Research
Every model at 00Capital is versioned, documented, and subjected to post-mortems. The goal is survivability — across regimes, not just peak CAGR.
How we work.
Every model at 00Capital is versioned, documented, and subjected to post-mortems. Backtests are stress-tested across regimes, not just optimized for maximum CAGR. The goal is survivability over bull runs and bear markets alike.
We treat each idea as an experiment. A hypothesis is formed, tested against historical regimes, stress-tested under edge cases, and either integrated or archived with a written record of why. Nothing ships without a documented failure mode.
The quant stack is deliberately modular — each phase can be swapped, upgraded, or disabled independently. ProjectR's 16-phase architecture is the direct result of learning what happens when you can't isolate a broken component in a monolithic system.
Changelog
ProjectR initialized. 16-phase modular architecture spec'd from scratch — direct response to ProjectV's monolithic design. Regime classifier wired from day one.
Soft direction gate implemented: 75% signal penalty for mildly negative
trends, hard exclusion below −2%. n_long increased to 10.
Sharpe improved from 0.71 → 0.91. CAGR baseline: 9.2%.
ML feature set rebuilt with orthogonal signal engineering: regime duration, transition probability, cross-sectional rank, mean reversion, volatility structure ratios. Lookahead bias shuffle test: PASSED.
ML layer wiring in progress. 30/70 ML/momentum blend target defined. 6-state GMM regime classifier (Bull Trend, Bull Volatile, Sideways, Bearish Volatile, Bear Trend, Crisis) fully operational.
Phase 9–16 execution in progress. 00Risk integration for live analytics. ML activation testing across regime-conditional states. Paper execution live with trade log.
Signal → Regime → Risk → Execution
Seven stages, each independently versioned. Every phase has a defined input contract, output contract, and failure mode.
Data Layer
Cross-asset ETF universe with macro proxies. Price-based features, volatility surface, and breadth indicators. Regime-conditioning features engineered at ingest time — not retrofitted.
Regime Engine
6-state GMM classifier. States: Bull Trend, Bull Volatile, Sideways, Bearish Volatile, Bear Trend, Crisis. Regime duration and transition probability are live inputs to the ML feature set — not post-hoc labels.
ML Layer
30/70 ML/momentum blend across regime states. Orthogonal signal features: regime duration, transition probability, cross-sectional rank, mean reversion, volatility structure ratios, regime-conditional excess return targets.
Risk Overlay
GARCH volatility forecast scales exposure. Monte Carlo simulation across thousands of paths estimates drawdowns, tail risk, and ruin probability. Kill-switches throttle the system when behavior exits expected ranges.
Sizing
Kelly-adjusted position sizing per regime state. Soft gate: 75% penalty for mildly negative trends. Hard exclusion below −2% trend threshold. n_long = 10 position cap applied before allocation.
Portfolio Construction
Modular 16-phase architecture. Exposure bands and drawdown constraints applied at the portfolio level. Regime-conditional rebalancing — turnover is minimized in stable regimes, accelerated in transition.
Execution & Analytics
Paper execution with full trade log. 00Risk integration delivers live VaR, Sharpe/Sortino/Calmar, factor decomposition, and per-regime attribution from the first backtest day.
Technical Brief
Select researchers and collaborators can receive a detailed ProjectR technical brief covering the full feature set, validation framework, and portfolio construction architecture.
Reach out via the Access page with "ProjectR Research" in your message. All materials are provided for research discussion only and do not represent an offer to manage capital.