Methodology.

Slice and minimum aggregation unit

The study's minimum unit is state × line × year. All metrics are built by aggregating from this grain, never on top of already-aggregated numbers. The choice isn't cosmetic: secondary reports that start from national totals lose the ability to investigate regional outliers, which is exactly where the information lives.

The universe covered in v1 is the top ten lines by direct premium in 2024 (SUSEP top-10) against all 27 Brazilian states, in the annual series 2022 → 2024. An honest subset · it covers roughly 74% of the market's direct premium (R$ 473 bn against R$ 639 bn for the complete SES/SUSEP universe).

Operational definitions

Coverage

The coverage metric used here is the share of national earned premium attributed to a specific state, as a percentage:

coverage_state = (earned_premium_state / earned_premium_national) × 100

Not to be confused with population coverage density (insurance per capita) or market penetration (premium / local GDP). These three are distinct metrics · v1 uses only the first for the choropleth map. The other two may appear in future versions as opt-in normalizations.

Loss ratio

Loss ratio is the classic incurred losses over earned premiums, runoff-adjusted:

loss_ratio = (incurred_loss / earned_premium) × 100

The runoff adjustment is critical: losses occur in one fiscal period but are reported and paid in subsequent periods. Without the adjustment, 2024 loss ratio appears artificially low because part of the losses hasn't materialized yet. V1 applies the runoff union · weighted union of losses reported up to 12 months after the event.

Cogroup

Insurance companies frequently operate as subsidiaries of holdings sharing a CNPJ root. For the study, cogroup is the weighted union of companies sharing the same CNPJ root (first 8 digits). HHI and CR4 are computed at cogroup level, not at legal entity level, to avoid underestimating concentration when the same group operates through two or three CNPJs.

Choropleth breaks

Map color scales use fixed quantile breaks per metric, not dynamic breaks. The reason is editorial: dynamic breaks make the same data look different across sections, breaking cross-reading. V1 breaks are:

For loss ratio, breaks explicitly ignore the RS/196 2024 outlier (L/P = 2.23). Including it in quantile computation would compress the entire palette into one extreme, making other states indistinguishable.

Primary sources

All metrics are derived exclusively from public SUSEP SES bases (Insurance Statistics System):

• Direct premium · SES premio_direto_mensal_uf_ramo
• Losses · SES sinistro_ocorrido_mensal_uf_ramo
• Active companies · SES cadastro_empresas (filtered to situacao = ativa)
• Line identification · SES de_para_ramos (2024 codification)

The raw CSVs, Python extraction and transformation scripts, and SHA-256 hashes of the snapshots used are available in the backstage repository.

§ procedência

fonte

SUSEP · SES (Insurance Statistics System)

extração

2025-04-20

pipeline

panorama-susep v1.0 · commit 1e9bbbb

licença

CC-BY 4.0 · MIT code

backstage repository ↗