type: insight tags: [fintech, lending, credit-quality, npl, underwriting, embedded-credit, scaling, risk-assessment] confidence: high created: 2026-03-31 updated: 2026-04-01 source: MELI stock-analysis 2026-03 source2: SE stock-analysis 2026-03 source3: DAVE earnings-review Q4_FY25 persona: bert provenance: legacy source_analysis_path: null source_paragraph_quote: null source_transcript_span: null source_loss_log_path: null

Simultaneous Credit Book Doubling + Declining NPLs Validates Underwriting Quality

Rapid credit book growth (>50% YoY) is typically treated as a risk flag — more volume often means lower underwriting standards. But when a credit book approximately doubles while NPLs simultaneously decline or hit all-time lows, it inverts this assumption: the underwriting model is working, credit scoring is accurate, and the expansion is quality-controlled rather than reckless. This is a positive signal, not a cause for concern. The inverse — fast growth + rising NPLs — is the danger pattern.

Evidence

Evidence (continued)

Implication

When evaluating any embedded-credit platform's risk profile, construct a 2x2: (book growth rate) × (NPL trend direction). The high-growth + declining-NPL quadrant is constructive; the high-growth + rising-NPL quadrant is existential risk. For platforms in the constructive quadrant, the street often prices in the credit risk as if growth alone implies deteriorating quality — creating valuation discounts that don't reflect actual credit data. Track this combination quarterly: as long as the credit book scales with stable-or-improving NPLs, the risk narrative is overblown. The signal breaks if NPL 15-90 rises more than 2pp QoQ during a period of book deceleration — that would indicate the good vintages are rolling off and newer, lower-quality cohorts are maturing. The pattern now holds across three companies (MELI, SE, DAVE), two geographies (LatAm, SEA, US), and three origination models (embedded e-commerce, SPayLater, standalone neobank EWA) — it is not MELI-specific, and it is not geography-specific.