This document outlines the research validation framework governing quantitative strategy development across Forticia and Qlumina. Candidate alpha factors advance sequentially through five empirical gates: (1) regime classification and macroeconomic state modeling, (2) causal factor-absence placebo testing against surrogate noise series, (3) Combinatorial Purged Cross-Validation (CPCV) with time-embargo buffers, (4) realistic limit order book matching with exchange fee schedules and queue depletion dynamics, and (5) live paper manifest staging. We reject synthetic market data flattery, enforcing rigorous historical out-of-sample data air-gaps.
The Five Institutional Decision Gates
Every candidate strategy model must advance sequentially through five empirical validation gates. Failure at any single gate results in immediate disqualification.
Stage 1 — Regime Space Classification
Financial markets are non-stationary, adaptive systems. Rather than evaluating candidate signals against static historical averages or simple trend indicators, market states are modeled dynamically. These state-space representations condition strategy behavior on real-time order-book queue depletion, cross-asset liquidity dispersion, and credit spreads.
Stage 2 — Causal Factor-Absence Placebo Testing
To address p-hacking and backtest Sharpe ratio inflation, candidate factors undergo counter-factual testing:
- Randomized Surrogate Controls: Time series are Fourier-transformed, and phase spectrums are randomized while preserving power spectral density. If a factor generates positive returns on phase-scrambled noise, it is rejected as a statistical artifact.
- Causal Factor Ablation: The candidate signal is isolated and ablated from the multi-factor set. If marginal predictive power does not decay to zero upon ablation, the factor possesses zero independent economic information.
- Lead-Lag Inversion: Verifying directional causal asymmetry by leading and lagging the signal against forward returns. Signals displaying bidirectional predictability are purged as lookahead-contaminated.
Stage 3 — Combinatorial Purged Cross-Validation (CPCV)
Standard k-fold cross-validation is flawed in financial time series due to serial autocorrelation. Our testing enforces:
- Combinatorial Path Splitting: Exhaustive walk-forward path generation across historical regimes.
- Purging: Complete removal of training labels that overlap with evaluation windows.
- Embargoing: Enforcing an empirical post-test time buffer to eliminate auto-regressive memory leakage.
Stage 4 — Realistic Microstructure Simulation
Surviving alpha models are re-simulated through the Blitz execution matching engine. The simulation incorporates prime broker clearing commissions (Clear Street), exchange regulatory fees, dynamic bid/ask crossing costs, and latency-induced fill degradation.
Stage 5 — Live Soak Staging
Before capital allocation, certified strategies undergo a multi-week live paper trading manifest soak period across staging accounts. This stage measures operational tracking between backtested signals and live broker API fills.
Empirical Data Quality & Out-of-Sample Integrity
Simulated market data introduces artificial smoothness that masks liquidity voids and fat-tailed drawdown distributions.
Strategy backtesting is conducted strictly across raw, discrete historical exchange ticks and authentic futures contract roll feeds. Out-of-sample data windows are strictly segregated to prevent researcher data snooping and curve-fitting.
Portfolio Risk & Reporting Console
Strategy telemetry integrates directly into our portfolio risk and allocator reporting console, providing clear institutional transparency:
- Real-time factor decomposition and stress testing across historical macroeconomic regimes.
- Automated generation of institutional due diligence reporting (DDQs).
- Continuous monitoring of strategy risk envelopes and margin consumption collars.