actudist 0.1.0

Actuarial probability distributions for P&C loss modeling: severity and frequency families, MLE fitting, layer statistics, goodness-of-fit.

actudist

Python library for P&C loss distributions. Severity and frequency families follow the parameterizations in Klugman, Loss Models. Every severity distribution exposes layer statistics (LEV, excess pure premium, ILF) as methods. MLE accepts right-censored and one- or two-sided truncated samples.

Installation

pip install actudist

From source:

git clone https://github.com/CosmikArt/actudist.git
cd actudist
pip install -e ".[dev]"

Quickstart

import numpy as np
from actudist import DistributionFitter, GoodnessOfFit
from actudist.severity.lognormal import Lognormal

rng = np.random.default_rng(42)
claims = Lognormal(mu=8.0, sigma=1.2).rvs(size=2000, random_state=rng)

fit = Lognormal()
fit.mle_fit(claims)
print(f"mu={fit.mu:.3f}  sigma={fit.sigma:.3f}")
print(f"AIC={fit.aic(claims):.1f}  BIC={fit.bic(claims):.1f}")
print(f"LEV(50000)={fit.limited_expected_value(50_000):.1f}")
print(f"EPP(50000)={fit.excess_pure_premium(50_000):.1f}")

fitter = DistributionFitter(["Lognormal", "Pareto", "Weibull"])
ranking = fitter.fit_and_rank(claims, criterion="aic")
for row in ranking:
    print(f"{row['name']:12s}  aic={row['aic']:.1f}")

gof = GoodnessOfFit(fit, claims)
result = gof.ks_test(n_boot=200, random_state=42)
print(f"KS stat={result['statistic']:.4f}  p={result['p_value']:.3f}")

Modules

• actudist.severity: 13 continuous distributions (Exponential, Pareto, Lognormal, Weibull, Gamma, Burr XII, Log-Logistic, Paralogistic and its inverse, Inverse Gaussian, Transformed and Inverse Transformed Gamma, Transformed Beta).
• actudist.frequency: Poisson, Binomial, Geometric, Negative Binomial, ZIP, ZINB.
• actudist.fitting: registries and DistributionFitter for AIC/BIC ranking.
• actudist.gof: KS and Anderson-Darling with parametric-bootstrap p-values, chi-squared, PP and QQ plots.
• actudist.layers: excess pure premium, ILF, finite-layer expected loss.

profile_likelihood_ci lives on the base class; call it on any fitted instance.

References

• Klugman, S. A., Panjer, H. H., & Willmot, G. E. Loss Models: From Data to Decisions (5th ed.). Wiley.
• Hogg, R. V. & Klugman, S. A. Loss Distributions. Wiley.
• Society of Actuaries. Exam STAM / FAM syllabus and study materials.

Contributing

Run pytest before sending a PR.

Author

Isaac López

MIT License. See LICENSE.