risksim 0.1.0

Portfolio-level loss simulation and aggregate treaty modeling in Python

risksim

risksim is a Python package for portfolio-level loss simulation and aggregate treaty modeling.

It is designed to sit on top of stochastic loss models that expose a .sample(size) method. In particular, it pairs naturally with lossmodels, where lossmodels handles loss-model generation and risksim handles portfolio composition, aggregate contracts, and simulation summaries.

Features

• combine multiple simulated loss components into a portfolio
• apply aggregate annual layers
• build simple multi-layer aggregate contract programs
• summarize gross, ceded, and retained loss
• compute simulation-based risk measures such as mean, VaR, and TVaR

Installation

From PyPI

pip install risksim

From source

git clone https://github.com/michaelabryant/risksim.git
cd risksim
pip install -e .

Optional companion package

risksim works with any model that implements:

sample(size: int = 1) -> np.ndarray

Many of the examples below use lossmodels, which provides aggregate actuarial loss models that integrate naturally with risksim.

If you want to run the lossmodels examples locally, install lossmodels as well.

Quick start

Any object with a .sample(size) method can be used inside a Portfolio.

import numpy as np

from risksim import AggregateLayer, Portfolio, PortfolioItem


class ConstantModel:
    def __init__(self, value: float) -> None:
        self.value = float(value)

    def sample(self, size: int = 1) -> np.ndarray:
        return np.full(size, self.value, dtype=float)

    def mean(self) -> float:
        return self.value

    def variance(self) -> float:
        return 0.0


medical = ConstantModel(200.0)
rx = ConstantModel(50.0)

portfolio = Portfolio(
    [
        PortfolioItem("medical", medical),
        PortfolioItem("rx", rx),
    ]
)

contract = AggregateLayer(
    attachment=100.0,
    limit=75.0,
    share=1.0,
    name="aggregate_xol",
)

result = portfolio.simulate(size=10_000, contract=contract)

print("net mean:", result.mean())
print("gross mean:", result.gross_mean())
print("ceded mean:", result.ceded_mean())
print("VaR 99%:", result.var(0.99))
print("TVaR 99%:", result.tvar(0.99))
print(result.summary())

Using lossmodels

A common workflow is to build aggregate loss models with lossmodels and then combine them with risksim.

from lossmodels.aggregate import CollectiveRiskModel
from lossmodels.frequency import Poisson
from lossmodels.severity import Lognormal

from risksim import AggregateLayer, Portfolio, PortfolioItem

medical = CollectiveRiskModel(
    Poisson(lam=2.0),
    Lognormal(mu=2.0, sigma=0.3),
)

rx = CollectiveRiskModel(
    Poisson(lam=1.0),
    Lognormal(mu=1.5, sigma=0.2),
)

portfolio = Portfolio(
    [
        PortfolioItem("medical", medical),
        PortfolioItem("rx", rx),
    ]
)

contract = AggregateLayer(
    attachment=50.0,
    limit=100.0,
    name="agg_xol",
)

result = portfolio.simulate(size=50_000, contract=contract)
print(result.summary())

Contract programs

risksim also supports simple multi-layer aggregate contract programs.

from lossmodels.aggregate import CollectiveRiskModel
from lossmodels.frequency import Poisson
from lossmodels.severity import Lognormal

from risksim import AggregateLayer, ContractProgram, Portfolio, PortfolioItem

line = CollectiveRiskModel(
    Poisson(lam=4.0),
    Lognormal(mu=3.5, sigma=0.7),
)

portfolio = Portfolio([PortfolioItem("line", line)])

program = ContractProgram(
    [
        AggregateLayer(attachment=50.0, limit=50.0, name="layer_1"),
        AggregateLayer(attachment=100.0, limit=100.0, name="layer_2"),
    ],
    name="two_layer_tower",
)

result = portfolio.simulate(size=20_000, contract=program)

print("gross mean:", result.gross_mean())
print("ceded mean:", result.ceded_mean())
print("retained mean:", result.retained_mean())
print("layer means:", result.layer_means())

Public API

from risksim import AggregateLayer, ContractProgram, Portfolio, PortfolioItem, SimulationResult

Core concepts

PortfolioItem

Wraps one simulated component with a name and optional weight.

Portfolio

Combines multiple simulated components into a total portfolio loss.

AggregateLayer

Applies a single aggregate annual contract to simulated gross losses.

For annual gross loss S, ceded loss is:

share * min(max(S - attachment, 0), limit)

with no cap if limit=None.

ContractProgram

Combines multiple aggregate layers applied to the same gross loss.

SimulationResult

Stores simulation outputs and provides convenience methods such as:

• mean()
• variance()
• std()
• var(q)
• tvar(q)
• prob_exceeding(threshold)
• summary()

If retained losses are present, the primary losses view is net/retained loss. Otherwise it is gross loss.

Example scripts

The examples/ directory contains runnable scripts:

• examples/basic_portfolio.py
• examples/aggregate_layer.py
• examples/contract_program.py

Run them with:

python examples/basic_portfolio.py
python examples/aggregate_layer.py
python examples/contract_program.py

Project scope

The current version focuses on:

• portfolio-level simulation from component loss models
• aggregate annual treaty application
• simple non-overlapping multi-layer programs
• simulation-based portfolio summaries

It does not yet model:

• dependence between components
• occurrence-based contracts requiring claim-level paths
• reinstatements
• capital allocation
• premium or pricing workflows beyond simulated summaries

Development

Run the test suite with:

python -m pytest -q

License

MIT