Quantitative Toolbox

Horizon’s Rust core includes a library of quantitative methods exposed to Python via PyO3. These cover state estimation, dependence modeling, market microstructure, execution optimization, pricing, and statistical testing.

All functions are called as hz.function_name() and run natively in Rust for performance.

State Estimation & Filtering

Module	What it does
Kalman Filters	Linear and unscented filters for noise removal, hedge ratios, spread tracking
Change Point Detection	Bayesian online detection (BOCPD) for regime shifts
Particle Filter	Sequential Monte Carlo for nonlinear, non-Gaussian dynamics

Dependence & Risk Modeling

Module	What it does
Copula Dependence	Bivariate copulas (Gaussian, Clayton, Gumbel, Frank) for tail dependence
Vine Copulas	C-vine and D-vine for 3+ market tail risk
Entropy Pooling	Meucci’s framework for blending views with market probabilities
Robust Portfolio	Worst-case optimization under parameter uncertainty
HRP & Denoising	Hierarchical Risk Parity + Marcenko-Pastur covariance cleaning

Market Microstructure

Module	What it does
Market Making	Avellaneda-Stoikov with inventory skew and competitive spread blending
HJB Market Making	Hamilton-Jacobi-Bellman PDE for optimal inventory-aware quoting
Queue Position	Fill probability and expected fill time for limit orders
Cross-Impact	Multi-market price impact estimation
Volatility Signature	Two-scale realized vol and optimal sampling frequency
ACD Durations	Inter-event timing analysis for informed trade detection
Microstructure Invariance	Kyle-Obizhaeva scaling laws for bet sizing and impact

Execution

Module	What it does
Optimal Execution	Garleanu-Pedersen and Almgren-Chriss execution scheduling
Optimal Stopping	Longstaff-Schwartz for optimal exit timing

Pricing

Module	What it does
Characteristic Function Pricing	Heston, Merton, Variance Gamma for binary options
Logit-Space Pricing	Black-Scholes framework for prediction markets
Belief-Volatility Surface	Streaming EM decomposition of diffusion and jumps
Breeden-Litzenberger	Risk-neutral density from options chains

Signal & Feature Construction

Module	What it does
Bars & Labeling	Information-driven bars + triple barrier labeling (de Prado)
Fractional Differentiation	Stationarity with memory preservation (de Prado)
Elastic Net Selection	Lasso/ridge/elastic net for feature selection
Alpha Research	Meta-labeling, feature importance, alpha decay
Lead-Lag Networks	Hayashi-Yoshida correlation and Granger causality
Graph Analysis	Correlation graphs, MST, community detection

Volatility & Statistics

Module	What it does
Volatility Suite	6 estimators: Parkinson, Garman-Klass, Yang-Zhang, EWMA, rolling
Bootstrap Methods	11 bootstrap variants + jackknife bias correction
Correlation Regime	Live correlation tracking with regime shift detection