Ensemble
Forecasting.
A single model has a single bias. Eltrove runs a panel of independent forecasters on every symbol, every cycle, then weights their votes by recent performance. No trade enters the book without conviction across models.
Why ensembles?
Markets reward humility. The strategies that survive over decades. Renaissance, Two Sigma, and AQR rely on combining many weakly-correlated signals, not betting on one perfect one. Eltrove inherits the same discipline.
Each model in the ensemble is trained on a different slice of history, a different feature set, or a different objective. When they all agree, the signal is strong. When they disagree, position size shrinks, or the trade is skipped entirely.
What's inside the ensemble
- Gradient-boosted predictors: short-horizon return forecasts from technical, microstructure, and cross-asset features.
- Transformer sequence models: patterns over multi-day price + volume windows.
- Reinforcement-learning policies: trained against historical regimes, optimized for risk-adjusted return rather than raw accuracy.
- Macro context layer: VIX, SPY trend, sector rotation, and catalyst proximity used as regime gates.
How a signal gets approved
Each model scores independently
For a given symbol, every model emits a directional score and a confidence value. No model sees what the others said.
Recent-performance weighting
Votes are scaled by each model's rolling out-of-sample accuracy. A model that's been hot in this regime carries more weight than one drifting.
Threshold + risk gate
Aggregate conviction must clear a regime-aware threshold. Position size scales with confidence. Below threshold = no trade, not a reluctant trade.
Hand-off to execution
Approved signals flow to the automated execution layer with sizing, stop, and exit logic already attached.
What you don't get
Eltrove is not a black-box stock picker shouting "BUY NOW." You'll see every model's vote, the aggregate conviction, the position size it produced, and why a trade was skipped when it was. The dashboard shows the work.