Uncertainty-Aware AI: Conformal Prediction versus Reinforcement Learning for Optimal Trade Execution

Reference implementation and reproducibility package accompanying the published study “Uncertainty-Aware AI: Conformal Prediction versus Reinforcement Learning for Optimal Trade Execution.”

This software release provides a reproducible implementation of the uncertainty-aware VWAP execution framework, including a self-contained execution simulator, conformal prediction and decision gating, VWAP-based execution, classical execution benchmarks, reinforcement-learning baselines, market-data processing, evaluation procedures, and scripts for reproducing the reported experiments.

The framework uses the conformal prediction interval as an explicit decision gate: the execution policy acts on the return forecast when the forecast magnitude exceeds a configurable multiple of the conformal uncertainty width, and otherwise falls back to VWAP tracking. The implementation supports systematic cost–risk evaluation across competing execution strategies.

The release is intended to support reproducible research, verification of the published results, and further development of uncertainty-aware trade execution methods.

Associated publication:
Irshad, A., & Biswas, S. (2026). Uncertainty-Aware AI: Conformal Prediction versus Reinforcement Learning for Optimal Trade Execution. Statistics, Optimization & Information Computing, 16(2), 1334–1349. DOI: 10.19139/soic-2310-5070-4159.

Software: Conformal VWAP Execution
Version: v1.0.0