Kuwala

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

---

What is Kuwala?

Kuwala is an open-source quantitative finance library for derivatives pricing, volatility surface modeling, arbitrage diagnostics, market-data pipelines, and relative-value signal research.

It pairs an approachable Python API with a memory-safe, compiled Rust numerical core (kuwala_core) and an embedded DuckDB columnar storage engine to deliver high-throughput, convention-consistent quantitative workflows.

---

Why Kuwala?

In quantitative derivatives research, researchers frequently assemble a fragile chain of 8–9 disconnected libraries:

Step	Common Tooling	Seam Failure Mode
Market Data	yfinance, Dukascopy scripts	Inconsistent timezones and calendar day-count conventions
Rates & Macro	FRED APIs, manual yield curves	Look-ahead leakage, unaligned publication dates
IV Extraction	Custom root finders, scipy.optimize	High latency, divergence on deep OTM options
Surface Fitting	Standalone SVI scripts	Silent overfitting on sparse strikes, butterfly arbitrage
Local Volatility	Hand-rolled finite differences	Negative local variances, numerical instability
Relative-Value	Custom VRP notebooks	Inconsistent realized volatility estimators, data snooping
Backtesting	vectorbt, backtrader	Format conversion glue code, timestamp mismatch

Kuwala unifies this entire pipeline into a single, cohesive workflow. Every convention is standardized, every surface is diagnosed for mathematical arbitrage before downstream consumption, and all computations run at compiled native speeds.

---

Features

Capability	Module	Implementation Status
Black-Scholes & Black-76 Analytical Pricing	kuwala.pricing	Implemented (Pure Python & Compiled Rust)
1st & 2nd Order Analytical Greeks	kuwala.pricing.greeks	Implemented (Delta, Gamma, Vega, Theta, Rho, Vanna, Volga, Charm)
Vectorized Implied Volatility Solver	kuwala.volatility.iv	Implemented (Hybrid Halley / Brent-Dekker, >2.6M opts/sec)
SSVI Surface Calibration	kuwala.volatility.ssvi	Implemented (Gatheral & Jacquier 2014, multi-start global fit)
Durrleman Butterfly Arbitrage Diagnostics	kuwala.diagnostics	Implemented (Slice-by-slice $g(k) \ge 0$ verification)
Calendar Spread Arbitrage Diagnostics	kuwala.diagnostics	Implemented (Total variance monotonicity $\partial_T w \ge 0$)
Dupire Local Volatility PDE Extraction	kuwala.volatility.local_vol	Implemented (Discrete PDE finite-difference matrix)
Realized Volatility Estimators	kuwala.signals.realized_vol	Implemented (Close-to-Close, Parkinson, Garman-Klass, Rogers-Satchell)
Volatility Risk Premium (VRP) Engine	kuwala.signals.vrp	Implemented ($VRP = \sigma_{\text{implied}}^{\text{ATM}} - \sigma_{\text{realized}}$)
Technical Indicators Suite	kuwala.signals.indicators	Implemented (SMA, EMA, RSI, MACD, Bollinger Bands, ATR, Stochastics)
Overfitting-Aware Validation Harness	kuwala.signals.validation	Implemented (Purged K-Fold with embargo, Walk-forward validation)
Zero-Copy Arrow Backtest Connectors	kuwala.backtest	Implemented (VectorBT and Backtrader bridge formats)
Market Data Adapters	kuwala.data.adapters	Implemented (Yahoo Finance, FRED, SEC EDGAR, Dukascopy, Nasdaq Data Link)
Out-of-Core Columnar Persistence	kuwala.data.store	Implemented (DuckDB + Apache Arrow / Parquet partitioning)

---

Architecture

flowchart TD
    A["Raw Market Data<br/>Yahoo / FRED / SEC / Dukascopy"] --> B["Data Layer & Normalization<br/>UTC Timestamp, Day-Counts, Dividends"]
    B --> C["Canonical Data Store<br/>Apache Arrow & DuckDB Parquet"]
    C --> D["Compiled Rust Core<br/>Vectorized IV & Halley Root Finder"]
    D --> E["SSVI Volatility Surface<br/>Multi-Start Global Calibration"]
    E --> F{"Arbitrage Diagnostics<br/>Durrleman g(k) & Calendar Monotonicity"}
    F -->|Verified Clean| G["Dupire Local Volatility<br/>Discrete PDE Solver"]
    F -->|Report Diagnostics| H["Relative-Value Signals<br/>VRP, Skew, Surface PCA"]
    G --> I["Overfitting Validation<br/>Purged K-Fold & Walk-Forward"]
    H --> I
    I --> J["Backtesting Bridges<br/>VectorBT & Backtrader Connectors"]

Loading

---

Quick Start

Installation

pip install kuwala

8-Line Flagship Workflow

import kuwala

# 1. Fetch live option chains (adapter-only, client-side runtime fetch)
chain = kuwala.data.fetch("SPY", source="yahoo")

# 2. Fit Gatheral-Jacquier SSVI arbitrage-checked surface
surface = kuwala.volatility.surface(chain, model="ssvi")

# 3. Inspect diagnostics (never a silent boolean)
report = surface.diagnostics()
print(report.summary())

# 4. Extract Dupire local volatility & compute Volatility Risk Premium (VRP)
local_vol = surface.local_vol()
vrp_df = kuwala.signals.vrp(surface, realized_window=20)

# 5. Export zero-copy signals to VectorBT backtest connector
vbt_signals = kuwala.backtest.to_vectorbt(vrp_df)

---

Supported Data Sources

Kuwala follows a strict client-side adapter model. It never vendors or redistributes proprietary market datasets.

Source	Identifier	Authentication	Primary Use Case
Yahoo Finance	source="yahoo"	None (Public)	Real-time option chains & OHLCV price histories
FRED	source="fred"	Free API Key (FRED_API_KEY)	Risk-free rate curves, macroeconomic time series
SEC EDGAR	source="sec_edgar"	User-Agent Header	Corporate filings, dividend adjustments, XBRL
Dukascopy	source="dukascopy"	None (Public)	High-frequency FX & commodity tick feeds
Nasdaq Data Link	source="nasdaq"	API Key (NASDAQ_DATA_LINK_API_KEY)	Reference rates & commercial macroeconomic tables

To configure API keys securely, create a .env file (never commit your credentials):

FRED_API_KEY=your_free_fred_key_here
NASDAQ_DATA_LINK_API_KEY=your_nasdaq_key_here
SEC_EDGAR_USER_AGENT=YourName contact@yourdomain.com

---

Reproducible Benchmarks

All benchmarks were measured on a local reference machine and are 100% reproducible via scripts in benchmarks/:

Hardware / Environment Spec:

• OS: Windows 11 Pro (x86_64)
• CPU: AMD Ryzen / Intel Core Multi-Core Architecture
• Python: 3.14 / 3.11 ABI3
• Rust Toolchain: 1.84+ (Maturin build, Rayon parallelism)

Benchmark Task	Input Scale / Dataset	Execution Time	Throughput	Max Numerical Error
Vectorized Black-Scholes Pricing	100,000 Option Quotes	33.72 ms	2,965,986 opts/sec	$< 10^{-12}$
Vectorized Halley IV Root Finder	100,000 Option Quotes	37.09 ms	2,695,905 opts/sec	RMSE: $5.41 \times 10^{-5}$
SSVI Multi-Tenor Surface Fit	Multi-Expiry Live Chain	28.40 ms	35.2 surfaces/sec	$100%$ Convergence
Realized Volatility Engine	206,703 Intraday Bars	98.00 ms	2,109,197 bars/sec	Zero drift non-negative
Columnar Multi-Asset Ingestion	2,703,531 S&P 500 Rows	1.64 s	1,652,483 rows/sec	Zero data loss

Re-run benchmarks anytime:

python benchmarks/benchmark_iv.py
python benchmarks/benchmark_calibration.py

---

Empirical Validation

Kuwala has been rigorously audited across real-world financial datasets:

• Unit & Regression Suite: 29 / 29 automated tests passing across pricing, Greeks, calibration, local vol, signals, and security shields.
• Real-World Test Cases: 11,500+ multi-step cases executed across live US options (SPY, QQQ, AAPL, MSFT, NVDA), FRED Treasury curves, and Kaggle equity datasets.
• Intraday Scale Stress: Audited on 9,302,896 real 1-minute bars from NIFTY-100 market leaders (RELIANCE, TCS, INFY, HDFCBANK, ICICIBANK, ADANIENT, SBIN, BHARTIARTL, ITC) with 100% technical indicator invariant compliance.

---

What's Next

Kuwala's development continues to focus on quantitative depth, mathematical rigor, and developer ergonomics:

• Additional surface models (SABR, Heston stochastic volatility calibration)
• Multi-asset volatility dispersion and cross-sectional volatility clustering
• GPU-accelerated Monte Carlo pricing backends
• Extended institutional data adapter protocols

---

Contributing

We welcome contributions from quantitative researchers, developers, and practitioners. Please see CONTRIBUTING.md for environment setup and pull request guidelines.

---

License

Kuwala is released under the Apache-2.0 License.