solvency2 for R

CRAN edition 0.5.1, derived from native R Golden Source 0.5.0 (commit 14b25dc). Maintainer: Dimitrios Geromichalos riskdatascience@web.de. See ?solvency2 for the package overview and documentation("DEVELOPMENT") for the bounded standard tests and optional exhaustive reference suite.

Automated package checks run on Linux (R release, oldrel-1 and devel), Windows and macOS. They include the reference tests, help examples and vignette. Click the status badge to inspect the individual platform results.

Native R tools for offline Solvency II analysis: 399 granular calculations, source-bound parameters and correlation matrices, dated rule profiles, synthetic reference portfolios, and archived reporting definitions and checks.

Use one formula in a spreadsheet replacement, combine components in your own valuation pipeline, inspect a regulatory parameter, or run a dependency graph. You do not need to adopt a monolithic application or install Python.

Start with a calculation

library(solvency2)

ctx <- Context(
  as_of = "2025-01-17", known_at = "2026-09-23", currency = "EUR",
  profile_id = "EU-S2-DOCUMENTS-2024-2025-v0.1"
)

charge <- intangible_risk(100, context = ctx)
charge$value
charge$sources
charge$details

Results retain their units, formula identifier, explicit context, source references, input/profile fingerprints and qualification limits. Monetary values use the stated currency, not implicit thousands; rates are decimal fractions unless the individual function documents another convention.

What you can use

• Valuation and technical-provision components, explicit discounting, risk-free rate calculations, adjustments, risk margin and reinsurance.
• Market, counterparty-default, life, health and non-life standard-formula components, including catastrophe geography, weights and correlation matrices.
• SCR aggregation, own-funds classification and eligibility, MCR, group and restricted-portfolio calculations, and prescribed USP numerical methods.
• Dated applicability and evidence checks, including the future profiles actually contained in the reference version. No rule set is selected from today’s date.
• Archived DPM catalogues, typed reporting expressions, closed-cell bindings, filing preconditions and activation evidence; deterministic XBRL draft assembly with currency, precision and explicitly selected rule checks.

The exact scope is available through coverage(), rule_profiles() and api_catalogue(). Every calculation has an R help page and an executable example. The catalogue preserves the Python reference’s domain paths and source signatures; use ?function_name or args(function_name) for the R calling convention.

Supply your own data

There is no runtime internet access, background update or market-data download. Obtain production curves, cashflows, holdings and qualifications yourself. DiscountCurve(), load_discount_curve_json(), load_discount_curve_csv() and discount_curve_from_spot_rates() provide explicit local input routes. best_estimate_with_curve() binds exact payment-date factors to supplied cashflows. No interpolation, extrapolation, calendar convention or currency conversion is silently selected. Archived rate APIs accept explicit market_data inputs.

Bundled market examples are invented and marked SYNTHETIC_DEMO; they are not observed production curves. Regulatory constants and source matrices are not synthetic.

Inspect or change a parameter

parameter("scalars/intangible", ctx)
parameter("matrices/health_nslt", ctx)

scenario <- ctx
scenario$overrides <- list(NumericOverride(
  "profile.json", list("scalars", "intangible", "value"),
  0.70, "Illustrative analyst sensitivity, not a regulatory amendment"
))
intangible_risk(100, context = scenario)$value
intangible_risk(100, context = ctx)$value  # installed baseline unchanged

Overrides are context-local, validated and audited with original values and reasons. JSON array paths use zero-based indices, also in R. Symmetric matrix overrides update the counterpart consistently. Version guards and nonnumeric regulatory qualifications cannot be bypassed with a numeric override.

Reproduce and combine

reference_cases() contains 12,382 fixed independent calculation cases and reference_portfolios() contains 227 integration portfolios. Additional synthetic reporting datasets are accessible with reference_datasets() and reference_dataset(). These include valid, boundary and deliberately rejected inputs; they are not all successful calculation examples.

p <- reference_portfolios()$portfolios[[1]]
result <- calculate(p$nodes, context = do.call(Context, p$context))
names(result$results)

An Engine() instance offers the same run() interface with an instance-local cache. Dependencies pass numeric values while provenance remains in node results. Demo provenance also propagates through dependent nodes.

Installation and documentation

From a checked-out source directory with R and the declared dependencies installed:

R CMD build solvency2-CARAN
R CMD INSTALL solvency2_0.5.1.tar.gz

The package requires R >= 4.1, jsonlite, digest, gmp, Rmpfr and xml2. Source builds of the last three may need the corresponding GMP, MPFR and libxml2 development libraries; binary R packages already provide the compiled dependencies. Python, NumPy and SciPy are not runtime dependencies.

Use help(package = "solvency2"), vignette("getting-started", package = "solvency2") and documentation() for local help. export_resources() writes verified JSON to an explicitly requested new directory; normal calculations do not write files.

Reference compatibility and boundaries

The immediate reference is the published native R solvency2 0.5.0, which implements the Python solvency2 0.5.0 reference. GOLDEN_SOURCE.json records its commit and source fingerprints. Data are preserved losslessly, including provenance and dated scope restrictions. R uses native GMP/MPFR arithmetic, stats::optim and stats::fft; numerical backend diagnostics, R input hashes and XML serialization hashes identify the implementation actually used rather than imitating Python runtime metadata.

The package implements the reference’s documented components. It does not develop an insurer’s internal model, perform actuarial projection on behalf of the user, grant supervisory approval, infer complete reporting obligations or submit filings. Externally supplied model results and qualifications remain explicit inputs. REFERENCE_COMPONENT, ANALYST_SCENARIO, SYNTHETIC_DEMO, PARTIAL and DRAFT_XBRL_INSTANCE describe distinct outcomes; none means regulatory approval. Unsupported or ambiguous situations remain visible through structured errors or review findings, never an invented default result.

License and attribution

Original code and invented demonstration data: RiskDataScience GmbH, GPL-3.0-or-later. Contact: riskdatascience@web.de. Regulatory and reporting extracts retain their separately identified reuse terms; see inst/THIRD_PARTY_NOTICES.md, inst/THIRD_PARTY_DATA_REGISTER.json and inst/licenses/. No observed historical market-data archive is included.