 13.5× (Mar-2026), with total assets +18.5% y/y against 8.4% loan growth. A meaningful share of the ROE recovery is balance-sheet expansion, not margin or cost improvement — which is the mechanical link to the CET1 decline and, in turn, to the equity raise.

(LCR and NSFR are not disclosed in the 1Q26 press release — Open Question.)

6.8 FY2026 guidance versus 1Q26 delivery

Bradesco’s own macro assumptions in the same document: Selic 12.50% (end-2026) and 9.50% (end-2027) from 14.75%; IPCA 4.3%/3.4%; GDP +1.5%/+2.0%; USD R$5.35/R$5.40.

INTERPRETATION — the composition of the beat matters more than the beat. Bradesco is running ahead on insurance and fees and behind on credit growth and net-of-provision margin — i.e. ahead where it already earns 22% and behind where it earns 13%. If that persists, group ROAE keeps grinding higher while the banking franchise stagnates, and the Bradsaúde separation logic gets stronger by the quarter. Note also that the R$42–48bn NII-net-of-provisions range is very wide (a 14% spread) for a single guided line, the low end is a low bar, and the run-rate is already under it.

6.9 Quality-of-earnings checklist

6.10 Verdict — Financial Quality

Bradesco’s economics do not improve with scale, and the recovery is narrower than the headline.

The nine-quarter recurring-earnings sequence is real, the ROAE climb from 10.0% (FY2023) to 15.8% (1Q26) is real, and the cost discipline — 28% of branches closed since 2023, 12-month cost-to-income down 260bp — is real. That is a competent operational turnaround by any standard.

But four things undercut it. First, the profit is in the insurer, not the bank. Insurance earns 21.6% on 28.5% of the equity and delivers 41% of the profit; the universal bank earns roughly 13% — below any defensible Brazilian cost of equity with the Selic at 14.75%. Second, the margin expansion is being bought with credit risk and leverage. Gross NIM +50bp but net NIM −10bp; cost of risk 3.5% (retail 5.5%); coverage down 22 points; the fastest-growing lines are vehicle, rural and MSME working capital, precisely where management now flags deterioration; and assets/equity has climbed from 11.0× to 13.5×. Third, the capital is thin and low-quality. CET1 fell 100bp in one quarter to 10.2%, R$59.9bn (34.5% of book equity) is deducted as prudential adjustments, and net DTAs of R$109.3bn equal 61% of book equity. Fourth, the tax line is a live risk — a ~20% recurring rate against 45% statutory means roughly a third of recurring net income is a tax-structuring artefact.

The 29-July capital increase is the honest summary of all four. A bank raising up to R$10bn at roughly book value (a blended R$16.53 against a corrected book of R$17.01), two days before a quarterly print, with the controllers underwriting R$8bn and the dividend accelerated so shareholders can fund their own dilution, is not a bank with surplus capital and a solved problem. It is a bank that has stabilised, that has a genuinely excellent insurance business inside it, and that needs equity to keep growing and to consume its tax credits.

Financial quality: improving from a low base, materially better than 2023, and still structurally inferior to Itaú on every operating measure that matters — efficiency (49.2% vs ~36–39%), ROE (15.8% vs 24.8%), asset quality (4.2% vs 1.9%), and capital quality.

7. Capital Allocation

7.1 Who actually controls Bradesco, and why it matters for every distribution decision

Bradesco is not a widely-held bank. It is controlled through a pyramid whose apex is a charitable foundation, and that fact explains more about its capital allocation than any strategic document.

Fundação Bradesco runs 40 school units across all 26 states, educated >42,000 basic-education students in 2025, and spent R$1.4bn against roughly R$10bn invested over ten years — the 20-F calls it “the largest private social investment project in the country.”

INTERPRETATION — this is the single most important structural fact about Bradesco’s capital allocation, and it cuts both ways.

It biases the company toward distribution. Fundação Bradesco’s 31.53% economic stake generated a claim on roughly R$4.57bn of FY2025’s R$14.5bn gross JCP, against an annual charitable budget of R$1.4bn — the foundation receives about 3.3× its annual spending need from the dividend. A perpetual, tax-advantaged owner whose mandate requires predictable annual funding has a structural preference for a high, stable, monthly JCP stream over retention. That is precisely the policy Bradesco has run since 1970 (monthly dividends) and 1997 (monthly JCP) — and it explains why the payout was allowed to reach ~59% in the 2023 earnings trough rather than being cut. Cutting the JCP starves the foundation.

It does not automatically imply extraction. The same over-funding means the foundation can comfortably recycle JCP into a rights issue at essentially zero opportunity cost — which is exactly what the July structure asks it to do. But note one asymmetry: the capital-increase filing’s tax table carves out “legal entity shareholders who are exempt from such taxation, who will receive the declared amounts” — so an immune or exempt foundation plausibly receives the R$6.5bn of accelerated JCP gross, while minorities and ADR holders receive it net of 17.5% withholding. OPEN QUESTION: confirm Fundação Bradesco’s IRRF status. If exempt, the round-trip is materially cheaper for the controlling block than for the float — worth roughly R$0.4bn on the controllers’ ~R$2.4bn JCP share.

And the company itself flags the governance risk. The 20-F’s own risk factor states that all non-independent directors “are associated with Fundação Bradesco and decisions in relation to our policy towards acquisitions, divestitures, financings or other transactions could be made by Fundação Bradesco and our Board of Directors” — a concession that strategic capital allocation may be decided in the foundation’s interest rather than the share classes’. Only 4 of 11 directors are independent. The ADR (BBD) is over the non-voting preferred.

7.2 The distribution record — counter-cyclical, and that is the problem

The bylaws mandate a minimum of 30% of adjusted net income; FY2025 paid 52.47%.

A correction to the received wisdom, and it matters. The ADR cash stream looks wildly unstable — $0.501 (2019), $0.138 (2020), $0.191 (2021), $0.086 (2022), $0.339 (2023), $0.148 (2024), $0.306 (2025) — and is commonly read as “the dividend has been halved twice in six years.” The declared BRL stream says otherwise: gross PN dividends per share ran R$1.005 / R$1.008 / R$1.114 / R$1.114 / R$1.436 across 2021–2025 — flat, then up, never cut. The ADR swings are two other things: a 7-to-12-month declaration-to-payment lag shifting cash across calendar years, and BRL depreciation. Only FY2020’s cut was real, and it was imposed by CMN Resolution 4.820/20, not chosen. The memo should say both: the declared distribution has been stable and rising; the dollar cash an ADR holder actually banked has been volatile and lumpy.

That lag is itself worth noting as a financing tool. JCP declared 20-Mar-2025 was paid 31-Oct-2025; declared 18-Dec-2025, paid 31-Jul-2026. Bradesco takes the corporate tax deduction on declaration and holds the cash, interest-free, for most of a year. The July action accelerates two 2026 declarations into 15-Sep-2026 — the first time the lag has been shortened, and it was shortened specifically to fund the rights issue.

INTERPRETATION — and here is the actual indictment. Over FY2021–FY2025 Bradesco declared R$56.5bn gross / R$48.3bn net against R$99.6bn of cumulative IFRS net income — a 57% gross payout. It sustained that payout straight through the 2023 earnings collapse, running ~59% in the year recurring profit fell to R$16.3bn and ROE to ~10%, because the foundation-controlled structure makes a JCP cut close to unthinkable. Over exactly the same window, CET1 fell to 10.2%, assets/equity rose from 11.0× to 13.5×, and RWA grew to R$1,152bn. Then, in July 2026, the company asked shareholders for R$10bn back.

That is the sequence: distribute through a trough you could not afford out of the capital account, re-lever, and recall the capital eighteen months into the recovery. The July raise is not the mistake. The five years of distribution policy that made it necessary is.

7.3 The July 2026 capital increase — defensible on its own terms

Three observations the headlines miss.

First, the deal is priced at book, not below it. The universally quoted “R$17.64 against R$19.43 of book = 0.91×” rests on the wrong share count (Section 10.1). On the corrected 10.57bn shares, book is R$17.01 and the blended issue price of R$16.53 is 0.97× of it. Because rights trade on B3, economic dilution to a non-participating holder is roughly 0.3% (theoretical ex-rights R$18.71 against a cum-rights R$18.77), not the 5.7% share-count figure. (Bradesco’s own filing claims 3.40% maximum dilution — a figure this analysis cannot reproduce on any construction. Flagged as an open question.)

Second, roughly half the “new” equity is money the bank was already going to pay out. Per existing share the subscription costs R$0.883/ON and R$1.009/PN; the accelerated JCP delivers R$0.586/ON gross (66% of the cost) and R$0.483 net (55%). Net new external capital is therefore closer to R$3.5bn than R$10bn — and the controllers’ R$8bn “firm commitment” is itself partly self-funded from their own JCP entitlement.

Third, the real driver is tax, and it is rational. R$10bn of permanent equity widens the JCP deduction base — worth roughly R$360m a year of corporate tax saved — and buys the tangible-capital headroom needed to consume R$109bn of deferred tax assets (Section 6.5). That is why JP Morgan framed it as “a first step to create a virtuous cycle” and BTG as “a key prerequisite.” The market’s reaction was mild: −2.2% on the day, fully recovered within two sessions.

On Marathon’s capital-cycle test the raise reads amber, not red. Capital is leaving Brazilian banking — 7,252 branches closed in a decade, fintech licensing applications down from ~15/month to ~2/month. A bank raising equity into a withdrawal phase is not the classic top-of-cycle warning. But the timing sits badly against management’s own words: eleven weeks earlier the CEO called capital “so much more comfortable” at 12.7% pro-forma (Section 8.2).

7.4 Buybacks — competent tactically, irrelevant strategically, about to be reversed

Bradesco does run buyback programmes and does cancel shares: 29.5m cancelled in 2022, 16.3m in 2023, 50.2m approved in February 2025, and a new programme on 7-May-2025 for up to 106,584,881 shares (~1.0% of capital) running to 8-Nov-2026.

Cash actually spent, five years: FY2021 R$666.7m, FY2022 R$224.4m, FY2023 R$0, FY2024 R$568.7m, FY2025 R$222.6m — a total of R$1,682.4m.

The execution deserves credit where it is due: the December-2024/February-2025 tranche bought 31.1m shares for R$348.6m at an average of ~R$11.2 — at the exact five-year low. The ADR bottomed on 18 December 2024 and has roughly doubled since. That is genuinely good timing.

But the scale makes it irrelevant. R$1.68bn of buybacks over five years is 3.0% of the R$56.5bn distributed as JCP over the same period and under 1% of today’s market capitalisation. The May-2025 authorisation for 1.0% of capital is only ~6% used with three months left to run. And the bonus-adjusted share count, which fell just 1.1% over six years (10,691m in 2020 to 10,570.7m at Mar-2026), is about to rise +4.5% to 11,175.6m. Bradesco is not a share-count-reduction story and should never be modelled as one. Itaú, on a comparable payout, has run materially larger repurchase programmes — this is a genuine differentiator, not a rounding error.

7.5 M&A — ten years, three decisions that matter, and no divestitures at all

The 20-F states flatly: “There have been no recent divestitures.” Across the entire 60-month window Bradesco has sold nothing. Every corporate action has been an acquisition, an increase in an existing stake, or an internal reorganisation. That is a finding in itself: there is no evidence of portfolio pruning, and Bradsaúde is a listing, not a sale.

(a) HSBC Brasil (2015–16) — US$5.186bn / ~R$16bn cash. Verdict: value-destructive, and the largest single reason the decade produced nothing.

Goodwill recognised: R$4,222m, of which R$2,302m was amortised as a non-recurring item in 2017, plus a R$255m impairment in 2019. Run the Greenwald test on what was bought: ~5m clients, ~850 branches and ~R$46bn of loans, into a bank that already had ~4,500 branches. It bought scale in a business where Bradesco already had scale and no cost advantage — the definition of a value-destroying acquisition under Greenwald, because incremental scale creates value only where it lowers unit costs behind a barrier to entry. The evidence on whether it did is dispositive and is Bradesco’s own: the efficiency ratio is 49–50% against Itaú’s 36–39%, and the bank has since closed 58% of its branches with no loss of loan market share — including most of what it bought. A network you can halve without consequence was not worth R$16bn. The counterfactual is brutal: R$16bn deployed into Bradesco’s own shares in mid-2016 would have bought ~10% of the company. The ten-year ADR total return is −0.9%.

(b) Cielo take-private (2024) — ~R$2.15bn of Bradesco’s money. Verdict: value-destructive, and the least defensible recent decision.

The unified tender offer cleared at R$5.82 on 14-Aug-2024, acquiring ~736.9m of 902.2m shares for ~R$4.3bn total and lifting Bradesco’s equity interest from 31.41% to 50.72%. It implied ~R$15.8bn for 100% of Cielo, roughly 9× annualised earnings. Cielo is not consolidated — it is equity-accounted inside an “Others” line of R$11,147m that also contains Fleury and Banco John Deere, so its carrying value is not separately disclosed.

The cash return is the damning number. On a ~50.7% stake in an asset the buyers themselves valued at R$15.8bn — i.e. ~R$8.0bn of value — Bradesco extracted R$124.0m of dividends in FY2025, down 18.2% from R$151.5m in FY2024. That is a 1.55% cash yield against a 14.75% Selic. Even on the equity-method earnings claim the implied return is ~10–11%, below any defensible Brazilian cost of equity and below Bradesco’s own group ROE.

And the industry direction is against it. The standard read on Brazilian acquiring — structurally deteriorating and permanently compressed by PIX — holds up on the evidence. Cielo’s ~R$870–898bn of 2025 TPV represents ~26–27% of a market it once dominated, and CIEL3 fell more than 80% from its 2015 peak to the R$5.82 take-out. The sharpest version of the criticism is the funding point: bank-owned acquirers price aggressively because their prepayment book is funded with cheap deposits. That is not a moat — the causation runs the other way. The bank is subsidising the acquirer’s pricing, and the acquirer returns 1.55% of cash on the capital employed. Marathon’s capital cycle says capital should be leaving acquiring. Bradesco added to it. The right trade was to sell the 31.41% stake to Banco do Brasil, not to buy nineteen more points of it.

© Bradsaúde / Odontoprev (Feb–May 2026). Verdict: value-creative — the one genuinely good decision in a decade.

The reorganisation consolidated Bradesco Saúde, Mediservice, Odontoprev, Atlântica Hospitais (3,600 beds), the health clinics and the ~25% Fleury stake into Odontoprev, renamed Bradsaúde S.A. (B3 Novo Mercado, SAUD3, first trade 2026-05-05). Bradesco 91.35%, former Odontoprev holders 8.65%. Terms were set by an independent Odontoprev committee with a Citigroup fairness opinion, and the exchange ratio was confirmed at closing without adjustment — the best available evidence that minorities were not squeezed. Starting scale: R$52bn revenue, R$3.6bn net profit, ~24% ROE, >13m beneficiaries; +~250bp to pro-forma CET1.

This is the mirror image of HSBC and Cielo. Instead of paying cash for scale it does not need, Bradesco took the group’s only asset that passes the Greenwald test and made it separately priceable at essentially zero cash cost, while banking 250bp of regulatory capital. The caveats are real — 91.35% is retained so nothing has been monetised, the 8.65% float is thin as a price signal, and Brazilian health is fragmented (649 operators, none above ~8%) so the advantage is verticalisation rather than concentration. It is an unlock in principle, not yet in cash. But it is the right kind of decision, and it is the first one in ten years.

The rest of the record: BAC Florida Bank (~US$500m, completed Oct-2020) — small, coherent, and never separately disclosed since, so unassessable. The ~25% Grupo Fleury stake, accumulated on-market 2020–21 — probably loss-making on a mark-to-market basis given FLRY3’s de-rating, but strategically defensible inside the health vertical. The Atlântica Hospitais build-out and the Atlântica D’Or joint venture with Rede D’Or — coherent verticalisation, prices undisclosed. Banco John Deere (50%, completed 10-Feb-2025) — sensible in principle, but consolidated directly into agribusiness exactly where 1Q26 credit deterioration is now flagged.

Ranked, best to worst: (1) Bradsaúde 2026 — value-creative; (2) the Atlântica/Fleury health verticalisation — coherent but unverifiable; (3) BAC Florida — neutral; (4) Cielo 2024 — value-destructive, buying into a structurally compressed profit pool; (5) HSBC Brasil 2015–16 — the largest and the worst.

7.6 Investment intensity — R$42.6bn for 2.2 points of efficiency ratio

Bradesco has spent roughly R$42.6bn on technology, systems and infrastructure across the transformation window. Data-processing and communications expense is running +29.8% y/y. The 12-month cost-to-income ratio has moved from 52.2% to 50.0% over two years — 2.2 percentage points — and the risk-adjusted efficiency ratio has deteriorated from 76.1% to 76.9%.

INTERPRETATION. The spending is not obviously wrong — the bank genuinely had to re-platform, and the digital-client migration (19m → 28m, targeting 40–50m) is the mechanism behind closing 28% of the branches. But the return so far is thin, and management’s own 40%-by-2028 target requires roughly five times the achieved pace. Branch-closure savings are being redeployed into technology, not banked, which is a defensible strategy and an undeniable drag on the near-term efficiency ratio. It also means the cost story cannot be underwritten as a self-funding one.

7.7 Incentives — paid on absolute profit, with no return metric at all

FACT. Variable compensation for both the Board and the executive officers keys off “accumulated net income” — absolute profit. There is no ROE link, no efficiency-ratio link, and no tangible-book-value-per-share link. There is no option plan and no performance-share plan. The approved 2026 maximum aggregate compensation is R$1,443m, up 16.4% y/y.

INTERPRETATION — this is misaligned in exactly the wrong direction for the decision the company just took. A management team paid on absolute net income is rewarded for growing the balance sheet and issuing equity, because both raise the numerator, and is not penalised for enlarging the denominator. That is a precise description of the last five years: assets/equity from 11.0× to 13.5×, R$16bn spent on HSBC, R$2.15bn on more Cielo, R$1.68bn of buybacks against R$56.5bn of distributions, and now +5.7% more shares. A 16.4% increase in the approved compensation cap, in a year whose guidance implies less than one point of ROE improvement, is not a coincidence — it is the incentive working as designed. The absence of any per-share or return-based metric is the single clearest governance criticism available on the public record.

7.8 Insider behaviour — the apparent buying is contractual

Bradesco is a foreign private issuer, so there is no long Form 4 history — but it began filing Section 16 forms in April 2026 and files monthly CVM Instrução 358 disclosures as 6-Ks. Those give a usable picture.

On the surface it looks bullish: R$197m of gross insider buying. Strip out the mechanics and it inverts. R$191.8m of that total is the compulsory conversion of variable pay into restricted shares, executed as six single-day block events — a compensation settlement, not a discretionary purchase. Netting to discretionary trades only: R$5.5m bought against R$38.2m sold — net selling of R$32.7m, and the selling accelerated into the rally. The controlling holding companies have not bought a share on the open market in twenty months.

The one clean signal in the other direction: director Júlio César Bueno bought 110,400 shares on the open market (code P) on 1 July 2026 — four weeks before the capital increase was announced. Against that, two directors sold 176,138 shares in a five-day window in late May/early June. A 14-July disposal by an executive officer is coded J (“other”) — an internal transfer or plan settlement, not an open-market sale — and should not be characterised as opportunistic without more evidence.

(The sample is small and there is no prior history to establish a base rate. Treat as weak evidence, and note that Brazilian issuers’ CVM Resolution 44 trading-window rules do much of the work a 10b5-1 plan would.)

7.9 Verdict — Capital Allocation

No. Management has not allocated capital intelligently. The record over a decade is: R$16bn paid for scale in a business with no cost advantage, most of which has since been closed; R$2.15bn added to a merchant acquirer whose profit pool the central bank is expropriating, now returning 1.55% in cash; R$42.6bn of technology investment for 2.2 points of efficiency ratio; R$1.68bn of buybacks against R$56.5bn of distributions; nothing sold in five years; and a payout sustained at ~57% of earnings straight through a trough the capital account could not afford — followed by a R$10bn call on shareholders eighteen months into the recovery.

The incentive structure explains it. Variable pay is keyed to absolute accumulated net income, with no ROE, efficiency or per-share metric anywhere in the scheme, under a board on which only 4 of 11 directors are independent and whose non-independent members the company itself describes as “associated with Fundação Bradesco.” A management paid to grow profit rather than returns will buy scale, will not sell anything, and will not shrink the share count. That is exactly what happened.

Two things should be said in fairness. The buyback execution in December 2024 was genuinely well-timed, if trivially small. And Bradsaúde is a good decision — the right asset, made separately priceable, at no cash cost, with 250bp of regulatory capital banked and a proper independent-committee process. The last two years are meaningfully better than the previous eight.

But one good decision does not make a record, and the July capital increase is best understood not as a mistake in itself — it is defensible, near book, and rationally motivated by the JCP base and the DTA stock — but as the bill arriving for the five years of distribution policy that preceded it.

8. Changes and Headwinds — Last Two Years

8.1 The timeline

8.2 The two contradictions that matter

Everything above is context. Two items are not context — they are direct conflicts between what management said and what the company did or discloses, and they belong in any assessment of this board’s credibility.

First: the capital reassurance. On 7 May 2026 the CEO told the market that pro-forma CET1 was 12.7%, would end the year “close to 12.7%, perhaps a little higher,” and described it as “so much more comfortable.” Eleven weeks later, on 29 July 2026, Bradesco announced up to R$10bn of new equity at roughly book value (a blended R$16.53 against R$17.01 of book, Section 10.1) — and did so while a 106.6m-share buyback authorisation from May 2025 sat unused. Either the comfort was overstated in May or something changed materially in Q2 that has not yet been disclosed. Both possibilities are unattractive, and the second one has a resolution date: 5 August 2026.

Second: the cost-of-capital claim reverses itself inside one quarter. The Q4-25 milestone that the entire turnaround narrative rests on was “ROAE of 15.2% exceeding our cost of capital for the first time.” On the Q1-26 call, the same CEO said: “we can guarantee this return of 15.8% against a cost of capital that should range close to 18%.”

This must be handled carefully. The sentence appears in an interpreted-English transcript, in an exchange where the questioner had just anchored the discussion on 18% as an ROE figure, and it may simply be garbled. It cannot be treated as a clean admission. But it is on the record; it is directionally consistent with this memo’s own independent arithmetic (banking-ex-insurance ROE ~13% against a 14.75% Selic); and if it is accurate rather than a transcription artefact, it inverts the single most-cited datapoint in the bull case. At a 15.8% ROE against an 18% cost of equity, the Gordon-justified P/B is roughly 0.81× — below where the stock trades today. This is the highest-priority question for the 5-August call.

A third, smaller item in the same register: the 40%-by-2028 efficiency target, stated explicitly by the CFO in February, was softened by May to “a low cost-to-income ratio.” And the fully-digital-client target moved from 40m (February) to 50m (May) with no change in the underlying client base.

8.3 Verdict — Changes and Headwinds

On balance these changes strengthen the operating business and weaken the investment thesis. That is not a contradiction; it is the point.

The turnaround is real and should be credited without hedging: nine consecutive quarters of recurring earnings growth, 28% of branches removed with no loss of loan market share, a genuinely cleaner book (secured mix +3.8pp, restructured −20% y/y, problem assets −28%), and in Bradsaúde a well-executed value crystallisation worth roughly R$42bn of listed market value and ~250bp of CET1 that also happens to be the capital fix management said it did not need.

Against that: the JCP shield narrowed on 1 January 2026 and the tax reform escalates from 2027; Resolution 4,966 makes growth expensive at the front end and removed the ability to smooth; the credit cycle turned in precisely the lines Bradesco chose to grow; the customer base is smaller than in 2022; and the July raise directly contradicted the May reassurance.

And the single largest positive is macro, not management. Copom’s easing from March 2026 toward Bradesco’s house-forecast 12.50% end-2026 and 9.50% end-2027 would simultaneously relieve household debt service, reverse the expensive funding-mix shift, and restore loan demand. The thesis from here is therefore materially more a rates call than a franchise call — which is itself an indictment, because a business with a durable advantage does not need the policy rate to fall 500bp to earn its cost of capital. Note too that the market does not share the house view: the BCB Focus survey of 20 July 2026 puts the Selic at 14.00% end-2026 and 12.00% end-2027 — 150bp and 250bp more hawkish than the forecast underpinning Bradesco’s own guidance.

9. Risk Analysis

9.1 The risk matrix

9.2 The three risks that actually matter

Sixteen rows is a checklist; three of them carry the thesis.

Risk 2 — the banking franchise never earns its cost of equity — is the structural one. Everything else is cyclical or macro. This one asks whether a 49%-efficiency, 4.2%-NPL, share-losing mass-retail bank can be turned into something that clears a 14–17% Brazilian CoE. The evidence is genuinely two-sided: the credit book has been repaired (secured mix +3.8pp, restructured −20% y/y, problem assets −28%), and Bradesco was Itaú’s peer fifteen years ago on Deutsche Bank’s stress screens — so the gap is a decade-old execution failure, not a structural inferiority. But the cost line, which is where the gap actually lives, has moved 2.2pp in two years against a target requiring roughly 5× that pace, on a unionised base indexed at 5.68% a year.

Risk 3 — Selic normalization — is the one nobody is discounting. It is the rare risk that hits both halves of the company simultaneously and in the same direction. A falling Selic compresses the record 15.17pp system spread, deflates 36% of insurance operating income, and shrinks the float return that produces most of Brazilian insurance’s structurally high ROE. The offsetting benefits — cheaper funding as the savings-to-time-deposit mix reverses, and a genuine fix to household debt service at record-low unemployment — are real and arrive with a lag. The bank and the insurer are not diversifying against each other; they are both long the Brazilian real rate. That is the single most under-appreciated correlation in this company.

Risk 1 — the credit cycle — is the near-term one, and the 5 August print is the test. System credit is at its worst configuration on record and Bradesco’s coverage has been falling into it. The specific things to watch in Q2: whether coverage drops another 5pp, whether MSME NPLs keep rising, whether cost of risk holds at or above 3.5%, and whether the rural/agribusiness deterioration management flagged in Q1 has broadened.

A note on what is not on this list. There is no governance-expropriation risk of the Petrobras or Banco do Brasil kind: Bradesco is privately controlled, and the state-interference discount applied to Brazilian state-linked issuers is generally judged rational rather than a mispricing. Bradesco’s discount to Itaú is an operating discount, not a governance discount — and an operating gap is, in principle, fixable. That distinction frames the entire investment question.

10. Valuation Discussion

No price target and no recommendation appear in this section. What follows is embedded-expectations and scenario analysis: what the current price requires to be true.

10.1 Where the stock trades — and a share-count correction that changes the headline

Start with a correction, because it inverts the most widely-repeated claim about this stock.

Every mainstream data vendor consulted for this memo — and, initially, this memo’s own working papers — carries a Bradesco share count of roughly 9.25 billion, which produces a book value per share of R$19.43 and the seductive headline “Bradesco trades at 0.95× book — below book value.” That is wrong, and it can be disproved arithmetically from Bradesco’s own capital-increase filing.

The 29-July 6-K sets a preemptive-rights ratio of 5.721967934% of existing holdings, issuing 302,876,396 ON + 301,976,357 PN = 604,852,753 new shares. Dividing the new shares by the ratio gives the existing share count: 604,852,753 ÷ 0.05721967934 = 10,570,712,034. Bradesco’s own FY2025 20-F independently reports 10,577,012,028 shares. The true count is ~10.57bn — roughly 14% higher than the vendor figure — split ~5.293bn ON and ~5.277bn PN.

Restated on the correct count, at $3.61 per ADR / R$18.43 per BBDC4, BRL/USD 5.1053:

Pro-forma for the capital increase (604.9m new shares = +5.72%, not +6.5%; R$10bn proceeds): book value per share moves to R$16.98 — essentially flat, −0.2%. And the deal is priced far closer to book than the headline suggests: the blended issue price is R$16.53 (not the R$17.64 PN price everyone quotes), which is 0.97× IFRS book and 1.007× BR GAAP book. Bradesco issued equity at approximately book value, not at a 9% discount to it. The PN tranche went at 1.04× IFRS book; the ON tranche at 0.91×. Because the rights trade on B3, economic dilution to a non-participating holder is roughly 0.3% (theoretical ex-rights price R$18.71 against a cum-rights R$18.77), not the 5.7% share-count figure. (Bradesco’s own filing claims 3.40% maximum dilution — a number this analysis cannot reproduce on any construction. Flagged in Open Questions.)

This correction runs against the bull case, and it should be stated as such. Bradesco is not a bank trading below its book value. It is a bank trading at roughly one times book — and, as Section 10.4 shows, that is almost exactly what management’s own guidance is worth.

(Data cautions that materially change the answer, all of them discovered the hard way on this engagement: ROIC.ai’s share count, return_com_eqy, eps and enterprise_value are all wrong for this issuer — it misreads the ON/PN dual-class structure as preferred capital, deducting R$12.4bn of total shareholder remuneration as “preferred dividends,” and its share count understates the true figure by ~14%. Enterprise value is meaningless for a bank in any case. AZI’s valuation_index book-value-per-share input is wrong by roughly 6×, contaminating both its P/B percentile and its composite; only the P/E percentile (54.0) and P/S percentile (26.3) are usable. Notably, the raw StatusInvest screen — showing 1.09× P/B — was closer to the truth than any of the “reconciled” figures built on vendor share counts.)

10.2 Own-history context — the finding that cuts against the “it’s cheap” reflex

(These are vendor-computed ratios on a consistent internal methodology. Because that methodology embeds the same understated share count identified in Section 10.1, every level in the table is understated by roughly 14% — the 2019 year-end figure is nearer 2.1× and today’s is 1.08×. The percentile ranking below is unaffected, because a constant scaling factor is percentile-invariant; the absolute levels should be read as an index, not as literal multiples. Flagged in Open Questions.)

Twelve-year median year-end P/B on this index: 1.18×, against today’s 0.949 — −20% versus that median. But the percentile depends entirely on which regime you benchmark against:

INTERPRETATION — and this is the most important own-history fact, because it is not the bullish one. Bradesco is cheap against the bank it used to be (6th percentile of 2014–2019) and upper-quartile against the bank it has actually been since the credit blow-up (76th percentile of 2022–2025). Whether the pre-2020 regime is the right comparison is the entire debate — and Section 3 and Section 4 of this memo conclude it is not. Demand-deposit share fell 11.0% → 6.6% in three years; the client base is smaller than in 2022; PIX and Open Finance have legislated away the captivity that supported the old multiple; and 58% of the branches that supported the old franchise are gone. A bank that has structurally lost its transactional franchise does not get to be re-benchmarked against its own 1.8× pre-PIX multiple. Framed correctly, Bradesco has already re-rated 46% off the end-2024 trough (0.65× → 0.95×) and sits in the upper quartile of its post-crisis range. The easy re-rating has happened.

One further tension worth naming: the Brazilian 10-year real yield (NTN-B 2035) has widened from ~7.3% at end-2025 to 8.16% on 27-July-2026. The discount rate went up ~90bp during 2026 while the multiple went up 46% off the trough. Both can only be right if the ROE expectation rose by more than the discount rate.

10.3 Cross-sectional comparison — and the arithmetic on proportionality

¹ BB Seguridade and Caixa Seguridade are capital-light distribution vehicles — underwriting capital sits in JVs, so their P/B and ROE are structural artefacts and are not comparable to a balance-sheet underwriter like Bradesco Seguros. Use their P/E only.

Is the discount to Itaú proportionate to the ROE gap? Three tests, in ascending order of correctness — and the share-count correction narrows the answer considerably.

1. Naive ratio-of-ratios (wrong, but it is what most people do). P/B ratio 1.02/2.40 = 42.5%; ROE ratio 15.8/24.8 = 63.7%. On this test the discount still looks more than proportionate.
2. Gordon-correct relative multiple (the right test). Justified P/B = (ROE − g)/(CoE − g) is linear in ROE with an intercept, not proportional. At any common (CoE, g): (0.158 − 0.065)/(0.248 − 0.065) = 50.8%. Bradesco should trade at ~51% of Itaú’s multiple; it trades at 42.5% (dual-class) or 45.2% (PN convention). The discount is ~11–16% wider than the ROE gap alone justifies — a real gap, but roughly half the size the uncorrected numbers implied.
3. Market-implied cost of equity (invert Gordon at g = 6.5%) — the cleanest statement:

This forces a decision the memo must make explicitly, and the corrected numbers make it sharper — not softer. Taking management’s recurring ROAE at face value, the market demands roughly 95–150bp more cost of equity from Bradesco than from Itaú — a real but moderate execution premium, and essentially the same premium the market charges Santander Brasil (16.0%), which says this is the “non-Itaú Brazilian private bank” discount, not a Bradesco idiosyncrasy.

But feed in the audited IFRS ROE of 13.6% and the implied cost of equity falls to 13.0–13.5% — below Itaú’s 14.14%. On the audited numbers Bradesco is not cheap relative to Itaú at all; it is modestly expensive. That is the reverse of the consensus framing, and it survives only if you believe the audited number is the right one.

The Lead’s call, stated plainly: neither number should be capitalised alone, and the honest anchor is ~14.5–15% ROE. The recurring number legitimately strips genuine one-offs, but it also excludes branch-restructuring provisions that have recurred in each of the last two years, and it sat 26% above the book number in 1Q26; the IFRS number is the audited truth but carries tax-line noise that ran in Bradesco’s favour through FY2023–25. What readers must take away is that the entire “Bradesco trades at an unjustified discount to Itaú” argument depends on (a) believing the recurring number rather than the audited one and (b) using a share count that is 14% too low. Correct either and the discount mostly disappears. Correct both and it inverts.

Banco do Brasil at 0.64× book on an 8.2% ROE remains the useful floor reference: it shows what a Brazilian bank earning 8–10% is worth, and it is roughly where Bradesco traded in 2024.

10.4 Embedded expectations — the market is making no forecast at all

Cost of equity, built rather than borrowed. With the Selic at 14.25–14.75%, importing Itaú’s ~14.5% uncritically will not survive scrutiny. Four independent builds:

A genuine tension worth naming: builds A and C disagree by 430bp, because a Brazilian investor’s actual risk-free alternative (an NTN-B at IPCA+8.16%, or a 10y nominal at 14.77%) already embeds ~600bp of country and fiscal risk premium that the Damodaran build then adds again on a US base. Worse, both banks’ stated ~15% hurdles imply an equity risk premium of barely 0.2–1.5pp over the local 10-year nominal — which is not a credible ERP. It means these banks are running new-investment hurdles that are, in substance, below the cost of buying their own government’s paper. ASSUMPTION adopted: CoE = 15.0% nominal BRL, with sensitivity across 13.0–17.0%. Growth g = 6.5% nominal — deliberately below the retention-implied 7.9%, to reflect a bank losing share in every primacy metric.

And now the elegant part — which the share-count correction makes cleaner, not messier. Solving (ROE − g)/(CoE − g) = P/B for ROE across a 300bp range of g moves the implied ROE by barely 15 basis points. That is not a coincidence: at P/B ≈ 1.0, justified P/B equals 1 exactly when ROE = CoE, independent of g. And Bradesco now trades at 1.02× book on the economically correct dual-class market capitalisation — almost exactly on that diagonal.

INTERPRETATION — the market is not making an ROE forecast at all. It is making a statement of near-perfect indifference: “whatever this bank’s cost of equity turns out to be, Bradesco will earn approximately that, and no more, forever.” At a 15.0% cost of equity the price implies ~15.1% ROE on the dual-class basis, or ~15.7% on the PN convention. Compare that to what management is actually promising: FY2026 guidance implies ~15.5% ROE, and 1Q26 delivered 15.8%.

That is the single most important sentence in this section: the market is already paying, almost exactly, for management to hit its own guidance. Bradesco is priced at mediocrity achieved and permanently sustained — no failure, no success, and no unpaid-for upside from the plan working as advertised.

The sensitivity grid — justified P/B, g = 6.5%.

Implied ADR price (US$), same grid, at the corrected BVPS of R$17.01 (= $3.332/ADR) and USD/BRL 5.1053. Today’s price is $3.61.

What the market is underwriting, stated plainly — and this is where the correction bites hardest. Read across the grid, $3.61 is consistent with 15.5% ROE at a 14.7% cost of equity, or 16.0% at 15.3%, or 17.0% at 16.2%. Every one of those pairs sits on the ROE ≈ CoE diagonal, and every one of them requires Bradesco to deliver at or above the 15.5% it is currently guiding to.

Put concretely, against the four scenarios:

That table is the whole investment case, and its skew is unattractive. Management delivering exactly what it has promised is worth roughly today’s price — a 2% discount to it. The most constructive mainstream sell-side case, which is explicitly framed as a ceiling, is worth +19%. A stall back to the 2020–22 average — not a crisis, just a failure to sustain the improvement — is worth −40%. You are being asked to accept a −40% downside against a +19% realistic upside, with the base case already in the price.

Justified-P/B band as an analytical output, not a target: on the plausible envelope of 13.0–17.5% ROE against a 14.0–16.0% cost of equity, the defensible band is roughly 0.62×–1.40×, an implied ADR range of roughly $2.07–$4.66. The width of that band is the honest measure of how little this analysis pins down without a view on the ROE path — but note that today’s $3.61 sits in its upper half.

10.5 Sum-of-the-parts — the variant perception lives here

A single blended multiple misprices a group that is ~41% insurance earnings at ~22% ROE and ~59% banking earnings at ~12–13% ROE.

Method A — mark health at its own listed price. Bradsaúde (B3: SAUD3) is now separately listed and separately priced: market cap R$46.55bn at ~R$15.92 on 2.923bn shares (31-Jul-2026), marked by the sell side at 9.2–9.8× 2026E earnings.

Triangulation. Two independent routes corroborate the insurance mark: the Gordon-justified value of the insurance arm on its own 22.3% ROE at a 14% insurer CoE and 7% growth is 2.19× its equity = R$108.2bn ≈ 10.7× FY2025 earnings; a flat 10× FY2025 earnings gives R$100.7bn. Three routes land at R$98–108bn. That is unusually tight and gives the sum-of-the-parts real weight. Sanity check: the Bradsaúde stake alone is 25% of Bradesco’s entire market capitalisation — from an asset that did not exist as a separate line item six months ago.

The reverse cut — what the market implies for insurance if the bank is fairly valued:

The SOTP answer — and here the share-count correction removes what would otherwise have been this memo’s headline variant perception. Mark the insurance and health assets at the prices the market itself is already paying — including SAUD3’s own listed R$46.6bn — and the residual banking franchise is valued at ~R$85bn: 0.65× its own book, 5.2× its own earnings, and an implied perpetual ROE of 12.0% (0.74×, 6.0× and 12.8% on the PN convention).

On the uncorrected 9.25bn share count, that stub came out at 0.47–0.56× book and a 10.5–11.2% implied ROE — a dramatic, quotable “the market is giving you Brazil’s second-largest private bank at half of book” claim. It does not survive the correction, and it should not be made.

What survives is quieter and more useful: the market is pricing the banking stub at an implied perpetual ROE of 12.0–12.8%, against the ~12.4% the banking segment actually earns today. In other words, the market is valuing the bank at almost precisely what it currently produces — no recovery credit, no decline assumed. That is the same conclusion Section 10.4 reached at the group level, arrived at independently from the segment side, and the two agreeing is a genuine cross-check rather than a coincidence.

The honest reading is therefore that both halves are priced about right, not that a good bank is hiding inside a cheap holding company. The insurance leg is marked at roughly 10× earnings — in line with Porto Seguro and BB Seguridade and with Bradsaúde’s own listed price — and the banking leg is marked at what it earns. Anyone hoping for a sum-of-the-parts unlock should note the reverse cut in the table above: to get the bank to a “fair” 0.70× book you have to mark the insurer down to 7.9× earnings, a ~20% discount to its listed peers. There is no arithmetic in which both legs are cheap simultaneously.

The single most valuable disclosure Bradesco could make, and does not, remains a segment-level capital allocation and a standalone banking ROE — because every number in this section is derived rather than reported.

10.6 Three normalisations the screens do not show

(a) Selic normalisation of the crown jewel. R$2,301m of the R$6,384m (36.0%) of 1Q26 insurance operating income was financial results on R$482.8bn of assets at a 14.75% Selic.

(ASSUMPTION: financial results scale with the Selic level on the shareholder-attributable invested position. Range shown deliberately.) A full normalisation takes ~80bp off group ROAE — all of it from the crown jewel — and drops the guided 15.5% to roughly 14.7%, back to or below the cost of equity. Nobody discloses underwriting-only insurance ROE, so this haircut is invisible in every screen. And note the asymmetry: Bradesco’s guidance rests on its house forecast of 12.50%/9.50%, while BCB’s Focus survey says 14.00%/12.00% — 150–250bp more hawkish.

(b) The tax shield — the largest single un-priced risk. A recurring effective rate of 20.3% against 45% statutory is a shield of R$2,111m per quarter = R$8.4bn a year = 31% of recurring net income.

At $3.61 the market is not pricing any tax normalisation at all. Move the recurring rate into the 29.5–32.5% band Itaú has already guided to — without touching a single operating line — and justified P/B falls to 0.75–0.80×, an implied $2.48–$2.67, i.e. −26% to −31%. (On the uncorrected share count this looked like −16% to −21%; the correction makes it worse, because there is no below-book cushion to absorb it.) This is the cleanest downside catalyst in the entire build because it requires no macro deterioration, no credit event and no competitive loss. The corporate-level JCP deduction survives, so the 45% column is a tail; but the 30% column is not a tail — it is where a directly comparable peer has already guided.

© Book quality, done fairly. I do not use “hard book” (ex-intangibles and ex-DTA, ~R$4.68/share) as the valuation anchor — that is an unfair standard against peers running the same structures, and net DTAs are a real, regulator-sanctioned asset. The CET1-basis comparison is fair, because it applies the same regulatory haircut to everyone:

*Itaú’s CET1 amount is derived from its disclosed 12.3% ratio and an estimated RWA — flagged as an assumption in Open Questions.

Bradesco’s book converts to regulatory capital at 65 cents on the real against Itaú’s ~79 — a genuine 14-point quality gap driven by the DTA and prudential deductions. But honesty requires noting it narrows the relative discount only modestly: Bradesco trades at 42% of Itaú’s book multiple and ~52% of its CET1 multiple. Book quality explains perhaps a fifth of the gap; execution and ROE explain the rest. Post-raise, P/CET1 of 1.23× is the single most flattering multiple in this memo — and it should be read as evidence that the capital problem is being solved, not that the earnings problem is.

10.7 Scenarios to end-2028

Mechanics: start from the corrected pro-forma post-raise BVPS of R$16.98, compound book at ROE × (1 − payout) for 2.75 years, apply an exit P/B derived from terminal ROE and scenario CoE, translate at the scenario FX, and add the cumulative JCP/dividend stream. All returns are ADR total returns from $3.61.

Probability-weighted total return: +25.0% over 2.75 years = +8.5% p.a. — against a cost of equity of ~15%. (On the uncorrected share count this read +41.7% / +13.5% p.a. The correction removes roughly a third of the expected return, and it is the single reason this memo’s conclusion is not more constructive.)

Two controls reframe the whole thing, and they matter more than the scenarios:

First: this is a multiple story, not an earnings story. Hold the multiple constant and the bear case still returns +16%, because book compounds at ~6% and the cash yield is ~7–8%. The entire downside is a de-rating — and a de-rating requires the market to conclude that ROE < CoE permanently, the same conclusion it held at 0.65× book eighteen months ago and has since abandoned. Symmetrically, hold the multiple and the bull case is only +55%, not +99%: ~45% of the bull return is re-rating, not earnings. Anyone underwriting this stock is primarily underwriting a change in the market’s cost-of-equity assumption and only secondarily underwriting Bradesco’s P&L.

Second: the currency is a large part of the outcome. The bear case moves from −38% to −30% on a flat FX, and the bull from +99% to +97% — the currency giveth mainly on the downside here, because the bear scenario is the one that assumes a 6.00 real. For an ADR holder the return decomposes into three near-independent bets — earnings, multiple, and the real — and with a Country:Brazil factor beta of 1.53 (higher than Itaú’s 1.43) into an October general election, the third has the fattest tails. The BRL has already appreciated ~10% from the 2025 average and ~20% from the 2025 worst level. That tailwind has been collected.

A tail not in the table: a repeat of 2022–23 — a credit blow-up on the Americanas scale plus mass-retail deterioration, taking ROE to 8–10% with the BRL back to its December-2024 record of 6.31 — produces roughly $1.45–$1.65, i.e. −55% to −60%. That is not a fantasy. It happened once in this exact cycle, in a book that today is growing fastest in vehicle (+25.4%), rural (+17.5–24.0%) and MSME working capital (+16.3%) — the three lines where management is currently flagging deterioration, against a system where non-earmarked NPLs are at a record 6.2% and unsecured personal credit defaults at 14.2%.

10.8 What the market is pricing correctly, and incorrectly

Correctly priced:

1. The banking franchise has no durable advantage and earns below its cost of capital. The 0.65–0.74× implied stub and 12.0–12.8% implied perpetual banking ROE — against the ~12.4% the segment actually earns — are a fair reading of a business with a 49.2% efficiency ratio against 36.2%, a 4.2% NPL against 1.9%, demand-deposit share down from 11.0% to 6.6% in three years, and a client base smaller than in 2022.
2. The insurance and health assets deserve a premium and are now separately visible. SAUD3’s R$46.6bn listed mark is a real, tradeable price.
3. The equity raise is dilutive. The stock fell 2.2–2.6% on 30 July despite uniformly positive sell-side commentary. A blended issue price of R$16.53 — 0.97× IFRS book, and requiring a ~16.2% return on new capital merely to be EPS-neutral, above the top of management’s own ROE guidance — deserved that.
4. The Itaú discount exists for a reason — a ~95–150bp implied cost-of-equity premium is a proportionate judgement on the reliability of the earnings, not an error. And on the audited IFRS numbers there is no discount at all.

Potentially mispriced — and note that three of the four run bearish:

1. The tax line is not in the price at all. A move to Itaú’s guided rate implies −16% to −21% with no operating deterioration. The largest identifiable un-priced risk in the build.
2. The insurance leg’s rate dependence is invisible. No screen shows the ~80bp of group ROAE that a Selic normalisation removes, because underwriting-only ROE is not disclosed.
3. The convexity around ROE = CoE is under-appreciated in both directions. At ~1.02× book the stock is priced precisely at indifference: two consecutive quarters above 16% ROAE moves justified P/B to ~1.12× (+10%); two quarters below 13% moves it to ~0.76× (−25%). There is no valuation cushion in either direction — which is the definition of a price with no margin of safety, and the skew is negative.
4. The own-history read is being told backwards. Consensus (9 Buys, 0 Sells; average price target R$23.17 / ADR ~$4.40) is anchored on “below book is cheap versus 1.8× in 2019.” Two things are wrong with that. First, the stock is not below book at all once the share count is corrected — it is at ~1.02×. Second, on the post-2021 regime that actually reflects the PIX/Open Finance world, today’s reading is the 76th percentile, not a bargain. If there is a consensus error in the bull direction, this is it — and it is being made on an arithmetic error, not just a judgement call.

11. Variant Perception

11.1 What consensus believes

Sell-side coverage is 9 Buys, 0 Sells, with an average price target of R$23.17 (~$4.40 per ADR, +22%). Independent commentary has migrated over 2026 from “no longer broken, too early to get aggressive” (January) to “the ROAE spread is turning positive again” (July). Zacks upgraded to Rank #2 (Buy) on 14 April 2026 — three days before the five-year price high.

The consensus argument runs: Bradesco is a cheap, self-help turnaround in a structurally high-ROE banking market. Nine consecutive quarters of earnings growth, ROAE up 4.1 points in two years, 28% of the branch estate removed, a genuinely cleaner credit book, an excellent insurance business now separately visible in Bradsaúde, an 8% dividend yield, and a Selic-cutting cycle just beginning — all available at below book value versus 1.8× in 2019 and roughly half Itaú’s multiple.

Almost every element of that is true. The conclusion still does not follow, for four reasons — and three of them are arithmetic rather than judgement.

11.2 Where consensus is offsides

(1) The stock is not below book. The share count is wrong by 14%. Every vendor screen carries ~9.25bn shares; Bradesco’s own 20-F reports 10,577,012,028, and the July capital-increase filing confirms 10,570,712,034 arithmetically from the 5.721967934% preemptive ratio. Corrected, book value per share is R$17.01, not R$19.43, and the stock trades at ~1.02× book on the dual-class market capitalisation — at book, not below it. The single most repeated fact in the bull case is a data error. (Note that the raw StatusInvest screen showing 1.09× was closer to the truth than any “reconciled” figure built on vendor share counts, including this memo’s own working papers before the correction.)

(2) The market is already paying for management to hit its guidance — precisely. At ~1.02× book the Gordon identity collapses to ROE = CoE independent of growth. At a 15% cost of equity the price implies ~15.1–15.7% ROE. Management guides FY2026 to ~15.5% and delivered 15.8% in 1Q26. There is no unpaid-for upside from the plan working as advertised. Run the four scenarios and the skew is unattractive: a stall back to the 2020–22 average of 12% is −40%; management’s own guide is −2%; JP Morgan’s explicitly-framed ceiling of ~17.5% is +19%; Itaú-like 20.5% is +52%. You are underwriting −40% downside against a +19% realistic upside with the base case in the price.

(3) The own-history anchor is being read against the wrong regime. “0.95× versus 1.8× in 2019” is the wrong comparison, twice over: the level is wrong (Section 1 above), and the benchmark is wrong. Against 2014–2019 today sits at the 6th percentile; against 2022–2025 — the regime that actually reflects the PIX and Open Finance world — it is the 76th percentile, upper quartile. A bank whose demand-deposit share fell from 11.0% to 6.6% in three years, whose client base is smaller than in 2022, and which has closed 58% of the branches that supported the old franchise, does not get re-benchmarked to its own pre-PIX multiple. Bradesco has already re-rated 46% off the December-2024 trough. The easy re-rating has happened.

(4) The two largest identifiable risks are not in the price at all, and both are bearish. The tax shield is worth R$8.4bn a year — 31% of recurring net income — and rests on JCP deductibility plus a R$109bn deferred-tax-asset stock. Move the recurring effective rate to the 29.5–32.5% band Itaú has already guided to, with no operating deterioration whatsoever, and justified P/B falls to 0.75–0.80×: −26% to −31%. Separately, 36% of insurance operating income is float carry at a 14.75% Selic; normalising to the 9.5% Bradesco itself forecasts takes ~80bp off group ROAE, dropping the guided 15.5% to ~14.7% — back below the cost of equity. Neither appears in any screen, because underwriting-only insurance ROE is not disclosed anywhere.

11.3 Where consensus may be offsides in the bull direction

Honesty requires the reverse case, and there are three real ones.

The credit cycle is testing underwriting, not cost — and that favours the incumbent. System NPLs are at all-time highs in every series, but that runs against the challengers: Nubank’s 90+ NPL of 6.5% exceeds the system’s 4.7%, and its shares are down roughly a third in 2026. This is the one cycle in which a 90-year-old bank with a full credit-bureau history and a 60.8% secured book out-competes an eight-year-old app. Commentary describing Brazilian credit as “benign” has been reading Itaú’s book, not the system’s — but the corollary is that the disruption narrative is also being over-extrapolated.

The regulator is not uniformly anti-incumbent, and the market has not noticed the second half. PIX and Open Finance are genuinely expropriatory. But BACEN has simultaneously made entry far harder — fintech licensing applications have fallen from ~15/month to ~2/month, and Resolution 494/2025 pulled the regularization deadline forward by more than three years — while the CSLL restructuring cuts banks from 22% to 20% and steps payment institutions and credit fintechs up toward 15–20% by 2028. The tax arbitrage that let the attackers undercut banks after tax is being narrowed by statute.

The gap to Itaú is an operating gap, not a governance gap — and operating gaps are, in principle, fixable. Fifteen years ago Deutsche Bank’s 2011 stress screens ranked Bradesco the most capital-resilient of Brazil’s big three and named it a preferred global name alongside Itaú. Unlike Petrobras or Banco do Brasil, Bradesco is privately controlled, so the state-interference discount that is generally judged rational rather than a mispricing does not apply here. Closing half the efficiency gap to Itaú is worth roughly R$6bn pre-tax — more than the entire 2023-to-2025 earnings recovery. That option is real, it is large, and on today’s multiple it is unpriced.

11.4 The five assumptions that matter most

Note assumption 5: management’s guidance rests on a rate forecast the market does not share — and assumptions 4 and 5 point in opposite directions for the insurer, because the rate cut that fixes the credit book also deflates the float income.

11.5 The positioning read

The factor model is unusually informative here, and it says the debate is not being expressed in the tape at all.

Bradesco is not priced as a bank. It is priced as Brazil. Its dominant factor loading is Country: Brazil at +1.53 — 2.2× its Market beta of 0.69 — against Value at +0.11 and Momentum at +0.04, effectively nil. Thirteen of its fifteen factor-nearest peers are Brazilian, and the closest matches are a fuel distributor (Ultrapar) and three Brazil index ETFs, ahead of Itaú. Roughly two-thirds of factor-explained risk is the country; and with an R² of 0.482 against 33.9% total realised volatility, idiosyncratic volatility of 21.2% means about half of BBD’s variance is neither Brazil nor any factor — it is Bradesco’s own execution. That residual is precisely what a fundamental thesis has to underwrite, and it is unusually large for a $38bn bank.

Nothing here is a crowded trade in either direction. Short interest is 0.4% of shares outstanding (~1.3 days to cover) even after a 245% spike in March — there is no short base to squeeze and no crowded-short risk. Momentum loading is nil and Quality is −0.02, so this is neither a momentum crowd waiting to unwind nor a quality compounder. The five-year Sharpe is 0.006 and the ten-year Sharpe is −0.056: over a decade this stock has paid nothing for 43% annualised volatility and a 76% peak-to-trough drawdown. Only the trailing twelve months look good (Sharpe 1.18).

And the relative gap has stopped widening without beginning to close. The entire 165-point five-year underperformance versus Itaú was created between Q2-2022 and Q1-2025; since the December-2024 low BBD has matched Itaú to within 1.3 points (+125.3% vs +124.0%). But the BBD/ITUB price ratio is 0.427 today against a 0.383 trough and 1.03 five years ago — less than 8% of the relative de-rating has closed. The market has stopped punishing Bradesco further; it has not begun paying for re-convergence.

The final positioning irony: the style regime is the friendliest it has been in a year for exactly this kind of stock — Value +16.5% (z +1.92) and DividendYield +17.8% (z +1.75) over 252 days, the two strongest factors tracked — and BBD is not capturing it, because its actual carrier factor, Country: Brazil, is down ~6% over 126 days (z −1.02) and Sector: Financials is the weakest 252-day factor at −12.6% (z −1.44). The stock is not fighting its own fundamentals. It is fighting its country. Whether that is the setup or the warning, the factor model cannot say.

12. Fact vs. Interpretation

The distinction matters more than usual on this name, because several of the memo’s load-bearing claims — the ~13% banking ROE above all — are derived from primary disclosure rather than reported by the company. Readers should know exactly which is which.

13. Open Questions

These are the things I could not resolve from public sources. Several are load-bearing, and the first three would materially change the analysis.

Disclosure gaps that the company could close tomorrow and chooses not to

1. What is the underwriting-only ROE of Bradesco Seguros? 36% of 1Q26 insurance operating income was financial results at a 14.75% Selic, and the group’s own forecast is 9.50% by end-2027. Nobody — not the company, not the sell side — has published an insurance ROE normalised for a 9.5% policy rate. This is the single most valuable unpublished number for this thesis, because it determines what 41% of group profit is actually worth. Related: what is Bradesco Saúde’s own loss ratio against the 82% industry figure?
2. What is the capital allocated to each segment? The ~13% banking ROE that anchors this memo is derived, using Bradesco Seguros’ consolidated equity (R$49.5bn) as a proxy. The 1Q26 disclosure also shows R$23.6bn for the regulated insurance/pension/capitalization companies — a very different denominator. A Basel-RWA-based allocation would be a useful cross-check, and the company could simply publish the number.
3. What is the pension-sales attrition rate in municipalities where a branch was closed? 93% of pension sales are branch-originated and that dependence has risen every year, while 58% of branches have gone. This is the central strategic tension of the group and there is no public disclosure quantifying it at all.
4. What is Bradesco’s principality / primary-bank rate, its monthly-active-client count, and its average client age? Nubank discloses an 83% monthly activity rate. Bradesco discloses none of the three. Its 103.9m monthly transactions against ~110m customers implies roughly 0.94 transactions per customer per month versus Nubank’s ~2.5.
5. Is the affluent build net-new share or internal re-labelling? Bradesco has disclosed 3.1m “upgrades” in 2025 and a target of 4.7m Prime/Principal clients for 2026, but no net-new-affluent-client figure — a conspicuous omission for the programme it presents as its main growth engine.
6. LCR and NSFR are not disclosed in the 1Q26 press release; they should be pulled from the Pillar 3 / Risk Management report.

Numbers I could not verify from primary sources

7. The R$4.9bn Americanas provision (4Q22) is not named, quantified or attributed to Americanas anywhere in the FY2022 or FY2023 20-F. The only trace is a FY2023 20-F line referring to “some specific large corporate clients.” The R$4.9bn figure circulating universally in the press and in this memo’s price-action timeline is press- and IR-sourced, not filing-confirmed (the EDGAR copies of the 4Q22–4Q24 press releases are image-only). Cite it accordingly.
8. FY2020–FY2022 recurring net income and ROAE are secondary-sourced from the Brazilian financial press and should be re-verified against the archived quarterly releases before being relied on.
9. FY2025 IFRS net income is stated as R$23,925m in the 20-F and R$23,673m in aggregator data — a R$252m gap, most likely total versus parent-attributable. Do not mix bases across periods; the R$1.5bn (FY2024) and R$0.6bn (FY2025) divergences between the IFRS and managerial figures are larger still.
10. The SUSEP Jan–Apr 2026 premium ranking reportedly shows Porto (R$7.9bn) ahead of Bradesco Seguros (R$7.78bn), but the line scope is unclear (possibly property/liability only). Not relied on in the verdicts above.

Live regulatory and legal uncertainties

11. What is BCB’s committed launch window and pricing model for PIX Garantido / parcelado, and does it permit issuer-set credit spreads? This is the highest-impact unquantified regulatory risk in the thesis — it targets parcelamento, the fattest remaining consumer-credit pool.
12. What Open Finance credit-portability volumes have actually executed since February 2026? No reliable public series exists. Consent counts (128m) measure challenger acquisition activity, not consumer switching.
13. How much of the R$11.1bn IRPJ/CSLL 2012–2015 CDI-interest case was extinguished by the 1Q26 PTI settlement, and how much remains live? The Dec-2025 balance sheet predates the charge; the next disclosure point is the 2Q26 print or the FY2026 20-F.
14. What recurring effective tax rate is management guiding to for FY2026–27, and how much of the FY2023–25 negative IFRS rate was FX-hedge-driven (reversible) versus JCP/DTA-structural? Itaú has guided 29.5–32.5% for 2026. Bradesco has not. This is the question to ask on the 5-August call.
15. Why did 31 Form 3s and 6 Form 4s appear in 2026 after zero in 2021–2025? Foreign private issuers are exempt from Section 16 under Rule 3a12-3(b). Either Bradesco has lost FPI status — which would bring 10-K and DEF 14A obligations and materially better disclosure — or it began voluntary compliance. No 6-K explains it. Worth an IR call, and a genuinely useful development if it is the former.

The one that expires in four days

16. Q2 2026 prints on 5 August 2026. The specific things that would move this analysis: whether coverage falls another 5pp; whether MSME and rural NPLs keep rising; whether cost of risk holds at or above 3.5%; whether the loan book accelerates off its flat 1Q26 to make the +8.5–10.5% guide reachable; whether NII net of provisions clears the R$42bn low end it is currently running below; and what management says about the effective tax rate and the use of the R$10bn.

14. What Must Be True

Each proposition below carries a falsification test — a specific, observable, dated event that would prove it wrong. Where the next test date is the 5 August 2026 print, it is named.

14.1 For the bull case

Bull 1 — ROE settles durably at 17.5% or better, not 15.5%. This is the load-bearing assumption; nothing else in the bull case matters if it fails. At ~1.02× book the market already pays for 15.1–15.7%, so the entire return above the dividend depends on beating management’s own guide. It requires the cost line to move — closing half the efficiency gap to Itaú is worth ~R$6bn pre-tax, more than the whole 2023–25 earnings recovery.

Falsification test: four consecutive quarters to 2Q27 in which the 12-month efficiency ratio fails to fall below 47.0% (from 49.2% today), or in which recurring ROAE fails to print above 16.5% in any quarter. Either outcome caps ROE near guidance and the stock is worth roughly what it costs.

Bull 2 — the recurring effective tax rate holds near 20%. Roughly a third of recurring net income is a tax-structuring artefact resting on JCP deductibility and a R$109bn DTA stock. The corporate deduction survived LC 224/25 — that is the bull’s strongest ground.

Falsification test: Bradesco guides, or reports for any two consecutive quarters, a recurring effective tax rate above 27% — i.e. approaching the 29.5–32.5% Itaú has already guided to. That alone implies −26% to −31% with no operating deterioration. First observable on 5 August 2026: management has never given effective-tax-rate guidance, and being asked to is now unavoidable.

Bull 3 — the credit book is genuinely repaired and the 2025–26 vintages do not roll. The de-risking is real: secured mix +3.8pp to 60.8%, restructured book −20% y/y, problem assets −28%, Stage 1+2 at 92.9%, new-Stage-3 coverage up to 118%.

Falsification test — the sharpest single test in this memo, and it resolves in four days: in the 2Q26 print (5 August 2026), coverage falls a further 5 percentage points or more below 161.0% while MSME NPLs rise above 4.3% and cost of risk holds at or above 3.5%. That combination would establish the coverage decline as P&L smoothing rather than mix, and would put the fast-growing vehicle, rural and FGI/FGO vintages on a rolling trajectory.

Bull 4 — Bradsaúde is monetised, not just listed. The health platform is worth an estimated R$40–50bn against a group market capitalisation of ~R$183bn, and the 91.35% stake alone is ~25% of it. Today nothing has been sold.

Falsification test: eighteen months from the 5 May 2026 listing (i.e. by November 2027) no further Bradsaúde equity has been sold down, no special distribution has been made from it, and the SAUD3 float remains below 15%. At that point the “unlock” is a presentational reorganisation, not a capital-allocation event.

Bull 5 — the Selic falls toward Bradesco’s house view of 9.5% by end-2027. This relieves household debt service at record-low unemployment, reverses the expensive savings-to-time-deposit funding shift, and restores loan demand.

Falsification test: the BCB Focus median for end-2027 Selic remains at or above 12.00% through year-end 2026. It is 12.00% today — 150–250bp more hawkish than the assumption underpinning Bradesco’s own guidance — so this test is currently failing, and management has not reconciled the gap.

14.2 For the bear case

Bear 1 — the banking franchise never earns its cost of equity, and the market eventually reprices to that. Banking-ex-insurance ROE is ~13.1% against a Brazilian cost of equity plausibly 14–17%. The market currently implies a perpetual banking ROE of 12.0–12.8% for the stub — i.e. it already believes this.

Falsification test: Bradesco discloses a standalone banking-segment ROE (or segment capital allocation) above 15%, or delivers two consecutive quarters in which group ROAE exceeds 17% without the insurance contribution rising as a share of group profit. Either would show the bank itself, not the insurer, generating the improvement.

Bear 2 — the growth is in the wrong places and the 2025–26 vintages are the next problem. The fastest-growing lines are vehicle (+25.4%), rural (+17.5–24.0%) and MSME working capital (+16.3%) — precisely where management flags deterioration — and the MSME “share gain” sits in government-guaranteed FGI/FGO lines growing +81% y/y whose grace periods are expiring.

Falsification test: by 4Q26, rural and MSME NPLs are flat or lower than March 2026 levels (4.3% MSME) while those books continue to grow above 12% y/y, and management stops describing agribusiness as “still not back to normal.” Add: Bradesco discloses rural growth ex-Banco John Deere and it is above 12%.

Bear 3 — the insurance ROE is substantially rate carry that mean-reverts. 36% of 1Q26 insurance operating income was financial results at a 14.75% Selic; the industry health loss ratio is 82% against a ~70% break-even, meaning Brazilian health insurance in aggregate underwrites at a loss and survives on float.

Falsification test: Bradesco discloses an underwriting-only insurance ROE above 15%, or insurance ROAE holds above 20% across two consecutive quarters in which the Selic has fallen by 200bp or more. Absent disclosure, watch the insurance financial-results line as a share of operating income: sustained below 25% with ROAE still above 20% would falsify this.

Bear 4 — capital allocation does not improve, and the July raise is not the last one. The record is R$16bn on HSBC, R$2.15bn more of Cielo returning 1.55% in cash, R$1.68bn of buybacks against R$56.5bn of distributions, nothing sold in five years, and variable pay keyed to absolute net income with no return metric.

Falsification test: within eighteen months Bradesco does two of the following three: (a) adds an ROE, efficiency-ratio or per-share metric to the executive variable-compensation scheme; (b) sells or materially reduces the Cielo stake; © executes the unused 106.6m-share buyback authorisation in full. Any two would evidence a genuine change in capital discipline rather than a single good decision.

Bear 5 — the valuation offers no cushion, so the return is a rates-and-currency bet in disguise. At ROE ≈ CoE there is no margin of safety in either direction; ~45% of the bull return is re-rating rather than earnings; and with a Country:Brazil factor beta of 1.53 into an October general election, the currency is a large share of the outcome.

Falsification test: the BBD/ITUB price ratio rises above 0.55 (from 0.427 today, against a 0.383 trough and 1.03 five years ago) on the back of a disclosed operating improvement rather than a broad Brazil rally — i.e. BBD outperforms ITUB by more than 25% over four quarters while the Brazil country factor is flat or negative. That would demonstrate the market repricing Bradesco specifically, which is the one thing it has declined to do for twenty months.

14.3 The single question

Strip everything else away and the investment case reduces to one testable proposition, which both sides should be able to agree on:

Is ~15.5% ROE the ceiling of this franchise, or its floor?

At ~1.02× book the market says ceiling — and prices it accordingly, with no credit for the plan and no penalty for its failure. The bull case requires it to be a floor. Everything in Section 14.1 and Section 14.2 is a way of finding out which, and the first instalment arrives on 5 August 2026.

15. Source Appendix

(see Appendix B of the combined report)

APPENDIX A — Standard Diligence Questionnaire

Banco Bradesco S.A. (NYSE: BBD / B3: BBDC4, BBDC3) · Report date 2026-08-01

A standard due-diligence questionnaire applied to Bradesco, supplementing the analysis above. Answers are labelled FACT / INTERPRETATION / ASSUMPTION where it matters. Where a question does not map to a bank-and-insurer business model, the correct sector analog is given instead of forcing the generic answer.

General

What thoughtful questions have other investors asked about this company?

The recurring institutional questions, from the Q4-2025 and Q1-2026 calls, sell-side commentary and independent coverage, cluster into five:

1. “Is 15% the ceiling?” BofA’s Mário Pierry pushed management directly on the 4Q25 call about why FY2026 guidance implies less than one point of ROE improvement after a year that delivered +4.1pp. This is the single most-asked question and management has not answered it with a bridge.
2. “Is the coverage-ratio decline a release?” Coverage fell 183.1% → 161.0% over twelve months on a flat provision balance and an 8.4%-larger book. Analysts have asked whether this is mix (Stage 3 down, secured up) or P&L smoothing. Management points to new-Stage-3 coverage rising to 118% from 103%, which is a real rebuttal but not a complete one.
3. “Why raise equity now?” The 29-July R$10bn increase — at roughly book value, two days before a quarterly print, from a bank whose CEO had described capital as “so much more comfortable” eleven weeks earlier. The stated rationale is offensive (fund growth into a Selic-cut cycle); the timing invites the defensive reading.
4. “What is Bradsaúde actually worth, and will you sell more of it?” 91.35% retained with an 8.65% float; estimated at R$40–50bn against a group market cap of ~R$175bn; BTG marks it at 10–12× earnings versus Bradesco’s own ~7.6×. Investors want to know whether this is a value-realization programme or a one-off capital manoeuvre.
5. “Is the 40%-by-2028 efficiency target real?” The 12-month ratio has moved 52.2% → 50.0% in two years. Nobody outside the company appears to underwrite 40%.

INTERPRETATION: the questions the market is not asking are more interesting — nobody is pressing on the R$109.3bn deferred-tax-asset stock (61% of book equity), on the ~20% recurring effective tax rate against a 45% statutory rate, or on how much of the 21.6% insurance ROE is float carry at a 14.75% Selic. Those are the three places the reported numbers are most fragile.

Cyclicality and the Nature of Earnings

Are earnings at a cyclical high or low?

Both, in different places — which is the crux. FACT: group recurring ROAE of 15.8% (1Q26) is well below the 18.1% of FY2021 and far below the pre-2015 high-teens-to-20% era, so on a company basis earnings are recovering from a trough, not at a peak. But FACT: the Brazilian system earned a record R$255bn in 2025 at a 16.76% ROE on banking spreads of 15.17pp — the widest since the ICC series began in 2013 — with the Selic at 14.25–14.75%. INTERPRETATION: the company is at a cyclical low relative to its own history and the industry is at a cyclical high relative to its own. Those two facts partly cancel, and the resolution matters enormously: if system spreads compress as the Selic falls to a forecast 9.50% by end-2027, Bradesco has to deliver its ROE recovery into a shrinking pool.

Driven by the external environment or internal actions?

Genuinely mixed, and the split is quantifiable. Internal: the credit-book repair is real and self-generated — secured mix +3.8pp y/y to 60.8%, restructured portfolio −20% y/y, problem assets −28%, Stage 1+2 at 92.9%, 28% of branches closed since 2023. External and not repeatable: the market-NII hole from 2023 (−R$312m in 1Q23) has been fully repaired and now contributes only ~3% of NII; the record system spread; the float income at a 14.75% Selic. And a third category, which is neither: leverage. Assets/equity went 11.0× (FY2020) → 13.5× (Mar-2026). A meaningful share of the ROE recovery is balance-sheet expansion, which is why CET1 fell 100bp in a quarter and why the equity raise followed.

How stable are revenues?

More stable than the earnings, and the composition is improving. FACT: IFRS total revenue grew every year except FY2023 (82.4 → 102.4 → 108.8 → 98.3 → 105.3 → 117.8 R$bn, FY2020–FY2025). Net income was far more volatile (15.8 → 23.2 → 21.2 → 14.3 → 17.3 → 23.7), because the swing factor is provisions, not revenue. Recurring revenue quality is decent: 53.3% NII, 30.4% fees, 16.2% insurance — with insurance and pension technical provisions of R$455.2bn and R$1,489bn of AuM providing genuinely annuity-like income. The two fee lines in structural decline (checking accounts −6.9% y/y, collections and payments −7.0%) are the PIX-exposed ones.

Outlook for products/services? How big will this market be — growing, shrinking, domestic or international?

Almost entirely domestic Brazil (the international footprint — Miami/BAC Florida, Luxembourg, Bradesco Argentina — is immaterial to the thesis). The pools diverge sharply:

Business Quality and Competitive Moat

Is the industry getting more or less competitive?

More, and the pressure is coming from the regulator rather than from competitors. PIX (free, instant, central-bank-owned) destroyed transfer fees and demand-deposit float; Open Finance reached 128m active consents by January 2026 — the world’s largest — and added executable credit portability in February 2026; Crédito do Trabalhador opened private payroll lending onto a portable national rail. Bradesco’s own filings measure the result: demand-deposit share 11.0% → 6.6% in three years, private payroll share 12.0% → 6.6% in one year.

But there is a genuine counter-current that is under-appreciated. BACEN has made entry harder — fintech licensing applications have collapsed from ~15/month to ~2/month, Resolution BCB 494/2025 pulled the regularization deadline for unauthorized payment institutions forward from December 2029 to May 2026, and BACEN itself reports “accelerated movements of mergers, capitalizations and voluntary exits.” Simultaneously the CSLL restructuring taxes the challenger cohort up toward bank rates by 2028 while cutting the bank rate from 22% to 20%. The tax arbitrage that let payment institutions undercut banks is being narrowed by statute.

INTERPRETATION: competition is intensifying in quality while thinning in quantity. Five or six well-capitalized digital winners with permanent balance sheets — Nubank (deploying R$45bn in Brazil in 2026), Mercado Pago, C6 (43% ROE, JPMorgan 46%), Inter, PicPay — is arguably a worse environment for Bradesco than a hundred underfunded startups.

How profitable is the business (ROIC, ROE)?

ROIC is not a meaningful metric for a bank; ROE and ROA are the correct analogs, and for the insurance segment, ROAE plus the combined ratio.

INTERPRETATION, and it is the memo’s central claim: the group number is a blend that hides a 22%-ROE insurer and a ~13%-ROE bank. Management’s declaration at 4Q25 that ROE had exceeded cost of capital “for the first time” is a group-level statement that does not survive disaggregation.

How profitable is the industry — how many competitors, what barriers to entry?

Structurally very profitable: Brazilian banks averaged a 17.4% ROE over 16 years against 11.02% for US institutions. Concentration is high (top five = 59.3% of assets, 63.4% of deposits) and falling only slowly (RC4 credit share 60.6% in 2019 → 57.9% in 2024), with leakage going to credit cooperatives and non-bank institutions as much as to neobanks. Barriers to entry at the industry level are real and rising: BACEN authorization, capital rules, deposit-insurance and settlement access, and the collapse in licensing applications. The industry passes Greenwald’s barriers-to-entry test. Bradesco fails Greenwald’s market-share-stability test inside it.

Can the business be easily understood?

The banking half, yes. The insurance half, no — and the opacity is material rather than incidental. Bradesco does not disclose: capital allocated to each segment (so the 13% banking ROE is derived, not reported); the split of insurance ROE between underwriting and float income; Bradesco Saúde’s own loss ratio against the 82% industry figure; average client age; a principality or primary-bank rate; a monthly-active-client count; or the pension-sales attrition rate in municipalities where a branch was closed. Several of the most important numbers in this analysis had to be derived rather than read.

Can it be undermined by foreign low-cost labour?

No — banking and insurance in Brazil are domestically licensed and locally regulated, and the customer relationship is onshore. The correct analog of the low-cost-labour threat here is the low-cost-digital-operator threat, and on that measure the answer is emphatically yes: Bradesco’s derived cost-to-serve is ~US$9–10 per client per month against Nubank’s disclosed ~US$0.80 — a roughly 12× gap — while its revenue per client is only ~1.8× Nubank’s ARPAC.

Do brands matter?

In insurance, consortia and among older/interior demographics, yes — management itself cites brand as its stated competitive advantage in consortia and pensions. In banking, the brand is not converting to pricing power: Nubank is the strongest banking brand in Brazil by Brand Finance’s brand-strength index, Bradesco’s total client base is smaller today (74.3m) than in 2022 (77.1m), and the fee lines most exposed to competition are shrinking ~7% a year. A brand is a moat only if it supports price.

What is the nature of competition?

Three distinct games. In mass retail it is a pure cost game and Bradesco is losing it. In wholesale, affluent and corporate it is a relationship-and-balance-sheet game where Bradesco holds share (MSME share 14.3% → 16.6%; large-corporate NPL 0.2%). In insurance and health it is a scale-capital-and-distribution game with genuine regulatory barriers, where Bradesco is the leader at 22.8% share — though Porto Seguro matches its ROE (22.7% FY2025) at a fraction of its scale, so Bradesco is bigger, not better.

Customers’ switching costs?

Real only where a contract, not a habit, binds: payroll/INSS direct-credit arrangements, mortgages (360-month average term, 51.2% stock LTV), corporate cash management, and — the strongest — insurance and pension policies with surrender penalties and tax lock-ins (R$385bn of the group’s R$455bn of technical provisions, 93% branch-originated). For the mass-retail client, post-PIX and post-Open-Finance, switching costs are essentially zero. The captive book is far smaller than the 74m-client headline implies, and its centre of gravity sits inside the insurance company.

Financial Condition and Balance Sheet

Assets not fully recognized on the balance sheet?

Yes, and it is the most important hidden value in the group. Bradesco Seguros and the Bradsaúde health platform are carried at book, not at market. Bradsaúde — created February 2026 by consolidating Bradesco Saúde, Odontoprev and Atlântica Hospitais, listed on B3’s Novo Mercado as SAUD3 on 2026-05-05 with 91.35% retained — brings R$52bn of revenue, R$3.6bn of net profit, a 24% ROE and >13m beneficiaries, and is estimated at R$40–50bn against a group market capitalization of roughly R$175bn. BTG marks it at 10–12× earnings versus Bradesco’s own ~7.6×. Also unrecognized at fair value: the 25% stake in Grupo Fleury and Atlântica Hospitais’ 3,600 beds. INTERPRETATION: the Bradsaúde listing exists precisely to make this visible, which is management conceding the point.

Off-balance-sheet liabilities?

The standard bank set — loan commitments, guarantees and standby letters of credit — plus the material Brazilian-specific item: labour, civil and tax contingencies, of which the “possible but not provisioned” tax portion is typically large for a Brazilian bank. (Quantification is carried in the memo’s filings-sweep material; the FY2025 20-F contingent-liabilities note is the source of record.) Also relevant: insurance technical provisions of R$455.2bn are on-balance-sheet but are a claim on future underwriting experience, and their adequacy depends on actuarial assumptions the company does not fully disclose.

How conservative is the accounting?

Mixed, and trending in opposite directions on the two most important lines.

More conservative: CMN Resolution 4,966 (effective 1-Jan-2025) replaced the 1999-vintage incurred-loss model with an IFRS-9-equivalent expected-loss model — provisions are now model-driven rather than rating-table-driven, which removes managerial discretion to smooth. That is a quality-of-earnings improvement (and a reported-volatility increase). It also breaks comparability across 1-Jan-2025 — no uncorrected five-year provision, NPL or coverage trend through that date is valid.

Less conservative: coverage fell 22pp in twelve months (183.1% → 161.0%) on a flat provision balance against an 8.4%-larger book, at the precise moment MSME NPLs rose 60bp and cost of risk jumped 50bp q/q to 3.5%. Roughly 60% of that decline is defensible mix (Stage 3 down 8.0% → 7.1%, secured share up 3.8pp, restructured book down 20%); roughly 40% reads as cushion release.

And management’s presentation is consistently flattering. Three different net-income numbers exist for 1Q26 — recurring R$6,811m, book R$5,030m, IFRS R$5,178m. Management leads with recurring everywhere, and the headline “+16.1% y/y, ninth consecutive quarter of growth” is the recurring number. On the book number, 1Q26 net income fell 13.3% y/y. Separately, 4Q25 carried a −R$661m branch-restructuring provision excluded from recurring earnings — so recurring earnings exclude the cost of the restructuring that produces the cost savings that flatter recurring earnings.

And the equity itself is low-quality. Net deferred tax assets of R$109.3bn are 61% of book equity; prudential adjustments of R$59.9bn are 34.5% of stated equity and are deducted from CET1. Stripping intangibles (R$26.6bn) and net DTAs leaves “hard” equity of roughly R$43bn — about 24% of stated book value. The reported 13.6% IFRS ROE is earned on a capital base that is substantially an accounting claim on future tax payments.

How CapEx-hungry is the business?

Not capital-expenditure-hungry in the industrial sense — the correct analog for a bank is regulatory capital consumption plus technology investment, and on both measures the answer is “more than it looks.”

• Regulatory capital: RWA grew from R$1,035.9bn (Mar-25) to R$1,152.5bn (Mar-26), +11.3%, against loan growth of 8.4% — inflated by an operational-risk RWA jump of R$22.4bn (+19.5% q/q) on a regulatory change. CET1 fell 100bp in a single quarter to 10.2%, and the July raise adds ~90bp.
• Technology: data processing and communications expense rose +29.8% y/y in 1Q26 and is the explicit strategy. Branch-closure savings are being redeployed into technology, not banked — which is why the efficiency ratio has moved only 2.2pp in two years despite closing 28% of branches.
• The honest summary: the business needs equity to grow. It just raised R$10bn to prove it.

Capital Allocation and Management

(See memo Section 7 for the full treatment; the answers below summarise it.)

How much free cash flow does the business generate?

“Free cash flow” is not a meaningful concept for a bank — operating cash flow is dominated by changes in deposits, loans and trading assets, and is not a claim on shareholders. The correct analogs are (a) recurring net income, (b) capital generated above the regulatory minimum, and © distributable earnings. On that basis: FY2025 recurring net income was R$24.7bn, of which R$12.4bn (~50%) was distributed as dividends and JCP. But capital generation was insufficient to fund growth: CET1 still fell 100bp in 1Q26 and the company raised up to R$10bn of new equity in July 2026. INTERPRETATION: Bradesco is currently distributing roughly half its earnings and simultaneously issuing equity — which is, in net terms, a bank whose organic capital generation does not cover its growth ambition.

Significant acquisitions recently? Buying back shares? Issuing large amounts of new shares to insiders? Compensation policy? Motivations of management?

(Full detail in memo Section 7. In summary:)

Acquisitions. The decade’s three defining decisions, ranked: Bradsaúde (2026) — value-creative, the only good one, consolidating the group’s best assets into a separately-listed R$46.6bn platform at no cash cost while banking ~250bp of CET1; Cielo take-private (2024) — ~R$2.15bn of Bradesco’s money to lift its stake from 31.41% to 50.72% of a merchant acquirer whose profit pool PIX is expropriating, and which returned R$124.0m of cash in FY2025, a 1.55% yield against a 14.75% Selic, down 18.2% y/y; and HSBC Brasil (2015–16) — US$5.186bn / ~R$16bn cash for scale in a business where Bradesco already had scale and no cost advantage, generating R$4.2bn of goodwill, R$2.3bn amortised in 2017, and a network of which 58% has since been closed with no loss of market share. The 20-F states flatly: “There have been no recent divestitures.” Nothing has been sold in five years.

Buybacks. Immaterial. R$1,682m of cash across five years — 3.0% of the R$56.5bn distributed as JCP over the same period. The bonus-adjusted share count fell just 1.1% over six years (10,691m → 10,570.7m) and is about to rise +4.5%. Execution was tactically good — 31.1m shares for R$348.6m at ~R$11.2, at the exact five-year low in December 2024/February 2025 — but a 106.6m-share authorisation from May 2025 sits ~6% used with three months to run. Bradesco is not a share-count-reduction story.

Issuance. The July 2026 increase issues 604,852,753 shares (+5.72% on the corrected 10.57bn count) at a blended R$16.53 — 0.97× the corrected book value of R$17.01, i.e. at book, not below it. Because rights trade on B3, economic dilution to a non-participant is ~0.3%. Controlling shareholders committed R$8bn firm, and R$6.5bn of already-declared JCP was accelerated to the settlement date so holders can self-fund — net new external capital is closer to R$3.5bn than R$10bn.

Compensation. Variable pay for both board and executives keys off “accumulated net income” — absolute profit, with no ROE, efficiency-ratio or per-share metric anywhere in the scheme. There is no option plan and no performance-share plan. The approved 2026 maximum is R$1,443m, up 16.4% y/y, in a year whose guidance implies less than one point of ROE improvement. This is misaligned in precisely the wrong direction: a management paid on absolute profit is rewarded for growing the balance sheet and issuing equity.

Motivations and control. The ADR is over the non-voting preferred. Control runs through the Cidade de Deus / Fundação Bradesco pyramid — the foundation holds 31.53% of total capital, its managing board is composed of Bradesco’s own directors and long-serving officers, and only 4 of 11 directors are independent. The foundation’s 31.53% stake generated a claim on ~R$4.57bn of FY2025 gross JCP against an annual charitable budget of R$1.4bn — roughly 3.3× its spending need. That is the structural reason the payout was sustained at ~59% straight through the 2023 earnings trough rather than cut, and it is the best single explanation of Bradesco’s capital allocation. The 20-F itself concedes the risk: decisions on “acquisitions, divestitures, financings or other transactions could be made by Fundação Bradesco and our Board of Directors.”

Insider behaviour. Gross insider buying of R$197m looks bullish until you strip out R$191.8m of compulsory conversion of variable pay into restricted shares. Discretionary trades net to R$5.5m bought against R$38.2m sold — net selling of R$32.7m, accelerating into the rally. The controlling holdcos have not bought a share on the open market in twenty months. One genuine counter-signal: director Júlio César Bueno bought 110,400 shares on the open market on 1 July 2026, four weeks before the raise.

Valuation and Market Data

Is the stock an ADR, MLP, or K-1 issuer?

ADR — and the share-class detail matters. NYSE: BBD is an ADR over the preferred (BBDC4) shares at a 1:1 ratio (parity set March 2004); NYSE: BBDO is the ADR over the common (BBDC3). The preferred is non-voting. Bradesco is a foreign private issuer: it files 20-F annually and 6-K for quarterly results and material events — there is no 10-K, no 10-Q, no 8-K and no Form 4 insider-transaction stream. No MLP, no K-1. Brazilian withholding applies to distributions, and JCP withholding rose from 15% to 17.5% on 1 January 2026.

Dividend policy?

Monthly interest on capital (JCP) plus periodic dividends and top-ups. FY2025 total shareholder remuneration was R$12.4bn (FY2024 R$9.0bn; FY2023 R$7.5bn) — roughly a 50% payout of recurring earnings, up from 46% in 2023–24. FY2025 dividends plus JCP were R$1.339 per share ≈ $0.264 per ADR, and the trailing-twelve-month cash yield is 8.15% on a $3.61 ADR.

But the yield is a variable JCP stream, not a coupon, and it has been halved twice in six years. Cash per ADR by calendar year: 2019 $0.501 · 2020 $0.138 · 2021 $0.191 · 2022 $0.086 · 2023 $0.339 · 2024 $0.148 · 2025 $0.306 · 2026 YTD $0.151. Cumulative 2019–2025 distributions of $1.709/ADR ≈ 47% of today’s price — and total return since January 2020 is still −40.7%. The payout is visibly pro-cyclical: cut in the 2022 credit blow-up and again in the 2024 trough, then rebuilt. Note also the R$6.5bn of JCP accelerated to 15 September 2026 so holders can recycle the cash into the July capital increase — elegant, but it does not change the economics of a below-book issue.

How profitable is the business?

(Answered above under Business Quality.) Group recurring ROAE 15.8%; insurance 21.6%; banking ex-insurance ~13.1%; audited IFRS ROE 13.6% (FY2025).

Is net income diverging from cash from operations?

The conventional test does not apply cleanly to a bank. The meaningful divergence tests for this issuer are three, and Bradesco fails or partly fails all three:

1. Recurring vs. statutory net income. 1Q26 recurring R$6,811m against book R$5,030m — a 26% gap, from PTI tax-settlement adherence. Management leads with recurring.
2. Reported earnings vs. cash taxes. The IFRS effective tax rate was negative in FY2023, FY2024 and FY2025 (−13.8% in FY2025) and then flipped to +33.2% in 1Q26. A ~20% recurring rate against a 45% statutory rate means roughly R$2.2bn per quarter — about 32% of recurring net income — is a tax-structuring artefact resting on JCP deductibility and a R$109.3bn net DTA stock.
3. Earnings vs. capital generation. The clearest divergence of all: nine consecutive quarters of recurring earnings growth, and CET1 fell 100bp in a single quarter while the company raised R$10bn of fresh equity. Reported profit is not converting into distributable capital at the rate the headline implies.

Risks and Downside

What factors would cause the stock to decline?

In rough order of probability × impact: (1) a Q2/Q3 credit print showing coverage falling further while MSME and rural NPLs keep rising; (2) evidence that ROE stalls at 15–16% and the efficiency ratio does not move, confirming the market’s implied “mediocrity achieved” expectation as a ceiling rather than a floor; (3) Selic normalization compressing both the record 15.17pp system spread and the 36% of insurance operating income that is float carry; (4) further narrowing of the JCP tax shield; (5) an election outcome that reopens fiscal, rate and sector-intervention risk, or simply a BRL reversal from ~5.07 toward the R$5.35–5.40 house forecast, which hits the ADR without touching a single operating line; (6) a second capital raise.

Risk of a catastrophic loss?

Low but not zero, and the historical template exists. The realistic severe case is not insolvency but a multi-year de-rating with a halved dividend — precisely the 2022–24 experience: a −52% peak-to-trough total-return drawdown (April 2022 → December 2024), a payout cut from $0.339 to $0.148 per ADR, and two separate single-day idiosyncratic losses above 15% (−18.6% on 9-Nov-2022; −15.5% on 7-Feb-2024). The ten-year total return is −0.9% cumulative with a −75.6% ten-year maximum drawdown and a five-year Sharpe of 0.006. This security has demonstrated the capacity to destroy half an investor’s capital while remaining entirely solvent.

The genuine tail risks: a Brazilian sovereign or fiscal crisis (the deficit breached 7% of GDP in 2025); a systemic corporate-credit event of the Americanas type (which cost R$4.9bn in a single provision in 4Q22); or a DTA impairment, since net DTAs at 61% of book equity depend on sustained future taxable profits.

Chance of a total loss?

Very low. Bradesco is a systemically important, 90-year-old institution with R$2,435bn of assets, R$180bn of equity, a comfortable loans-to-funding ratio of 85.0%, a Total (BIS) capital ratio of 14.9% against an 8.00% CET1 minimum, diversified insurance earnings, and a controlling shareholder that has just demonstrated its willingness to inject R$8bn of fresh equity. Brazilian regulators have a demonstrated forbearance record. In Deutsche Bank’s 2011 two-year severe-recession stress test Bradesco screened as the most capital-resilient of Brazil’s big three. The correct framing is de-rating and dilution risk, not solvency risk.

Recent News and Events

Has the business environment changed recently?

Yes, in four distinct ways within eighteen months:

1. Accounting. CMN Resolution 4,966 (1-Jan-2025) replaced a 26-year-old incurred-loss provisioning regime with an expected-loss model — breaking comparability and front-loading credit recognition into the worst household-credit cycle on record.
2. Tax. JCP withholding rose to 17.5% and the base was narrowed (1-Jan-2026); CSLL was restructured (April 2026) — cutting banks from 22% to 20% while stepping the fintech cohort up toward 15–20% by 2028.
3. Rates. The Selic peaked at 15.00% (June 2025) and Copom began cutting in March 2026 — 14.75% (18-Mar), 14.50% (29-Apr), 14.25% (June) — with the easing cycle signalled to be near its end as inflation breached the top of the target band. Bradesco’s house forecast is 12.50% end-2026 and 9.50% end-2027.
4. Competition. Open Finance credit portability went live (February 2026); Nubank overtook Bradesco in the BACEN customer ranking (January 2026); PIX Automático extends to salary accounts from October 2026.

Significant acquisitions?

Bradsaúde (February 2026) — the consolidation of Bradesco Saúde, Mediservice, Odontoprev and Atlântica Hospitais into a single listed health platform (B3 Novo Mercado, SAUD3, first trade 2026-05-05), with 91.35% retained and an 8.65% float, adding +250bp to pro-forma CET1. And Cielo (2024) — the joint take-private with Banco do Brasil of the merchant acquirer. (Both assessed in memo Section 7.)

Change in accounting policies?

Yes — Resolution 4,966, above. This is the single most important accounting change in Brazilian banking since 1999 and it invalidates any uncorrected multi-year credit trend line spanning 1 January 2025.

Recent changes — new markets, facilities, management?

• Management: Marcelo Noronha became CEO in January 2024 and launched a five-year (2024–28) transformation plan in February 2024, developed with McKinsey — modernise technology, delayer the hierarchy (three layers removed), rebuild commercial segmentation, restore credit quality, restore profitability to cost of capital. The market’s immediate verdict was a −15.5% single-day ADR move on 7-Feb-2024 against Itaú’s −1.4%, reading a long and expensive fix.
• Facilities: branches down from 2,695 (2023) to 1,938 (Mar-2026) — −28%; own ATMs down 36% across 2023–2025; headcount down to 82,095. 4Q25 carried a −R$661m restructuring provision for the programme.
• New segments: Bradesco Principal launched November 2024 (high-net-worth, ~70 offices, 320k clients), alongside a re-tooled Prime (2.3m clients) and 3.1m affluent “upgrades” in 2025 with +1.5m planned for 2026.
• Capital: the 29-July-2026 capital increase of up to R$10bn at R$15.43/ON and R$17.64/PN, with an R$8bn controlling-shareholder commitment and a record date of 4 August 2026 — announced two days before this report date.
• And the immediate next event: Q2 2026 results print on 5 August 2026, four days after this report date. Everything in this appendix is pre-print.

APPENDIX B — Source Appendix

Banco Bradesco S.A. (NYSE: BBD / B3: BBDC4, BBDC3) · Report date 2026-08-01 All sources accessed 2026-08-01 unless otherwise stated.

Sources are grouped by authority: primary regulatory filings first, then company disclosure, then regulators and statistical agencies, then market data, then secondary press and third-party research. Where a figure in the memo rests on a secondary source, that is stated explicitly in the body.

B.1 Primary regulatory filings — SEC (CIK 0001160330)

Bradesco is a foreign private issuer: it files Form 20-F annually and Form 6-K for quarterly results and material events. There is no 10-K, 10-Q or 8-K. Section 16 filings (Forms 3/4) appeared for the first time in April 2026 (see B.1.4).

B.1.1 Annual reports on Form 20-F. All available from SEC EDGAR.

Canonical URL pattern: https://www.sec.gov/Archives/edgar/data/1160330/.... FY2025: .../000129281426001753/bbdform20f_2025.htm.

B.1.2 Quarterly results (Form 6-K).

• 1Q26 press release, furnished 2026-05-07 — .../000129281426002869/bbdpr1q26_6k.htm. The single most-used document in this memo: NII split, asset-quality tables, expanded loan portfolio, fee lines, Bradesco Seguros income statement, operating expenses, Basel III capital table and CET1 bridge, funding, service network, FY2026 guidance, “Economic Perspectives” (Selic/IPCA/GDP/FX house forecasts).
• 1Q26 Report on Economic and Financial Analysis (“the book”), 6-K 2026-05-07 — .../000129281426002862/bbdbook1q26_6k.htm. Ratings table (p.39).
• 4Q25 / FY2025 press release, furnished 2026-02-06 — .../000129281426000261/bbdpr4q25_6k.htm. FY2025 managerial revenue lines; recurring-to-book reconciliation and the non-recurring item table.

B.1.3 Material-event 6-Ks.

B.1.4 Section 16 filings. 31 Forms 3 (April/May 2026) and 6 Forms 4 — the first ever; zero in 2021–2025. Foreign private issuers are exempt under Rule 3a12-3(b), and no 6-K explains the change. Transactions used in Section 7.8. Flagged as an open question.

B.1.5 Corpus. The full trailing-60-month filing history (2021-08-01 onward) was enumerated and reviewed from SEC EDGAR.

B.2 Company disclosure — transcripts and investor material

• Q4-2025 earnings call, 2026-02-06 — full transcript retrieved and archived. Source of the “ROAE of 15.2% exceeding our cost of capital for the first time” statement and the 40%-by-2028 efficiency target. Corroborating public coverage: Investing.com transcript, https://www.investing.com/news/transcripts/earnings-call-transcript-banco-bradesco-q4-2025-sees-revenue-beat-stock-down-93CH-4491239; Seeking Alpha, https://seekingalpha.com/article/4867236-banco-bradesco-s-a-bbd-q4-2025-earnings-call-transcript.
• Q1-2026 earnings call, 2026-05-07 — not available in the ROIC.ai transcript corpus (verified: the tool returns “No earnings call is available for Q1 2026”), and absent from Seeking Alpha, StockAnalysis and Motley Fool. The full ~13,700-word transcript (prepared remarks plus a ten-question Q&A) was retrieved from the AlphaSpread archive and archived. Source of the two most consequential management quotes in this memo: the CET1 “so much more comfortable” passage and the “15.8% against a cost of capital that should range close to 18%” exchange with Goldman Sachs. Both are flagged in the body as interpreted-English transcript text that should be put to management directly.
• No standalone 2024-02-07 investor-day transcript exists publicly; the Q4-2025 call is the de-facto strategy document for the five-year plan.
• Bradesco Investor Relations, https://banco.bradesco/ri — Central de Resultados.
• Bradesco Integrated Report 2017 (HSBC goodwill amortisation).

B.3 Regulators, central banks and statistical agencies

Banco Central do Brasil (BCB / BACEN)

• Nota de Política Monetária e Operações de Crédito, June 2026 — SFN credit stock R$7.356tn, +9.7% y/y, 55.8% of GDP.
• Copom statements and interest-rate decisions — https://www.bcb.gov.br/en/monetarypolicy/copomstatements, .../interestrates. Selic path.
• PIX statistics — https://dadosabertos.bcb.gov.br/en/dataset/pix.
• Open Finance — https://www.bcb.gov.br/en/pressdetail/2462/nota.
• Drex / Digital Brazilian Real — https://www.bcb.gov.br/en/financialstability/digital_brazilian_real.
• Relatório de Economia Bancária / Relatório de Estabilidade Financeira (RC4 concentration, 2H2024), as reported by Poder360, https://www.poder360.com.br/poder-economia/bancoes-detem-579-das-operacoes-de-credito-no-pais/.
• System NPL series (May 2026): total 4.7%, non-earmarked 6.2%, unsecured personal 14.2%, vehicle 6.5%, private payroll 7.9%.
• BCB Focus survey, 20-Jul-2026 — Selic 14.00% (end-2026) / 12.00% (end-2027); IPCA 5.15% / 4.20%. Via https://conteudos.xpi.com.br/economia/boletim-focus-ipca-2026-segue-em-queda-mesmo-com-alta-do-petroleo-20-07-2026/.

SUSEP / ANS / CNseg / Fenaprevi — insurance and health market data.

• SUSEP boletins via Revista Apólice, https://revistaapolice.com.br/2026/05/susep-setor-arrecada-r-1061-bilhoes-ate-marco/ and .../2026/07/susep-divulga-boletim-com-dados-do-mercado-ate-maio-de-2026/; SUSEP, https://www.gov.br/susep/pt-br/central-de-conteudos/noticias/2026/julho/....
• CNseg 2025 sector data (R$764.5bn) via Revista Apólice, https://revistaapolice.com.br/2026/03/setor-segurador-movimenta-r-7645-bilhoes-em-2025/; CNseg, https://cnseg.org.br/noticias/....
• ANS beneficiary and operator data (53.08m beneficiaries, 668 operators, 82% loss ratio) via Conjur, https://www.conjur.com.br/2026-jun-12/brasil-tem-668-operadoras-de-plano-de-saude-e-53-milhoes-de-beneficiarios/; ANS economic-financial data, https://www.gov.br/ans/pt-br/assuntos/noticias/numeros-do-setor/....
• IESS VCMH medical-cost-inflation series, https://iess.org.br/taxonomy/term/586; 2026 VCMH of 15.1% via Bradesco Saúde Empresas, https://www.bradescosaudeempresas.com.br/blog/reajuste-plano-de-saude-empresarial-2026.

Other Brazilian official sources

• Ministério do Trabalho e Emprego, Crédito do Trabalhador, Jan-2026 — https://www.gov.br/trabalho-e-emprego/pt-br/noticias-e-conteudo/2026/janeiro/credito-do-trabalhador-fortalece-inclusao-financeira-e-supera-r-101-bilhoes-em-operacoes.
• Casa Civil, Novo Desenrola, May-2026 — https://www.gov.br/casacivil/pt-br/assuntos/noticias/2026/maio/...; Agência Brasil, https://agenciabrasil.ebc.com.br/economia/noticia/2026-05/desenrola-20-renegocia-quase-r-1-bilhao-em-dividas-diz-durigan.
• IBGE/PNAD unemployment via TradingEconomics.
• Senado Notícias, May-2026 — https://www12.senado.leg.br/noticias/infomaterias/2026/05/dividas-em-recorde-assombram-as-familias-brasileiras.
• DIEESE Síntese Especial nº23, May-2026 — https://www.dieese.org.br/sinteseespecial/2026/sinteseEspecial23.html.
• FGV/IBRE Boletim Macro nº176, Mar-2026 — https://portalibre.fgv.br/sites/default/files/2026-03/2026-03-boletim-macro-1.pdf.
• CNC Peic household-indebtedness survey (81.6% indebted, 29.9% unable to pay, Jun-2026).
• FEBRABAN spread-composition study, Oct-2025 — https://cmsarquivos.febraban.org.br/Arquivos/documentos/PDF/Composição%20Spread_Site.pdf; FEBRABAN, https://portal.febraban.org.br/noticia/4253/pt-br.
• OECD Economic Outlook 2026/1, Brazil chapter — https://www.oecd.org/en/publications/2026/06/oecd-economic-outlook-volume-2026-issue-1_8be0dba6/full-report/brazil_f971a312.html.

Tax and accounting legislation

• LC 224/2025 (26-Dec-2025) — CSLL restructuring effective 2026-04-01; JCP withholding 15% → 17.5%. Via Econet, https://blog.econeteditora.com.br/novas-regras-jcp-base-calculo-irrf-2026/; B3/Bora Investir, https://borainvestir.b3.com.br/noticias/empresas/entenda-aumento-da-taxacao-do-jcp-aprovado-no-senado-e-impacto-para-empresas-e-investidores/.
• MP 1.303/2025 — CSLL ladder for payment institutions and fintechs. Via Genial, https://analisa.genialinvestimentos.com.br/acoes/b3/mp-1303-25-aumento-de-impostos-adquirentes-financeiras-capitalizacao-bolsas-e-fintechs-sao-os-mais-afetados/; FENACON, https://fenacon.org.br/noticias/novas-aliquotas-da-csll-para-o-setor-financeiro-passam-a-valer-em-abril/.
• EC 132/23, LC 214/2025, LC 227/26 — consumption-tax reform; financial-services rates 10.85% (2027) → 12.50% (2033). Per the FY2025 20-F Item 3.D.10.01-07.
• Lei 14.789/2023 (JCP base narrowing, effective 2024); Lei 14.689/2023 (voto de qualidade).
• Ministry of Finance Ordinance 1,384/2024 — Programa de Transação Integral (PTI).
• CMN Resolução 4.966/21 and BCB Resolution 352 — expected-loss provisioning, effective 2025-01-01. Via KPMG Brasil, https://kpmg.com/br/pt/home/insights/2025/04/instituicoes-financeiras-harmonizacao-ifrs-9.html; Deloitte, https://www.deloitte.com/br/pt/services/audit-assurance/analysis/cinco-pilares-resolucao-4966.html.
• CMN Resolução 4.820/20 — the 2020 distribution cap.
• Resolution BCB 494/2025 — payment-institution regularization deadline. Via Finsiders Brasil, https://finsidersbrasil.com.br/regulamentacao/pedidos-de-abertura-de-fintechs-no-bc-caem-com-novas-regras/.
• CVM Resolução 44 (trading windows); CVM Instrução 358 (insider disclosure); CVM Opinion 35 (related-party transactions).
• Lei 6.404/76 Art. 170 Section 1-III — the statutory basis for the capital-increase pricing.
• STF ruling of 2025-05-24 on the Bresser/Verão/Collor economic plans, with a 24-month adhesion window from 2025-06-03.

B.4 Market and quantitative data

• AZI price history — https://azitrading.com/controls/download-data.php?t=BBD (and t=ITUB), full history through 2026-07-31; archived. Source for the five-year event map, total returns, EMAs, beta, alpha, realised volatility and the ADR dividend series. AZI valuation_index percentile ranks were pulled and only pe_percentile (54.0) and ps_percentile (26.3) used — the pb_percentile and composite are contaminated by a book-value-per-share input wrong by roughly 6×.
• FactorsToday — /api/stock-loadings/{BBD,ITUB}, /api/leaderboard/, /api/stock-info/, /api/stock-specific-vol/, /api/related-stocks/, /api/factor-returns/historic and /intraday, all pulled 2026-08-01. Methodology at https://www.factorstoday.com/about. Third-party statistical estimates, not primary. Note: rs_peak reads −99.97% for BBD and −99.91% for ITUB — both are artefacts of corrupt 2001–02 values in the adjusted series and are not used.
• ROIC.ai MCP — get_income_statement, get_balance_sheet, get_profitability_ratios, get_per_share_data, get_valuation_multiples, get_enterprise_value, get_company_news, list_earnings_calls. Third-party aggregated data, not primary. Four documented failures on this issuer, all disclosed in the body: (i) the share count is understated by ~14%, which is the origin of the “trades below book” error; (ii) return_com_eqy and eps deduct total shareholder remuneration as “preferred dividends,” misreading the ON/PN dual-class structure; (iii) enterprise_value is meaningless for a bank; (iv) get_company_news returns nothing for NYSE:BBD — the bare ticker BBD is required, and returns only nine thin items for the window.
• StatusInvest (B3 snapshots, 31-Jul-2026 closes) and stockanalysis.com (ADR data, 2026-07-31) — peer comparison table. Note: the StatusInvest screened P/B of 1.09× proved closer to the corrected figure than any vendor-share-count “reconciliation.”
• ANBIMA indicative curve — NTN-B 2035 real yield 8.16% (27-Jul-2026), via https://fidcs.com.br/rendafixa-publico/ntn-b; end-2025 comparison via InfoMoney/Quantum.
• TradingEconomics — Brazil 10-year nominal government bond yield 14.77% (31-Jul-2026), https://tradingeconomics.com/brazil/government-bond-yield.
• Damodaran country risk premia — https://pages.stern.nyu.edu/\~adamodar/New_Home_Page/datafile/ctryprem.html. The page served the January-2026 vintage (Brazil total ERP 7.47%) when fetched. A July-2026 figure of 7.74% appears in third-party summaries and is not verified against the source workbook.
• FX: https://www.exchangerates.org.uk/USD-BRL-spot-exchange-rates-history-2025.html and https://www.exchange-rates.org/exchange-rate-history/usd-brl-2025.
• Short interest: 41,838,178 shares as of 2026-03-13, +245.3% m/m, 0.4% of shares outstanding — https://www.defenseworld.net/2026/03/27/banco-bradesco-sa-nysebbd-short-interest-update.html.
• Credit ratings: Moody’s, S&P and Fitch, per the 1Q26 Report on Economic and Financial Analysis ratings table; Fitch outlook revision to Stable, 2025-12-12, via https://www.marketscreener.com/news/fitch-revises-brazilian-lender-bradesco-s-rating-outlook-to-stable-affirms-idrs-at-bb-ce7d50d8da8df52c.
• RankingsLatAm — Brazilian bank asset shares, March 2026, https://rankingslatam.com/blogs/industry-news/brazil-banking-and-financial-institutions-ranking-by-total-assets-market-leaders-and-asset-concentration-analysis-in-2026.
• BACEN customer rankings (Q4-2025 and Q1-2026) via Brazil Stock Guide, https://brazilstockguide.com/insights/brazil-banks-customer-ranking-2026/, and OpenPR, https://www.openpr.com/news/4508985/brazil-retail-banking-market-report-2026-nubank-surpasses.

B.5 Peer and competitor primary sources

• Itaú Unibanco 1Q26 Form 6-K (Brazil 12-month efficiency 36.2%) — https://www.stocktitan.net/sec-filings/ITUB/6-k-itau-unibanco-holding-s-a-current-report-foreign-issuer-c2c7481a9db6.html.
• Santander Brasil 1Q26 Form 6-K — https://www.stocktitan.net/sec-filings/BSBR/6-k-banco-santander-brasil-s-a-current-report-foreign-issuer-6802d7095f59.html.
• Nu Holdings FY2025 results (2026-02-25), https://international.nubank.com.br/company/nu-holdings-ltd-reports-fourth-quarter-and-full-year-2025-financial-results/; Q1-2026 (2026-05-14), https://www.businesswire.com/news/home/20260514553909/en/Nu-Holdings-Ltd.-Reports-First-Quarter-2026-Financial-Results; FY2025 Form 20-F; and the R$45bn 2026 Brazil investment announcement, https://international.nubank.com.br/company/nubank-to-invest-r-45-billion-in-brazil-in-2026/.
• Banco Inter Form 6-K 1Q26 (SEC CIK 0001864163).
• Banco do Brasil 1Q26 MD&A and FY2026 guidance cut, via Seu Dinheiro, https://www.seudinheiro.com/2026/empresas/lucro-do-banco-do-brasil-bbas3-cai-no-1t26-e-roe-rentabilidade-recua-balanco-miql/, and Safra, https://oespecialista.safra.com.br/banco-do-brasil-guidance-1t26-bbas3/.
• Cielo OPA edital summary — https://www.b3.com.br/data/files/0A/A5/48/68/98CB09105FE89209AC094EA8/Resumo%20do%20Edital%20-%20OPA%20Cielo%20_CIEL_.pdf.
• Porto Seguro, BB Seguridade and Caixa Seguridade results via Genial Analisa, InfoMoney and A Revista; Sincor-SP Ranking das Seguradoras 2025, https://sincor.org.br/wp-content/uploads/2026/05/ranking_das_seguradoras_2025_web.pdf.

B.6 Comparative and framework research

(Peer and sector research used for cross-read, comparative framing and cost-of-equity anchors — never as a substitute for primary research on Bradesco itself. Peer figures cited in the body are sourced to the peers’ own filings, listed in B.5.)

Comparative work drawn on for this report covers Itaú Unibanco (the efficiency and ROE gap, PIX and Open Finance quantification, and the Brazilian Gordon-growth framework), Nu Holdings (cost-to-serve, ARPAC, activity rate and the share-taking mechanism), Santander Brasil (Brazilian cost-of-risk benchmarks), Credicorp (LatAm bank comparison), and PagSeguro and StoneCo (Brazilian acquiring structure, the Cielo take-private, and the bank-owned-acquirer funding-subsidy argument used in Section 7.5). Note that this report explicitly corrects the common characterisation of Brazilian credit quality as “benign” — true of Itaú’s book, false of the system (Section 3.8).

Third-party sell-side and reinsurance research consulted for structural framing only, all 2011-vintage or older and never used as current data:

• Deutsche Bank, Global Banking Industry Primer, 26-Sep-2011 (LatAm analyst Mario Pierry). Used for: the structural high-spread / high-NPL / high-provision Brazilian model; the two-year severe-recession stress-test methodology in which Bradesco screened as the most capital-resilient of Brazil’s big three and was named a preferred global name alongside Itaú (Section 4.5); the pre-fintech loan-mix baseline; the BNDES adverse-selection problem; the consumer debt-service ceiling; and the observation that Brazilian system capital was already “inflated by a large volume of deferred tax credits.”
• Deutsche Bank, Banking 101, 11-May-2011. Empirical anchors and analytical rules. Negative finding recorded: it contains no justified-P/TBV formula, no DuPont decomposition and no cost-of-equity algebra — it is not a valid citation for the Gordon identity.
• Swiss Re, An introduction to reinsurance (undated; internal evidence dates it ~1995–96). Training text, used for insurance-framing vocabulary only.

Recorded for completeness: no prior published research on Bradesco specifically was available to this analysis, and no Bradesco earnings-call transcript was available from the usual aggregators for Q1-2026. This is fresh coverage. All transcripts were sourced externally (B.2).

B.7 Secondary press and third-party research

(Used for event dating, corroboration and management-commentary context. Every material claim resting on these has been checked against a primary source where one exists; where none exists, the memo says so.)

Event dating and market reaction

• Bloomberg, “Bradesco (BBDC) Posts Biggest Tumble Since 1998, Stung by Bad Loans,” 2022-11-09.
• Reuters/Yahoo Finance, “Brazil lender Bradesco’s profit plunges 76% as bad loan provisions near $3 bln,” 2023-02-09.
• CNN Brasil, “Bradesco provisiona R$ 4,9 bilhões para cobrir 100% de exposição à Americanas,” 2023-02-09. The R$4.9bn Americanas provision is press- and IR-sourced: it is not named, quantified or attributed to Americanas anywhere in the FY2022 or FY2023 20-F (the EDGAR copies of the relevant press releases are image-only). Flagged in Open Questions.
• NeoFeed, “O plano de Marcelo Noronha para mexer os ponteiros no Bradesco” — the five-year plan presented 2024-02-07.
• Rio Times, “From Crisis to Transformation”; Bloomberg Línea, 2024-11-21.
• MercoPress/Rio Times on the Copom easing cycle, https://en.mercopress.com/2026/03/18/brazil-s-central-bank-starts-easing-cycle-cuts-selic-rate-to-14.75.
• Motley Fool, “Stock Market Today, Jan. 27,” 2026-01-27.
• Money Times, 2026-07-30 — JP Morgan and BTG commentary on the capital increase, https://www.moneytimes.com.br/aumento-de-capital-do-bradesco-bbdc4-e-primeiro-passo-para-um-ciclo-virtuoso-diz-jp-morgan-acoes-caem-ate-3-lmrs/.
• Seu Dinheiro, on JP Morgan’s view that Bradesco’s ROE “stalls at ~17%, below peers,” https://www.seudinheiro.com/2026/empresas/rentabilidade-do-bradesco-bbdc4-deve-emperrar-em-17-abaixo-dos-rivais-aposta-jp-morgan-miql/.

Sector and competitive context

• Seu Dinheiro on the private-payroll race, https://www.seudinheiro.com/2026/economia/a-corrida-dos-bancos-pelo-consignado-privado-e-o-credito-bom-para-pessoas-fisicas-mlim/; and “Bancões na corrida do ouro” on the affluent segment.
• PaymentsCMI; EBANX via PYMNTS, https://www.pymnts.com/digital-payments/2026/ebanx-predicts-pix-captures-half-of-brazil-online-sales-by-2028/; ABECS card data.
• TI Inside on Open Finance consents, 2026-01-22, https://tiinside.com.br/en/22/01/2026/open-finance-brasil-lidera-ranking-global-com-128-milhoes-de-consentimentos-ativos/.
• Matera, “Agenda Evolutiva do Pix 2026–2027,” https://www.matera.com/br/blog/entenda-agenda-evolutiva-do-pix/; Estado de Minas on the 2026 PIX roadmap.
• Rio Times on C6 Bank, Nubank downgrades and Brazilian household debt.
• SpaceMoney/Economatica (record R$255bn system profit, 16.76% system ROE); Metrópoles (big-five R$107bn); Folhapress via Diário do Comércio, 2026-05-04 (spreads, 16-year ROE comparison).
• Portal do Cooperativismo Financeiro, Oct-2025 — 7,252 branches closed in a decade.
• Cuántico VP LatAm VC Report 2026; LAVCA Q1-2026 data.
• Brand Finance, “Nubank is strongest banking brand in Brazil.”
• Sindicato dos Bancários SP on the 2011 loss of the Banco Postal concession to Banco do Brasil.
• Sonho Seguro, Jul-2026, on the SUSEP property/liability premium ranking. Line scope unclear; not relied on in any verdict.
• Rio Times, “Brazil’s Bradesco Builds a Health Giant Worth Up to $9 Bn”; InfoMoney and Suno on SAUD3 sell-side marks.
• Exame, “Bradesco paga R$16 bilhões e conclui compra do HSBC Brasil”; Agência Brasil, Aug-2015.
• InfoMoney, “Cielo vai fechar capital após OPA de R$4,3 bi.”
• AS/COA Poll Tracker, https://www.as-coa.org/articles/poll-tracker-brazils-2026-presidential-election; Polymarket, https://polymarket.com/event/brazil-presidential-election.

B.8 Analytical frameworks

• Bruce Greenwald & Judd Kahn, Competition Demystified — the three genuine advantage types (supply/cost, demand/captivity, economies-of-scale-plus-captivity), the market-share-stability test, and barriers to entry as dominant. Applied throughout Section 3 and Section 4.
• Edward Chancellor (ed.), Capital Returns (Marathon Asset Management) — supply-side capital-cycle analysis and the asset-growth anomaly. Applied in Section 3.10, Section 5.2 and Section 7.3.
• Both accessed via the repository skill investment-research-frameworks.

B.9 Known data limitations, stated plainly

1. The share count. Every mainstream vendor consulted carries ~9.25bn shares. The true figure is ~10.57bn, provable from the 5.721967934% preemptive ratio and confirmed by the 20-F. This error is the origin of the widely-repeated “Bradesco trades below book” claim, and correcting it materially changed this memo’s valuation conclusion. It is the single most important data caution in this appendix.
2. The ~13% banking-ex-insurance ROE is derived, not reported. Bradesco does not disclose segment capital; the derivation uses Bradesco Seguros’ consolidated equity as the proxy. A Basel-RWA-based allocation would be a useful cross-check.
3. Underwriting-only insurance ROE is not disclosed anywhere. The Selic-normalisation haircut in Section 10.6 is an assumption-driven estimate.
4. Resolution 4,966 breaks comparability at 2025-01-01. No uncorrected multi-year provision, NPL or coverage trend spanning that date is valid.
5. FY2020–FY2022 recurring net income and ROAE are secondary-sourced from the Brazilian financial press.
6. The Americanas provision is not verifiable from EDGAR (item B.7).
7. The historical P/B series in Section 10.2 carries the same share-count question; its percentile ranking is scale-invariant and therefore robust, but its absolute levels are understated by ~14%.
8. Itaú’s Mar-2026 RWA and CET1 amount are estimated, not independently pulled, in the P/CET1 comparison.
9. Bradesco’s retained Bradsaúde stake of 91.35% could not be independently confirmed from the SEC corpus; the B3 SAUD3 ticker change is not in a Bradesco 6-K.
10. Bradesco’s own filing claims 3.40% maximum dilution from the capital increase — a figure this analysis cannot reproduce on any construction.
11. Two management quotes from the Q1-2026 call are interpreted-English transcript text and are flagged as such in Section 8.2 rather than treated as clean admissions.
12. Q2 2026 reports on 2026-08-05, four days after this report date. The entire memo is pre-print off Q1 2026.