Drawdown-Aware Deep Learning for Algorithmic Trading During Market Stress

This paper presents a drawdown-aware deep learning framework for algorithmic trading under adverse market conditions. The proposed approach integrates LSTM-based return prediction with dynamic position sizing, volatility filtering, and drawdown control mechanisms. Unlike traditional models that focus solely on return maximization, this framework prioritizes capital preservation during high-risk periods. The system is evaluated using Bitcoin high-frequency market data, comparing AI-driven strategies against baseline buy-and-hold approaches. Experimental results demonstrate improved drawdown stability and competitive risk-adjusted performance, highlighting the potential of combining deep learning with risk-aware trading logic.