Using Jev Model Judgments in a Perpetual Futures Trading Strategy
Summary
The article describes integrating Jev, a model that returns choices, scores, or probabilities, into a Binance USDT-margined perpetual strategy. It feeds the model order-book data, recent trades, short-term price changes, and volatility, while local code retains control of positions, order size, constraints, and execution. The strategy asks separate questions about likely price direction and magnitude, post-fill adverse movement, and the chance a maker order will fill. It combines those outputs with different probability thresholds for opening and reducing positions.
The author describes adapting forecast bands to recent volatility, monitoring prediction calibration with classification accuracy and Brier scores, and tracking price movement after fills to assess execution quality. The article reports that the strategy is running live on one asset, but offers no quantified trading performance. It cautions that a simulated loss figure needs verification, model confidence is not a win probability, and fill estimates cannot know true queue position. Hard stop-losses, daily loss limits, and liquidation-distance checks are still missing, so the implementation has material risk controls to complete.
Key ideas
- The strategy separates model judgments about direction, post-fill adverse movement, and maker-order fill likelihood.
- Local code handles positions, order sizing, exchange constraints, and whether model outputs qualify for action.
- Forecast movement bands and repricing distances are adjusted using recent volatility estimates.
- Opening trades require stronger directional evidence than reducing an existing position.
- Prediction scores and post-fill price movement are tracked, but neither alone establishes net profitability.
- The described implementation lacks hard stop-losses, daily loss limits, and liquidation-distance checks.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.