finhjb 0.1.2

This is a Python library for solving Hamilton-Jacobi-Bellman (HJB) equations using JAX. It provides a framework for numerical solutions to dynamic optimization problems in finance and economics.

FinHJB

FinHJB is a Python library for solving one-dimensional Hamilton-Jacobi-Bellman (HJB) problems with JAX.

It provides a structured workflow for:

• defining model parameters, boundaries, policy rules, and residual equations,
• solving with policy iteration,
• searching for optimal boundaries,
• running parameter sensitivity analysis.

The package is built around typed interfaces (AbstractModel, AbstractPolicy, AbstractBoundary, AbstractParameter) and a high-level orchestrator (Solver).

Features

• JAX-based numerical core with optional JIT compilation
• Flexible policy updates via @explicit_policy and @implicit_policy
• Multiple boundary search methods (gauss_newton, lm, broyden, lbfgs, bisection, hybr, broyden1, krylov)
• Built-in continuation workflow for parameter paths (sensitivity_analysis)
• Serializable result containers (Grid, Grids, SensitivityResult)

Installation

This project requires Python >=3.13.

Install from source in editable mode:

pip install -e .

If you use uv:

uv sync

To include the development dependency group with uv:

uv sync --group dev

Quick Start

The minimal workflow has 4 pieces:

1. Parameter: immutable model parameters
2. Boundary: state/value boundaries
3. Policy: policy variable initialization and update rules
4. Model: HJB residual and optional boundary conditions

Then pass them to Solver.

from dataclasses import dataclass

import jax.numpy as jnp
from jaxtyping import Array

import finhjb as fjb


class Parameter(fjb.AbstractParameter):
  r: float = 0.06
  sigma: float = 0.10
  theta: float = 1.5


class PolicyDict(fjb.AbstractPolicyDict):
  investment: Array


@dataclass
class Boundary(fjb.AbstractBoundary[Parameter]):
  @staticmethod
  def compute_v_left(p: Parameter) -> float:
    return 0.9

  @staticmethod
  def compute_v_right(p: Parameter, s_max: float) -> float:
    return 1.0 + 0.1 * s_max


@dataclass
class Policy(fjb.AbstractPolicy):
  @staticmethod
  def initialize(grid: fjb.Grid, p: Parameter) -> PolicyDict:
    return PolicyDict(investment=jnp.full_like(grid.s, 0.1))

  @staticmethod
  @fjb.explicit_policy(order=1)
  def update_investment(grid: fjb.Grid) -> fjb.Grid:
    grid.policy["investment"] = jnp.maximum(1e-6, 0.5 * grid.v / grid.dv)
    return grid


@dataclass
class Model(fjb.AbstractModel[Parameter, PolicyDict]):
  @staticmethod
  def hjb_residual(v, dv, d2v, s, policy, jump, boundary, p):
    inv = policy["investment"]
    return -p.r * v + (s - inv) * dv + 0.5 * p.sigma**2 * d2v


parameter = Parameter()
boundary = Boundary(p=parameter, s_min=0.0, s_max=0.2)
model = Model(policy=Policy())

config = fjb.Config(
  pi_method="scan",
  pi_max_iter=50,
  pi_tol=1e-8,
)

solver = fjb.Solver(
  boundary=boundary,
  model=model,
  policy_guess=True,
  number=500,
  config=config,
)

state, history = solver.solve()
print(state.converged, state.best_error)
df = state.df

Boundary Search

If your model provides boundary_condition() targets, you can search for boundaries directly:

search_state = solver.boundary_search(method="bisection", verbose=True)
print(search_state.optimal_boundary)
print(search_state.best_error)

Supported methods:

• gauss_newton
• lm
• broyden
• lbfgs
• bisection
• hybr
• broyden1
• krylov

Sensitivity Analysis

Run path-following continuation on one parameter:

import jax.numpy as jnp

result = solver.sensitivity_analysis(
  method="hybr",
  param_name="sigma",
  param_values=jnp.linspace(0.05, 0.20, 10),
)

print(result.df.head())

Save and Load

Main objects support pickle-based save/load:

• Grid.save(path)
• Grids.save(path)
• SensitivityResult.save(path)
• fjb.load(path)

state.grid.save("single_grid")
loaded = fjb.load("single_grid")

Public API

Top-level exports include:

• Config
• Solver
• Grid, Grids, ImmutableBoundary
• AbstractBoundary, BoundaryConditionTarget
• AbstractModel, AbstractParameter
• AbstractPolicy, AbstractPolicyDict
• explicit_policy, implicit_policy
• LinearInitialValue, QuadraticInitialValue
• load

Examples

Two complete examples are available in:

• src/example/BCW2011Liquidation.py
• src/example/BCW2011Hedging.py

They show full model construction, solver configuration, and boundary search.

Documentation

Additional guides are in docs/:

• docs/getting-started.md
• docs/modeling-guide.md
• docs/solver-guide.md
• docs/api-reference.md