Actionable insights — Is Nu Holdings An Interesting Stock?

1. Before scoring a bank, strike the metrics that invert — and name the replacement for each

1. Go through the standard worksheet line by line and ask whether the item is a cost or a revenue for this business. For a bank, interest is income and deposits are raw material.
2. Drop the tests that invert: interest coverage (interest is revenue), net debt / FCF (deposits are "debt" you want more of), ROIC (the invested-capital base is mostly customer money). Downgrade goodwill/assets if growth is organic.
3. Substitute a bank-native metric for each: loan-to-deposit for liquidity, CET1 for solvency, 90+ day NPLs for credit quality, the efficiency ratio for capital intensity, ROE for capital allocation.
4. Keep the thresholds that still make sense (ROE >20%, SBC <10% of net income, net margin >10%) and write down the regulatory minimum next to each capital ratio so it is judged as a multiple of the floor.
5. Record the swap in the write-up, so a later reader does not compare a bank's scores against a software company's.

• Fintechs that are not deposit-funded (marketplace lenders, BNPL) — the bank metrics don't fit them either; funding cost and warehouse lines matter instead.
• A score that stays comparable across the archive only if the worksheet shows which rows were substituted.

2. Read a lender's safety through three numbers — and check each against its own chart

1. Liquidity — loan-to-deposit ratio. Loans ÷ deposits. Lower means more room for a run; above ~100% the bank is funding loans with something other than deposits.
2. Solvency — CET1. Common equity and retained earnings ÷ risk-weighted assets. Express it as a multiple of the local minimum, not as a raw percentage.
3. Credit — 90+ NPLs. Loans more than 90 days overdue ÷ total loans. Judge the trend and compare with the customer segment served: a sub-prime-leaning book will run higher than a prime bank and still be healthy if it is flat.
4. Pull the quarterly series, not a single print, and note the last period each chart covers.
5. Reconcile the number in the text with the latest point on the chart. A mismatch means one of them is stale.

• A rising LDR is the natural consequence of a lender growing its book faster than deposits — the post itself says Nu "will need to retain more and more capital." Track it quarterly.
• NPLs that are flat only because the loan book is growing fast (the denominator effect). Compare with loan growth; if growth slows, the ratio rises mechanically.

3. Test a branchless moat with cost-to-serve, referral share and the efficiency-ratio trend

1. Find the monthly cost to serve one customer and compare it with incumbents'. A multiple (not a percentage) gap is the moat claim.
2. Find the share of new customers acquired organically (referral / word of mouth). High organic share means acquisition cost stays low as the base scales.
3. Confirm both in the reported numbers: the efficiency ratio (operating expenses ÷ revenue) should fall while revenue compounds, and the customer count should keep growing without a matching rise in costs.
4. Benchmark the efficiency ratio against local incumbents, not against global banks.
5. Look for the flattening point — where the ratio stops improving tells you when operating leverage is used up.

• The flattening already visible: 29.9% → 24.7% (one-off) → 28.3% → 27.7% across 2025. The easy leverage is done; the next leg needs revenue per customer.
• Cost-to-serve advantages erode when a well-funded rival copies the model — the post's named risk is exactly that (Mercado Pago in Mexico).

4. Put the company's own multiple history next to the incumbents' — the premium is the thesis

1. Chart the forward P/E against its own average. Check the start date of that average — a short listing history makes "below average" a weak signal.
2. On the same chart, add the two or three largest domestic incumbents.
3. State the premium explicitly (company multiple ÷ incumbent multiple).
4. Ask what has to stay true to keep that premium: growth rate, efficiency gap, credit quality. Each is a trigger to re-rate toward the incumbents.
5. Price the downside: what the stock is worth at a blended incumbent multiple on today's earnings.

• The direction of the gap: since December 2022 Nu's multiple fell 63% while Itaú's rose 29%. The premium has been compressing for four years.
• At Itaú's 8.7x, today's forward earnings would support roughly half the current price — that is the size of the bet on continued growth.

5. Reconcile the reverse-DCF starting earnings with the multiple before trusting the implied growth

1. For a bank, use net income (less SBC) as the cash-flow proxy — FCF is distorted by loan growth and deposit flows.
2. Cross-check the year-1 input: market cap ÷ forward P/E should approximately equal the NTM net income you entered.
3. Also check it against LTM net income. An NTM figure below LTM for a company growing 30%+ is almost certainly a stale estimate.
4. Re-run the reverse DCF with the reconciled input and report both implied growth rates.
5. Note what share of value sits in the terminal value; above ~60%, the answer is mostly a statement about year 11 onwards.

• Refreshed write-ups: when price and multiple charts are updated but the model inputs are not, the three valuation methods stop describing the same company.
• A worksheet row labelled "FCF" fed with net income — harmless if intended, but it should say so.

6. Choose the margin denominator for a lender on purpose

1. Identify which revenue line the data vendor uses: gross interest and fee income, net interest income plus fees, or revenue after credit-loss provisions.
2. Compute net margin on at least two of them.
3. Use revenue after provisions only when comparing lenders with similar credit risk — it quietly nets bad-debt costs out of the denominator.
4. Label margin charts precisely: net interest margin (interest spread ÷ earning assets) and net income margin are different quantities.
5. Apply the same >10% threshold to the most conservative version before calling the test passed.

• A high-yield consumer lender's margin after provisions will look best exactly when provisions are low — i.e. late in a credit cycle.

7. Under "skin in the game", read the votes as well as the equity

1. Record the founder's economic stake and voting stake separately.
2. Compute the gap (votes ÷ economics). Above ~2x, minority shareholders cannot remove management or block a transaction.
3. Check the domicile of the listed holding company and what shareholder protections it carries.
4. Treat high alignment (a large economic stake) and low accountability (super-voting shares) as two separate scores rather than one.
5. Look for sunset clauses on the dual-class structure.

• The combination this archive has already flagged once — founder governance as a single-cause conviction downgrade (KPG.AX, April 2026) — can arrive without warning when control is uncontestable.