bsde-cfft-sv 0.0.0

MATH5030: BSDE-CFFT API aligned with BSDE-CFFT-Method-For-Stochastic-Volatility-Models-main.

bsde-cfft-sv

BSDE-CFFT option pricing tools for Black-Scholes, Heston, and GARCH-diffusion models.

What Problem It Solves

This project provides a compact Python API for pricing European call options and computing sensitivities under several diffusion models using BSDE-CFFT methods.

The package is useful when you want to:

• validate a 1D BSDE-CFFT solver against the analytic Black-Scholes formula,
• price options in 2D stochastic-volatility models such as Heston,
• study a GARCH-diffusion volatility model with the same numerical interface,
• compare numerical accuracy against benchmark prices,
• test how grid resolution, damping, and time discretization affect the results.

Implemented methods:

• price_black_scholes_1d(...): 1D BSDE-CFFT validation problem with analytic Black-Scholes comparison,
• price_heston_2d(...): 2D BSDE-CFFT solver for the Heston stochastic-volatility model,
• price_garch_2d(...): 2D BSDE-CFFT solver for a GARCH-diffusion stochastic-volatility model,
• grid_damping_sensitivity(...): helper for sensitivity studies on grid and damping parameters.

Main parameters:

• S0: initial asset price,
• K: strike price,
• r: risk-free interest rate,
• T: maturity,
• sigma: volatility in the Black-Scholes model,
• kappa, theta, xi, rho, V0: Heston model parameters,
• mu, a, b, c, rho, V0: GARCH-diffusion model parameters,
• N: 1D spatial grid size,
• Nx, Nv: 2D grid sizes in log-price and variance,
• L, Lx, Lv: truncation widths of the computational domain,
• n_steps: number of backward time steps, with dt = T / n_steps,
• alpha, alpha_x, alpha_v: damping parameters used in the FFT-based scheme,
• v_boundary: boundary handling mode in the variance direction.

In practice:

• larger grids such as bigger N, Nx, Nv usually improve spatial accuracy but cost more,
• larger n_steps refines the time discretization but increases runtime,
• damping parameters such as alpha_x can materially affect stability and accuracy.

Installation

Install from PyPI:

pip install bsde-cfft-sv

Install locally from the repository:

git clone https://github.com/ranshaodedatuimaomao/MATH5030.git
cd MATH5030
python3 -m pip install -e .

Requirements:

• Python >=3.10
• numpy
• scipy
• pandas
• pyfeng>=0.3

Quick Start

Minimal Black-Scholes example:

from bsde_cfft_sv import price_black_scholes_1d

result = price_black_scholes_1d(
    S0=100.0,
    K=100.0,
    sigma=0.2,
    T=1.0,
    N=4096,
    n_steps=1000,
)

print(result.price)
print(result.delta)
print(result.analytic_price)
print(result.abs_price_error)

Minimal Heston example:

from bsde_cfft_sv import price_heston_2d

result = price_heston_2d(
    S0=100.0,
    K=100.0,
    V0=0.04,
    Nx=96,
    Nv=48,
    n_steps=700,
)

print(result.price)
print(result.delta)
print(result.z_x, result.z_v)
print(result.reference_price)

API Reference

Return Types

OneDResult

Field	Type	Description
price	float	Numerical option price
delta	float	Numerical delta
analytic_price	float	Closed-form Black-Scholes price
analytic_delta	float	Closed-form Black-Scholes delta
abs_price_error	float	Absolute error versus analytic price
abs_delta_error	float	Absolute error versus analytic delta

TwoDResult

Field	Type	Description
price	float	Numerical option price
delta	float	Numerical delta
z_x	float	BSDE sensitivity in the log-price direction
z_v	float	BSDE sensitivity in the variance direction
reference_price	float | None	Benchmark price when available
abs_price_error	float | None	Absolute error versus the benchmark

price_black_scholes_1d(...) -> OneDResult

Prices a European call in the Black-Scholes model and compares the numerical result with the analytic formula.

Parameters:

Parameter	Type	Description
S0	float	Initial spot
K	float	Strike
r	float	Risk-free rate
sigma	float	Constant volatility
T	float	Maturity
L	float	Log-price truncation width
N	int	Number of spatial grid points
n_steps	int	Number of backward time steps
alpha	float	Damping parameter

price_heston_2d(...) -> TwoDResult

Prices a European call in the Heston model using the 2D BSDE-CFFT solver.

Parameters:

Parameter	Type	Description
S0	float	Initial spot
K	float	Strike
r	float	Risk-free rate
kappa	float	Mean-reversion speed of variance
theta	float	Long-run variance
xi	float	Vol-of-vol
rho	float	Correlation
V0	float	Initial variance
T	float	Maturity
Nx, Nv	int	Grid sizes in log-price and variance
Lx, Lv	float	Truncation widths
n_steps	int	Number of backward time steps
alpha_x, alpha_v	float	Damping parameters
v_boundary	str	Variance-boundary treatment
fft_workers	int | None	Number of FFT workers

price_garch_2d(...) -> TwoDResult

Prices a European call in the GARCH-diffusion model using the same numerical interface.

Parameters:

Parameter	Type	Description
S0	float	Initial spot
K	float	Strike
r	float	Risk-free rate
mu	float	Drift parameter kept for model compatibility
a	float	Mean-reversion speed
b	float	Long-run variance level
c	float	Vol-of-vol coefficient
rho	float	Correlation
V0	float	Initial variance
T	float	Maturity
Nx, Nv	int	Grid sizes in log-price and variance
Lx, Lv	float	Truncation widths
n_steps	int	Number of backward time steps
alpha_x, alpha_v	float	Damping parameters
v_boundary	str	Variance-boundary treatment
fft_workers	int | None	Number of FFT workers

grid_damping_sensitivity(...) -> dict[str, list[dict[str, float | str | None]]]

Runs lightweight sensitivity studies:

• Heston sensitivity with respect to Lv,
• GARCH sensitivity with respect to alpha_x.

Key parameters:

Parameter	Type	Description
Nx, Nv	int	2D grid sizes
heston_n_steps	int	Time steps for Heston runs
garch_n_steps	int	Time steps for GARCH runs
Lx	float	Log-price domain width
V0	float	Initial variance
heston_Lv_values	tuple[float, ...]	Tested Lv values
garch_alpha_values	tuple[float, ...]	Tested damping values

License

MIT

Demo Notebook

Notebook:

• notebooks/Demo.ipynb

Open in Colab: