QuantNova.jl

Appearance

Validation ​

This page documents how QuantNova is validated for numerical correctness. It focuses on how to reproduce checks rather than reporting static numbers.

What We Validate ​

  • Black-Scholes parity and edge cases (e.g., T → 0, σ → 0, deep ITM/OTM).

  • Greeks accuracy versus analytical formulas.

  • Monte Carlo sanity checks versus analytical prices where available.

  • Yield curve bootstrapping and bond analytics consistency.

  • SABR / Heston calibration convergence and stability on known inputs.

  • Statistical metrics (Sharpe, confidence intervals) against reference formulas.

How to Reproduce ​

1) Core Test Suite ​

Runs the unit tests that cover correctness and edge cases:

bash
julia --project=. -e 'using Pkg; Pkg.test()'

Key test files (see test/):

  • test/instruments.jl

  • test/ad.jl

  • test/interest_rates.jl

  • test/montecarlo.jl

  • test/statistics.jl

  • test/calibration.jl

2) Accuracy Benchmarks ​

We keep targeted accuracy scripts under:

  • benchmarks/accuracy/black_scholes_parity.jl

  • benchmarks/accuracy/heston_reference.jl

Run a specific script:

bash
julia --project=. benchmarks/accuracy/black_scholes_parity.jl

3) QuantLib Cross-Checks (Optional) ​

If you have QuantLib installed, you can compare outputs directly using:

bash
julia --project=. benchmarks/comparison/quantlib_comparison.jl

This script validates prices and Greeks and reports differences.

Notes on Units and Conventions ​

  • Theta: QuantNova returns theta per year. Some libraries return per day.

  • Vega/Rho: QuantNova reports per 1% change (scaled by 0.01), which matches most practitioner conventions.

When comparing results, always align conventions first.

Interpreting Results ​

Validation focuses on:

  • Exact identities (e.g., put-call parity).

  • Small absolute error for analytical comparisons.

  • Consistency across implementations (e.g., AD vs analytical).

If any check fails, open the relevant test file and review assumptions (inputs, conventions, or units).