Increase Alpha: Performance and Risk of an AI-Driven Trading Framework

Battle Count: 140

Highly relevant to quantitative trading. The paper presents a complete end-to-end AI-driven trading framework from signal generation through portfolio construction and execution. It demonstrates practical deployment with real operational costs, cloud infrastructure details, and out-of-sample performance metrics. The focus on low-correlation alpha generation, risk-adjusted returns (Sharpe >2.5), and regime robustness directly addresses core quantitative trading concerns. The walk-forward design and stress testing through volatile 2025 markets add practical credibility.

The described infrastructure is moderately complex: daily inference on 814 equities, multi-horizon signal generation, grid search across 2,280 scenarios per ticker-signal pair, Azure AKS containerized parallel computation (20 pods, 10 VMs), and quarterly walk-forward portfolio rebalancing. However, the core model architecture is proprietary and cannot be replicated. The operational cost (~$95-100/day AWS, ~$17K total infrastructure) and cloud setup are well-documented. The execution parameter optimization and portfolio construction logic are described in sufficient detail for a practitioner to implement an analogous system.

The model architecture and feature definitions are explicitly described as proprietary and treated as a black box. The paper states it outlines the general overview without disclosing core underlying concepts. While the evaluation framework, data pipeline, and execution logic are described, the actual model cannot be reproduced. Data source (eodhd.com) is a paid subscription service. No code or model weights are released.