SMVaddadi/propensity-data-eiopa-solvency2

eiopa_solvency2 — mirror

A convenience mirror of a public dataset, pushed by propensity.data.hf_mirror so that benchmark runs do not depend on a flaky upstream origin.

This is a mirror, not a grant of rights

Copyright and licence remain with the original source named below. Mirroring moves bytes; it does not and cannot change the terms they arrived under. If you intend to redistribute this further, or to use it commercially, read the original terms — not this card, and not the license: field above, which is our best reading and may be wrong.

Provenance

• Source: EIOPA Solvency II Insurance Statistics, premiums/claims/expenses (QRT template S.05.01). Source: EIOPA - European Insurance and Occupational Pensions Authority, https://eiopa.europa.eu/ . REQUIRED DISCLAIMERS, because this is a derivative and EIOPA's legal notice requires them verbatim: 'This document has been drafted using material downloaded from EIOPA website'; 'EIOPA does not endorse this publication and in no way is liable for copyright or other intellectual property rights infringements nor for any damages caused to third-parties through this publication'. If incorporated into anything sold, buyers must be told it is obtainable free of charge from EIOPA. The permission does not extend to third-party material. WHAT THIS IS, measured 2026-11-10: four arms. solo_quarterly 57,705 x 16 (2016 Q3-2026 Q2, 40 periods), solo_annual 27,528 x 16 (Y2016-Y2025), group_quarterly 1,570 x 16 (2018 Q2-2026 Q1, 32 periods), group_annual 927 x 16. Long format, label_columns ['value'], entity reporting_country, time period_end. Here because it is the only openly-licensed EEA-wide insurance premiums/claims/expenses panel in the sweep, and it publishes the Net Combined Ratio and Net Expense Ratio directly rather than derived. NINE MEASURED CAVEATS. (1) The entity axis is a COUNTRY, not an insurer -- there is no undertaking id, and business_type has exactly 2 values (Life/Non-Life), so there is no line-of-business split. (2) solo_quarterly's 31 reporting_country values are 30 countries PLUS an 'EEA' aggregate measured as their EXACT sum (ratio 1.0, min=mean=max across all 40 periods, on two independent item codes), sharing the same id space -- filter is_eea_aggregate before any entity-disjoint split. solo_annual has 30 and NO EEA row, so the two solo arms disagree about whether the total is in the file and neither can be assumed from the other. (3) THE QUARTERLY ARMS ARE YEAR-TO-DATE CUMULATIVE and nothing in the file says so: 279 of 279 full solo country-years are monotone Q1<Q2<Q3<Q4, median Q4/Q1 3.257. A naive quarter-on-quarter growth rate reads a ~70% collapse at every Q4-to-Q1. (4) 'item' is a row LABEL, not a metric: 16 labels carry 50 item codes quarterly and 22 carry 101 annually, so use metric_key or item_code -- pivoting on item collides 26,947 of 57,705 rows. (5) 'value' mixes EUR amounts (up to 815,227) with two 2-decimal fraction ratios; see value_kind. Both ratios are Non-Life only -- there is no Life combined ratio. (6) Net is an EXACT identity over four sibling codes (max absolute error 0.000271), so a wide pivot leaks its own label; and the published Total technical expenses recovers the suppressed Other expenses + Balance pair in 1,172 of 1,237 cases. (7) 'Changes in other technical provisions' is empty for EVERY entity in the last 10 quarters, so a wide pivot plus dropna() deletes the most recent 2.5 years. (8) The quarterly and annual arms are NOT spliceable: joining Q4 onto the annual arm gives 10,499 jointly-populated cells and ZERO of the 42 shared item codes agree exactly -- only 24.1% match to 1e-6, 4,200 differ by >1%, 1,749 by >10%, and they are different extraction vintages. (9) There is NO UNITED KINGDOM row in any arm in any period, including 2016 Q3 through 2020 Q4, despite EIOPA's own page implying pre-2021 coverage -- and the EEA aggregate is UK-free back to the start, so a pre/post-Brexit comparison off this release measures nothing. The amounts' UNIT IS NOT VERIFIED (magnitudes fit EUR millions; EIOPA states no unit) and is not rescaled here. The whole historical series is revised every release; date_of_extraction is the vintage stamp and the arms carry different ones. WHAT THIS IS NOT: not an insurer, policy or line-of-business panel; no adjudication outcome, no exposure denominator, no price. No target is constructed -- in particular the combined ratio is NOT thresholded at 100%, because no EIOPA publication supplies that cut, and the ratios are 2-decimal fractions where 90 observations sit strictly above 1.00 and 29 exactly at it, so '>1' versus '>=1' would move the positive class by 32% on rounding alone.
• Origin URL: https://www.eiopa.europa.eu/tools-and-data/insurance-statistics_en
• Loader: propensity.data.public_loaders.load_eiopa_solvency2
• Extract: full pull from origin
• Reported source: eiopa_solvency2
• Rows pushed: 47,793
• Columns: 16

Licence and redistribution

• Licence as found: other (eiopa-legal-notice-attribution)
• Redistribution verdict: open
• Basis: EIOPA, reproduction authorised with attribution. Read 2026-11-10 in prose from the rights holder's own Legal notice at https://www.eiopa.europa.eu/legal-notice_en (HTTP 200, 207,240 bytes), section "Reproduction of information and documents from EIOPA website", verbatim: "Reproduction of information and documents from the Authority's website is authorised without prior permission, provided that the user acknowledges the Authority as the source: 'Source: EIOPA - European Insurance and Occupational Pensions Authority, https://eiopa.europa.eu/'." THREE CONDITIONS TRAVEL, all from the same page and all applicable here. (a) A loader IS a derivative, so the twofold disclaimer is required verbatim: "This document has been drafted using material downloaded from EIOPA website"; "EIOPA does not endorse this publication and in no way is liable for copyright or other intellectual property rights infringements nor for any damages caused to third-parties through this publication". (b) If incorporated into anything sold, buyers must be told it is obtainable free from EIOPA. (c) The permission does not extend to third-party material. There is also a carve-out for EIOPA tools needing a special Licence Agreement, which does not apply: these are plain CSVs on the statistics page. No non-commercial bar, no share-alike. THE PROGRAMME'S OWN CONTRADICTION, RESOLVED: one lane recorded EIOPA's legal notice as containing "no reuse clause" while another read it as permissive. The permissive reading is correct. Note https://www.eiopa.europa.eu/copyright_en is HTTP 404 -- there is no separate copyright page, the legal notice is it. Also recorded: https://register.eiopa.europa.eu/ is HTTP 403 anonymously (179 bytes), so the DCAT-advertised register distributions are unreachable; the working paths are eiopa.europa.eu/assets/insurance-statistics/.csv. And data.europa.eu's DCAT record labels this the European Commission reuse notice (Decision 2011/833/EU) -- but EIOPA is an agency, not the Commission, and publishes its own notice, which governs. The terms happen to be similarly permissive so the verdict is unchanged, but the catalogue was citing the wrong instrument. No personal data: country-level aggregates.

Citation

Source: EIOPA - European Insurance and Occupational Pensions Authority, https://eiopa.europa.eu/ -- Solvency II Insurance Statistics, premiums, claims and expenses (template S.05.01). This dataset has been drafted using material downloaded from EIOPA website; EIOPA does not endorse this publication and in no way is liable for copyright or other intellectual property rights infringements nor for any damages caused to third-parties through this publication.

Loading

from propensity.data.hf_mirror import load_mirrored

result = load_mirrored("eiopa_solvency2", org="SMVaddadi")
result.frame.head()

load_mirrored falls back to the original loader (origin, then its own offline replica) if this mirror is unreachable, so it is safe to depend on.