Skip to content

Native quantitative engines

ASRQuant keeps its finance-specific numerical logic inside the package. Pricing, curve construction, rate risk, calibration and related derivative analytics are not delegated to an external quantitative-finance pricing engine.

General-purpose scientific libraries such as NumPy, SciPy, pandas and statsmodels remain part of the numerical stack where appropriate. They provide numerical primitives, array operations, data structures and econometric routines; they do not replace ASRQuant's finance-model implementations.

Validation philosophy

Native implementations are validated through contracts that remain meaningful independently of any vendor or third-party finance library:

  • Analytical identities — e.g. European-option put-call parity.
  • Inverse / round-trip checks — e.g. price -> implied volatility -> repriced option.
  • Curve invariants — discount-factor positivity/ordering where the convention implies it, quote repricing and key-rate partition checks.
  • Limiting and boundary cases — explicit treatment of non-identifiable implied-volatility boundaries and degenerate inputs.
  • Deterministic conventions — explicit day-count, calendar, smoothing and PCA-orientation semantics.
  • Regression fixtures — confirmed defects receive tests that prevent silent reintroduction.
  • Reproducibility controls — deterministic seeds, experiment fingerprints and artifact lineage where stochastic workflows are involved.

Native validation benchmark

The repository includes benchmarks/native_engine_contracts.py. It exercises selected model identities using only ASRQuant and the scientific core. It is a validation aid, not a claim that the covered contracts exhaust every model risk.

PYTHONPATH=src python benchmarks/native_engine_contracts.py

A successful run reports the selected contracts as PASS. The package test suite remains the release gate and provides broader coverage.

Scope boundary

Native implementation does not imply that ASRQuant reproduces every instrument, convention or market feature found in large institutional libraries. The design goal is different: finance-specific behavior should be inspectable, testable and extensible directly in ASRQuant, with assumptions and conventions made explicit rather than hidden behind an external pricing engine.