A Learnable Wavelet Transformer for Long-Short Equity Trading and Risk-Adjusted Return Optimization

Researchers developed WaveLSFormer, a wavelet-based Transformer model that directly generates market-neutral long/short trading portfolios from financial time series data. The AI system achieved a 60.7% cumulative return and 2.16 Sharpe ratio across six industry groups, significantly outperforming traditional ML models like LSTM and standard Transformers.

• →WaveLSFormer combines learnable wavelet decomposition with Transformer architecture for intraday trading optimization.
• →The model directly outputs long/short positions rather than predictions, training end-to-end on trading objectives with risk management.
• →Testing on five years of hourly data showed consistent outperformance across multiple industry sectors and random seeds.
• →The system achieved 60.7% cumulative returns with a 2.16 Sharpe ratio, demonstrating strong risk-adjusted performance.
• →The approach addresses key challenges in algorithmic trading including noise, non-stationarity, and cross-sectional asset dependencies.