Quantitative Research Engine
Divergia operates on a fully deterministic quantitative core. We eliminate arbitrary thresholds by ranking public equities relative to their S&P 500 peers using cross-sectional factor models, combined with dynamic cash flow modeling and an autonomous multi-agent audit.
1. Cross-Sectional Factor Model
Traditional screeners filter using static rules (e.g., 'P/E < 15'), which fail to adapt to different sector economics or changing market regimes. Divergia evaluates the full S&P 500 universe (~300+ eligible tickers post sector filter) relative to each other. We compute cross-sectional z-scores across four key dimensions: Quality (TTM ROIC, capital efficiency), Value (FCF yield, margin of safety), Growth (TTM FCF growth, fundamental momentum), and Risk (balance sheet leverage, structural volatility). Weighting is locked in production: 35% Quality, 25% Value, 25% Growth, and 15% Risk.
2. Adaptive Financial Valuation
A single DCF model cannot value a software company, a REIT, and a bank. The engine detects the sector nature of each ticker and dynamically shifts models. Financial services and banking institutions are valued using Dividend Discount (Gordon Growth) and Residual Income models. Real Estate Investment Trusts (REITs) are modeled using Funds From Operations (FFO) metrics. Operating corporations use a Discounted Free Cash Flow (DCF) model with dynamic cost of capital (WACC) calculations, yielding Bull, Base, and Bear scenarios.
3. T-90 publication lag constraint
Backtest results are often inflated by look-ahead bias (using data that was not yet publicly known at the rebalance date). Divergia applies a strict T-90 publication lag: every transaction and valuation in the 2019-2025 backtest uses financial statement metrics only after 90 days have elapsed since the fiscal-quarter close. In the official 2026-08-10 measurement base (production parity plus a determinism contract: strict freeze, lake-pinned regime inputs, and a historized risk-free rate), DE_ALPHA records +6.70pp daily-reconstructed, annualized point-in-time alpha versus SPY. The provenance is inseparable: this is a measurement-base re-pin built on coherence repairs, not a new engine improvement.
4. Data Sources & Pipeline Hygiene
Our pipeline ingests raw reports from public filings (SEC EDGAR), macroeconomic indices (FRED), and SimFin. We enforce strict vendor-corruption guards: impossible corporate rows (such as cash flows exceeding total assets or extreme OCF/NI mismatches) are quarantined at ingestion. Multiple fallbacks prevent blind spots while preserving data integrity.