How the models are built, validated, and run.
No black boxes. Every rule below is the actual rule the models trade against — not a simplified summary for marketing purposes.
1. Fixed rule sets, published before the fact
Every strategy is fully specified in code: entry triggers, exit triggers, position sizing, and hedge conditions. Signals are generated from the previous session's close and published before the next session opens. There is no discretionary override.
2. Trend and momentum, not prediction
The models don't forecast price. They react to trend and momentum in the underlying (Nasdaq-100, S&P 500, 20Y Treasury yield, gold) and size into or out of leveraged exposure based on rules that held up across multiple market regimes in backtest.
3. Anti-overfitting validation
Before a model goes live it runs through a validation battery: walk-forward out-of-sample windows, parameter perturbation (does performance collapse if a threshold moves 10%?), and regime-split testing (2008, 2020, 2022 handled separately from the rest of the sample). Pass/fail counts are published on every strategy page.
4. Volatility-targeted sizing
Leveraged legs (TQQQ, SSO) scale position size against realized volatility rather than using a fixed weight, so exposure comes down mechanically in high-volatility regimes instead of relying on a stop being hit.
5. Real financing costs, real instruments
Backtests model the actual daily financing drag of leveraged ETFs (Fed funds + typical spread), not a frictionless synthetic return. Where a leg trades a real ETF (TQQQ, SSO, TMF), that ticker is what the signal is written against.