UQ-LOB: Uncertainty-Aware Limit Order Book Mid-Price Forecasting

UQ-LOB adds an encoder-agnostic uncertainty quantification module to pretrained limit order book models using attentive neural processes conditioned on recent realized outcomes. The module provides calibrated Gaussian distributions for displacement regression and categorical distributions for price direction, enabling selective prediction via predicted signal-to-noise ratios across 5.2 billion LOB events.