Variant Perception

Where We Disagree With the Market

Consensus has decided the May 14 reset already priced the Q1 credit-cost crack, and is now positioned for a "buy the dip" recovery into the Q2 print: 18 of 21 ratings sit at Buy with the median 12-month target clustered between $17.60 and $19.87 against a $12.07 spot. The sharpest disagreement: the market treats Nu's 81%-of-CDI funding cost as a structural moat, but Inter & Co's FY2025 20-F discloses a smaller digital peer funded at 65.3% of CDI — a 160 bps disclosure-level gap that means the deposit-cost edge is narrower than the bull case (and most sell-side models) embeds. Two further gaps follow: consensus reads the 15.4% expected-credit-loss coverage held exactly flat through 56% loan-book growth as analytical reserve discipline, while the forensics evidence reads it as a denominator artifact that has to break in Q2; and consensus models Mexico as a 5-year-lagged Brazil cohort replica, while the competitive evidence shows Mercado Pago — which already earns 36% ROE off captive e-commerce traffic Nu cannot reach — is the load-bearing constraint on that thesis. All three resolve on the August 13–14 Q2 print or the disclosures it forces.

Variant Perception Scorecard

Reading the score. Consensus clarity scores high (78) because there is a tight cluster of ratings — 18 Buys / 2 Holds / 1 Sell with PTs concentrated $17.60–$19.87 — and a clean "buy the dip on credit normalization" narrative across the most recent CICC, UBS, Susquehanna, Morgan Stanley notes. Evidence strength scores 68 because two of the three disagreements rest on disclosure-level peer data (Inter 20-F cost of funding; Mercado Pago Mexico monetization) and one rests on the exact-decimal stability of the 15.4% coverage ratio across two fiscal years — strong but not unfalsifiable. Variant strength sits at 62, not higher, because we are not betting against the franchise: we are pricing risks the consensus appears to underweight, on a name where the bull engineering case (30% ROE, 27% revenue growth, $3.5B FCF) is largely correct.

Consensus Map

The Disagreement Ledger

Disagreement 1 — Deposit-cost moat is narrower than priced. Consensus would say: Nu's 81% of CDI funding is structurally cheaper than the incumbents' 100%+ and supports a multi-year NIM premium; the deposit base ran past $40B because the franchise is sticky. Our evidence says: Inter — smaller, digital, Brazil-only, also Open-Finance-enabled — discloses 65.3% of CDI in the same period; Nu's ratio has crept up from the high-70s as the book scaled; and long-term debt tripled to $4.4B with $2.3B of net wholesale inflow exactly when management is talking about deposit strength. If we are right, the market has to concede that 100-160 bps of the assumed NIM premium is competitive headroom Nu could be giving up rather than a structural floor. The cleanest disconfirming signal: Nu's cost of funding drifting back toward 75-78% of CDI without a corresponding Selic move, with wholesale-borrowing share flat or down.

Disagreement 2 — 15.4% ECL coverage is mechanical, not analytical. Consensus would say: management held coverage flat because the underlying credit profile is unchanged; the Q1 spike was Mexico seasoning + mix and risk-adjusted NIM will rebuild to 10.5% in H2 2026. Our evidence says: a coverage ratio held to one decimal across two fiscal years and 56% portfolio growth is exactly the disclosure shape that precedes a forced uplift when a cycle turns; the Q1 2026 provisions jump (+31% QoQ to $1.72B), the 89-bp jump in 15-90 NPL to 5.0%, and Lahrech's departure as the architect of the credit book nine months earlier are coherent signals of a regime change, not a seasonal print. If we are right, Q2 forces a coverage uplift through the P&L exactly when consensus expects stabilization. The disconfirming signal: Q2 provisions ≤$1.6B with 90+ NPL ≤6.6% and coverage stable at 15.4% without strain.

Disagreement 3 — Mexico is a structurally different game. Consensus would say: Mexico has 15M customers, the #1 issuer of new credit cards, and the cohort ARPAC math will work like Brazil with a 3-5 year lag. Our evidence says: Mercado Pago is already 36% ROE / 39% revenue growth in Mexico from a captive e-commerce flywheel Nu cannot replicate from a banking license; Nu's own Q1 2026 management commentary blamed Mexico loan-book seasoning for the provisioning spike, meaning Mexico is paying for itself with credit losses before it pays for itself with ARPAC; and the Moat tab itself concedes the data-network advantage does not transport cleanly across countries. If we are right, Mexico settles at 18-22% ROE rather than replicating Brazil 30% — the difference between a 5x and 3x book multiple as the customer-base mix shifts away from Brazil. Disconfirming signal: Mexico-specific risk-adj NIM disclosure showing positive contribution and improving credit cost across two consecutive prints, with Mercado Pago Mexico ARPAC not pulling further ahead.

Disagreement 4 — Stable-metric headline mask is a governance discount. Consensus would say: $0.80 cost-to-serve, 15.4% ECL coverage, "above 85" NPS, and a stable Managerial P&L bridge are positive operational signals from a high-quality franchise. Our evidence says: each of these numbers is held to a rounding convention that flatters consistency, the Managerial P&L carries only KPMG limited assurance (a deliberately lower bar than reasonable assurance), the CEO monetized $500M of stock in August 2025 with no public 10b5-1 disclosure surfaced in the research, and the President/COO who built the credit book departed five months later. If we are right, the priced governance discount is 0 and the analytical discount should be 5-10% — small on its own, but it compounds the magnitude of any thesis-breaker above. Disconfirming signal: the Managerial P&L bridge to IFRS staying consistent across Q1-Q3 2026 prints with no definition drift, and the cost-to-serve / ECL coverage lines breaking their rounding bands without P&L surprise.

Evidence That Changes the Odds

How This Gets Resolved

What Would Make Us Wrong

Take the deposit-cost disagreement first, because it carries the most evidence and the most downside if we are wrong. Nu over-funds its loan book by $14B ($41.9B deposits versus $27.7B loans) and runs 1.49x cash-and-securities cover on deposits — both well above incumbent ratios. If the 81% of CDI reflects management choosing to over-fund into a high-Selic environment rather than competitive ceiling pricing, then the 16 percentage-point gap to Inter is not headroom Nu is giving up; it is excess liquidity Nu can roll into higher-yielding assets as Selic falls. If two more quarterly prints show Nu's cost of funding drifting down toward 78% of CDI as Selic eases while wholesale-borrowing share stays flat, the variant view dies on the disclosure tape — and the bull base case of NIM expansion on funding-cost compression becomes the more probable path.

Take the ECL coverage disagreement next. The Bull case for 15.4% being analytical is genuinely defensible: under IFRS 9, expected credit loss is a forward-looking weighted-probability calculation, and management's stated position is that the 56% portfolio growth was funded by new-vintage cohorts with credit quality identical to seasoned cohorts. If 90+ NPL has already peaked at the Q3 2024 7.0% reading (as Q1 2026's 6.5% suggests), then a flat coverage ratio against a falling delinquency rate is not mechanical — it is reserving stability despite improving credit. The cleanest signal that we are wrong: 90+ NPL declining below 6.0% in Q2 with no coverage uplift. That would prove the 15.4% line is an outcome, not a denominator.

The Mexico disagreement is the most contested. The variant view rests on Mercado Pago's reported metrics, but those numbers aggregate seven LatAm geographies and Nu has actually overtaken Mercado Pago as the #1 issuer of new credit cards in Mexico (6.6M holders, June 2025). If Mexico ROE in years 4-5 from launch matches Brazil's year 4-5 — and the underlying credit data so far supports it — Mexico becomes the second engine the bull case requires and the variant compression assumption (18-22% Mexico ROE) is too pessimistic. The cleanest disconfirming signal: standalone Mexico ROE disclosed at 22%+ in any FY2026 quarter.

And finally, on the governance overlay: Vélez's Aug 2025 $500M sale could be a tax-loss-harvesting structure tied to the 2026 Mexican CNBV license-approval timeline, his $356M voluntary 2022 CSA giveback is a counter-signal that few founders have ever matched, and the Managerial P&L bridge to IFRS has been published cleanly in Q1 2026 with no definition drift in the line items the reader cares about. If that bridge stays consistent for three more prints and the cost-of-funding ratio holds, the governance discount we are flagging may have been priced wrong on the negative side.

The first thing to watch is the Q2 2026 print the week of August 13-14, 2026 — specifically whether provisions for credit losses come in at or below $1.6B with 90+ NPL stable to declining and ECL coverage held at 15.4% without strain. That single print resolves more of the variant view than the next nine months of news flow combined.