Variant Perception
Where We Disagree With the Market
The sharpest disagreement: consensus FY2026 EPS of $0.89 is mechanically too high because it extrapolates a one-quarter 8.7% effective tax rate that cannot persist into a Brazilian fintech regime where the corporate rate is stepping from ~40% to 45%. Sell-side carries a $19.50 median target on that EPS at ~17× forward; correct the denominator for a structural 22–27% ETR and the same multiple maps to roughly $0.70–0.75 of EPS — consistent with a $13–15 scenario, close to where the stock trades.
The market's 35% drawdown from $18.83 has been blamed on a credit-cycle scare, but the second disagreement is that the load-bearing 10-year variable is deposit cost as a percentage of CDI through a Selic normalization cycle, and consensus models assume the 88%-of-CDI funding edge holds as a flat dollar number rather than halving when Selic moves from 15% toward 9–11%.
The third disagreement is institutional rather than analytical: short interest at 2.9% of float says the de-rate is forced LatAm/EM redemption flow, not directional bears, which makes the marginal seller a price-insensitive holder rather than an informed one. Each resolves on observable data — Q2 2026 prints the tax rate, the BCB Copom path prints the funding-moat compression mechanism, and 13F flow into Q3 prints whether the marginal seller stays redemption-driven.
Variant Perception Scorecard
Variant Strength (0-100)
Consensus Clarity (0-100)
Evidence Strength (0-100)
Time to resolution: 3–6 months (ETR), 12–24 months (funding).
The variant is strong but not maximal. Consensus is unusually clear — an 18/2/1 buy ratio with a $19.50 median target and a published FY26 EPS of $0.89 leaves little ambiguity about the market's implied model. The evidence base is solid: the Q1 2026 8.7% ETR is a disclosed fact, the Brazilian corporate tax step-up is regulator-confirmed, and the 88%-of-CDI funding base is in supplemental disclosure. What pulls the score below 80 is that the EPS-denominator critique is partially front-run by the stock at $12 already, and the funding-moat critique resolves over years rather than quarters — so the variant is correct but not maximally monetizable in a 90-day window.
The single highest-conviction disagreement. Consensus FY26 EPS of $0.89 silently anchors on Q1's 8.7% effective tax rate, which was driven by Brazilian interest-on-equity (juros sobre capital próprio) benefits and is not a sustainable run-rate. The Brazilian fintech corporate rate is stepping progressively from ~40% to 45% in 2026. At a structural ETR of 22-27%, the same pre-tax base produces EPS of roughly $0.70-0.75 — meaning the median target of $19.50 implicitly carries a 23-25x forward P/E, not the 17x figure consensus advertises.
Consensus Map
The Disagreement Ledger
Disagreement 1 — the EPS denominator. Consensus would say FY26 EPS at $0.89 is the right number because growth, operating leverage, and a 50%+ ROE on retained earnings carry the line. Our evidence disagrees because the 8.7% Q1 ETR is a one-quarter interest-on-equity event that cannot annualize — the Brazilian fintech statutory rate is rising to 45%, and management has signalled the structural rate "normalizes higher." If we are right, the market would have to concede that the same $19.50 sell-side target carries a 23-26x forward multiple, not 17x, and the "obvious upside to consensus" framing collapses to a fair-value-around-here framing. The cleanest disconfirming signal is Q2 2026 reported ETR — a print at or above 20% forces the cut; a sustained print below 15% would force the bear to explain how the structural tax rate avoided the regime that hit every Brazilian fintech peer.
Disagreement 2 — the funding moat is regime-dependent. Consensus would say the deposit franchise is the structural moat that makes the multiple defensible — 88% of CDI, sticky, primary-relationship, holds through Open Finance. Our evidence disagrees because the moat is a percentage, not a dollar — at Selic 9-11% the absolute funding gap halves even if the percentage holds, and the long-term thesis tab is explicit that this is the single load-bearing test for the 10-year case. If we are right, sell-side targets that implicitly assume 25%+ through-cycle ROE need to be reset to 18-22%, which compresses the multiple to 4-5x P/TBV — implying a $9-11 stock, not $19.50. The cleanest disconfirming signal is the quarterly deposit-cost-percentage-of-CDI disclosure once Selic cuts begin; if it stays in the 87-91% band on a $50B+ deposit base, the moat survived regime change and the bear loses this leg.
Disagreement 3 — the marginal seller is technical, not informed. Consensus would say a 35% drawdown from peak is the market doing its job — the bears have arrived. Our evidence disagrees: short interest at 2.9% of float and 3.41 days-to-cover is not consistent with directional bears writing the credit-cycle thesis. The de-rate looks like LatAm/EM redemption flow plus Brazilian political risk repricing, layered on a fundamental Q1 disappointment that gave the rotation an excuse. If we are right, a Q2 print that even partially refutes the credit-cycle read produces an asymmetric rally because there is no informed short cover waiting to absorb the move. The cleanest disconfirming signal is short interest spiking above 5% of float before Q2 — that would tell us informed bears have actually positioned and the technical-seller interpretation is wrong.
Disagreement 4 — the disclosure framework erosion is non-trivial. Consensus is taking the Managerial P&L at face value because net income reconciles. Our evidence is that the framework is on ISAE 3000 limited assurance only, FY24 and FY23 have not been re-presented, and $243M of FY25 development spend was capitalized rather than expensed under IAS 38. None of this is fraud, and Forensics keeps the file in the "Watch" band rather than "Elevated" — but the comparability and audit machinery is materially weaker than at a US domestic, and a PM paying 7x tangible book is entitled to a 100-200 bps multiple haircut for the asymmetry. The cleanest disconfirming signal is the FY25 20-F: a clean three-year re-presentation under the Managerial framework with no new non-IFRS add-backs would remove this concern.
Evidence That Changes the Odds
How This Gets Resolved
Two signals resolve fast, two resolve over a multi-year window. The tax-rate and Q2 NPL prints arbitrate in the next 90 days. The deposit-cost-percentage-of-CDI through Selic normalization and the Mexico ARPAC trajectory are 2027-2028 questions. The 20-F re-presentation is a single-event resolution on disclosure quality. None of these signals are "better execution" or "time will tell" — each is a specific, observable line item in a known filing date.
What Would Make Us Wrong
The strongest disconfirming evidence is the implied FY25 ROA of 4.58% versus Itaú's 1.47% — the single cleanest moat fact in the file. That gap is not a peak-Selic artifact; even if Selic normalization compresses it by 30-40%, Nu still earns ~3.0% on assets versus 1.0-1.5% at incumbents on roughly half the leverage. If our EPS-denominator critique is right but the ROA gap closes by less than 100 bps under Selic normalization, then the through-cycle ROE clears 22-25% rather than 18-22%, and the bull's $22 target on $1.20 of 2027 EPS becomes the right anchor — not consensus's $19.50 and not the bear's $8. We are betting that consensus is paying for permanence on a partly cyclical edge; that bet loses if the edge is more structural than cyclical.
The second place we are most likely wrong is the short-interest interpretation. Short interest data is reported with a two-week lag, and the cohort could already be building. If a fresh print shows short interest above 5% of float, the "technical seller" interpretation breaks — the de-rate becomes thesis-driven and the asymmetric-rally setup we are betting on disappears. We would also be wrong if the LatAm/EM derate stays intact through Q2 regardless of fundamentals; in that case even a clean Q2 print gets sold and the variant pays out on a longer horizon than we think.
The third risk is that the interest-on-equity (juros sobre capital próprio) regime extends longer than the corporate-rate step-up suggests. Brazilian tax policy is politically negotiated. If the JCP benefit is preserved through the October 2026 election and into 2027, the 8.7% ETR could persist for 3-4 quarters rather than 1, and consensus EPS at $0.89 could prove more durable than the variant assumes. The bear-case-on-the-variant is that we are right on the structural rate but wrong on the path, and the EPS revision we expect lands in 2027 rather than 2026 — which materially weakens the 90-day monetization case.
The fourth risk is the Mexico optionality call. We have flagged it as priced at zero, which is favourable for our variant. But Mexico break-even is one quarter, and the competitive density there (BBVA, Banorte, Plata, Ualá, Revolut, Mercado Pago) is materially higher than Brazil 2014-2018. If Mexico ARPAC stalls below $10 through 2027 even with continued customer growth, the "second engine" framing fails to compound, and the variant view of "optionality undervalued at zero" loses its kicker — the case has to win on Brazil alone.
The first thing to watch is the Q2 2026 effective tax rate, because it is the single line item that mechanically arbitrates whether consensus FY26 EPS of $0.89 survives the print or gets cut by 15-20% in a week.