Kuwala

A Unified, Arbitrage-Checked Quantitative Options & Volatility Research Library

What is Kuwala?

Kuwala is a high-performance quantitative finance library for derivatives pricing, arbitrage-free volatility surface modeling (SSVI, Dupire PDE), market data pipelines, high-frequency microstructure aggregation, and relative-value alpha signal research.

It pairs an approachable Python 3 research API with:

• Compiled Rust Core (kuwala_core): Memory-safe PyO3 numerical kernels with Chebyshev rational approximations (>2.2M ops/s).
• Ultra-Low Latency C++20 Engine (kuwala_cpp): Standalone SIMD-optimized engine achieving >11.9M pricing ops/s and >63.9M ticks/s.
• Julia Scientific Research Package (julia/): Pure Julia mathematical module with LLVM JIT vectorization reaching >12.3M ops/s.
• Scala / JVM Big Data Module (scala/): Primitive array batch pricing on Temurin JDK 17 / HotSpot C2 reaching >13.3M ops/s with near-zero parity error ($7.11 \times 10^{-15}$).
• Embedded Out-of-Core Lakehouse: Zero-server Hive-partitioned Parquet storage queried via embedded DuckDB and Apache Arrow.

Empirical Graphs & Visualizations

1. Arbitrage-Free Volatility Smile Gradients

Multi-tenor SSVI implied volatility smiles across expiries $T \in [0.08, 5.0\text{Y}]$. Demonstrates continuous log-moneyness skew, absence of butterfly arbitrage ($g(k) \ge 0$), and monotonic variance accumulation across maturities.

2. 3D Volatility Surface Topology

Continuous 3D volatility surface $(k, T) \mapsto \sigma(k, T)$ calibrated under Gatheral & Jacquier (2014) power-law formulation, providing a mathematically guaranteed arbitrage-free grid for discrete Dupire local volatility extraction.

3. Multi-Language Execution Throughput Comparison

Benchmark throughput measured across Pure Python, Compiled Rust (kuwala_core), Native C++20 (kuwala_cpp), Julia 1.12.7, and Scala 3.9.0 on identical 1,000,000-option batches.

4. Out-of-Core Columnar Storage Performance (DuckDB vs. Pandas)

Direct Parquet query benchmarking on 1,000,000 rows. Embedded DuckDB achieves a 3.2x query speedup (0.1472s vs. 0.4706s) while maintaining a bounded near-zero heap memory footprint (<0.1 MB RAM) via memory-mapped predicate pushdown.

Core Capabilities (v0.2.0)

Reproducible Multi-Language Benchmarks (1M Options)

Measured on Windows 11 (x86_64, AMD/Intel Multi-Core) across identical Black-Scholes pricing vectors:

Comprehensive Engineering Audit & Technical Reports

All documentation is backed by empirical logs, raw market data, and verified test executions:

• AUDIT_BASELINE.md: Baseline state, toolchain inventory, and fixed imports.
• ROADMAP_UMBRELLAS.md: Foundation 0.1 Umbrella (v0.1.0 → v0.5.0) vs Quant Engine 0.2 Umbrella (v0.5.0 → v1.5.0).
• BENCHMARK_REPORT.md: Cross-language throughput tables, cold vs warm distributions (p50/p95/p99).
• REAL_WORLD_VALIDATION.md: Live market data validation on SPY, QQQ, AAPL, MSFT options and FRED curves.
• NUMERICAL_VALIDATION.md: 100,000 Put-Call parity invariant checks, SSVI Durrleman condition, and Dupire grid convergence.
• GS_QUANT_RED_TEAM.md: Hostile head-to-head competition audit against Goldman Sachs' gs-quant.
• CPP_LOW_LATENCY.md: Standalone C++20 engine architecture, SIMD benchmarks, and 63.98M ticks/s aggregator.
• JULIA_INTEGRATION_REPORT.md: Julia 1.12.7 verification report (6/6 tests passing in 0.3s, 12.37M ops/s).
• SCALA_INTEGRATION_REPORT.md: Scala 3 & Temurin JDK 17 report (13.33M ops/s, Arrow bridge).
• R_INTEGRATION.md: R 4.6.0 research package interface, Arrow datasets, and ggplot2 gradient smiles.
• Q_KDB_INTEGRATION_REPORT.md: Transparent BLOCKED status report due to proprietary KX Systems commercial license.
• ZERO_COPY_AUDIT.md: Critical analysis of memory copies across PyO3 and DuckDB boundaries.
• MEMORY_AUDIT.md: Profiling memory retention across 1M option batches and out-of-core scans.
• CI_CD_AUDIT.md: CI/CD test isolation, packaging, and credential security.
• KNOWN_LIMITATIONS.md: Scope boundaries, Dupire input requirements, and licensing constraints.
• BREAKAGE_REPORT.md: Log of bugs and edge cases caught and remediated during hostile testing.
• REPRODUCIBILITY_REPORT.md: Step-by-step commands to reproduce every benchmark and test suite.
• VERSION_0.1_TO_1.5_VALIDATION.md: Feature-by-feature lifecycle matrix from v0.1 to v1.5.
• RELEASE_DECISION.md: Formal Go/No-Go release readiness certification for v0.2.0.
• REAL_IV_VALIDATION.csv: 4,013 real option quotes with nanodollar repricing error ($2.88 \times 10^{-9}$ median).

License

Kuwala is released under the Apache-2.0 License.