ratingmodels 0.1.0__tar.gz

ratingmodels-0.1.0/.gitignore

@@ -0,0 +1,66 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# Python build artifacts
+build/
+dist/
+*.egg-info/
+.eggs/
+*.egg
+
+# Virtual environments
+.venv/
+venv/
+env/
+ENV/
+
+# Test and coverage outputs
+.pytest_cache/
+.coverage
+coverage.xml
+htmlcov/
+.tox/
+.nox/
+
+# Type checker / linter caches
+.mypy_cache/
+.pyright/
+.ruff_cache/
+
+# Jupyter notebooks
+.ipynb_checkpoints/
+
+# Local environment files
+.env
+.env.*
+*.local
+
+# IDE / editor files
+.vscode/
+.idea/
+*.swp
+*.swo
+.DS_Store
+
+# Logs
+*.log
+
+# Temporary files
+tmp/
+temp/
+*.tmp
+
+# Data files generated locally
+*.csv
+*.xlsx
+*.xls
+*.parquet
+*.feather
+*.pkl
+*.pickle
+
+# Keep example datasets only if intentionally committed
+# !examples/*.csv
+# !examples/*.xlsx

ratingmodels-0.1.0/CHANGELOG.md

@@ -0,0 +1,55 @@
+# Changelog
+
+All notable changes to this project are documented here. The format follows
+[Keep a Changelog](https://keepachangelog.com/), and the project adheres to
+[Semantic Versioning](https://semver.org/).
+
+## [0.1.0] - 2026-06-29
+
+Initial release.
+
+### Changed
+- **Credibility consolidated into `actuarialpy`.** Trend and credibility are
+  shared ecosystem primitives; `actuarialpy` is the home for credibility. The
+  credibility math is no longer duplicated here -- `ratingmodels.credibility`
+  and `blend` are thin adapters over `actuarialpy` (`full_credibility_claims`,
+  `limited_fluctuation_z`, `credibility_weighted_estimate`, and
+  `BuhlmannStraub.from_frame`). The public `ratingmodels` API and results are
+  unchanged. This removes the risk of the two Bühlmann-Straub implementations
+  drifting apart, and `actuarialpy` now uses the general unbiased estimator
+  (handling unequal period counts). Adds a dependency on `actuarialpy>=0.32.0`
+  and drops the direct `scipy` dependency.
+
+### Added
+- **Credibility** (`credibility`): `full_credibility_standard`,
+  `limited_fluctuation_credibility`, `buhlmann_credibility`, and empirical
+  `buhlmann_straub` with exposure weights.
+- **Trend** (`trend`): midpoint-to-midpoint trend factors, date helpers, and
+  utilization / unit-cost decomposition.
+- **Manual rating** (`manual_rate`): `ManualRate`, `manual_pmpm`,
+  `aggregate_demographic_factor`.
+- **Experience rating** (`experience_rate`): `ExperienceRate`, `pool_claims`,
+  `expected_excess_charge`.
+- **Rate build-up** (`buildup`): typed steps (`start`, `multiply`, `add`,
+  `segment_multiply`, `checkpoint`), an `evaluate` engine and `BuildUp` fluent
+  builder producing a reconciling breakdown, and `participation_blend` /
+  `combine_streams` for combining par/non-par and medical+drug streams.
+  `ManualRate` is reimplemented as a thin layer over the engine and gains
+  `breakdown()` / `steps()`.
+- **Base rate & off-balance** (`base_rate`): `base_rate_from_experience`,
+  `average_relativity`, `off_balance_factor`, `rebalance_base_rate`.
+- **Retention & loading** (`loading`): `RetentionLoad` (fundamental insurance
+  equation gross-up), `gross_rate`, `permissible_loss_ratio`. `ManualRate`,
+  `ExperienceRate`, and `RateIndication` accept an optional `retention` that
+  overrides the single-loss-ratio path with the full expense/profit build-up.
+- **Blending & indication** (`blend`, `indication`): `blend` and the
+  `RateIndication` orchestrator with build-up and loss-ratio methods.
+- **Rate-change decomposition** (`decomposition`): `decompose_rate_change`
+  with multiplicative and percentage-point contributions and an explicit
+  residual.
+- **GLM relativities** (`relativity`): `GLMRelativities` (Poisson / Gamma /
+  Tweedie via in-package IRLS), `FactorTable`, `one_way_relativities`.
+- **Constraints & renewal** (`constraints`, `renewal`): caps, floors, banding,
+  rounding, corridors; `renew` and `member_level_renewal`.
+- **Synthetic data** (`datasets`): `sample_claims`, `sample_rating_data`.
+- Full pytest suite (54 tests) and MkDocs Material documentation.

ratingmodels-0.1.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 OpenActuarial
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

ratingmodels-0.1.0/PKG-INFO

@@ -0,0 +1,230 @@
+Metadata-Version: 2.4
+Name: ratingmodels
+Version: 0.1.0
+Summary: Actuarial pricing and rate-indication tools for experience-rated insurance portfolios.
+Project-URL: Homepage, https://github.com/OpenActuarial/ratingmodels
+Project-URL: Documentation, https://openactuarial.github.io/ratingmodels/
+Project-URL: Repository, https://github.com/OpenActuarial/ratingmodels
+Project-URL: Issues, https://github.com/OpenActuarial/ratingmodels/issues
+Author: OpenActuarial
+License: MIT
+License-File: LICENSE
+Keywords: actuarial,credibility,glm,health-insurance,insurance,pricing,rate-indication,rating
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Financial and Insurance Industry
+Classifier: Intended Audience :: Science/Research
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3 :: Only
+Classifier: Topic :: Office/Business :: Financial
+Classifier: Topic :: Scientific/Engineering :: Mathematics
+Requires-Python: >=3.9
+Requires-Dist: actuarialpy>=0.32.0
+Requires-Dist: numpy>=1.22
+Requires-Dist: pandas>=1.4
+Provides-Extra: dev
+Requires-Dist: pytest-cov; extra == 'dev'
+Requires-Dist: pytest>=7; extra == 'dev'
+Provides-Extra: docs
+Requires-Dist: mkdocs-material>=9; extra == 'docs'
+Requires-Dist: mkdocstrings[python]>=0.24; extra == 'docs'
+Description-Content-Type: text/markdown
+
+# ratingmodels
+
+**Actuarial pricing and rate-indication tools for experience-rated insurance portfolios.**
+
+Part of the [OpenActuarial](https://github.com/OpenActuarial) ecosystem.
+
+`ratingmodels` covers the group rating workflow — the step that turns experience
+analysis and loss modeling into an actual rate. Where the rest of the ecosystem
+explains *what happened* and models *loss risk*, this package answers the central
+pricing question: **what rate should we charge, and why did it change?**
+
+```
+actuarialpy   -> experience analysis (PMPM, loss ratios, trend, completion)
+ratingmodels  -> pricing and rate indications      <-- this package
+lossmodels    -> loss-distribution modeling
+risksim       -> portfolio Monte Carlo simulation
+extremeloss   -> extreme-value tail estimation
+```
+
+## What it does
+
+- **Credibility** — limited fluctuation (square-root rule), Bühlmann, and
+  empirical Bühlmann-Straub with exposure weights.
+- **Trend** — midpoint-to-midpoint factors; utilization / unit-cost split.
+- **Manual rating** — base rate × relativities, loaded to a charged rate.
+- **Experience rating** — pooling of large claims, trend, pooling charge,
+  benefit/demographic adjustments, loading.
+- **Rate build-up** — an ordered, auditable evaluator (multiply / add-dollar /
+  segment-conditional) with labeled subtotals and a reconciling breakdown, plus
+  par/non-par participation blending and medical+drug combining. Supplies the
+  *grammar* of a manual build-up; the factor values stay yours.
+- **Base rate & off-balance** — indicated base loss cost from book experience
+  (base × relativities reproduces book losses); off-balance correction and
+  base rebalancing when relativities are revised.
+- **Retention & gross-up** — charged rate from the fundamental insurance
+  equation (loss & LAE, flat fixed expense, percent-of-premium loads, profit),
+  with the target loss ratio as an *output*, not an input.
+- **Blending & indication** — credibility-weighted blend; build-up and
+  loss-ratio indication methods.
+- **Rate-change decomposition** — multiplicative and percentage-point
+  contribution-to-change with an explicit residual.
+- **GLM relativities** — Poisson / Gamma / Tweedie GLMs fit by IRLS, so factors
+  are estimated *jointly* (correcting for correlation between rating variables)
+  rather than one-way. No statsmodels dependency — the IRLS is in-package.
+- **Constraints & renewal** — rate caps/floors, banding, rounding, corridors,
+  and member-level re-rating.
+
+Dependencies are `numpy`, `pandas`, and `actuarialpy` (which supplies the shared credibility primitives; see below).
+
+## Install
+
+```bash
+pip install ratingmodels
+```
+
+From source:
+
+```bash
+git clone https://github.com/OpenActuarial/ratingmodels
+cd ratingmodels
+pip install -e ".[dev]"
+pytest
+```
+
+## Quick start
+
+```python
+import ratingmodels as rm
+
+# --- experience side -------------------------------------------------------
+capped, excess = rm.pool_claims(group_claims, pooling_point=250_000)
+exp = rm.ExperienceRate(
+    incurred_claims=4_200_000,
+    exposure=96_000,            # member-months
+    trend_annual=0.075,
+    trend_years=1.5,            # experience midpoint -> rating midpoint
+    pooled_excess=excess,
+    pooling_charge_pmpm=4.00,
+    target_loss_ratio=0.85,
+)
+
+# --- manual side -----------------------------------------------------------
+man = rm.ManualRate(
+    base_pmpm=480,
+    factors={"area": 1.05, "industry": 0.97, "tier": 1.10},
+    target_loss_ratio=0.85,
+)
+
+# --- credibility and indication -------------------------------------------
+z = rm.limited_fluctuation_credibility(n=96_000, n_full=120_000)
+
+ind = rm.RateIndication(
+    experience_claims_pmpm=exp.claims_pmpm(),
+    manual_claims_pmpm=man.claims_pmpm(),
+    credibility=z,
+    current_rate=560,
+    target_loss_ratio=0.85,
+    trend_total_factor=exp.trend_factor(),
+    benefit_factor=1.00,
+    demographic_factor=1.01,
+)
+
+print(f"indicated rate   : {ind.indicated_rate():.2f}")
+print(f"indicated change : {ind.indicated_rate_change():+.2%}")
+
+# why did the rate move?
+print(ind.rate_change_decomposition().to_frame())
+
+# apply a renewal cap
+action = rm.renew(current_rate=560, indicated_rate=ind.indicated_rate(), cap=0.15)
+print(f"proposed (capped): {action.proposed_rate:.2f} ({action.proposed_change:+.2%})")
+```
+
+### Rate build-up with an audit trail
+
+```python
+import ratingmodels as rm
+
+med_par = rm.evaluate([
+    rm.start("Par Base Claim Cost", 941.63),
+    rm.add("$30 specialist copay", -11.44),
+    rm.multiply("Rating Region", 1.083),
+    rm.checkpoint("Medical Par Base Claim Cost"),
+])
+med_par.value          # final running total
+med_par.breakdown      # DataFrame: step, operation, label, operand, running_total
+
+# blend in-/out-of-network, then add the drug stream
+med = rm.participation_blend(med_par.value, nonpar=1478.56, participation_rate=0.90)
+total = rm.combine_streams({"Medical": med, "Drug": 323.67})
+total.value            # feeds into trend / credibility / retention
+```
+
+The package supplies the build-up *grammar*; you supply the factor values
+(cost-sharing, age/sex, area, ...) from your filed tables. `ManualRate` is a
+thin shortcut over this engine, so `ManualRate(...).breakdown()` returns the
+same audit trail.
+
+### Base rate and retention
+
+```python
+import ratingmodels as rm
+import pandas as pd
+
+# indicated base loss cost from book experience (off-balance method)
+book = pd.DataFrame({
+    "exposure": [24_000, 18_000, 30_000, 24_000],
+    "area":     [1.00, 1.20, 0.90, 1.05],
+    "tier":     [1.00, 1.10, 0.95, 1.25],
+    "loss":     [11_750_000, 11_400_000, 12_300_000, 15_200_000],
+})
+base = rm.base_rate_from_experience(book, "exposure", "loss",
+                                    factor_cols=["area", "tier"])
+base.base_loss_cost        # average loss cost / average relativity
+
+# gross claims up to a charged rate; the loss ratio falls out
+retention = rm.RetentionLoad.from_items(
+    fixed_expense_pmpm=22.0,
+    variable_items={"commission": 0.03, "premium_tax": 0.023, "aca_fees": 0.005},
+    profit_margin=0.03,
+)
+retention.gross_rate(540.0)            # charged rate
+retention.implied_loss_ratio(540.0)   # target loss ratio (an output)
+
+# rebalance the base when relativities are revised (hold level, then +8%)
+rm.rebalance_base_rate(current_base=base.base_loss_cost,
+                       current_avg_relativity=1.0928, new_avg_relativity=1.12,
+                       overall_change=0.08)
+```
+
+### GLM relativities
+
+```python
+import ratingmodels as rm
+from ratingmodels.datasets import sample_rating_data
+
+df = sample_rating_data(n=20_000)
+model = rm.GLMRelativities(family="poisson").fit(
+    df, response="claims", predictors=["area", "industry", "tier"],
+    exposure="exposure",                 # enters as a log offset
+    base_levels={"area": "A"},           # optional; defaults to modal level
+)
+print(model.base_value_)                 # fitted base frequency
+print(model.relativities_["industry"])   # relativity per level, base = 1.0
+```
+
+## Scope and honest limitations
+
+This is a modeling and workflow toolkit, not filed rate software. It does not
+manage rate filings, store filed factor tables with effective dating, or enforce
+state-specific rating rules. The pooling-charge helper is a simple group-level
+estimate; a production charge is normally derived book-wide or from an EVT tail
+model (see `extremeloss`). All bundled data in `ratingmodels.datasets` is
+synthetic and carries no assumptions.
+
+## License
+
+MIT. See [LICENSE](LICENSE).

ratingmodels-0.1.0/README.md

@@ -0,0 +1,198 @@
+# ratingmodels
+
+**Actuarial pricing and rate-indication tools for experience-rated insurance portfolios.**
+
+Part of the [OpenActuarial](https://github.com/OpenActuarial) ecosystem.
+
+`ratingmodels` covers the group rating workflow — the step that turns experience
+analysis and loss modeling into an actual rate. Where the rest of the ecosystem
+explains *what happened* and models *loss risk*, this package answers the central
+pricing question: **what rate should we charge, and why did it change?**
+
+```
+actuarialpy   -> experience analysis (PMPM, loss ratios, trend, completion)
+ratingmodels  -> pricing and rate indications      <-- this package
+lossmodels    -> loss-distribution modeling
+risksim       -> portfolio Monte Carlo simulation
+extremeloss   -> extreme-value tail estimation
+```
+
+## What it does
+
+- **Credibility** — limited fluctuation (square-root rule), Bühlmann, and
+  empirical Bühlmann-Straub with exposure weights.
+- **Trend** — midpoint-to-midpoint factors; utilization / unit-cost split.
+- **Manual rating** — base rate × relativities, loaded to a charged rate.
+- **Experience rating** — pooling of large claims, trend, pooling charge,
+  benefit/demographic adjustments, loading.
+- **Rate build-up** — an ordered, auditable evaluator (multiply / add-dollar /
+  segment-conditional) with labeled subtotals and a reconciling breakdown, plus
+  par/non-par participation blending and medical+drug combining. Supplies the
+  *grammar* of a manual build-up; the factor values stay yours.
+- **Base rate & off-balance** — indicated base loss cost from book experience
+  (base × relativities reproduces book losses); off-balance correction and
+  base rebalancing when relativities are revised.
+- **Retention & gross-up** — charged rate from the fundamental insurance
+  equation (loss & LAE, flat fixed expense, percent-of-premium loads, profit),
+  with the target loss ratio as an *output*, not an input.
+- **Blending & indication** — credibility-weighted blend; build-up and
+  loss-ratio indication methods.
+- **Rate-change decomposition** — multiplicative and percentage-point
+  contribution-to-change with an explicit residual.
+- **GLM relativities** — Poisson / Gamma / Tweedie GLMs fit by IRLS, so factors
+  are estimated *jointly* (correcting for correlation between rating variables)
+  rather than one-way. No statsmodels dependency — the IRLS is in-package.
+- **Constraints & renewal** — rate caps/floors, banding, rounding, corridors,
+  and member-level re-rating.
+
+Dependencies are `numpy`, `pandas`, and `actuarialpy` (which supplies the shared credibility primitives; see below).
+
+## Install
+
+```bash
+pip install ratingmodels
+```
+
+From source:
+
+```bash
+git clone https://github.com/OpenActuarial/ratingmodels
+cd ratingmodels
+pip install -e ".[dev]"
+pytest
+```
+
+## Quick start
+
+```python
+import ratingmodels as rm
+
+# --- experience side -------------------------------------------------------
+capped, excess = rm.pool_claims(group_claims, pooling_point=250_000)
+exp = rm.ExperienceRate(
+    incurred_claims=4_200_000,
+    exposure=96_000,            # member-months
+    trend_annual=0.075,
+    trend_years=1.5,            # experience midpoint -> rating midpoint
+    pooled_excess=excess,
+    pooling_charge_pmpm=4.00,
+    target_loss_ratio=0.85,
+)
+
+# --- manual side -----------------------------------------------------------
+man = rm.ManualRate(
+    base_pmpm=480,
+    factors={"area": 1.05, "industry": 0.97, "tier": 1.10},
+    target_loss_ratio=0.85,
+)
+
+# --- credibility and indication -------------------------------------------
+z = rm.limited_fluctuation_credibility(n=96_000, n_full=120_000)
+
+ind = rm.RateIndication(
+    experience_claims_pmpm=exp.claims_pmpm(),
+    manual_claims_pmpm=man.claims_pmpm(),
+    credibility=z,
+    current_rate=560,
+    target_loss_ratio=0.85,
+    trend_total_factor=exp.trend_factor(),
+    benefit_factor=1.00,
+    demographic_factor=1.01,
+)
+
+print(f"indicated rate   : {ind.indicated_rate():.2f}")
+print(f"indicated change : {ind.indicated_rate_change():+.2%}")
+
+# why did the rate move?
+print(ind.rate_change_decomposition().to_frame())
+
+# apply a renewal cap
+action = rm.renew(current_rate=560, indicated_rate=ind.indicated_rate(), cap=0.15)
+print(f"proposed (capped): {action.proposed_rate:.2f} ({action.proposed_change:+.2%})")
+```
+
+### Rate build-up with an audit trail
+
+```python
+import ratingmodels as rm
+
+med_par = rm.evaluate([
+    rm.start("Par Base Claim Cost", 941.63),
+    rm.add("$30 specialist copay", -11.44),
+    rm.multiply("Rating Region", 1.083),
+    rm.checkpoint("Medical Par Base Claim Cost"),
+])
+med_par.value          # final running total
+med_par.breakdown      # DataFrame: step, operation, label, operand, running_total
+
+# blend in-/out-of-network, then add the drug stream
+med = rm.participation_blend(med_par.value, nonpar=1478.56, participation_rate=0.90)
+total = rm.combine_streams({"Medical": med, "Drug": 323.67})
+total.value            # feeds into trend / credibility / retention
+```
+
+The package supplies the build-up *grammar*; you supply the factor values
+(cost-sharing, age/sex, area, ...) from your filed tables. `ManualRate` is a
+thin shortcut over this engine, so `ManualRate(...).breakdown()` returns the
+same audit trail.
+
+### Base rate and retention
+
+```python
+import ratingmodels as rm
+import pandas as pd
+
+# indicated base loss cost from book experience (off-balance method)
+book = pd.DataFrame({
+    "exposure": [24_000, 18_000, 30_000, 24_000],
+    "area":     [1.00, 1.20, 0.90, 1.05],
+    "tier":     [1.00, 1.10, 0.95, 1.25],
+    "loss":     [11_750_000, 11_400_000, 12_300_000, 15_200_000],
+})
+base = rm.base_rate_from_experience(book, "exposure", "loss",
+                                    factor_cols=["area", "tier"])
+base.base_loss_cost        # average loss cost / average relativity
+
+# gross claims up to a charged rate; the loss ratio falls out
+retention = rm.RetentionLoad.from_items(
+    fixed_expense_pmpm=22.0,
+    variable_items={"commission": 0.03, "premium_tax": 0.023, "aca_fees": 0.005},
+    profit_margin=0.03,
+)
+retention.gross_rate(540.0)            # charged rate
+retention.implied_loss_ratio(540.0)   # target loss ratio (an output)
+
+# rebalance the base when relativities are revised (hold level, then +8%)
+rm.rebalance_base_rate(current_base=base.base_loss_cost,
+                       current_avg_relativity=1.0928, new_avg_relativity=1.12,
+                       overall_change=0.08)
+```
+
+### GLM relativities
+
+```python
+import ratingmodels as rm
+from ratingmodels.datasets import sample_rating_data
+
+df = sample_rating_data(n=20_000)
+model = rm.GLMRelativities(family="poisson").fit(
+    df, response="claims", predictors=["area", "industry", "tier"],
+    exposure="exposure",                 # enters as a log offset
+    base_levels={"area": "A"},           # optional; defaults to modal level
+)
+print(model.base_value_)                 # fitted base frequency
+print(model.relativities_["industry"])   # relativity per level, base = 1.0
+```
+
+## Scope and honest limitations
+
+This is a modeling and workflow toolkit, not filed rate software. It does not
+manage rate filings, store filed factor tables with effective dating, or enforce
+state-specific rating rules. The pooling-charge helper is a simple group-level
+estimate; a production charge is normally derived book-wide or from an EVT tail
+model (see `extremeloss`). All bundled data in `ratingmodels.datasets` is
+synthetic and carries no assumptions.
+
+## License
+
+MIT. See [LICENSE](LICENSE).

ratingmodels-0.1.0/docs/api/base_loading.md

@@ -0,0 +1,5 @@
+# Base rate & retention
+
+::: ratingmodels.base_rate
+
+::: ratingmodels.loading

ratingmodels-0.1.0/docs/api/buildup.md

@@ -0,0 +1,3 @@
+# Build-up engine
+
+::: ratingmodels.buildup

ratingmodels-0.1.0/docs/api/credibility.md

@@ -0,0 +1,3 @@
+# Credibility
+
+::: ratingmodels.credibility

ratingmodels-0.1.0/docs/api/decomposition.md

@@ -0,0 +1,3 @@
+# Rate-change decomposition
+
+::: ratingmodels.decomposition

ratingmodels-0.1.0/docs/api/indication.md

@@ -0,0 +1,5 @@
+# Indication & blending
+
+::: ratingmodels.indication
+
+::: ratingmodels.blend

ratingmodels-0.1.0/docs/api/rates.md

@@ -0,0 +1,5 @@
+# Manual & experience rating
+
+::: ratingmodels.manual_rate
+
+::: ratingmodels.experience_rate

ratingmodels-0.1.0/docs/api/relativity.md

@@ -0,0 +1,3 @@
+# Relativities (GLM)
+
+::: ratingmodels.relativity

ratingmodels-0.1.0/docs/api/renewal.md

@@ -0,0 +1,5 @@
+# Constraints & renewal
+
+::: ratingmodels.constraints
+
+::: ratingmodels.renewal

ratingmodels-0.1.0/docs/api/trend.md

@@ -0,0 +1,3 @@
+# Trend
+
+::: ratingmodels.trend

ratingmodels-0.1.0/docs/changelog.md

@@ -0,0 +1,55 @@
+# Changelog
+
+All notable changes to this project are documented here. The format follows
+[Keep a Changelog](https://keepachangelog.com/), and the project adheres to
+[Semantic Versioning](https://semver.org/).
+
+## [0.1.0] - 2026-06-29
+
+Initial release.
+
+### Changed
+- **Credibility consolidated into `actuarialpy`.** Trend and credibility are
+  shared ecosystem primitives; `actuarialpy` is the home for credibility. The
+  credibility math is no longer duplicated here -- `ratingmodels.credibility`
+  and `blend` are thin adapters over `actuarialpy` (`full_credibility_claims`,
+  `limited_fluctuation_z`, `credibility_weighted_estimate`, and
+  `BuhlmannStraub.from_frame`). The public `ratingmodels` API and results are
+  unchanged. This removes the risk of the two Bühlmann-Straub implementations
+  drifting apart, and `actuarialpy` now uses the general unbiased estimator
+  (handling unequal period counts). Adds a dependency on `actuarialpy>=0.32.0`
+  and drops the direct `scipy` dependency.
+
+### Added
+- **Credibility** (`credibility`): `full_credibility_standard`,
+  `limited_fluctuation_credibility`, `buhlmann_credibility`, and empirical
+  `buhlmann_straub` with exposure weights.
+- **Trend** (`trend`): midpoint-to-midpoint trend factors, date helpers, and
+  utilization / unit-cost decomposition.
+- **Manual rating** (`manual_rate`): `ManualRate`, `manual_pmpm`,
+  `aggregate_demographic_factor`.
+- **Experience rating** (`experience_rate`): `ExperienceRate`, `pool_claims`,
+  `expected_excess_charge`.
+- **Rate build-up** (`buildup`): typed steps (`start`, `multiply`, `add`,
+  `segment_multiply`, `checkpoint`), an `evaluate` engine and `BuildUp` fluent
+  builder producing a reconciling breakdown, and `participation_blend` /
+  `combine_streams` for combining par/non-par and medical+drug streams.
+  `ManualRate` is reimplemented as a thin layer over the engine and gains
+  `breakdown()` / `steps()`.
+- **Base rate & off-balance** (`base_rate`): `base_rate_from_experience`,
+  `average_relativity`, `off_balance_factor`, `rebalance_base_rate`.
+- **Retention & loading** (`loading`): `RetentionLoad` (fundamental insurance
+  equation gross-up), `gross_rate`, `permissible_loss_ratio`. `ManualRate`,
+  `ExperienceRate`, and `RateIndication` accept an optional `retention` that
+  overrides the single-loss-ratio path with the full expense/profit build-up.
+- **Blending & indication** (`blend`, `indication`): `blend` and the
+  `RateIndication` orchestrator with build-up and loss-ratio methods.
+- **Rate-change decomposition** (`decomposition`): `decompose_rate_change`
+  with multiplicative and percentage-point contributions and an explicit
+  residual.
+- **GLM relativities** (`relativity`): `GLMRelativities` (Poisson / Gamma /
+  Tweedie via in-package IRLS), `FactorTable`, `one_way_relativities`.
+- **Constraints & renewal** (`constraints`, `renewal`): caps, floors, banding,
+  rounding, corridors; `renew` and `member_level_renewal`.
+- **Synthetic data** (`datasets`): `sample_claims`, `sample_rating_data`.
+- Full pytest suite (54 tests) and MkDocs Material documentation.