Financial Language Models as Applied Artificial Intelligence Systems for News-Based Trading under Market Frictions

Battle Count: 0

Highly relevant as it directly addresses the gap between academic NLP accuracy and practical trading feasibility by incorporating market frictions, transaction costs, liquidity constraints, and execution timing into the evaluation framework.

High complexity due to the end-to-end nature of the MFAST framework, requiring integration of data pipelines, model fine-tuning, calibration, portfolio construction logic, cost modeling, and statistical validation. Requires significant computational resources (A100 GPUs) for the best-performing models.

The paper provides a detailed replication package structure, including configuration templates, public-data scripts, model checkpoint identifiers, random seeds, and deterministic output tables. It also includes a public-data replication arm using GDELT and Yahoo Finance/Stooq data to ensure auditability without proprietary data.