bs-pricer

A small, production-ish Black-Scholes pricer in modern C++20 with a clean API, unit tests, and CI.

Build

cmake -S . -B build -DCMAKE_BUILD_TYPE=Release -DBS_BUILD_TESTS=ON
cmake --build build -j
ctest --test-dir build --output-on-failure

CLI

./build/bs_cli --type call --S 100 --K 100 --r 0.05 --sigma 0.2 --T 1

Library API

• price_call(const Params&), price_put(const Params&)
• greeks_call(const Params&), greeks_put(const Params&)
• Params{S,K,r,sigma,T} with validate()
• norm_pdf, norm_cdf helpers

Numerics

see docs/NUMERICS.md for formulas, units, and tolerances. Summary:

• $C = S N(d_1) - K e^{-rT} N(d_2), \quad P = K e^{-rT} N(-d_2) - S N(-d_1)$
• Greeks: $\Delta, \Gamma, \text{vega}, \Theta, \rho$
  (vega per $1.0 , \sigma$, theta per year)

Why C++20?

Feature richness (e.g., constexpr, library polish) with excellent toolchain maturity on Ubuntu 22.04 CI.

Benchmark

Built in Release on an Azure Standard B1s (1 vCPU, 1 GiB), gcc (Ubuntu 11.x), CMake Ninja.

./build/bs_bench --type call --iters 300000

$Time: 10.34 ms \space | \space ns/eval: 34.47 \space | \space evals/s: ~29.0M$

./build/bs_bench --type put --iters 300000

$Time: 10.64 ms \space | \space ns/eval: 35.45 \space | \space evals/s: ~28.2M$

Numbers vary with hardware/load; CI builds the bench but does not run it.

Roadmap

• Stage-2: Monte Carlo engine + variance reduction
• Stage-3: SIMD/AVX + cache-aware layouts + microbenchmarks