Skip to content

CHANGELOG.md

Latest commit

 

History

History
81 lines (63 loc) · 5.94 KB

CHANGELOG.md

File metadata and controls

81 lines (63 loc) · 5.94 KB
Raw

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[1.1.0] - 2026-03-28

Added

  • Wang Transform MLE (oracle3.pricing.wang_mle): full maximum likelihood estimator implementing the core model from Yang (2026), "Pricing Prediction Markets: Risk Premiums, Incomplete Markets, and a Decomposition Framework"

    • Pooled and hierarchical estimation with analytic gradients
    • Three SE estimators: Fisher, sandwich robust, Liang-Zeger clustered sandwich
    • Numerical Hessian with eigenvalue regularization and BLAS chunk-size safety
    • Model comparison (LR tests, AIC/BIC, pseudo-R²)
    • Empirical priors: Polymarket λ=0.166, Kalshi=0.187, Metaculus=0.287, Manifold=-0.218

  • Probabilistic fair value engine (oracle3.pricing): full pricing module grounded in Yang (2026)

    • Three distortion families: probit (Wang 2000), dual power (Denneberg 1994), proportional hazard
    • Exact empirical coefficients: λ_i = 0.259 - 0.072·ln(1+V) + 0.143·ln(1+D) - 0.477·|p-0.5|
    • Time-varying model: γ₁=-0.156·τ + γ₂=0.074·τ², half-life 33-77% of contract lifetime
    • Volume-stratified alpha targeting: >$10K volume → λ≈0 (premium competed away); $500-$10K = sweet spot
    • Hybrid calibrator: batch MLE (historical data) + streaming EWMA (live trading) with hierarchical shrinkage
    • Premium lifecycle tracker with polynomial decay fitting and optimal entry timing
    • Contract microstructure scorer (volume, spread, duration, extremity, book depth)

  • Model sensitivities (Greeks) (oracle3.pricing.greeks): analytic derivatives of the Wang model

    • dp/dλ, dp/dp*, premium ratio, Kelly fraction, edge decay rate
    • Favorite-longshot bias proven as theorem: overpricing ratio monotonically decreasing in p*
    • Batch computation for portfolio-level risk decomposition

  • Model-informed Kelly sizing (oracle3.trading.sizing): position sizing with Wang-derived edge

    • Kelly criterion with model edge, confidence scaling, and volume-tier gating
    • Automatic skip of very-high-volume markets where premium is already competed away
    • Inspired by three-tier sizing approaches in agent-native prediction market systems

  • Edge-weighted capital allocator (oracle3.trading.allocator): multi-strategy budget allocation

    • Risk-adjusted scoring (PnL / |drawdown|) with 30-day exponential time decay
    • Premium-alpha strategy bonus, reserve capital, per-strategy caps
    • Graceful degradation: performance-weighted → equal → minimum budgets

  • Correlation-aware risk manager (oracle3.risk.correlation_risk_manager): correlated exposure limits

    • EWMA rolling correlation estimation on price returns
    • Effective exposure via correlation matrix quadratic form
    • Concentration ratio gating, stale correlation decay

  • Fair value divergence strategy v2 (FairValueStrategy): model-driven alpha with exact coefficients

    • Uses Yang (2026) hierarchical model for per-contract λ estimation
    • Kelly-optimal sizing from model Greeks
    • Volume-tier targeting: focuses on medium-liquidity alpha sweet spot

  • Premium decay strategy (PremiumDecayStrategy): timing-based premium lifecycle alpha

  • Distortion-based validation tools: estimate_risk_premium, cross_platform_premium_test, favorite_longshot_test in oracle3.market.validation

  • 79 new tests (total: 633) covering MLE recovery, Greeks, sizing, allocation, strategies

1.0.0 - 2026-03-09

Added

  • 8 constraint-based & statistical arbitrage strategies: cross-market, exclusivity, implication, conditional, event-sum, structural, cointegration spread, and lead-lag — each with formal invariant, fee-aware edge, cooldown windows, and audit trail
  • Market relation graph: persistent knowledge graph (~/.oracle3/relations.json) with lifecycle management (discovered → validated → deployed → retired) and quantitative validation (Engle-Granger cointegration, ADF stationarity, OLS hedge ratio, OU half-life, Pearson correlation, lead-lag detection)
  • SpreadExecutor: safe multi-leg execution with automatic LIFO unwind on partial fills — no naked positions
  • Engine control server: Unix socket runtime control (pause/resume/stop/killswitch) without process restart
  • Strategy portfolio registry: lifecycle tracking (paper → live → retired), health checks, Kelly capital allocation
  • 8 on-chain agent capabilities: cross-market arbitrage, on-chain risk manager, on-chain signal source, MEV protection (Jito), agent reputation, multi-agent pipeline, flash loan arbitrage, atomic multi-leg trader
  • AI-powered trading with OpenAI Agents SDK, LiteLLM multi-provider support, and 8 built-in agent tools
  • Solana integration: native transaction signing, on-chain trade logging via Memo program, Jito bundle submission, Solana Blinks
  • Multi-exchange support: Solana/DFlow (SPL tokens), Polymarket (CLOB API), Kalshi (REST API)
  • Live trading dashboard at /live with 8 feature cards, equity chart, execution pipeline animation, and pause/resume/e-stop controls
  • Classic terminal dashboard at / for headless environments
  • Risk management: dual-layer validation (local limits + Solana simulateTransaction), max drawdown monitoring, daily loss limits, kill switch
  • Backtesting engine with DFlow episode replay (parquet format)
  • Coinjure matching pipeline: cross-platform market relation discovery (implication, exclusivity, complementary) with resolution filter, volume filter, keyphrase pre-filter, confidence sizing, and tag coverage
  • CLI (oracle3) with commands for market browsing, paper/live trading, engine control, reputation, blinks, trade logs
  • CI/CD: pytest (553 tests), ruff, mypy, codespell, MkDocs documentation site
  • Interactive demo script (demo.sh)