Kalman Filter
Linear-Gaussian state-space estimator; used for token-level price/velocity tracking.
The mathematics I actually ship — filters, factor research, volatility models, options pricers, risk metrics, microstructure. Implemented in Rust and Python across kokoro-alpha-lab and kokoro-pricing-service.
Linear-Gaussian state-space estimator; used for token-level price/velocity tracking.
Sigma-point nonlinear filtering with automatic warm-up for non-Gaussian dynamics.
Sequential importance resampling with distribution snapshots for multimodal posteriors.
Robust state-space filter with measurement gating and velocity inflation under adversarial noise.
Distribution-free calibrated prediction intervals around any point forecaster.
Frequency-domain decomposition for cycle detection and harmonic extraction.
Multi-scale time-frequency decomposition with regime-change detection.
Analytic signal yielding instantaneous phase and frequency.
Nonparametric trend isolation via L1-penalised smoothing.
Conditional variance forecasting with grid-search fit and variance targeting.
Exponentially-weighted variance estimator for short-horizon vol.
Barndorff-Nielsen & Shephard decomposition isolating jumps from realised variance.
Multi-scale rescaled-range analysis distinguishing trend persistence from mean reversion.
OLS-based stationarity test with MacKinnon critical values.
Quantile loss estimators from both fitted normals and empirical distributions.
Tail-mean loss beyond the VaR threshold.
Peak-to-trough underwater analysis on equity curves.
Risk-adjusted return with √252 scaling; the standard hurdle metric.
Aggregated Δ / Γ / Vega for options inventory.
Closed-form European pricer plus analytical Δ / Γ / Θ / Vega / ρ.
Implied-volatility extraction via gradient descent on Vega.
Discrete binomial pricer with backward induction for American features.
GBM path simulation with Box-Muller draws and confidence intervals.
Unconstrained (analytical) and constrained (projected gradient / SLSQP) weight allocation.
Risk-return locus swept across the target-return space.
Log-optimal bet sizing for repeated edges.
L2-based ΔBid − ΔAsk with level-improvement logic; classified into directional pressure.
Volume-weighted and outlier-robust price blends across DEX sources.
Depth-profiled execution cost from quote-API ladder sweeps.
Baseline / excitation / decay cascade model for event clustering.
Cholesky-sampled dependence with heavy-tailed variants for joint scenarios.
Biased-distribution reweighting for rare-event variance reduction.
Negatively-correlated path pairing and proxy-variable correction.
Uniform bucket coverage for Monte Carlo integrators.
Euclidean clustering with cohesion / separation quality scoring.
Covariance-weighted distance for anomaly and regime discrimination.
Logarithmic Market Scoring Rule — the cost-function AMM for prediction markets.
Twenty-four distinct factors, each its own crate with bespoke calibration — not re-skinned momentum.