Using Jev for Perpetual Trading Decisions and Maker-Order Evaluation

Summary

This document describes an experiment integrating Jev’s structured judgments into a Binance USDT-margined perpetual strategy. The model receives order-book state, recent trades, short-term price changes, and volatility, while positions, order sizes, and trading limits remain local to the program. Instead of asking for one buy, sell, or wait decision, the system separates the task into forecasts of return ranges, scores of adverse-selection risk for each side, and estimates of whether a passive order may fill.

The outputs inform entry thresholds and quote placement for Maker orders. The document discusses changing fixed return boundaries for volatile markets and describes validation through forecast comparisons, order records, and post-fill price movement. These observations help diagnose model judgments and execution, but post-fill movement excludes fees and is not net performance evidence. Fill estimates cannot know actual queue position, and Maker orders may remain unfilled or execute just before adverse moves. The described implementation also lacks a complete hard stop, daily-loss breaker, and liquidation-distance check.

Key ideas

• Separate direction and magnitude forecasts from fill probability and post-fill adverse-selection judgments.
• Keep account positions and execution constraints in program logic rather than delegating them to the model.
• Use volatility-aware return intervals when fixed thresholds do not fit fast-moving markets.
• Evaluate forecasts and post-fill price paths separately, and compare fill estimates with actual order records.
• The experiment is not evidence of profitability and has material risk-control gaps.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.